Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Essays on policies to create jobs, improve health, and reduce corruption
(USC Thesis Other)
Essays on policies to create jobs, improve health, and reduce corruption
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
1
ESSAYS ON POLICIES TO CREATE JOBS, IMPROVE HEALTH,
AND REDUCE CORRUPTION
by
Yunsun Li
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(Economics)
December 2015
Copyright 2015 Yunsun Li
2
Table of Contents
List of Tables ........................................................................................................................... 4
Acknowledgements ............................................................................................................... 5
1. What Can Be Learned about the Employment Effects of Labor Market
Deregulation from Hypothetical Questions to Firms? Evidence from the Enterprise
Surveys .................................................................................................................................... 6
1.1. Introduction ............................................................................................................. 7
1.2. Relevant Literature ................................................................................................ 10
1.3. Data and Analytical Framework ......................................................................... 18
1.3.1. The Enterprise Surveys and our sample ..................................................... 18
1.3.2. Definition of Variables .................................................................................. 19
1.3.3 Descriptive Statistics ...................................................................................... 24
1.3.4 Analytical Framework ................................................................................... 25
1.4 Empirical Results ................................................................................................... 28
1.4.1 Main Results: Employment Change and the Testing of Hypotheses ..... 28
1.4.2 Allowing for Heterogeneous Effects of Perceived Labor Rigidities and
Regulatory Enforcement on Employment Changes ............................................... 34
1.4.3 Effects of Rigidity of Labor Regulations and their Enforcement on the
Firms’ Perceptions of the Seriousness of Labor Regulations as an Obstacle to
Business. ........................................................................................................................ 41
1.4.4 Tests of Robustness ........................................................................................ 44
1.5 Concluding Remarks ............................................................................................ 46
2. Improving Coverage and Utilization of Maternal and Child Health Services in Lao
PDR: Impact Evaluation of the Community Nutrition Project ..................................... 51
2.1 Introduction ........................................................................................................... 52
2.2 The Project with Two Demand-Side Interventions .......................................... 55
2.3 Data and Project Outcome ................................................................................... 64
2.4 Methods .................................................................................................................. 73
2.4.1 Difference-in-Differences .............................................................................. 73
2.4.2 Matched Difference-in-Differences ............................................................. 77
2.5 Results ..................................................................................................................... 78
2.5.1 Main Results ................................................................................................... 78
2.5.2 Associated Results ......................................................................................... 79
3
2.6 Heterogeneity......................................................................................................... 89
2.7 Conclusion .............................................................................................................. 93
3. Corruption, Governance Institutions and the Extent to Which Governance
Institutions Can Reduce the Pervasiveness of Corruption: Evidence from Firms
Perceptions and Behavior ................................................................................................... 95
3.1. Introduction ........................................................................................................... 96
3.2. Relevant Literature ................................................................................................ 98
3.3. Data, Sample, Variable Definitions and Estimation Procedure .................... 104
3.3.1. Measurement of the Outcomes of Interest ............................................... 104
3.3.2. Data for Measures of the Explanatory Variables ..................................... 107
3.3.3. Estimation Procedure .................................................................................. 109
3.4. Empirical Results ................................................................................................. 110
3.4.1. Baseline Results: How Do the Governance Indexes affect Firms’
Perceptions of Corruption and Likelihood of Engaging in It? ............................ 110
3.4.2. Results of Heterogeneous Specifications: What Role Do Firm
Characteristics Play in the Corruption- Governance Relationship? ................... 114
3.4.3. Effects of Control Variables ........................................................................ 121
3.5. Conclusion ............................................................................................................ 123
Bibliography ....................................................................................................................... 125
Appendix 1 .......................................................................................................................... 133
Appendix 2 .......................................................................................................................... 141
Appendix 3 .......................................................................................................................... 154
4
List of Tables
Table 1.1 ................................................................................................................................. 21
Table 1.2 ................................................................................................................................. 29
Table 1.3 ................................................................................................................................. 30
Table 1.4 ................................................................................................................................. 31
Table 1.5 ................................................................................................................................. 32
Table 1.6 ................................................................................................................................. 35
Table 1.7 ................................................................................................................................. 37
Table 1.8 ................................................................................................................................. 38
Table 1.9 ................................................................................................................................. 39
Table 1.10 ............................................................................................................................... 42
Table 1.11 ............................................................................................................................... 48
Table 2.1 ................................................................................................................................. 57
Table 2.2 ................................................................................................................................. 68
Table 2.3 ................................................................................................................................. 79
Table 2.4 ................................................................................................................................. 81
Table 2.5 ................................................................................................................................. 83
Table 2.6 ................................................................................................................................. 84
Table 2.7 ................................................................................................................................. 86
Table 2.8 ................................................................................................................................. 87
Table 2.9 ................................................................................................................................. 88
Table 3.1 ............................................................................................................................... 107
Table 3.2 ............................................................................................................................... 111
Table 3.3 ............................................................................................................................... 113
Table 3.4-1 ........................................................................................................................... 116
Table 3.5-2 ........................................................................................................................... 117
5
Acknowledgements
First of all, I feel very much grateful to Prof. Jeffrey Nugent, for being my advisor and
committee chair. Without much previous research experience, I had benefited a lot
from Prof. Nugent’s extremely helpful advice and guidance during my PhD study.
My research would not be ever possible without his knowledge and insight which
have always been encouraging me to take on new steps while his professionalism and
experience makes sure I am in a right direction. Words become powerless to express
my thankfulness for his guidance all along the way.
I would also like to extent my sincere thanks to Prof. Cheng Hsiao, who
introduced numerous different and powerful applied econometric tool to me in early
stages of my PhD study. His feedback and comments on various econometric issues
of my research have been greatly helpful. I am very thankful to Prof. Yu-Wei Hsieh,
Prof. Shui-Yan Tang, Prof. Anant Nyshadham and Prof. Guilbert Hentschke for
kindly serving on my dissertation and qualifying exam committees and generously
providing helpful feedback, and in the cases of Professor Hsieh and Tang to do so on
quite short notice.
Moreover, the completion of this dissertation benefited tremendously from my
co-authors of the second chapter, Jeffery Tanner and Ryotaro Hayashi at the World
Bank. It is all because of them I got the privilege of analyzing a unique nutrition
program with real-world impact. I will never forget the days working with them:
learning from their expertise, testing new models and troubleshooting problems, all
deemed important for me to become a better researcher.
Lastly, I feel most indebted to my wife, Jing, for her endless support and
encouragement with me in the last few years. Special dedication to my family.
6
1. What Can Be Learned about the Employment Effects
of Labor Market Deregulation from Hypothetical
Questions to Firms? Evidence from the Enterprise
Surveys
7
1.1. Introduction
The financial crisis that has hit much of the world since 2008 has, for some countries
at least, gave rise to considerably higher unemployment rates than have existed since
the 1930s. As a result, policy makers of the world are more in need of research on
alternative means to increase employment than they have been for a long time. This
is especially true in those developing countries where young workers (aged 19–39)
constitute unusually large percentages of the labor force and the unemployment rates
among them have reached unprecedented levels (30% or more) according to ILO
(2011).
1
Some believe that these developments have even triggered the revolutions
and political instability that have occurred in some countries in recent years. Since
debt burdens, credit constraints and trade deficits are often sufficiently severe as to
rule out the use of Keynesian monetary and fiscal policies in treating these
unemployment problems in such countries, many policy makers are searching for
alternative types of policies for dealing with their youth unemployment problems.
2
Labor market reforms represent one such alternative approach. It is frequently argued
that firms might be more willing to hire additional workers if it would be easier to
hire them on short term contracts or even on long term ones if the costs of dismissing
them later on were not so high. Advocates argue that, if such reforms were successful,
it is quite possible that unemployment rates could be lowered, that efficiency, exports
and growth could be increased, and business cycles moderated (Hopenhayn and
Rogerson, 1993; Lagos, 2006 and Belot et al., 2007).
Yet, at the same time, since the very clauses in the labor laws that give rise to rigidity
also provide protection to employed workers against the prospects of future job loss,
1
This is because of the well-known result that it is the willingness of firms to hire young workers entering the
labor force that is most discouraged by restrictions both on firing permanent workers and on hiring workers on
non-permanent short-term contracts.
2
In the case of the Southern European countries as members of the Euro zone, and other regions with common
currencies, the use of independent monetary policy is also ruled out.
8
their elimination may also raise serious social concerns about the risk-increasing and
adverse distributional implications of such policies. Under such circumstances
deregulation of labor might increase unemployment and lower income and aggregate
demand. While this would not be a serious problem if unemployment insurance
programs were in place, unfortunately in most developing countries, such schemes
are either not in place or not functioning well (e.g., Angel-Urdinola and Leon-Solano,
2013).
Labor markets are typically so multi-tiered and complex and the effects of labor
deregulation so diverse as to make it virtually impossible to derive clear-cut
predictions about the net effects of deregulation on employment from theoretical
models. Yet, considering the very limited experience in developing countries with
labor market reforms, there is a dearth of empirical studies on the effects of such
reforms. Even in the rare (often crisis-driven) situations, in which existing labor laws
are changed in developing countries, many other policies or circumstances may have
changed at the same time, making it very difficult to separate the effects of changes in
labor regulations in empirical studies from those arising from these other
simultaneous changes (Campos, Hsiao and Nugent, 2010). While field experiments
are increasingly being used to evaluate the effects of many different kinds of policy
treatments, the controversial nature of any such changes and discriminatory charges
that would be likely to arise from treating different firms in the same region or
industry in different ways, undoubtedly greatly limiting the applicability of
randomized control trials to labor laws.
The purpose of this study is to take advantage of suitable firm-level data taken from
the World Bank’s Enterprise Surveys undertaken between 2002 and 2009 for some 89
countries in as many as three different survey rounds. These surveys contain one
unique question of relevance to evaluating the effects of labor deregulation: “If you
could change the number of regular full-time workers your firm currently employs
without any restrictions (i.e., without seeking permission, making severance
9
payments, etc.) what would be your optimal level of employment as a percent of your
existing workforce?”
3
We also take advantage of the fact that the labor regulations in
the countries surveyed vary considerably in both the degree of rigidity (as measured
objectively in existing rigidity indexes) across countries, and the enforcement of these
regulations vary considerably even within countries across industries and locations
and as a result so, too, do the subjective perceptions of firms concerning the severity
of labor regulations as an obstacle to their business.
While hypothetical questions to assess the effects of labor deregulation have their own
limitations (as discussed below), this approach greatly mitigates the methodological
problems that plague more common empirical methods of assessing such effects at
the macro level (via cross country regressions of aggregate labor market outcomes on
labor rigidity indexes, or after-the-fact evaluations of the effects of changes in labor
regulations). Our primary empirical strategy is to link firm assessments of the
employment changes that would result from complete deregulation of labor to the
seriousness that the firm attaches to labor and other regulations as obstacles to
business. This is because these assessments could reflect not only the rigidity of labor
regulations (de jure) but at least in part also the degree of enforcement of these
regulations in the vicinity of the firm. We also allow for some interactions between
the two and other relevant firm, industry and country characteristics. We attempt to
reduce endogeneity in the use of such measures by relating each firm’s answer to the
employment change question, not to its own subjective evaluations of the importance
of labor regulations as an obstacle to the firm’s business and of the extent of
enforcement, but rather to the responses of other firms in the same location and
industry (i.e., excluding the evaluation of the firm itself).
Then, in the final stage of the analysis, we attempt to relate the firm’s perception of
the seriousness of labor regulations as an obstacle to their business back to
3
While numerous additional Enterprise Surveys have been carried out since 2009, unfortunately they did not
include this hypothetical question concerning the employment effects of removing all labor regulations.
10
independent measures of the rigidity of those regulations (de jure) and alternative
proxies for their enforcement to test the validity of our assumption that the perception
measures would reflect both the rigidity of these regulations and their enforcement.
Our results provide strong support for the proposition that labor deregulation would
affect employment decisions quite considerably and more strongly in countries in
which existing labor laws are more restrictive, in situations where these regulations
are better enforced and in firms that are smaller, younger, and in when the overall
unemployment rate is lower. The estimated effects of deregulation, however, are by
no means all positive. Dismissals could largely or possibly even completely offset job
creation in very large firms. In view of the heterogeneity of effects for different firm
types, it is quite clear that no single reform would maximize employment increases
for all firms. Indeed, the overall effects are quite heterogeneous across different
industries and firm types, suggesting that there may be no uniform set of labor
regulations (or deregulation) that would maximize employment.
The remainder of the chapter is organized as follows. Section II presents a review of
literature that deals with one or another of all the ingredients of our study. Section III
identifies the analytic framework and data sources and presents some descriptive
statistics on the variables identified. Section IV presents the empirical results and
Section V the conclusions, including implications for policy and future research.
1.2. Relevant Literature
Virtually all economists and practitioners have increasingly come to recognize the
importance of transaction costs in both the design and enforcement of contracts and
in analyzing their effects. Transaction costs are especially prominent in long term
labor contracts in which labor effort may be costly to monitor and current
observations on such effort may not be very dependable predictors of the quantity
and quality of such effort over the future. Similarly, workers may find it hard to know
11
how well to trust employers in rewarding them over time. Various other kinds of
risks, uncertainties and information problems in long term employment contracts are
also likely to appear in a prominent way and complicate the choices among contracts
and matching particular workers with particular employers. As a result of the
resulting informational asymmetries, moral hazard and adverse selection behavior
can easily arise, leading to market failure. Labor regulations may arise to overcome
these asymmetries and help the labor market function properly, but on the other hand
it is quite possible that such regulations can lead to excessive rigidity and limit the
variety of available contractual forms relative to the variety that might be considered
necessary to fit the variety of tasks in which labor is employed.
While labor market regulations are but one of the many institutions that can affect the
efficiency and growth of employment, their effects are likely to be among the most
direct and potentially useful ways of mitigating the challenging youth and other
unemployment problems that exist in many parts of the developing world. For
reasons stated in the introduction, the choice between greater regulation of labor
markets and deregulation is a very controversial one, in large part because of the
important distributional consequences of changes in either direction.
The complexity of theoretical analyses that would be needed to arrive at conclusions
relevant to deregulation of labor in the real world has virtually ruled out the
possibility of deriving clear-cut conclusions from theoretical models. This means that
such evaluations are largely left to empirical analyses. However, considering the
multiplicity of elements contained in any set of labor regulations and the many other
institutional and economic conditions that may affect any particular country or
industry setting, and interdependencies among different sectors, skill types and
locations, reliable and clear-cut empirical assessments are also not easily achieved.
Yet, since in most countries, the main set of labor regulations are set at the national
level, and much progress has been made in coding the many different clauses in the
labor laws and regulations so as to construct overall indexes of the rigidity of the labor
12
laws, attempts to examine such effects on various labor market or productivity
outcomes have been growing. At least until the late 1990s, however, the availability
and use of such indexes were limited to OECD countries
4
.
For non-OECD countries, and especially developing countries, something of a
breakthrough came in 2004 with the publication of Botero et al. (2004) based on the
attempt of these authors to code the labor laws and regulations in effect in the late
1990s for a total of 85 countries and to construct separate indexes of rigidity with
respect to hiring, firing and working conditions and their aggregation into a single
overall index of employment rigidity. Indeed, based on their cross-section of 85
countries, they demonstrated that their employment rigidity index had significant
negative effects on male labor force participation and significant positive effects on
the overall, male, and youth unemployment rates, the latter generally believed to be
the group most adversely affected by rigid labor regulations.
Following that study, Heckman and Pages (2000, 2004), Muravyev (2014) and the IMF
(Aleksynska and Schindler, 2011) have constructed somewhat different but related
indexes of the rigidity of labor regulations for Latin American countries, transition
economies and additional developed and developing countries, in most cases with
considerable time coverage that was lacking in Botero et al. (2004). The largest set of
comparable labor rigidity indexes (which are also those most closely related to those
of Botero et al. (2004) discussed above) are those constructed by the Doing Business
Surveys for 2006–2014.
5
None of the indexes can claim to be truly comprehensive since
each has necessarily to focus on some of the clearest clauses and regulations that are
deemed to be important, codable and relatively objective. All of the above sources of
4
See, for example, Addison and Grosso (1996), Addison and Teixeira (2003), Allard (2005), Arpaia et al. (2007),
Blanchard and Wolfers (2000), Blanchard and Portugal (2001), Deakin et al. (2007), Nicoletti et al. (2000), and
Nickell (1997). The effects on labor market outcomes in these studies were somewhat more mixed.
5
The International Labor Office (especially, Rama and Artecona, 2002 and Forteza and Rama 2006) have
constructed a quite different index based on the internationals labor standards and conventions that the
countries have signed onto. This and other indexes of labor rights (Kucera 2002, Greenhill, Moseley and Prakash
2009) are quite different than the rigidity index under study here.
13
indexes of labor law rigidity are based on what the laws, regulations and standards
say (i.e., de jure). The regulations in practice (de facto) can differ quite substantially
from de jure ones and much of the difference may be due to transactions and
enforcement costs. Indeed, it is often when regulations are very costly to enforce, or
enforcement efforts extremely weak, that the differences between de jure and de facto
regulations are likely to be large.
Since it is primarily at the country level where such differences in labor laws and
regulations appear, most existing studies of the effects of these labor regulations use
cross country differences to explain cross country differences in various labor market
outcomes as in Botero et al. (2004). However, in some exceptional cases, such as in
India where the labor regulations also vary across states, the effects of variations in
labor regulations have also been examined across states and over time within the
country (e.g., Besley and Burgess 2004, Bhattacharjea 2006, and Adhvaryu et al. 2013).
In particular, the study by Besley and Burgess (2004) showed that the outcomes (such
as on industrial employment or value added of formal firms) were generally better
and improving in those states in which the labor regulations were moving in a pro-
employer direction (implying less rigidity) than in a pro-worker direction (thereby
consistent with Botero et al. 2004).
6
Following the critique of Besley and Burgess (2004)
by Bhattacharjea (2006), Adhvaryu et al. (2013) focused on firing restrictions alone and
pointed to the need for controlling for local supply and demand shocks which, if not
accounted for, might bias the estimates of the effects of labor regulations. They
examined the effects of one type of shock, namely rainfall shocks, which they argued
were exogenous and would work through income and demand at the local (district)
level, leading to the prediction that the effects of the labor regulations would vary also
between firms in districts receiving positive rainfall shocks and those receiving
6
They showed, however, that, while employment of formal firms covered by the law decreased when labor
regulations in the state got more rigid or pro-labor, employment in informal firms increased. The more pro-labor
labor regulations exerted negative not only on formal employment but also were not confined to employment
but also to labor productivity, investment, and labor force participation.
14
negative shocks. Moreover, they showed that the rainfall shocks served as a
satisfactory instrument for the demand shocks (affecting income and hence demand
which would in turn affect employment indirectly but not directly through higher
wage rates) emanating from increased agricultural output whose influence on labor
and firm outcomes they were trying to estimate. They also showed that these effects
were restricted to those firms of sufficient size as to be covered by the labor law. Once
again, the results showed that industrial employment was more sensitive to positive
weather shocks in states with less rigid firing rules.
The aforementioned outcome of greater rigidity of labor regulations leading to greater
informality among firms as they attempt to avoid the extra costs derive from these
rigidities of the law has been confirmed in many contexts. To do so, however, as
demonstrated by Elbadawi and Loayza (2008), firms have to operate at very small
scale, and often without sufficient access to capital to adopt efficient technology or
sufficient capital to raise labor productivity. The extent to which it pays to do this may
depend, however, on the degree of enforcement, norms with respect to corruption,
and other factors that can compound the problems of empirically estimating the
effects of labor regulations on labor market outcomes. None of these studies has fully
come to grips with the demonstration by Haltiwanger et al. (2008) that the effects of
labor regulations on employment and other variables may vary by industry not only
because the regulations might vary by industry (something that is actually not
common) but more importantly because firms in some sectors and environmental
circumstances face greater unionization and may have greater need to adjust to
supply and demand shocks than firms in other sectors and circumstances.
To capture the effects of factors that might differ, not only between industries, but also
between firms in the same industry, several studies including the present one make
use of the aforementioned (1) firm-level perceptions of labor laws as an obstacle to the
firm’s operations and growth and (2) firm-level assessments of how much and in what
direction the firm’s employment would be changed as a result of complete
15
deregulation of existing labor laws as revealed in the aforementioned Enterprise
Surveys (and their predecessors, the Investment Climate Surveys). Dollar et al. (2005);
Carlin, Schaffer and Seabright (2006); Carlin and Seabright (2009) have all shown that
these subjective perceptions could also be affected by various kinds of whims and
other influences that might have nothing to do with labor regulations making for
paradoxical findings
7
. Yet, for the specific case of the firm-level perceptions of labor
regulations as an obstacle to doing business taken from these same Investment
Climate Surveys, Pierre and Scarpetta (2006) demonstrated that the indicators of
perceived seriousness of this obstacle were positively and significantly related to the
country-specific scores on the index of rigidity of labor regulations constructed from
the laws themselves.
8
Not surprisingly, a number of studies have investigated the
relation between the rigidity in labor regulation to firm productivity. The results of
these studies have been rather mixed.
9
Quite naturally, one of the reasons why labor market regulations have not received
much attention in these studies making use of these firm level surveys is that labor
regulations are seldom ranked very high among the various obstacles to firm
operations by the firm managers.
10
Although also not dealing explicitly with the labor
market obstacles, Ayyagari et al. (2008) identified important interdependencies
7
For example, based on data in four Asian clothing exporting countries Dollar et al. (2005) showed that, while
the telecom access obstacle (measured by delays in getting a phone line, was seldom mentioned as of above-
average importance by the managers, it was the most important among several different institutional indicators
in explaining firm productivity.
7
Carlin and Seabright (2009) and Carlin, Schaffer and Seabright (2006)
contribute to the resolution of some of these apparent paradoxes by pointing to either reversed causality or more
often the ability to substitute private goods for inadequate access to public goods.
8
Also relevant to the present study is that they showed that the effect of the rigidity of the labor regulation index
was stronger in firms that were either expanding or contracting employment and in those that were innovators.
9
In a more recent study that uses the same Investment Climate Surveys but is limited primarily to MENA
countries Kinda et al. (2011) do include labor regulations both as a separate institutional obstacle to business
potentially affecting firm level inefficiency and as part of a broader Government-Business relations measure.
While, by itself, the severity of labor regulations does not have a significant effect on technical inefficiency -
either positive or negative- in any of the eight sectors studied, when appearing as part of the Government-
Business Relations aggregate obstacle, it turns out to have significant positive effects on inefficiency in both
garments and metal and machinery sectors.
10
Typically, the higher ranked obstacles are the much studied access to finance and access to various utilities,
both cases in which selection and endogeneity issues may be much more serious than in the case of labor
regulations.
16
among the various institutional obstacles and as a result showed that a relatively low-
ranked obstacle could have a strong influence on firm behavior and performance in
part through its indirect influences. Nugent (2012) showed that the seriousness of
corruption obstacle was very positively and significantly associated with the labor
regulations obstacle in Egypt, Syria and especially Morocco (where labor inspectors
have to sign off on each firm’s adherence to the country’s labor laws).
Another related paper is Fiori et al. (2007). Using panel data for OECD countries over
the period 1980–2002, these authors showed that the effects of product market
liberalization on employment are stronger when labor market regulations are
particularly strict. This would imply a certain degree of substitutability between the
two types of reforms on employment creation. Moreover, they also showed that
product market liberalization tends to promote labor market deregulation.
The four recent studies that to our knowledge come closest to the present study in the
use of this source of firm survey data are Kaplan (2009), Bhaumik et al. (2012), Seker
(2012) and Almeida and Aterido (2008). Kaplan (2009) made use of country averages
of firm-specific responses to the same hypothetical question about the employment
changes that would be generated by complete elimination of labor regulations as we
do. In his case these were taken from the Enterprise Surveys for 14 Latin American
countries and, since it was only the cross country differences that were exploited, the
regression results were based on but 14 observations. The results showed that on
average this hypothetical, but complete, elimination of the labor regulations would
result in a significant but relatively modest increase in employment of a little over 2%,
though somewhat larger in the case of very small firms (with less than 20 employees)
and firms in countries with more rigid labor laws (as measured by the perception-
based index of labor rigidity from the Fraser Institute’s Economic Freedom of the
World.
17
Bhaumik et al. (2012), on the other hand, examined the effects of a measure of labor
market flexibility from Botero et al. (2004), a forerunner to the indexes of employment
law rigidity constructed in Doing Business, as well as other institutional indicators on
firm-level productivity across countries and especially within countries across
different types of firms based data from the Enterprise Surveys. They did this for a
single industry, textiles and clothing, in nine developing countries, each of which
afforded a sufficient number of firm-specific observations. This industry was a
fortuitous choice because it is a sector believed to reflect comparative advantage for
developing countries and also because it is a sector in which worker turnover is high,
and labor market regulations relevant. Hypothesizing that the effects of labor market
rigidity (or in their case flexibility) would vary across different types of firms, they
found that indeed small and otherwise disadvantaged firms were negatively affected
by labor market deregulation.
Following Melitz (2003) suggesting that the most efficient and profitable firms are
more likely to export, and Helpman and Itskhoki (2010) suggesting that labor market
rigidities could impede exports, Seker (2012) hypothesized that labor market rigidity
is more likely to impede the ability of export in those sectors for which job turnover
is more important but which is impeded by labor regulations. Seker used of the
Enterprise Surveys of 26 Transition countries from Central and Eastern Europe and
Central Asia between 2002 and 2005 to provide rather robust evidence in support his
hypothesis.
Yet, as should be clear from the above, these studies did not fully take advantage of
the intra- country, cross-firm, industry and regional differences in the data. Kaplan
(2009) took advantage only of cross-country differences and Seker (2012) measured
net job creation at the industry and country level and then used this (along with other
firm and industry level controls) to explain variation across firms in their ability to
export. None of these studies has made reference to enforcement of labor and other
regulations which might vary both between and within countries. More recently,
18
however, Almeida and Aterido (2008) advocated and pioneered the use of an
enforcement strategy similar to one used in this paper. The innovation in this
approach is to recognize that the regulations that actually affect firms are not the
regulations as they exist in the law statutes (de jure) but rather the way they work in
practice (de facto). They developed proxies for enforcement of national level
regulations that would vary across firms within countries and demonstrated the
usefulness of firm-specific responses to other questions in the Enterprise Surveys of
66 countries in constructing such proxies. Their application, however, was to labor
training instead of employment as in this paper. They showed that enforcement
interacted with the index of labor laws had quite different effects on labor training
than did enforcement by itself (indeed the two effects were of opposite direction).
11
1.3. Data and Analytical Framework
1.3.1. THE ENTERPRISE SURVEYS AND OUR SAMPLE
As noted earlier, the main data source comes from the World Bank’s ongoing
Enterprise Surveys (ES). One major advantage of the ES data is its wide country
coverage, especially for developing countries where previously such data had been
so scarce. Despite the inclusion of a few OECD countries such as Germany, Portugal
and Spain, ES data are available for more than 120 countries (mostly developing and
transition economies) since 2002. Another desirable feature is that the Enterprise
Surveys adopt a common survey instrument for countries with such different
11
This approach, focusing on the link between regulations and their enforcement, was motivated by
an earlier study for Brazil by Almeida and Carneiro (2011) which was facilitated by rich administrative
information on the allocation of inspectors of various sorts by municipality and fines imposed. Their
choice of Brazil was interesting because Brazil is known as a country with very rigid labor (and other)
regulations but very limited enforcement. The result has been the creation of a very large informal
sector which obeys few of the country’s regulations. Their results showed that greater enforcement led
to lesser reliance on informal workers within a given firm (but also lower productivity, average wages,
profits, capital per worker, and technology but not lower employment). To mitigate doubts about the
validity of the assumption that enforcement is exogenous, they managed to identify satisfactory
instruments and showed the 2SLS with instruments to be even stronger than those without them.
19
institutional, geographical and economical differences such that global comparisons
can be more easily made, both across-countries and over time.
Moreover, the ES include detailed and representative
12
firm-level data on a rich array
of (a) firm characteristics and performance measures, (b) perceptions of
owners/managers on the business environment in which the firm operates (including
enforcement)and most importantly (c) the firm’s response to the following relevant
question in evaluating the effects of labor deregulation—“If you could change the
number of regular full-time workers your firm currently employs without any
restrictions (i.e., without seeking permission, making severance payments, etc.) what
would be your optimal level of employment as a percent of your existing workforce?”
From this question one can identify separately the (net) percentages of workers each
firm would like to add or dismiss if all existing labor regulations were removed.
While unfortunately this question was not included in all of the more than 120
countries to which ES was administered, we have been able to put together the
relevant data from ES for some 89 countries (in some cases for several different rounds
over time), many of which come from the “comprehensive” dataset
13
. The sample
used consists of more than 53,000 firms from the 89 economies between 2002 and 2009.
The list of countries and survey years is reported in Table A1 in Appendix 1.
1.3.2. DEFINITION OF VARIABLES
For this sample, we construct four different indicators of our employment change
measure as a dependent variable, based on that rather unique hypothetical question.
The first of these measures is simple a dummy variable for whether or not the firm
12
For most countries surveyed after 2006, the sampling scheme follows a stratified random selection. Within
each country (and survey year), the standard strata are industry/sector, firm size, and geographic location. For
details see Kuntchev, Ramalho, Rodriguez-Meza, and Yang (2013).
13
Sampling weight information for firms, however, is not available in some of of the countries and years
surveyed by ES and especially for those surveyed prior to 2005. Even in countries countries which do have
sampling weight information, they are sometimes missing for some firms in the sample.
20
indicated that there would be any change in its employment at all (Any Job Change).
The more important and accurate measures that will be used throughout much of the
analysis are those for the percentage changes. The first of these is “Net Job Change”,
which is defined as the net percentage increase (+) or decrease (−) in the number of
workers the firm world like to hire or fire relative to its current employment level if
all existing labor regulations were removed. The second of these measures in
percentage changes is “Net Job Creation”, which also measures the percentage change
in employment but is only defined for the sample observations where a net increase
in jobs is reported. By definition it is equal to “Net Job Change” for all non-negative
values, and missing otherwise. The third one is “Net Job Destruction” which is to be
equal to the absolute value of “Net Job Change” for all non-positive responses, and
missing otherwise.
Allow for separate assessments of the effects of deregulation of labor on the positive
and negative sides of the employment change, respectively
14
. Complete descriptions
of all three dependent variables are given in Table A2 in Appendix 1. Descriptive
statistics for all employment change measures are provided in Panel A of Table 1.1.
Note that only about half of the firms in the sample indicated that they would change
employment in any direction as a result of total deregulation, among which almost a
third indicated that they would increase employment in the event of deregulation.
From the mean for Net Job Creation one can see that, on average, the Net Job Creation
would be almost 15 percent and Net Job Destruction would be just under 7 percent.
14
Since both variables are non-negative and involve some large values, we also take natural logarithms of “net
job creation” and “net job destruction” to examine the robustness of results to exclusion of large outliers in both
variables.
21
Table 1.1
Chief among the explanatory variables used are the firm’s perception on the
seriousness of various obstacles to its business is evaluated by the following
Table 1 Summary Statistics
Panel A: Dependent Variables
N Mean Std. Dev. Min 10th
percentile
90th
percentile
Max
Any Job Change 53,698 0.51 0.50 0 - - 1
Net Job Change 53,698 6.84 42.5 -100 -15 30 600
Net Job Creation 43,526 14.6 41.9 0 0 30 600
Panel B: Labor Obstacle,
Enforcement Variables and
Productivity Shocks (including
original and transformation)
N Mean Std. Dev. Min 10th
percentile
90th
percentile
Max
ObstLabor 51,637 1.13 1.26 0 0 3 4
meanObstLabor 52,428 1.12 0.53 0 0.49 1.80 4
ManTimeReg 47,053 7.91 11.4 0 0 20 50
meanLog_ManTimeReg 50,005 1.47 0.62 0 0.61 2.24 3.93
ObstInformal 50,673 1.49 1.41 0 0 4 4
meanObstInformal_Low 53,489 0.54 0.19 0 - - 1
DaysPowerOut 45,679 20.0 58.3 0 0 40 365
Log_DaysPowerOut 45,679 1.43 1.59 0 0 3.71 5.90
DaysWaterOut 43,633 10.1 48.5 0 0 10 365
Log_DaysWaterOut 43,633 0.53 1.25 0 0 2.40 5.90
Panel C: Other Explanatory
Variables and Labor Law
Rigidity Indexes
N Mean Std. Dev. Min 10th
percentile
90th
percentile
Max
Years of operation (6–20 years) 51837 0.55 0.50 0 - - 1
Years of operation (>20 years) 51837 0.27 0.45 0 - - 1
Foreign owned 53698 0.10 0.30 0 - - 1
Government owned 53698 0.06 0.24 0 - - 1
Sole proprietorship 53698 0.22 0.41 0 - - 1
Partnership 53698 0.15 0.35 0 - - 1
Medium size 53681 0.34 0.47 0 - - 1
Large size 53681 0.26 0.44 0 - - 1
Agroindustry and food 53165 0.13 0.34 0 - - 1
Textiles and garments 53165 0.16 0.37 0 - - 1
Manufacturing 53165 0.36 0.48 0 - - 1
Service 53165 0.15 0.36 0 - - 1
Retail and wholesale 53165 0.11 0.31 0 - - 1
index O (overall labor rigidity) 53288 43.3 14.8 3 27 63 78
index H (difficulty of hiring) 53288 34.3 25.7 0 0 67 100
index F (difficulty of firing) 53288 46.4 23.3 0 20 80 100
Source: World Bank Enterprise Surveys and Doing Business Surveys (indexes O, H, F)
City level, country, year, effective tax rate, unemployment rate not shown
22
question—“Please tell us if any of the following issues are a problem for the operation
and growth of your business. If an issue poses a problem, please judge its severity as
an obstacle on a four-point scale.” The observed evaluation of the labor regulations
obstacle, as with all other obstacles
15
to the firm’s business, is treated as an ordered
response variable in the following categories “no obstacle”, “minor obstacle”,
“moderate obstacle”, “major obstacle” and “very serious obstacle”. These are coded
on a 0 to 4 scale
16
. Instead of using the firm’s own subjective rating (ObstLabor),
however, we define the seriousness of labor regulation obstacles for the firm as the
average ratings of all other firms in the same industry and location (meanObstLabor).
The purpose is to at least mitigate the endogeneity that might arise if unobservables
at the individual firm level might affect both the firm’s perceived labor obstacle and
its hypothetical employment change.
Additional sources of endogeneity could be unobservable shocks to firms’ business
and productivity activities which could affect the present level of employment and
therefore also the change in employment brought about by deregulation of labor.
Even if the use of meanObstLabor would adequately address the endogeneity issue
caused by unobservables at the individual firm level, such shocks could be universal
to firms in a particular group, i.e. country-industry-city level combination at which
level the averages of labor obstacles are calculated. Therefore, these shocks could still
simultaneously affect meanObstLabor and the firm’s employment decision. Although
it is not possible to identify and measure all unobserved shocks occurring at the group
level, the questionnaire provides information on both the occurrence or not, and the
extent of disruption for each of several sources of disruptions faced by firms. From
15
These are Telecommunications, Electricity, Transportation, Access to Land, Tax rates, Tax Administration,
Customs and Trade Regulations, Skills and Education of Available Workers, Business Licensing and Operating
Permits, Access to Financing, Cost of Financing, Economic and Regulatory Policy Uncertainty, Macroeconomic
Instability, Corruption, Crime, Anti-Competitive or Informal Practices, Legal System/Conflict Resolution. In one
version, largely for robustness purposes we include the average of all these scores (Average OtherObstacles).
16
In order to test for robustness, the four-point scale is further reduced to a 0, 1 dichotomous measure by
recoding values 3–4 to 1 and 0–2 to 0.
23
this we construct (1) a measure of the total extent of the disruptions of these sorts
measured by the number of days during the past year that the individual firm suffered
from service interruptions of any of the following: electricity, water, telephone
connection, and transportation,, and (2) separate measures of the numbers of days
without power or alternatively without water. Hence, the endogeneity problem can
be further mitigated by controlling for these shocks.
The ES questionnaire also enables us to create two proxies for the degree of labor law
enforcement: (1) the percentage of senior management's time spent on requirements
that are imposed by government regulations (including those on labor), and (2) a
dummy variable for a low score on Informal Sector constituting an obstacle to the
firm’s business. These measures are abbreviated as ManTimeReg, and Obstacle
Informal Low, respectively), though it is the averages for these measures for each
firm’s “cell” group (excluding the firm itself), labeled meanManTimeReg (or in
natural logs meanLog_ManTimeReg) and mean Obstacle Informal Low, that are used in
the analysis as explanatory variables. Even in the same environmental conditions, the
managers of different firms may differ quite substantially in their subjective views
about the seriousness of the constraints they are facing and their access to various
public goods. For this reason we construct a measure called “Average Obstacle” that
is the average of all the various obstacles identified in the ES other than the Obstacle
Labor.
In addition to the ES, another important data source is the World Bank’s Doing
Business database that provides us with codable indexes of labor law rigidity for the
year captured in each particular Enterprise Survey (usually the year before the survey
was actually carried out). In particular, three indexes are available for each country
covered, the overall rigidity index (Index O), and that for two quite distinct
components of the overall index, namely, indexes of rigidity in hiring (Index H) and
in firing (Index F). Each of these indexes has a range of 0–100 with 0 the least possible
rigidity and 100 the maximum possible. No country in the world is assigned a 100
24
score. Since the answers that firms give to the effects of removing all existing labor
regulations could also be affected by sudden shocks to which the individual firms
may have been subjected in the year of the survey, we also deem it important to
control for some specific shocks. While questions about several shocks are included
in several of the surveys and responses obtained, the only two of which were
responded to with any degree of completeness were water and power outages. In both
cases, their importance is measured in terms of the number of days without access to
power (DaysPowerOut) and water (DaysWaterOut), respectively. In addition, the
measures of all the other control variables including age of the firm, size and
ownership of the firm, industry, legal status, are also taken from the Enterprise
Surveys, with the exception of country-level variables such as tax rates (as control)
and de-jure labor index (used in secondary analysis reported in Table 1.10).
1.3.3 DESCRIPTIVE STATISTICS
Table 1.1 presents the summary statistics for the sample of firms to be analyzed in the
next section. The average, minimum and maximum values are shown, as well as the
10th and 90th percentiles of the distribution for continuous variables. Panel A features
all five measures of the firm’s decision on employment change. About half of all firms
would change from their current employment level in absence of any existing labor
regulations, while more than a third of the firms would add more workers than they
would let go. Notably, firms as a whole would add only about 7% of their current
work force in this event, but firms in the 90
th
percentile of net job creation report that
they would increase their employment by 30% and one firm would do so by 600%,
highlighting large dispersion especially at the right hand size of the distribution.
Descriptive statistics on key explanatory variables are presented in Panel B of Table
1.1. On average, firms say that Obstacle labor is only a “minor obstacle” (i.e.
ObstLabor = 1 on a 0–4 scale) but this varies quite considerably (represented by the
standard deviation relative to the mean). On average, almost 8 percent of managers’
time is spent on dealing with regulations (especially from tax and labor inspectors).
25
As for the two measures of supply outages, data is available for only about 80% of the
firms. Power outages are more frequent than water outages but in both cases there is
considerable variation in these indicators. Descriptive statistics on all other variables
used in the analysis are presented in Panel C of Table 1.1. As can be seen for most of
the variables deemed relevant to the model, there are 43,000 or more observations,
and for each such variable there is considerable variation within the sample.
1.3.4 ANALYTICAL FRAMEWORK
As noted above, the main objective is to establish a link between the percentage of
employment change by individual firms to meanObstLabor and proxy of labor law
enforcement in the vicinity of each firm, if all existing labor regulations were
eliminated. At the same time, it important to recognize the influence that might be
exercised on these responses by the degree of enforcement of regulations in the
vicinity of each firm in each different country. The importance of doing so was
stressed by Almeida and Aterido (2008) and Almeido and Carneiro (2011) who
demonstrated quite appropriately that the labor laws would have effect primarily
only when they are enforced. Thus, our strategy in assessing the employment effect
of labor deregulation also depends heavily on the introduction of the aforementioned
enforcement measures into the analysis. This means that the effective labor law
rigidity might appear only as part of an interaction term with one of the enforcement
variables.
Specifically, our empirical model can be formulated in two steps. In the first step, a
firm makes the decision whether to alter its employment level given the deregulation.
𝑊 𝑖 = {
1 𝑖𝑓 𝑎𝑛𝑦 𝑗𝑜𝑏 𝑐 ℎ𝑎𝑛𝑔𝑒 0 𝑖𝑓 𝑛𝑜 𝑗𝑜𝑏 𝑐 ℎ𝑎𝑛𝑔𝑒
𝑊 𝑖 = 𝛽 0
+ 𝛽 1
∗ 𝑚𝑒𝑎𝑛𝑂𝑏𝑠𝑡𝐿𝑎𝑏𝑜𝑟 (−𝑖 )𝑐 ,𝑖𝑛𝑑 ,𝑦𝑟 ,𝑐𝑖𝑡𝑦 + 𝛾 ′ 𝑋 𝑖 + 𝛿 1
′
𝑍 𝑐𝑡
+ 𝛿 2
′
𝐶𝑜𝑢𝑛𝑡𝑟𝑦 𝑐 + 𝛿 3
′
𝐼𝑛𝑑𝑢𝑠𝑡𝑟𝑦 𝑖𝑛𝑑 + 𝛿 4
′
𝑌𝑒𝑎𝑟 𝑦𝑟
+ 𝛿 5
′
𝐶𝑖𝑡𝑦 𝐿𝑒𝑣𝑒𝑙 𝑐𝑖𝑡𝑦 + 𝘀 𝑖 (1)
26
Then for firm which would indeed change employment, i.e. W = 1, the second step
involves estimation of separate coefficients on meanObstLabor for firms adding and
removing jobs.
𝐷 𝑖 = {
1 𝑖𝑓 𝑛𝑒𝑡 𝑗𝑜𝑏 𝑖𝑛 𝑐 𝑟𝑒𝑎𝑠𝑒 𝑖𝑠 𝑟𝑒𝑝𝑜𝑟𝑡𝑒𝑑 0 𝑖𝑓 𝑛𝑒𝑡 𝑗𝑜𝑏 𝑑𝑒𝑐𝑟𝑒𝑎𝑠𝑒 𝑖𝑠 𝑟𝑒𝑝𝑜𝑟𝑡𝑒𝑑
𝑌 𝑖 = 𝛽 0
+ 𝛽 1
∗ 𝑚𝑒𝑎𝑛𝑂𝑏𝑠𝑡𝐿𝑎𝑏𝑜𝑟 (−𝑖 )𝑐 ,𝑖𝑛𝑑 ,𝑦𝑟 ,𝑐𝑖𝑡𝑦 + 𝛽 2
∗ 𝐷 𝑖 ×
𝑚𝑒𝑎𝑛𝑂𝑏𝑠𝑡𝐿𝑎𝑏𝑜𝑟 (−𝑖 )𝑐 ,𝑖𝑛𝑑 ,𝑦𝑟 ,𝑐𝑖𝑡𝑦 + 𝛾 ′ 𝑋 𝑖 + 𝛿 1
′ 𝑍 𝑐𝑡
+ 𝛿 2
′ 𝐶𝑜 𝑢𝑛𝑡𝑟𝑦 𝑐 + 𝛿 3
′ 𝐼𝑛𝑑𝑢𝑠𝑡𝑟𝑦 𝑖𝑛𝑑 +
𝛿 4
′ 𝑌𝑒𝑎𝑟 𝑦𝑟
+ 𝛿 5
′ 𝐶𝑖𝑡𝑦 𝐿𝑒𝑣𝑒𝑙 𝑐𝑖𝑡𝑦 + 𝘀 𝑖 (2)
Hence, the parameter 𝛽 1
estimates the effect of meanObstLabor on percentage job
change firm for firms which would reduce employment level in net, while 𝛽 1
+ 𝛽 2
becomes the estimate for those increasing employment.
A single step regression can also be written in the following form
𝑌 𝑖 = 𝛽 0
+ 𝛽 1
∗ 𝑚𝑒𝑎𝑛𝑂𝑏𝑠𝑡𝐿𝑎𝑏𝑜𝑟 (−𝑖 )𝑐 ,𝑖𝑛𝑑 ,𝑦𝑟 ,𝑐𝑖𝑡𝑦 + 𝛽 2
∗
𝐸𝑛𝑓𝑜𝑟𝑐𝑒𝑚𝑒𝑛𝑡 (−𝑖 )𝑐 ,𝑖𝑛𝑑 ,𝑦𝑟 ,𝑐𝑖𝑡𝑦 + 𝛽 3
∗ 𝑚𝑒 𝑎𝑛𝑂𝑏𝑠𝑡𝐿𝑎𝑏𝑜𝑟 (−𝑖 )𝑐 ,𝑖𝑛𝑑 ,𝑦𝑟 ,𝑐𝑖𝑡𝑦 ×
𝐸𝑛𝑓𝑜𝑟𝑐𝑒𝑚𝑒𝑛𝑡 (−𝑖 )𝑐 ,𝑖𝑛𝑑 ,𝑦𝑟 ,𝑐𝑖𝑡𝑦 + 𝛾 ′ 𝑋 𝑖 + 𝛿 1
′ 𝑍 𝑐𝑡
+ 𝛿 2
′ 𝐶𝑜𝑢𝑛𝑡𝑟𝑦 𝑐 + 𝛿 3
′ 𝐼𝑛𝑑𝑢𝑠𝑡𝑟𝑦 𝑖𝑛𝑑 +
𝛿 4
′ 𝑌𝑒𝑎𝑟 𝑦𝑟
+ 𝛿 5
′ 𝐶𝑖𝑡𝑦 𝐿𝑒𝑣𝑒𝑙 𝑐𝑖𝑡𝑦 + 𝘀 𝑖 (3)
where 𝑌 𝑖 is the percentage employment change (including positive, zero, and negative
changes) that would be resulted from deregulation. As discussed above,
meanObstLabor and the enforcement proxy are calculated averages for these
measures for each firm’s “cell” group (i.e. the average of other firms excluding the
firm itself in a country, industry, year and city level combination, as indicated by the
subscript −𝑖 ). 𝑿 𝒊 is a vector of firm-level characteristics (e.g. firm’s age, size and
ownership) that may be expected to exert direct influences on employment change.
𝒁 𝒄𝒕
is a vector of variables at country level that could vary by different years, such as
tax rate. 𝘀 𝑖 is firm-specific error term. In addition, various fixed effects (country,
industry, survey year and level/status of city) are included to control for variations
at these levels.
A key to implementing the empirical strategy is to make use of both the perceived
severity of labor regulations as an obstacle to the firm’s business (meanObstLabor)
27
and the two alternative measures of local level of enforcement (meanManTimeReg and
mean Obstacle Informal Low) that we assume to be, as well as other relevant firm,
industry and country level controls, independent of the firm-specific error term 𝘀𝑖 .
Then as an additional stage of the analysis, we attempt to verify the validity of our
assumption that meanObstacle Labor (and ObstLabor) would reflect the separate effects
of the rigidity of labor regulations (IndexO, IndexH and IndexF) and the local-level
enforcement proxy, on which the meanObstLabor (ObstLabor) is regressed on. Other
components of our set of control variables that we consider as “key variables of
interest the effective tax rate, proxies for middle aged (6–20 years) and old (>20 years)
firms and sector (especially the labor-intensive and general highly competitive sector
Textiles and Garments). The definitions of all explanatory variables included along
with those of the dependent variables are given in Appendix 1 Table A2.
In the interest of simplicity and consistency, the focus in the main results presented is
on the parameter estimates estimated by OLS, standardized by means and standard
deviations. However, due to the specific nature of the dependent variables which in
some cases have zeros or 100 as bounds, for robustness we also make use of Probit
and Tobit estimates. Moreover, because of the relatively large number of zeros in the
different measures of the dependent variables and different possible explanations for
these zeros, we also experiment for robustness purposes with a number of alternative
ways of treating the zeros. These results are given in appendix tables.
H1: Employment change from deregulation of labor should be positively related to
both (a) the severity with which firms in the same country-industry-city group of the
firm view labor regulations as an obstacle to their business and (b) enforcement of the
existing labor regulations, and but on the other hand negatively related to both (c)
firm characteristics that would make it less likely that the firms would need to increase
their employment (such as larger firm size, or age of firm and (d) negative shocks that
might have been regarded as temporary (like power outage shocks).but which might
28
have already reduced the firm’s employment relative to that which the firm would
normally want.
H2: The perceived importance of meanObstLabor should be positively related to both
labor law rigidity and enforcement, and interactions between them (the latter to reflect
the extent to which they are substitutes or complements for each other).
1.4 Empirical Results
1.4.1 MAIN RESULTS: EMPLOYMENT CHANGE AND THE TESTING OF HYPOTHESES
Our empirical results for this hypothesis are given in Tables 1.2–1.5 below. Table 1.2
focuses on H1a, the effects of the perceived seriousness of labor regulations as an
obstacle to business (meanObstLabor) on firm’s decision to alter employment
(equation 1) together with both directions of job change, after controlling for firm
characteristics and fixed effects for country, industry, city-level and year (equation 2).
To keep the sample size as large as possible, the enforcement and shock variables are
excluded in all specifications in this table as they contain a moderate amount of
missing observations, while subsequent tables provide more detailed analysis with
the shock added. In all columns, the effects of (meanObstLabor) are positive and
statistically significant, although the magnitude of the coefficients are sensitive to
changes in controls for job decrease. For this reason, the subsequent tables will focus
on Net Job Change and Net Job Creation (equation 3). To facilitate interpretation of
the magnitudes of the effects of meanObstLabor (and other key explanatory variables
of a continuous nature in subsequent tables), the explanatory variable has been
standardized by its mean and standard deviation so that the estimated coefficient
represents the effect on the dependent variable of a one standard deviation increase
in the variable. Hence in this case, a one standard deviation increase in
meanObstLabor is estimated to generate a net increase in job creation of more than 5
percent on average for the firms which would report job increase under deregulation.
29
Table 1.2
Table 1.3 presents comparable results for Net Job Change and Net Job Creation when
the effects of the two alternative enforcement measures (meanLog_ManTimeReg and
meanObstInformalLow) and the shock variable Log_DaysPowerOut are added. Note
that the power outage shock variable has a highly significant and positive influence
on both outcomes (supporting H1d). The effects of the first enforcement measure
meanLog_ManTimeReg are (as expected) positive and significant effect on Net Job
Change and Net Job Creation, confirming H1b. Yet, the coefficient of the second
enforcement measure (meanObstInformaLow), while positive, is not statistically
significant for either of the dependent variables. It is understandable the coefficients
for Net Job Creation (column 4, 5 ,6) are somewhat smaller than those of in column 2,
3, 4 of Table 1.2, since Net Job Creation does not exclude the firms with zero job change
Stage 1: Probit - ME
(1) (2) (3) (4) (5) (6) (7)
0.032*** 5.016*** 5.794*** 5.444** 4.271* 6.145** 12.217***
(0.009) (1.368) (2.195) (2.132) (2.586) (2.957) (2.337)
Firm characteristics Yes No No Yes No No Yes
Country FE Yes No Yes Yes No Yes Yes
Industry FE Yes No Yes Yes No Yes Yes
Year FE Yes No Yes Yes No Yes Yes
City-level FE Yes No Yes Yes No Yes Yes
Observations 48,531 22,817 22,817 21,352 22,817 22,817 21,352
Adjusted R-sq - 0.019 0.113 0.153 0.019 0.113 0.153
*** Significant at the 1 percent level.
** Significant at the 5 percent level.
* Significant at the 10 percent level.
Table 2 Estimates of the Effects of Mean Obstacle Labor on Any Job Change and Net Job
Change (coefficients for both job increase and job decrease)
FE = fixed effect; ME = marginal effects; OLS = ordinary least squares
Any job change
Notes: meanObstLabor is standardized by its mean and its standard deviation, allowing us to interpret its
estimated coefficient as the effect of a one standard deviation increase in meanObstLabor. Standard errors
are clustered by country × industry × city-level × year. Firm characteristics include all variables in Panel C
of Table 1 except country-level indexes. Estimates of column (1) by logit and OLS yield identical
significance levels as probit.
meanObstLabor
Stage 2: OLS
Net job change
coefficients for increase coefficients for decrease
30
(i.e. include firms with both positive and zero job change) while the latter coefficients
are essentially estimated for firms who have positive job change.
Table 1.3
Naturally, since meanObstLabor is expected to reflect the effects of both the rigidity
of labor regulations and their enforcement, we might expect it to be correlated with
each of the enforcement variables.
17
To allow each of the enforcement variables to also
have both a separate direct effect on the Net Job Change in Table 1.4 and Net Job
Creation in Table 5 and interaction effects with meanObstLabor, in columns (3) and
(5) of both tables we now include all three of these variables at the same time. For
17
Indeed, the correlation coefficient between meanObstLabor and meanLog_ManTimeReg is 0.36 and that with
respect to MeanObstInformalLow is –0.48.
(1) (2) (3) (4) (5) (6)
4.141*** 3.650*** 4.332*** 3.520*** 3.035*** 3.841***
(1.046) (1.037) (1.091) (0.983) (0.985) (1.037)
2.779*** 2.778***
(0.711) (0.718)
0.551 0.937
(0.704) (0.721)
1.407*** 1.310*** 1.404*** 1.584*** 1.485*** 1.581***
(0.296) (0.288) (0.295) (0.312) (0.303) (0.311)
Firm characteristics Yes Yes Yes Yes Yes Yes
Country fixed effects Yes Yes Yes Yes Yes Yes
Industry fixed effects Yes Yes Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes Yes Yes
City-level fixed effects Yes Yes Yes Yes Yes Yes
Observations 41,968 41,763 41,959 35208 35051 35199
Adjusted R-sq 0.084 0.081 0.084 0.093 0.085 0.093
*** Significant at the 1 percent level.
** Significant at the 5 percent level.
* Significant at the 10 percent level.
Table 3 Effects of Perceived Seriousness of the Labor Regulations Obstacle and
Alternative Enforcement Measures on Net Job Change and Net Job Creation
Net Job Change Net Job Creation
Notes: meanObstLabor, meanLog_ManTimeReg, meanObstInfml_Low and
Log_DaysPowerOut are standardized in the same way as indicated in Table 2. Standard
errors are clustered at each level of country × industry × city-level × year. Firm
characteristics include all variables in Panel C of Table 1 except country-level indexes.
meanObstLabor
meanLog_ManTimeReg
meanObstInfml_Low
Log_DaysPowerOut
31
convenience of comparison columns (1), (2) and (4) in Table 1.4 are taken directly from
columns (1), (2) and (3) of Table 3 and these same columns in Table 1.5 are taken from
columns (4)–(6) of Table 1.3.
18
Table 1.4
18
Each of the explanatory variables measured in a continuous way, as opposed to being a dummy
variable, is standrdized such that its coefficient represents the effect of a one standrd deviation on the
dependent variable. Likewise, the interaction terms have been contructed by multipling the
standardized version of first variable in the pair with the standardized version of the second one.
(1) (2) (3) (4) (5)
4.141*** 3.650*** 2.177* 4.332*** 6.813***
(1.046) (1.037) (1.319) (1.091) (1.991)
2.779*** 2.914***
(0.711) (0.712)
0.601
(0.424)
0.551 0.356
(0.704) (0.730)
-0.983*
(0.527)
1.407*** 1.310*** 1.304*** 1.404*** 1.368***
(0.296) (0.288) (0.289) (0.295) (0.293)
Firm characteristics Yes Yes Yes Yes Yes
Country fixed effects Yes Yes Yes Yes Yes
Industry fixed effects Yes Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes Yes
City-level fixed effects Yes Yes Yes Yes Yes
Observations 41968 41763 41763 41959 41959
Adjusted R-sq 0.084 0.081 0.081 0.084 0.084
*** Significant at the 1 percent level.
** Significant at the 5 percent level.
* Significant at the 10 percent level.
Table 4 Estimates of the Effects of meanObstLabor, Enforcement with Interaction
Terms on Net Job Change
Net Job Change
Notes: meanObstLabor, meanLog_ManTimeReg, ObstLabor_TimeReg,
meanObstInfml_Low, ObstLabor_InfmlLow and Log_DaysPowerOut are standardized in the
same way as indicated in Table 2. Standard errors are clustered at each level of country ×
industry × city-level × year. Firm characteristics include all variables in Panel C of Table 1
except country-level indexes.
meanObstLabor
meanLog_ManTimeReg
ObstLabor_TimeReg
meanObstInfml_Low
ObstLabor_InfmlLow
Log_DaysPowerOut
32
Table 1.5
(1) (2) (3) (4) (5)
3.520*** 3.035*** 2.970*** 3.841*** 3.697***
(0.983) (0.985) (0.975) (1.037) (0.991)
2.778*** 2.894***
(0.718) (0.715)
0.584
(0.430)
0.937 0.838
(0.721) (0.746)
-0.544
(0.531)
1.584*** 1.485*** 1.480*** 1.581*** 1.562***
(0.312) (0.303) (0.303) (0.311) (0.309)
Panel B: Firm
characteristics (1) (2) (3) (4) (5)
-2.830*** -2.557*** -2.544*** -2.823*** -2.818***
(0.882) (0.872) (0.874) (0.879) (0.878)
-3.589*** -3.141*** -3.121*** -3.578*** -3.584***
(0.929) (0.900) (0.902) (0.926) (0.926)
-0.494 -0.418 -0.412 -0.547 -0.538
(0.684) (0.673) (0.673) (0.692) (0.692)
-1.584 -1.858 -1.854 -1.631 -1.615
(1.276) (1.275) (1.276) (1.286) (1.285)
2.450*** 2.103** 2.099** 2.427** 2.443**
(0.948) (0.928) (0.928) (0.949) (0.950)
1.115 0.888 0.882 1.116 1.148
(1.015) (1.006) (1.005) (1.014) (1.015)
-5.778*** -5.717*** -5.706*** -5.780*** -5.768***
(0.880) (0.857) (0.856) (0.880) (0.881)
-12.363*** -12.469*** -12.457*** -12.405*** -12.380***
(1.299) (1.306) (1.305) (1.302) (1.304)
-0.033 -0.041* -0.040 -0.037 -0.035
(0.025) (0.025) (0.025) (0.025) (0.025)
Partnership
Medium size
Large size
Effective tax rate
Net Job Creation Panel A: Labor law
rigdity and
Table 5 Estimates of the Effects of meanObstLabor, Enforcement with Interaction
Terms on Net Job Creation (showing key firm characteristics)
Years of operation in
country (6–20 years)
meanObstLabor
meanLog_ManTimeReg
ObstLabor_TimeReg
meanObstInfml_Low
ObstLabor_InfmlLow
Log_DaysPowerOut
Years of operation in
country (>20 years)
Foreign owned
Government owned
Sole proprietorship
33
From columns (3) and (5) in each of these tables it can be seen that the effects of
including the interactions are quite different between the two alternative measures of
enforcement but similar between the two tables for the two different dependent
variables. In particular, the effect of the interaction of meanObstLabor and the time
spent on regulation measure is positive but not significant in columns (3) of both
tables, while not affecting either the sign or statistical significance of this enforcement
measure by itself. On the other hand, when interacted with the second enforcement
measure meanObstInformalLow as in column (5) of both tables, the enforcement
measure by itself now has positive and non-significant effect, but those of the
interaction term negative (and marginally significant in Table 1.4 but not significant
in Table 1.5). In order to shed light on H1c, in Table 1.5 we report the parameter
estimates of several important and relevant firm characteristics. Clearly, consistent
with H1c above, older and larger firms, as well as firms with larger shares of
Table 5 (continued)
Panel C: Industry
dummies (1) (2) (3) (4) (5)
0.276 0.672 0.702 0.336 0.254
(1.298) (1.287) (1.285) (1.299) (1.285)
3.254** 3.956*** 3.920*** 3.248** 3.051**
(1.356) (1.386) (1.385) (1.350) (1.324)
0.498 0.875 0.947 0.434 0.386
(1.069) (1.080) (1.082) (1.069) (1.064)
-0.562 -0.268 -0.275 -0.775 -0.865
(0.990) (0.974) (0.974) (1.005) (0.992)
-2.360** -1.293 -1.441 -2.219** -2.247**
(1.038) (1.006) (0.999) (1.031) (1.028)
Country fixed effects Yes Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes Yes
City-level fixed effects Yes Yes Yes Yes Yes
Observations 35208 35051 35051 35199 35199
Adjusted R-sq 0.093 0.085 0.085 0.093 0.093
*** Significant at the 1 percent level.
** Significant at the 5 percent level.
* Significant at the 10 percent level.
Service
Retail and wholesale
Agroindustry and food
Manufacturing
Notes: meanObstLabor, meanLog_mantimereg, ObstLabor_TimeReg, meanObstInfml_Low,
ObstLabor_InfmlLow and Log_DaysPowerOut are standardized in the same way as
indicated in Table 2. Standard errors are clustered at each level of country × industry × city-
level × year. Firm characteristics include all variables in Panel C of Table 1 except country-
level indexes.
Textiles and garments
34
government in ownership and facing higher effective tax rates are less likely to report
increases in net job creation after deregulation of labor. On the other hand, firms
which are individually owned and in the labor intensive, and generally also highly
competitive, Textiles and Garments sector are more likely to do so.
1.4.2 ALLOWING FOR HETEROGENEOUS EFFECTS OF PERCEIVED LABOR RIGIDITIES
AND REGULATORY ENFORCEMENT ON EMPLOYMENT CHANGES
Especially since the results for the effects of deregulation on Net Job Creation in Table
1.5 revealed significant differences in Net Job Creation by size of firm, age of firm and
industry, this raises the question of the extent to which the effects of meanObstLabor
itself on Net Job Creation (as well as the other two measures of employment change)
may also vary by some of these same characteristics. To capture this, in Tables 1.6–1.9,
we present results obtained when interactions between meanObstLabor and our
primary measure for enforcement on the one hand, and various firm, industry and
country characteristics on the other, are included in the specifications. As in Tables 1.4
and 1.5, each of the interaction terms is constructed as the interaction between the
standardized values of each variable in the pair. As shown in columns (1) and (3) of
Table 1.6, the interactions of mean Obstacle Labor with the dummy for the oldest firms
(>20 years) have negative and significant effects on Net Job Creation and thus also a
negative and significant effect on Net Job Change. These point to a much smaller effect
of meanObstLabor on employment measures for the oldest firms compared to
youngest firms, e.g., the effect on Net Job Creation for the oldest firms being about
30% smaller.
35
Table 1.6
As shown in columns in Table 1.6, the interactions of the dummy for the oldest firms
with the enforcement measure Log Mean Management Time are also significant for
Net Job Change in column (2), but with no significant effect on Net Job Creation in
column (4). For Net Job Change, the effect of this enforcement measure for the oldest
group of firms is only slightly more than a half of that for the youngest firms. Notice
also that the direct effects of older firm dummies on the different employment
measures remain more or less the same way they were in Table 1.5.
Table 1.7 presents the corresponding results when interactions between
meanObstacle Labor and Log Mean Management Time, respectively, are interacted
instead with the Medium and Large Firm Size dummy variables. As shown in the first
(1) (2) (3) (4)
3.659** 2.045* 5.120*** 3.505***
(1.432) (1.086) (1.396) (0.944)
3.686*** 2.979***
(0.842) (0.795)
-1.207 -1.183
(0.802) (0.898)
-1.667* -1.789**
(0.864) (0.891)
-0.329 -0.099
(0.440) (0.469)
-1.694*** -0.491
(0.565) (0.591)
-3.781*** -3.091*** -3.037*** -2.309***
(0.795) (0.735) (0.899) (0.828)
-5.666*** -4.724*** -3.459*** -2.576***
(0.858) (0.775) (0.941) (0.837)
Firm characteristics Yes Yes Yes Yes
Country fixed effects Yes Yes Yes Yes
Industry fixed effects Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes
City-level fixed effects Yes Yes Yes Yes
Observations 48,793 48,550 39439 39254
Adjusted R-sq 0.124 0.123 0.099 0.091
Table 6 Heterogeneous Effect of meanObstLabor & Enforcement, by Firm's Age
Net Job Change Net Job Creation
meanLog_ManTimeReg
meanObstLabor
Years of operation in country
(>20 years)
ObstLabor X "dummy for
6–20 years"
ObstLabor X "dummy for
>20 years"
ManTimeReg X "dummy for
6–20 years"
ManTimeReg X "dummy for
>20 years"
Years of operation in country
(6–20 years)
36
four columns of the table in both cases, the interactions with Large Size have negative
and significant effects on both Net Job Change and Net Job Creation. Strikingly, for
Net Job Change in columns (1), the magnitude of the negative coefficients on the
interaction term is large enough to fully offset the direct positive effect of these
variables, i.e. for firms of large size there is no effect of meanObstLabor on Net Job
Change. In the case of Net Job Creation the sizes of these offsets do offset the positive
direct effects quite substantially but not completely in column (3). This means that the
overall positive employment effect of meanObstLabor on Net Job Creation is much
smaller for the largest size firms. It also implies that for small firms the effects of
meanObstLabor and enforcement on these employment change variables in the event
of labor deregulation would be even larger than the overall positive effect. It should
also be added that the interaction involving medium size dummy also has a negative
and significant effect on Net Job Creation in column (3).
37
Table 1.7
(1) (2) (3) (4)
4.292*** 2.046* 6.304*** 3.495***
(1.566) (1.086) (1.506) (0.944)
3.642*** 3.336***
(0.882) (0.865)
-1.329 -2.046*
(1.052) (1.159)
-4.424*** -5.835***
(1.509) (1.732)
-0.132 -0.271
(0.655) (0.723)
-1.910*** -1.769**
(0.729) (0.856)
-6.240*** -5.793*** -6.027*** -5.486***
(0.859) (0.723) (0.930) (0.800)
-11.710*** -11.592*** -11.713*** -11.745***
(1.053) (1.036) (1.197) (1.253)
Firm characteristics Yes Yes Yes Yes
Country fixed effects Yes Yes Yes Yes
Industry fixed effects Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes
City-level fixed effects Yes Yes Yes Yes
Observations 48,793 48,550 39439 39254
Adjusted R-sq 0.125 0.123 0.101 0.092
Table 7 Heterogeneous Effect of meanObstLabor & Enforcement, by Firm's Size
Net Job Change Net Job Creation
meanObstLabor
Medium Size
Large Size
meanLog_ManTimeReg
ObstLabor X Medium Size
ObstLabor X Large Size
ManTimeReg X Medium Size
ManTimeReg X Large Size
38
Table 1.8
(1) (2) (3) (4)
1.442 2.048* 3.515*** 3.451***
(1.164) (1.075) (1.024) (0.931)
3.188*** 3.052***
(1.203) (0.992)
2.079* 1.658
(1.182) (1.245)
0.887 0.038
(1.218) (1.249)
1.198 0.292
(1.228) (1.297)
1.003 0.520
(1.017) (1.056)
0.431 0.888
(1.305) (1.070)
-0.176 -0.656
(1.316) (1.078)
0.432 0.184
(1.244) (1.029)
0.168 -0.254
(1.192) (0.882)
-0.203 0.375
(1.090) (0.885)
-1.433 -0.916
(1.354) (0.989)
0.310 0.303 0.130 0.242
(1.372) (1.353) (1.240) (1.224)
3.086** 3.432** 3.270*** 3.601***
(1.323) (1.409) (1.237) (1.307)
0.649 0.712 0.748 0.918
(1.219) (1.242) (1.082) (1.079)
-0.187 -0.138 -0.089 0.135
(1.246) (1.190) (1.119) (1.028)
-1.248 -1.065 -1.708 -1.718
(1.364) (1.417) (1.159) (1.169)
Firm characteristics Yes Yes Yes Yes
Country fixed effects Yes Yes Yes Yes
Industry fixed effects Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes
City-level fixed effects Yes Yes Yes Yes
Observations 48,793 48,550 39439 39254
Adjusted R-sq 0.124 0.123 0.099 0.091
Retail and wholesale
ManTimeReg X Retail
Agroindustry and food
Textiles and garments
Manufacturing
ManTimeReg X TxlGmt
ObstLabor X Retail
ManTimeReg X Manu
ManTimeReg X Service
Service
ObstLabor X TxlGmt
ObstLabor X Food
ObstLabor X Manufacturing
ObstLabor X Service
ManTimeReg X Food
Table 8 Heterogeneous Effect of meanObstLabor & Enforcement, by Industry
Net Job Change Net Job Creation
meanObstLabor
meanLog_ManTimeReg
39
In Table 1.8 we present results obtained when the effects of meanObstacle Labor and
enforcement are allowed to interact with the five different industry group dummies,
specifically, food, textiles & garments, manufacturing, services and retail. Without
exception, these interactions are not statistically significant, indicating lack of
heterogeneous effect of meanObstLabor and enforcement on either of the
employment variables. Yet it should be noted that, as was shown in Table 1.6,
presumably because of its greater labor intensity and international competiveness, the
direct effects on Net Job Change and Net Job Creation of the dummy for Textiles and
Garments are positive and significant.
Table 1.9
In keeping with our interest in determining the extent to which labor deregulation
might help alleviate unemployment problems in countries with high unemployment
rates, Table 1.9 presents results in which a country and year-specific measure of the
existing unemployment rate (based on unemployment rates as percentages of the
(1) (2) (3) (4) (5) (6)
2.078* 2.786** 2.020* 3.565*** 4.172*** 3.512***
(1.111) (1.107) (1.099) (0.969) (0.972) (0.967)
3.352*** 3.823*** 3.074*** 3.280***
(0.806) (0.769) (0.753) (0.734)
-1.337** -0.659
(0.631) (0.564)
-3.316*** -1.556***
(0.679) (0.545)
-0.291 -0.603 1.297 -0.302 -0.343 0.578
(2.949) (2.874) (2.708) (1.620) (1.620) (1.574)
Firm characteristics Yes Yes Yes Yes Yes Yes
Country fixed effects Yes Yes Yes Yes Yes Yes
Industry fixed effects Yes Yes Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes Yes Yes
City-level fixed effects Yes Yes Yes Yes Yes Yes
Observations 45,920 46,163 45,920 36805 36990 36805
Adjusted R-sq 0.125 0.126 0.126 0.090 0.097 0.090
Table 9 Heterogeneous Effect of meanObstLabor & Labor Enforcement, by
Unemployment Rate
ObstLabor X Unemployment
rate
Unemployment rate
ManTimeReg X
Unemployment rate
meanObstLabor
meanLog_ManTimeReg
Net Job Change Net Job Creation
40
labor force data from The International Labor Office (ILO) and its interactions with
either meanObstLabor or Mean Management Time on Regulations are added to the
model for each of the three alternative job change measures. Because of missing
observations on unemployment rates for certain countries and years, the sample size
is reduced by well over 2,000 observations making the results no longer as comparable
with those in other tables as was the case in the preceding tables
19
. Nevertheless, from
all columns of the table it can be seen that, by itself, the unemployment rate at country-
year level has no significant effect on either Net Job Change or Net Job Creation. Note,
however, from the interaction terms in the third and fourth rows of Table 9 that the
effects of the interactions of the unemployment rate with the enforcement measure
are statistically significant in both cases and those with the meanObstLabor are
significant for Net Job Change. Other things equal, a one-standard-deviation increase
in unemployment rate
20
would reduce the positive effect of meanObstacle Labor on
Net Job Change by about 50 percent. In the case of the interactions of unemployment
rate and enforcement (ManTimeReg) the effects on Net Job Change and Net Job
Creation are both negative. In the case of Net Job Change and Net Job Creation, i.e.,
columns (3) and (6), these effects are large enough to offset 86% and 47% of the direct
effects, respectively.
Hence, while firms are overall more likely to say they would increase employment as
a result of complete deregulation of labor, their incentives to boost their employment
level as a result of the removal of labor regulations are significantly less related to the
current severity of labor regulations when unemployment rates are high. In general,
therefore, Tables 6–9 reveal considerable heterogeneity in the effects of the both
perceived rigidity of the labor regulations as an obstacle to business (presumably
reflecting both the regulations themselves and their enforcement) and again direct
19
The loss in observations without much of an effect on the results was the main reason why unemployment
rates were not included in the first place.
20
This standard deviation of this unemployment rate variable is for the studied sample only.
41
measures of enforcement across years of operation and size of firm groups as well as
with the national unemployment rate, but not so much by industry type.
All these results further support Hypotheses (1a–1c) and demonstrate that there may
be no unique combination of changes in regulations and their enforcement that would
be likely to maximize increases in employment in all industries, firm types and
classes. In general, it would seem that the largest increases in employment via
deregulation would be obtained in quite young firms of small size in countries where
labor markets are relatively tight, i.e., unemployment rates are relatively low. The
magnitudes of the indicated effects, which will receive additional attention in the
concluding section, however, are not always large.
1.4.3 EFFECTS OF RIGIDITY OF LABOR REGULATIONS AND THEIR ENFORCEMENT ON
THE FIRMS ’ PERCEPTIONS OF THE SERIOUSNESS OF LABOR REGULATIONS AS AN
OBSTACLE TO BUSINESS.
Then in the final stage of our analysis we turn to empirical testing of H2 and the
validity of our assumption that the measure meanObstLabor that has been featured
in the above analysis is a de facto measure that captures the combined effects of the
rigidity of labor regulations (de jure) and their enforcement. To this end, we next turn
to the results of Table 1.10 where the dependent variable is now the directly observed
Obstacle Labor instead of the more exogenous meanObstLabor used as an
explanatory variable in the previous tables. In this case we concentrate our attention
on our first enforcement measure, meanLog_ManTimeReg in all specifications and
each of the three alternative country-level indexes of rigidity of labor regulations , the
overall index (Index O), the index of rigidity in hiring alone (Index H) and that on
firing alone (Index F). The results with each of these pairs (introduced additively) plus
a common set of covariates are presented in columns (1), (3) and (5) respectively of
Table 1.10. In columns (2), (4) and (6) we add another term representing the interaction
between the relevant rigidity index and the enforcement measure.
42
Table 1.10
In each of the columns (2), (4) and (6), that include the interaction terms which we
hypothesized in H2 to be important for ObstLabor, we find the two independent
measures to have positive and highly significant effects on ObstLabor, thereby
providing strong support for H2. In each such case, however, the effect of the
interaction term itself is negative, indicating that the two determinants are net
substitutes for one another. In every case the adjusted R-sq is higher when the
interaction term is added and the magnitudes as well as the directions of the
coefficients of all the other control variables are robust across all the specifications in
the table.
(1) (2) (3) (4) (5) (6)
0.058*** 0.058*** 0.055*** 0.064*** 0.059*** 0.048***
(0.018) (0.015) (0.018) (0.016) (0.019) (0.017)
-0.017 -0.028**
(0.015) (0.012)
-0.110***
(0.012)
-0.031** -0.042***
(0.015) (0.015)
-0.064***
(0.014)
0.016 0.020
(0.015) (0.014)
-0.045***
(0.016)
Panel B: Other
selected covariates (1) (2) (3) (4) (5) (6)
0.499*** 0.499*** 0.500*** 0.501*** 0.500*** 0.497***
(0.010) (0.010) (0.010) (0.010) (0.010) (0.010)
0.076*** 0.070*** 0.078*** 0.074*** 0.077*** 0.082***
(0.013) (0.012) (0.013) (0.013) (0.013) (0.013)
0.118*** 0.103*** 0.120*** 0.114*** 0.118*** 0.122***
(0.018) (0.017) (0.018) (0.018) (0.018) (0.018)
0.040** 0.039** 0.040** 0.042*** 0.045*** 0.039**
(0.016) (0.015) (0.016) (0.015) (0.016) (0.016)
-0.098*** -0.076*** -0.108*** -0.096*** -0.102*** -0.111***
(0.024) (0.023) (0.024) (0.023) (0.024) (0.024)
ObstLabor X Index F
Government owned
Average obstacle
Foreign owned
Table 10 Relating Labor Obstacle Back to the Rigidity of Labor Regulation Indexes,
Enforcement and Firm & Industry Characteristics
Panel A:Main
covariates
ObstLabor
Years of operation in
country (6–20 years)
Years of operation in
country (>20 years)
Index O
meanLog_ManTimeReg
ObstLabor X Index O
Index H
ObstLabor X Index H
Index F
43
Consistent with the common finding (Carlin and Seabright 2009, Kaplan and Pathania
2010) that, even in the same environment, firm managers may differ significantly and
in consistent ways in the seriousness with which they view all obstacles to their
business is the relatively large positive effect of Average Obstacle on ObstLabor in
each column of the table. Other interesting results are that firms in countries with
higher effective tax rates are less likely to view the labor regulations as a less serious
obstacle to their business. The same is true for firms with higher shares of government
ownership.
21
Consistent with some of the earlier findings, firms which are older,
larger and in Textiles and Garments tend to view labor regulations as more serious
barriers to their business than other firms. These results indicate that the effects of
these variables on Net Job Creation are partly direct (i.e., independent of
21
This may be attributable to the fact that government- owned firms may be more highly subsidized in various
ways or face softer budget constraints.
Table 10 (continued)
(1) (2) (3) (4) (5) (6)
0.089*** 0.079*** 0.089*** 0.083*** 0.093*** 0.086***
(0.016) (0.015) (0.016) (0.016) (0.016) (0.015)
0.194*** 0.176*** 0.193*** 0.179*** 0.198*** 0.188***
(0.021) (0.020) (0.021) (0.021) (0.021) (0.020)
-0.002*** -0.002*** -0.002*** -0.002*** -0.002*** -0.002***
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
0.110** 0.121*** 0.104** 0.099** 0.107** 0.118**
(0.048) (0.044) (0.047) (0.046) (0.048) (0.048)
Country fixed effects No No No No No No
Year fixed effects Yes Yes Yes Yes Yes Yes
City-level fixed effects Yes Yes Yes Yes Yes Yes
Observations 46900 46900 46900 46900 46900 46900
Adjusted R-sq 0.285 0.294 0.286 0.289 0.285 0.287
*** Significant at the 1 percent level.
** Significant at the 5 percent level.
* Significant at the 10 percent level.
Medium Size
Large Size
Effective tax rate
Textiles and garments
Notes: ObstLabor, meanLog_ManTimeReg, indexo, o_ManTimeReg, indexh, h_ManTimeReg,
indexf, f_ManTimeReg are all standardized in the same way as indicated in Table 2. Standard errors
are clustered at each level of country × industry × city-level × year. Firm characteristics include all
variables in Panel C Table 1 except country-level indexes.
44
meanObstLabor) and partly indirect operating through the effects on ObstLabor
observed in this table.
1.4.4 TESTS OF ROBUSTNESS
Various steps in the above analysis have been based on choices of measures, ways of
dealing with data idiosyncrasies, specific choices of specifications and estimation
procedures that could have been done differently. In this section, therefore, we report
the results of attempts to test the robustness of the results to the use of alternatives in
a number of such respects. The results of these alternatives are available in an on-line
appendix but which are identified and summarized briefly in the following
paragraphs.
Appendix 1 Table A3 subjects the results of Table 1.3 for each of the three measures
of the employment change dependent variable to two changes are made to that table.
One is to add three relevant variables which were added in subsequent tables
(dummy variables for the firm having more than one plant and for having its own
website) and a third, the aforementioned measure for financial constraint (FinWcInt).
The other change is to add the variable Average Obstacle that had previously been
included only in Table 1.6 for ObstLabor. None of these three ancillary control
variables (FinWcInt, Website and Multiplant) has any significant effect on Net Job
Change and Net Job Creation. In contrast, average obstacle has positive and
significant effects on each of Net Job Change and Net Job Creation, which is
compensated for by slight reductions in the magnitudes of the effects of
meanObstLabor on both outcomes.
Appendix 1 Table A4 investigates the robustness of the estimates of the impact of
meanObstLabor and the first enforcement measure (meanLog_ManTimeReg) and
hence the validity of H1a and H1b to different ways of handling idiosyncrasies in the
data. One of these is the large number of zero values in the reported employment
changes. When the zero values are excluded, not surprisingly the estimates of the
45
positive effects of both meanObstLabor and the enforcement measure become even
larger. On the other hand, to treat the sensitivity of the results to possible outliers in
the Net Job Change measure, two different treatments are applied: (a) winsorizing the
top and bottom tails of the distribution
22
and (b) trimming (or eliminating these
observations in the tails altogether). Again, not surprisingly, in both these treatments
the magnitudes of the coefficients of these two explanatory variables are reduced, but
without affecting their statistical significance. In the last column of Table A4 we
examine the effect of using a quite different functional form, namely ordered probit
for the case in which the measures of Net Job Change are reduced to quintiles. Once
again, although the magnitudes of the effects of standardized meanObstLabor and
enforcement measures are affected, their strong statistical significance is retained.
Similar exercises (not shown but available on request) have been carried out using the
second enforcement measure meanObstInformalLow and similar results are obtained.
All of these robustness tests show that the positive effects of both the firm’s perception
of the seriousness of labor relations as an obstacle to their business and the local
measure of regulatory enforcement remain significant, thereby supporting H1a and
H1b. It is also true that the effects of the shock variables and certain firm
characteristics are robust to such changes, thereby demonstrating the robustness of
our findings with respect to H1c and H1d as well.
In Appendix Table A6-1, A6-2 and A6-3, we subject our estimates in Table 1.10 for
ObstLabor to various robustness checks. In the first of these tables, the dependent
variable used is our more exogenous measure meanObstLabor instead of ObstLabor
as in Table 1.6. Once again, the magnitudes of the rigidity indexes, their enforcement
and the interactions between the two are affected, but not their significance. In Table
A6-2 the equation for ObstLabor is used in Table 1.6 is re-estimated using the second
measure of enforcement, namely ObstInformalLow and in Table A6-3
22
In otherwords, all observations in the 95
th
to 100
th
percentile range are recoded as being equal to the value of in
the 95
th
percentile range.
46
meanObstLabor is estimated using this second enforcement measure. As was the case
in Tables 1.4 and 1.5 above for the Employment change dependent variables, our two
enforcement measures interact somewhat differently with the rigidity indexes, but
their significance remains. The explanatory power of the models with the second
enforcement measures tend to be higher than those with the first enforcement
measure. From the significance of the different individual effects of the other firms
and industry characteristics when this second enforcement measure is used, it is clear
that not only is the support for H2 robust but so too are the effects of the various firm
and industry characteristics on ObstLabor or meanObstLabor.
Lastly, throughout this study, standard errors are clustered at each combination of
country, industry, city level and survey year level (a total of 1,328 clusters). Since error
terms (unobservable characteristics) in regressions tend to be correlated for firms
sharing same observable characteristics and hence also with unobservable
characteristics (Moulton 1990), this could bias the estimated coefficients. For this
reason, two alternative clustering levels are tested (country × industry × survey year
and country × industry × city level), but neither of them causes any noticeable changes
to the statistical significance of our results.
1.5 Concluding Remarks
The results shown in Table 1.10 and Appendix tables A6-1 – A6-3 provide rather
strong support for the usefulness of the meanObstLabor measure generated from the
Enterprise Surveys of the World Bank as a de facto measure of the seriousness of labor
regulations as obstacles to the business of the individual firms. This measure also
captures the de jure rigidity of the labor regulations, together with each of two
standalone measures of enforcement of regulations. Moreover, by computing these
measures as averages in the firms’ specific industry and location (but excluding the
firm’s own response), we have also greatly limited the endogeneity of these measures
relative to earlier studies. These results, and especially their robustness to many
47
alternative specifications, measures and estimation procedures, provide relatively
strong support for H2, one of the key hypotheses tested in this paper.
Similarly, Tables 1.2–1.5 provide robust results showing that liberalizing labor
regulations either de jure or de facto (through less rigid enforcement) may serve as a
means of increasing employment. This may be an especially valuable way of
promoting employment in countries constrained by their institutional circumstances
from being able to boost employment and thereby reduce high youth unemployment
rates through deficit spending, exchange rate depreciation or other methods. The
results for the various control variables in Table 5 and those for the effects of various
interaction terms involving meanObstLabor and the enforcement measures in Tables
1.6–1.9 demonstrate considerable heterogeneity in the effects across firm types,
industry and existing national unemployment rates. These findings support H1a,
H1b, H1c and H1d suggest that numerous firm and industry characteristics also affect
the likelihood and magnitude of increases in employment through de facto
liberalization of labor regulations.
Particularly demonstrated in these results is that there will be certain firm types and
industries where such liberalization effects will be smaller or even negative effects.
Nevertheless, even in these respects the results seem quire robust across different
specifications, measures and estimation procedures, and suggest that firms will at
least moderately increase their employment levels (in the range of 2–5%) for a one
standard deviation increase in either the perceived rigidity of labor regulations or
their enforcement.
To give the reader a more specific set of effects of differences in the perceived rigidity
of labor regulations, on the one hand, and in their enforcement, on the other hand, in
Table 1.11 we present alternative estimates based on the results of Tables 1.5, 1.6, 1.7
and 1.9 for different firm sizes, firm ages (measured by years in operation, and
industries of the effects of complete deregulation of labor on each of our three
48
quantitative measures of employment change. In each case these estimates are taken
from the parameter estimates from a particular column in the particular table
identified in the notes of Table 1.11.
Table 1.11
The estimated effects of one standard deviation increases in the firm’s average
perceived de facto rigidity of labor regulations (meanObstLabor) on the employment
changes that would be brought about by deregulation are given for different firm
sizes, firm years of operation, and industries on the left side of this table. Similarly,
the corresponding estimates of the deregulation of labor effects in the case in which
enforcement is at a level of a one standard deviation above its average are given on
the right side of the table.
Take first the results for firm size. In the case of a one standard deviation increase in
the average perceived increase in rigidity of labor regulations, both Net Job Change
and Net Job Creation will be increased. In the case of increases in enforcement, the
estimates of Net Job Change are even greater than those for Net Job Creation. In the
former case, the estimates of that one standard deviation increase in MeanObstLabor
range from a little over 1 percent for large firms to almost 6 percent for small firms.
The estimated effects of deregulation being carried out in situations when
Net Job Change Net Job Creation Net Job Destruction
Size
Small 4.292 *** 5.856 *** 1.564 *** 3.642 *** 2.956 *** −0.685 *
Medium 2.963 ** 4.201 *** 1.237 ** 3.51 *** 2.689 *** −0.819 **
Large −0.132 1.119 1.251 ** 1.732 *** 1.463 ** −0.269
Firm Age
<=5 3.659 ** 4.828 *** 1.169 ** 3.686 *** 2.626 *** −1.060 ***
6 to 20 2.452 ** 3.814 *** 1.362 ** 3.357 *** 2.569 *** −0.788 **
20+ 1.992 * 3.45 *** 1.458 *** 1.992 ** 2.114 *** 0.181
Industry
Food and Agriculture 3.521 *** 4.984 *** 1.463 *** 3.012 *** 2.117 ** −0.896 *
Textiles and Garments 2.329 * 3.522 *** 1.192 *** 3.62 *** 2.795 *** −0.824 *
Manufacturing 2.64 3.909 *** 1.269 *** 3.356 *** 2.532 *** −0.824
Service 2.445 * 3.833 *** 1.388 *** 2.985 *** 2.99 *** 0.005
Retail 1.873 4.239 *** 2.366 *** 1.755 2.284 ** 0.176
Others 1.442 3.125 *** 1.683 *** 3.188 *** 2.638 *** −0.550
Net Job Change Net Job Creation Net Job Destruction
Effects of meanObstLabor Effects of enforcement
One standard deviation = 0.624 (difference in 25th and
75th percentile is 0.984)
One standard deviation = 0.527 (scale 0-4)
Table 11: Effects of meanObstLabor and Enforcement on Employment, by firm types
49
enforcement is one standard deviation above the average would result in Net Job
Creation increases are estimated to vary from 1.463 percent for large firms to almost
3 percent for small firms and to vary from 2.1 percent for firms in operation for more
than 20 years to a little over 2.6 percent for firms in operation for no more than 5 years.
From the corresponding columns for average firm size and age but different
industries, the estimates of the respective one standard deviation increases in
meanObstLabor vary only slightly from 3.1 percent to 4.9 percent and less still from
2.1 percent to about 3 percent when there is a corresponding increase in local
enforcement.
While these effects of deregulation in the face of differences in either the perceived
rigidity of labor regulations or their enforcement are shown to vary somewhat
according to the characteristics identified in Table 1.11, our attempts to identify other
significant interaction terms capturing heterogeneity in the effects of eliminating the
rigidity of labor regulations have largely failed. Together with all the robustness tests
carried out and reported I Section d above, the most important message to be obtained
from the results of this study are that loosening the rigidity of labor regulations in
countries with relatively rigid labor regulations would seem quite generally to yield
moderate increases in employment and to do so relatively quickly.
These estimates are of course not general equilibrium effects since they are based on
responses of individual firms without taking into account what other firms would
want to do. Clearly, if all firms expanded employment at the same time, this would
certainly raise wage rates and make it harder for firms to find the kinds of workers
they would like to hire. At the same time, however, this does not necessarily mean
that these partial equilibrium estimates are over- estimates of the general equilibrium
effects since the simultaneous increases in employment would also raise aggregate
output, income and thereby product demand as well.
50
Therefore, the employment-increasing effects of deregulation of labor regulations
presented in this paper may be considered complementary to the other existing
studies showing that labor deregulation can improve the ability of firms to deal with
shocks (Advaryu et al. 2013), to realize comparative advantage (Cunat and Melitz
2009), to export (Helpman and Itskhoki 2010, Seker 2012), to expand the formal sector
relative to the informal sector (Besley and Burgess 2004). Our findings are also
complementary to other studies showing that, without such liberalization in countries
with highly rigid regulations, such regulations are likely to remain an important
impediment to a variety of such adjustments (Agenor and El-Aynaoui 2003, Kpodar
2007). At the same time, in the face of the greater evidence in support of these
favorable effects of labor de-regulation, one should not overlook the relevance of other
kinds of benefits of regulation. For example, as suggested by Freeman (2004), greater
rigidity of labor regulation may contribute to lower income inequality.
At the same time, if more data were available, there would be important extensions
that could lead to further refined results and add additional insights. For example, as
the number of surveys with the relevant questions included increases, it would be
possible to examine how the results would differ with actual over time changes in
both the rigidity indexes constructed for the existing labor regulations and their
enforcement. It would also be useful to match data from these ES firm surveys with
corresponding data from worker surveys in order to determine the extent that the
employment responses to labor deregulation is affected by more detailed
characteristics of existing and potential workers. Finally, it could be very useful to
explore the use of either more detailed rigidity indexes of the various regulations to
see which elements matter most as opposed to simply the overall rigidity indexes
based on all the rules relevant to hiring or firing of workers or alternative measures
like those with respect to collective bargaining and labor-management relations
outside of hiring, firing and working hours. These extensions could possibly have the
effect of modifying the conclusions drawn.
51
2. Improving Coverage and Utilization of Maternal and
Child Health Services in Lao PDR: Impact Evaluation of
the Community Nutrition Project
52
2.1 Introduction
The fourth and fifth Millennium Development Goals (MDGs) on maternal and child
health (MCH) have proved to be the most difficult to attain. Even as Lao PDR
successfully transitioned out of the low-income tier of the World Bank’s classification
in 2011, the country is focusing on the health sector to further reduce poverty. The
National Growth and Poverty Eradication Strategy prepared by the Lao government
in 2004 identified health as one of four sectors of focus for eradicating poverty in the
medium term. Numerous efforts in line with this strategy have generated favorable
results. From 1995 to 2009, maternal mortality fell from 796 to 357 per 10,000 live
births, and infant mortality fell from 123 to 68 per 1,000 live births from 1995 to 2010–
2011. Comparing the Multiple Indicator Cluster Survey 2006 (MICS) with the Lao
Social Indicator Survey 2011–12 (LSIS), the percentage of deliveries at health facilities
increased from 17 percent to 38 percent, while home delivery dropped from 85 percent
to 59 percent during the same period. Overall, the proportion of births attended by
skilled health personnel doubled from 20 percent to 42 percent in five years. Despite
this progress, mothers in the neighboring countries of Vietnam and Cambodia are
attended at birth at far higher rates (92.9 percent and 71.7 percent), yielding better
progress on that indicator for MDG 5.
Geographical, sociocultural, and economic challenges have inhibited the progress.
Lao PDR is a landlocked, largely rural country with a variety of food taboos and
languages creating barriers among its diverse ethnic groups. These social
characteristics may influence service-seeking behaviors in relation to MCH. For
instance, many mothers believe that food restrictions in pregnancy lead to smaller
babies and easier (and more survivable) delivery (Phimmasone et al. 1996). Food
taboos can affect breastfeeding in addition to pregnancy (Holmes et al. 2007). The
rural poor are commonly faced with a lack of adequate quality and quantity of food
(Kounnavong et al. 2011). The LSIS shows that 26.6 percent of children are
underweight, and 44.2 percent are moderately or severely stunted. Alarmingly, these
53
rates have been unchanged for a decade, and the stunting and underweight
prevalence among children under age five are the worst in the Indochina region
(Kamiya 2011). The incidence of low birth weight is also likely to be an explanatory
factor in the region’s high neonatal mortality rates (Viengsakhone, Yoshida, and
Sakamoto 2010).
To remedy observed shortages in the number of trained health staff (Yamada,
Sawada, and Luo 2013) and low use rates of health services (WHO 2012), the Lao
government set out a new policy in 2008, in consultation with other multilateral and
bilateral donors, to address nutrition, skilled birth attendance, immunization, and
other MCH services. Although some progress has been made, certain MCH programs,
such as the one for national immunization, have not achieved the intended goals. Only
34 percent of children age 12–23 months received all recommended immunizations
before their first birthday according to the 2011 LSIS, despite remarkable
improvement over the 2006 MICS. Regarding preventable diseases, one out of 10
children in Lao PDR suffered from diarrhea in the two weeks prior to the 2011 LSIS
survey.
Although globally the bottom 40 percent of the population in terms of income
distribution made better progress than the other 60 percent on several MCH outcome
indicators (Wagstaff, Bredenkamp, and Buisman 2014), Lao PDR is considered to be
one of the most inequitable countries for the maternal, newborn, and child health
interventions (Barros et al. 2012). Of 54 developing countries, Lao PDR has the fifth
lowest rank in the composite coverage index, which is the weighted mean of the
coverage of eight essential interventions in the fields of family planning, maternity
care, child immunization, and case management. In particular, skilled birth attendant
coverage is the second lowest among countries included in the study, besting only
Ethiopia. Furthermore, skilled birth attendance (SBA) ranges from 5 percent in the
lowest wealth quintile to 80 percent in the highest (Barros et al. 2012).
54
The distribution of out-of-pocket expenditures for MCH services is also inequitable.
Although out-of-pocket health expenditures for the richest quantile in Lao PDR were
26 percent of monthly household expenses, out-of-pocket expenditure for the poorest
quintile of households accounts for 43 percent of monthly expenses (World Bank
2013). This high level of out-of-pocket expenditures is at least partially attributable to
health financing by the government, which accounts for only 41 percent of the total
health expenditures. Still, higher levels of government support of health service use
need not be long term. Short-term, demand-side subsidies—through conditional cash
transfers (CCTs), for example—can induce positive learning of the health services,
which in turn can lead to better take-up in the long run (Dupas 2014a, 2014b, and
2014c).
In this challenging context, potentially made more difficult by the 2007–2008 global
food crisis, the government of Lao PDR and the World Bank instituted the
Community Nutrition Project (CNP) in 2009 as a pilot program that consists of two
demand-side interventions. CCTs were complemented by community-based
nutrition (CBN) programs to directly give people incentives to use MCH services
offered by the public health center and to educate them on proper nutrition, hygiene,
and child health care.
This impact evaluation seeks to address whether the CNP can make causal claims to
have improved indicators related to its six stated project development objective
measurements for mothers and children under two years old: antenatal care visits,
institutional delivery, well-child checkups, breastfeeding, immunization, and
diarrhea/oral rehydration solutions. The results of quasi-experimental impact
evaluation methods indicate that although general effects for these outcomes are
mixed, the project shows improvements for the poorest 40 percent of the population.
Importantly, although this project was implemented as a pilot, it was done at a
relatively large (though subnational) scale and employs a real-world counterfactual.
55
Instead of being tested against no intervention at all, these results are relevant for
what nonbeneficiaries did in the absence of the program, including receiving benefits
from a mixed range of competing interventions in comparison areas (treatment areas
were still relatively clean from contamination, according to the World Bank
implementing team). This impact evaluation tries to address such practical issues
independently.
This evaluation further fills the evidence gap for MCH impact evaluation through
demand-side interventions. Although the body of impact evaluation evidence has
improved for demand-side interventions such as CCTs, a recent systematic review
finds only eight impact evaluations (six studies in Latin America and two studies in
South Asia), but no such impact evaluation evidence from Southeast Asia on the
causal relationship between CCT and MCH outcomes (Glassman et al. 2014).
Conditional cash transfers have the potential to reduce intergenerational transmission
of poverty and improve the uptake of health services, but there is limited and mixed
evidence on health and education outcomes (Bastagli 2011). Finally, although SBA is
included as one of the MDG indicators for monitoring progress, the impact evaluation
evidence base for SBA as an outcome is still slim (IEG 2013). This is the first impact
evaluation in Lao PDR on any CCT intervention or on SBA as an outcome.
2.2 The Project with Two Demand-Side Interventions
In the wake of the global food crisis, the World Bank provided US$2 million (US$1.63
million was spent) to “improve the coverage of essential maternal and child health
(MCH) services and improve mother and child caring practices among pregnant and
lactating women and children less than 2 years old” in the seven southern and central
provinces of Lao PDR (Trust Fund Grant Agreement covering the period of 2009–
2013). The EU provided an additional €1.44 million of support between 2010 and 2012.
The Community Nutrition Project (CNP) marked the first World Bank project
executed by a Lao line ministry instead of a project implementation unit led by
56
external consultants. The project was administrated by the Department of Hygiene
and Health Promotion in the Ministry of Health (MoH). Implementing agents
included public health centers and their staff for the conditional cash transfer (CCT)
components, and local nongovernmental organizations (NGOs)—or international
NGOs with a local presence—and the Lao Women’s Union (LWU) for the community-
based nutrition (CBN) component. To avoid duplicating the project in the northern
part of Lao PDR supported by the Asian Development Bank (ADB), 62 health centers
in seven out of eight central and southern provinces were selected for participation.
The ministry determined assignment purposively through somewhat opaque criteria.
Characteristics such as the district financial management capacity, health center
staffing, and service capacities were considered in selection.
23
These considerations
make finding comparable nonbeneficiary areas more difficult and allow for an
increased possibility of manipulating the selection of treatment areas.
The CNP bundles two demand-side interventions: CCT and a CBN education
program. The CCT program under CNP differed from many other CCT programs in
two ways. First, the CNP was not specifically means-tested or targeted beyond the
geopolitical targeting or the intervention areas. All pregnant mothers within a
selected health center’s service area were eligible. Although the CNP is implemented
in public health centers in rural areas, the CCT was not means-tested or otherwise
targeted, but was paid to all people—regardless of income—who directly satisfied the
transfer conditions (for example, birth at a health center). Second, the project was not
designed to deliver cash in regular time intervals. Instead, payments were intended
to be made at the time beneficiaries complied with the conditions after enrollment.
Even so, there was a significant backlog of beneficiaries receiving payments.
23
It is not clear what these characteristics are in reality, but according to the project appraisal document, “In selecting the
health centers, both health center characteristics (staffing, service capacity, etc.) and district financial management capacity
will be considered. A complete set of districts and health centers that meet minimum criteria will be identified, and pilot
areas will be selected from this group in ways that permit rigorous evaluation” (World Bank 2009, 5), The World Bank
country team indicated that there may have been slightly different criteria: i) health centers needed at least three health
center staff, ii) health centers should be located neither “too close” nor “too far” from the district hospital, and iii) health
centers should have quality and readiness to take up this program.
57
Information about conditions and entitlements were explained to beneficiaries during
enrollment and CBN activities. A full description of the conditions and transfers is in
table 2.1.
Table 2.1
CCT Incentive Structure
Source: IEG, The World Bank.
Notes: Two transportation charges are allowed to cover the eventuality of false labor. The scheme covers the cost
of transportation (in addition to the appropriate conditionality payment) in cases where a patient is transferred to
a superior medical facility. Round trip transportation is covered for institutional delivery for the patient and her
escort. In cases of a Caesarian section, KN160,000 is paid above the normal delivery benefit. 1 U.S. dollar is
approximately equivalent to KN 8,100 (as of December 2014).
By design, the transfer amount varied depending on the distance from the public
health center, ranging in increments of KN 10,000 from KN 50,000
24
to KN 70,000 for
each 3 kilometers from the health center to the beneficiary’s home.
25
The transfer level
is intended to cover six to 10 days of average consumption per capita for the bottom
40 percent of people in Lao PDR.
26
All conditioned actions received the same size of
24
The KN 50,000 is equivalent to 2.4 days’ earnings, in terms of the minimum daily wage for a 19-year-old worker or
apprentice in Lao PDR, based on the 2014 Doing Business report published by the World Bank. It is equivalent to about
US$6.50 in 2014.
25
With appropriate data, this characteristic could have been used to estimate the effect of incentive size using a regression
discontinuity identification strategy. Unfortunately, neither administrative nor global positioning system data are available
for analysis to determine distance from household to health center.
26
According to the Poverty Profile in Lao PDR (Poverty Report for the Lao Consumption and Expenditure Survey, 2012–
2013) prepared by Lao Statistical Bureau and the World Bank (Pimhidzai and others 2014), the average nominal monthly
consumption per capita in 2012–13 is KN 227,105.
Conditionality Payment frequency
Less than 3 km = KN 50,000
3 km–6 km = KN 60,000
More than 6 km = KN 70,000
Antenatal visit 4 times
Delivery Less than 3 km = KN 260,000
3 km–6 km = KN 280,000
More than 6 km = KN 300,000
Postnatal visit 1 payment after birth
Child 0–12 months checkup Monthly, up to 12 payments
Child 13–24 months checkup Monthly, up to 6 payments Same benefit as enrollment
Total benefit
Same benefit as enrollment
1 payment for delivery; up to
2 transport coverages
Same benefit as enrollment
Same benefit as enrollment
Enrollment at health center 1 payment
58
transfer except for institutional delivery, which merited four to five times the standard
payment in addition to the transportation cost because of other costs involved in
delivery. The cash incentive for delivery ranges from KN 260,000 to KN 300,000,
comparable to the median nominal monthly consumption per capita in a rural area
(KN 270,966), and more than the estimated KN 203,600 comprising the 2012–2013
national poverty line for monthly consumption (Pimhidzai et al. 2014). Furthermore,
given the high level of out-of-pocket expenditures for MCH services (World Bank
2013), beginning in 2012, beneficiaries were exempt from fees for MCH-related
services rendered at the public health centers except for those living in the
Bolikhamxay province. Public health centers received a fixed fee per quantity of
services provided to compensate for the cost. This user fee exemption was gradually
phased out from the end of March 2013 to the end of July 2013. The public health
center is also incentivized to provide more MCH services. Importantly, the CCT
program benefits accrue only for services rendered at the local government health
centers (except in the case of formal referrals to higher-level public facilities).
Activities performed in other health facilities, such as private clinics or government
district, provincial, or central hospitals, do not trigger cash transfer payments.
The second demand-side intervention is the CBN activities, which aimed to encourage
behavioral change and strengthen mutual support in improving nutrition for
children, and pregnant and lactating mothers. The CBN provided cascading training
for local female facilitators before the rollout. Beginning in 2012, regular village
meetings were organized by trained residents to teach and discuss nutrition and
MCH-related issues, including pregnancy and delivery. This educational activity
addressed food taboos and other health behaviors, such as breastfeeding, hygiene,
and the administration of an oral rehydration solution in the case of diarrhea, or even
seeking qualified care in the case of other communicable diseases. The CBN activities
also included sensitization about the CCT initiatives to encourage uptake of antenatal
care, institutional delivery, and postnatal and routine growth checkups. Health
59
Poverty Action, an international NGO for MCH services, facilitated the initial CBN
activities for about six months. In August 2012 the LWU took over the monitoring and
evaluation for the rest of the project. In practice, the CBN activities are taken from
United Nations Children’s Fund (UNICEF) and World Health Organization (WHO)
training on maternal and child nutrition through a participatory approach.
Some supply-side elements to the program were provided through a third CNP
component used mainly for project administration. Sizable administrative support for
overall project supervision at the public health center–level was provided to backstop
the MoH during the project. Regarding support to health providers, the third
component funded the five module training developed by the MoH and implemented
by the Department of Education and Research.
27
Community-based distributors were
trained by the end of 2011 to include essential MCH micronutrients. The facilities
(though not the staff) received a fixed fee from a schedule for each user fee–exempted
service rendered. This was intended primarily to compensate for lost user fees instead
of incentivizing the provision of health care. More training and supervision was
provided, as well as health supplies and commodities. In total, however, actual
supply-side support for health care provision was relatively small.
As a heavily demand-side project, the CCT and CBN components of the CNP relied
on a demand-side program logic. The first component of the project’s objectives,
expanding MCH service use, is largely accomplished through the CCT program. In
the treatment area, the CCT program is operated in the public health centers with
health center staff that is assumed to be both capable of providing quality MCH
services (after taking the multi-module training) and managing CCT funds. Mothers
learn about the program through CCT enrollment campaigns and CBN educational
activities, and they enroll and receive cash incentives at the public health center
27
The five module training covers areas such as basic emergency obstetric and newborn lifesaving skills; essential newborn
care; antenatal care and postnatal care; family planning and integrated management of neonatal and childhood illness. All
treatment health centers have at least one staff member who completed the training.
60
subject to participating in antenatal care, delivery, and/or children’s growth
checkups. Mothers are assumed to know about and attend the village meetings, and
they understand and change their nutritional and health-seeking behavior for the
better. It is also assumed that MCH practices could be improved through talks with
health center staff at the expanded MCH visits (and vice-versa). Most important, this
program logic essentially assumes that demand-side constraints can be overcome by
the financial incentives of the CCT and expanded knowledge of MCH practices
through the CBN component. On the supply side, regarding improving MCH-caring
practices, unpaid local village facilitators are assumed to be available and trainable,
and then reliably hold regular village meetings through CBN activities to raise
awareness of nutrition and available MCH services among mothers. The project
design also assumes that supply-side health services and resources at the public
health center, including the administrative capacity to implement the project, are
sufficient to meet the anticipated uptick in demand.
The CNP also faced many implementation challenges through supply-side
shortcomings, which began in the design stage. Despite some project administration
support, the CNP faced implementation delays because critical design details were
left for the implementation phase. For instance, the cash incentive structure and the
mechanism for distribution of the CCT payments were not initially determined.
Foreseeable challenges, such as the lack of easily accessible local banking services in
the project area, were not addressed until well into implementation; the eventual
solution—requiring CCT payments to be disbursed through public health centers—
placed significant added burdens on an already under-resourced staff and required
more protocols and training for transparency and accountability. As a result, the CCT
payments were not delivered in a timely manner. Two large CCT backlog clearing
campaigns were held in November 2012 and February 2013. Those enrolled in the
program reported an average delay of three months for the enrollment payment. The
61
average delay for payment for delivery was only slightly shorter. This backlog may
have cost the project credibility with subsequent and potential beneficiaries.
Beyond payment delays, compliance with protocols was a significant challenge. In the
treatment areas, approximately 30 percent of households that were eligible for the
CCT program did not enroll in it, even though 94 percent of mothers or guardians
were aware of the program’s existence in their villages. Even among the enrolled, 25
percent of beneficiaries did not receive the enrollment transfer. Payment for
institutional delivery was not made for 28 percent of beneficiaries who fulfilled the
conditions of the transfer. The amount of the transfer seemed to be consistent with the
protocol on enrollment, but more than 65 percent of beneficiaries did not receive even
the minimum transfer amount for delivery (KN 260,000). Furthermore, although the
program’s exemption of user fees reduced out-of-pocket costs in the treatment area,
there were still many people who reported paying for fee-exempted services.
Implementation of the CBN and CCT was uneven. The two demand-side
interventions started at about the same time in January 2012, but the CCT was
implemented in earnest only for 11 months from July 2012 to June 2013. For most
beneficiaries, the CBN activities started earlier than the CCT. Still, there may have
been some anticipation effects where potential beneficiaries changed their health-
seeking behavior after finding out about the possibility of the CCT through the CBN
meetings.
28
All the project activities described above are summarized in figure 2.1.
Imperfect project rollout requires reporting the effects for multiple age groups.
Project-targeted beneficiaries did not receive any tangible treatment for the first few
years after the World Bank approved the project in 2009. This implies that children
age 0–11 months and their mothers received the full set of CNP bundled interventions,
and mothers of 12–17 month-olds received a bundle of services weighted more toward
CBN activities and user fee exemptions during their pregnancy and delivery. It
28
The possibility of these anticipation effects disallows a clean disaggregation of the CNP versus CBN effects.
62
follows that children age 18–23 months at the time of the survey were least likely to
have benefited from the CNP at or before birth. Because of this, most estimation
results in this impact evaluation are reported for both the overall age group (0–23
months) and the age group most likely to receive the full benefit of the program (0–11
months).
Intervention spillovers—households in comparison areas benefitting from the
intervention when they should not have—is possible. The only real barrier to such
spillover for comparator households is the cost of additional travel required to a (more
distant) intervention health center. This would result in a downward bias, likely
making the effect estimates reported here a lower bound on the true effect.
Similarly, nonproject “contamination” of comparison areas may have occurred if
other organizations provided some benefits similar to the CNP. The data did not
support the view that treatment areas were contaminated. However, some
comparison areas were the subject of similar interventions from outside
organizations, notably the Luxembourg Agency for Development Cooperation and
the Lao Red Cross. Although these organizations may have intervened in comparison
areas anyway, the delay in CNP implementation prolonged the period of the vacuum
of services in the comparisons areas. This delay, combined with the lack of a sector-
wide approach (SWAP) in health, increased the likelihood of contamination.
29
As a
result, a similar proportion of eligible mothers in both the treatment and comparison
areas seem to have benefited from a user fee exemption. Furthermore, in the
Bolikhamxay and Champasak provinces, about one-fifth of mothers in the
comparison areas who have a child under two reported enrolling in some type of CCT
scheme; many of these mothers reported that they received a transfer.
30
29
It must be noted that this contamination of the comparison areas receiving benefits is almost certainly a positive outcome
for those living in the affected areas.
30
Without a SWAP or broader donor coordination, it is difficult to verify whether these reports accurately indicate the
existence and disbursal of CCTs in comparison areas or a social desirability bias, but anecdotal evidence from the World
63
Even so, the contamination of comparison and potential spillover of health services in
the data imply that the program effects estimated in this paper are lower bound
estimates for the program compared with a scenario in which no additional benefits
are available to the comparison group. Alternatively, the results may be interpreted
in relation to the de facto counterfactual, or what did happen in the absence of the
program, rather than the theoretical (and unrealistic) counterfactual of nothing
happening in the intervention areas if not for the CNP. Some alternative interventions,
proxied by those observed in the comparison areas, almost certainly would have
taken place in the CNP intervention areas even if the CNP had never been
implemented.
With respect to CBN, 56 percent of people in the comparison area responded that there
are regular village meetings to discuss nutrition and health issues. This proportion is
less than the 83 percent figure in the treatment area, but it could underestimate the
effect of CBN. Yet CBN in the treatment area had slightly more active participation
with better understood messages than in the comparison area. Implementation
variations also existed by province. To the best of our knowledge, no activities similar
to the CNP were undertaken in CNP target areas by the World Bank or other donors.
The World Bank did provide supply-side assistance for five of the seven intervention
provinces through the Health Sector Improvement Project, though it did not provide
CCT or CBN-type services during the implementation of CNP. The national expanded
program on immunization was also implemented in both the treatment and
comparison areas. These activities could have interacted with CNP activities in
uneven ways.
Despite these spillover and contamination issues, results in the effect estimates
reported in this study are not significantly undermined. Instead, the counterfactual
becomes what did happen in reality in the absence of (and in place of) the program.
Bank indicates that other aid agencies with intervention designs somewhat similar to the CNP were likely operating in these
areas.
64
Therefore, the evaluation is measuring the effectiveness of a real-world project against
what happened in observably identical areas. Estimates reported here are thus lower
bounds of program effectiveness against a counterfactual of no interventions
whatsoever in comparison areas.
2.3 Data and Project Outcome
According to the Community Nutrition Project (CNP) 2010 baseline survey
documentation, the baseline survey design included calculations of minimum
detectable effect across proxies of the 11 original project development objective (PDO)
indicators using the 2005 Lao Reproductive Health Survey or the 2006 Multiple
Indicator Cluster Survey. These calculations took into account the multi-level survey
design and intra-cluster correlation. Calculations based on power of 0.80 and two-
sided significance at the 5 percent level yielded an optimal sample size of 3,000
households.
The health center sample was generated through the following steps. First, the
Ministry of Health (MoH) purposively selected 62 health centers (based on loosely
defined criteria) to receive the intervention from all health centers in the seven central
and south provinces designated as eligible for the intervention. Second, the World
Bank country team randomly selected 20 health centers for data collection from
among those 62. Third, after this sampling of the treatment health centers, the country
team identified 20 non-treatment health centers as comparators using Stata‘s
“nnmatch” algorithm to select treatment and comparison health center pairs. This
process used characteristics of the health center,
31
demographics,
32
and the geographic
31
Health center characteristics used include total number of health center staff, whether the health center had a laboratory
for examination, whether it used a computer, and whether the health center used electricity.
32
Demographic characteristics include total population size, poverty rate, and dominant ethnic group. The dominant ethnic
group was defined as more than 70 percent of the population belonging to one of six ethnic groups (Lao, Khmuic, Katuic,
Bahnaric Khmer, Vietic, and Hmong). In cases where there was no dominant ethnic group in a given health center service
area, a best match of the individual ethnic groups was used.
65
characteristics
33
of its service area to generate the best single matching comparator for
each health center.
34
The random sample of 20 intervention health centers was not stratified by province.
As a result, health centers from only six of the seven intervention provinces entered
the random sample. In the search for comparators, the country team identified health
centers in five of the seven intervention provinces (Xekong province was completely
excluded from the sample, and Attapeu province was included only for the treatment
sample).
35
The sampling protocol then called for a random sample of five villages per health
center and 15 households with a child under two years old per village (see appendix
B for a further description of the conceptual sampling framework). In cases with fewer
villages and/or households than prescribed in the protocol, data were collected from
additional households in neighboring villages within the same health center’s service
area. If no such villages were available, additional households were collected from the
existing villages. Instances of supplementing the sample by more than three
households from other villages were rare.
36
The mapped locations of the public health
centers and their corresponding sampled villages are seen in figure 2.1.
33
The geographic characteristics are difficulty of terrain measured through range of elevation, and mean travel time to the
district town.
34
In cases where two of the selected health centers happened to have the same match, the best of the two fits was selected,
and the second best was taken for the rest of the health centers.
35
The reader is again referred to appendix A for a list of the treatment and matched comparator health centers and
corresponding districts.
36
In only 4 percent of villages were there both (i) a fewer number of both baseline and endline households than the protocol
required, and (ii) the number of household in the endline was smaller than the baseline by more than three households.
66
Figure 2.1
Location of Public Health Centers and Associated Villages
Source: IEG, World Bank
Note: HC = health center; Vil = village.
For both the treatment and comparison areas, the survey instrument collected data at
three levels: health centers, villages, and households. The baseline was collected from
April to June 2010, before project implementation. The follow-up was collected from
May to July 2013, just before the project closing in September 2013. Because the
intervention was designed to help children under age 24 months and mothers, and
because the households with a child of that age would be different in the endline
versus the baseline, the two rounds of data collection constituted a health center–level
and village-level panel, and a repeated cross section at the household level.
67
Consistency between baseline and endline surveys was maintained by hiring the same
local survey firm for data collection in both rounds.
As a result, the baseline has 41 health centers, 207 villages, and 2,979 households.
Compared with the national average in the Lao Social Indicator Survey 2011–12 and
other surveys in a similar period, the baseline survey population has a lower
education level of the household head, more non-Lao Tai ethnicity, a lower utilization
rate for antenatal care and delivery services, less health insurance coverage, improved
sanitation facilities, and fewer monthly household expenditures (World Bank 2013).
Because of the Nam Theun 2 Hydropower Project (NT2), one health center in the
comparison area (Sabkam health center) was flooded during the course of the CNP
project implementation, and so was dropped from the analysis. The endline dataset,
therefore, covers 40 health centers, 201 villages, and 3,269 households.
Along with the Sabkam health center, two more health centers (Nongboua from the
treatment group and Sob One from the comparison group) were dropped because the
villages in the service areas of those two public health centers received relocation
benefits and compensation from NT2 during CNP implementation.
37
As a result, data
from 19 treatment and 19 comparison health centers and their respective villages (191
total panel villages) and households (2,864 children in households at baseline and
3,101 children in households at endline) are used in the analysis.
Analysis of the household dataset uses probability weights to make it representative
at the health center level. The probability weight is calculated based on the inverse of
the product of the proportion of villages selected for the survey sample from the
health center service area and the proportion of households selected from a sampled
37
Two small villages (Mamonluek and La Vern) were also dropped because they were not sampled at endline (though they
were at baseline). The Mamonluek village (in Kimae Health Center) and La Vern village (in Phabang Health Center) have
six and seven households; dropping them does not materially affect results. Also, two households do not have health center
information; because the intervention indicator cannot be generated for these households, they are also dropped.
68
village.
38
The proportion of selected households is calculated using the roster of
program-eligible households in each specific village, rather than from the village
population or all households in the village.
The CNP originally defined 11 PDOs. After a formal restructuring by the World Bank,
six main indicators were retained: i) antenatal care provided by skilled health
personnel, ii) child delivery at a health facility, iii) DPT3 immunization before the first
birthday, iv) attending a monthly growth checkup, v) breastfeeding within one hour
after birth, and vi) receipt of oral rehydration solutions (ORS) during diarrhea. Exact
definitions are given in table 2.2.
39
Table 2.2
Project Development Outcome Indicators
Source: World Bank (2015)
Note: DPT = diphtheria, pertussis and tetanus; ORS = oral rehydration solutions; PDO = project development
objective.
38
𝑆𝑎𝑚𝑝𝑙𝑖𝑛𝑔 𝑊𝑒𝑖𝑔 ℎ𝑡 = [
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑉𝑖𝑙𝑙𝑎𝑔𝑒𝑠 𝑆𝑒𝑙𝑒𝑐𝑡𝑒𝑑 𝑓 𝑟𝑜𝑚 𝐻𝐶
𝑗 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑉𝑖𝑙𝑙𝑎𝑔𝑒𝑠 𝑖𝑛 𝐻𝐶
𝑗 ×
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝐻𝑜𝑢𝑠𝑒 ℎ𝑜𝑙𝑑𝑠 𝑆𝑒𝑙𝑒𝑐𝑡𝑒𝑑 𝑓𝑟𝑜𝑚 𝑉𝑖𝑙𝑙𝑎𝑔𝑒 𝑖 𝑖𝑛 𝐻𝐶
𝑗 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝐻𝑜𝑢𝑠𝑒𝑙𝑑𝑠 𝑖𝑛 𝑉𝑖𝑙𝑙𝑎𝑔𝑒 𝑖 𝑓𝑟𝑜𝑚 𝐻𝐶
𝑗 ]
−1
39
The ages of the mother and child are adjusted depending on the outcome being estimated for consistency with the Lao
Social Indicator Survey 2011–12. In other words, the analysis uses mother’s age between 15 and 49. As described in table
3.1, children age 0–11 months are used for all analyses of the formal project development objectives (PDOs), unless
otherwise stated, with the exception of PDO Indicator 3 (which uses children age 12–23 months) and PDO Indicator 6
(which uses children age 0–23 months).
PDO PDO indicators description
PDO1
Percent of women age 15–49 years who were attended at least once during pregnancy
in the past 12 months by a skilled health personnel
PDO2 Percent of women age 15–49 with a birth in the last 12 months delivered at a health facility
PDO3
Percent of children age 12–23 months receiving DPT3 before their first birthday
PDO4
Percent of women age 15–49 with a child age 0–11 months who attended at least one routine
monthly checkup in the past 12 months
PDO5
Percent of women age 15–49 with a live birth in the past 12 months who breastfed
within one hour of birth
PDO6
Percent of children age 0–23 months with diarrhea in previous two weeks who received ORS
69
Because important anthropometric outcomes have not improved much during the last
decade in Lao PDR, this report also examines stunting, underweight, and wasting for
eligible children. Effect estimates on additional outcomes related to each PDO are also
calculated to better understand behavioral changes; these outcomes are defined as
follows:
Antenatal Care (ANC): PDO1 refers to antenatal care from skilled health personnel.
This outcome (listed in the regression table in chapter 5 as Attended by health staff
vs. Attended by other persons + None) takes the value 1 if antenatal care is supported
by skilled health staff (either doctor, nurse/midwife, or auxiliary nurse) and zero
otherwise (antenatal care seen by either traditional birth attendant, village health
volunteer, family, friend, traditional healer, or nobody). PDO1 implicitly assumes that
the proportion of biological mothers who see someone for antenatal care during
pregnancy will be increased, and given that pregnant mothers have limited access to
health facilities other than the nearest public health center, this is the overall extra-
marginal effect of the CNP. Yet given the cash incentive is conditional on visits to the
public health center and not for visits to other facilities, health center visits are
examined against all other options (other institutional visit + non-institutional visit +
no ANC visit). These alternative options against which health center visits are
compared is then broken down to inspect whether there is a substitution effect (or
infra-marginal effect) of visits moving from other institutions
40
to health centers
(health center vs. other institutional) and/or an extra-marginal effect of individuals
coming to the health center who would not have had an institutional visit otherwise
(health center vs. non-institutional + no ANC visit).
Delivery: PDO2 is one of the most important outcomes—the proportion of women
who delivered their baby at the health facility. Although the incentive size for delivery
is far larger, it is intended to cover higher anticipated expenditures incurred for
40
The “other institutions” category includes government district hospitals, provincial hospitals or central hospitals; private
hospitals, clinics, or maternity homes; and hospitals or other institutions in Thailand or Viet Nam.
70
deliveries (for example, food and potentially higher transport costs due to the
uncertainty of time of travel), and be a stronger incentive to encourage more deliveries
at health centers. With the far stronger incentive for this action, and the robust
evidence that vouchers and CCTs are the interventions most likely to improve skilled
birth attendance (IEG 2013), the treatment effect on this outcome is expected to be
pronounced. The health facility is defined here as the public health center, but also
government hospitals (district, provincial, and central) and private hospitals, clinics,
and maternity homes. The PDO2 outcome (institutional delivery versus non-
institutional delivery) takes the value 1 if the delivery took place at any of these health
facilities, and zero if delivery occurred at home, in the forest or outdoors, or in a birth
structure. The infra-marginal effect (health center delivery versus other institutional
delivery) and other comparators (health center delivery versus non-institutional
delivery, health center delivery versus other institutional delivery + non-institutional
delivery) are also examined to better understand the health-seeking behavioral
change regarding delivery.
DPT3: Lao PDR is one of few countries where the prevalence of tetanus has yet to be
eliminated (Masuno and others 2009), even though the disease is fully preventable
through three doses of the diphtheria, pertussis, and tetanus (DPT) vaccination.
Accordingly, the project set the PDO3 indicator to increase the share of children age
12–23 months who receive the three doses of combined DPT-Hepatitis B (Hep B)-
Haemophilus influenza type B (Hib) vaccine before their first birthday. This indicator
implicitly assumes that all the vaccination and date of birth records are observable;
unfortunately, this is not the case, even when requesting records from the child’s
immunization card.
Although some respondents could show full immunization records on their
vaccination card, many responses on the number of DPT3 vaccinations were based on
recall, and no dates are recorded for those who gave this information from recall. The
PDO3 outcome takes the value 1 if children age 12–23 months received DPT-HepB-
71
Hib at least three times according to recall or vaccine card information provided by
their biological mother or guardian, and zero otherwise.
The proportion of children with an immunization card actually observed by
interviewers was also examined as a proxy for the household’s awareness about
vaccination and conscientiousness about their child’s health. In other words, we
assume that the household is more health conscious of their child if they could show
a vaccination card for the child.
Checkup: PDO4 examines the effect on monthly routine checkups for children, which
is incentivized through CCT for up to 12 visits in the first year of life and up to six
visits in the child’s second year (table 2.1). The outcome takes the value 1 if caregivers
indicate that the child went to at least one well-child checkup and zero otherwise (any
growth checkup versus none). The report also examines whether the program caused
an increase in repeat visits with an indicator for two or more visits.
41
Unlike the cases
of antenatal care and delivery, the location of the routine growth checkup and the
personnel who assisted with the checkup are not available in the questionnaire, so
those health-seeking behaviors cannot be examined for checkups.
Breastfeeding: WHO (2010) recommend that newborns begin breastfeeding within
one hour of birth, and this is the basis for PDO5. This outcome takes the value 1 if
children get breastfed within one hour of birth, and zero otherwise. The candidate
responses to the question “How long after delivery did you start breastfeeding
him/her?” differ slightly for baseline and endline. Instead of using the explicit term
“within one hour,” the endline questionnaire lists “immediately” as one of the
candidate responses to this question, which eventually takes the value 1 under the
PDO definition. Even though supplementary guidance is provided to code
“immediately” if the breastfeeding was provided within one hour of birth, the
41
Unfortunately, the structure of the baseline questionnaire does not allow treating the number of visits as a continuous
variable.
72
possibility that the endline interviewer might code “one hour” instead of
“immediately” cannot be ruled out if the biological mother breastfed her child within
one hour of birth. Consequently, a variable for breastfeeding within three hours of
birth was also developed.
Receive ORS with Diarrhea: The final indicator, PDO6, examines the proportion of
children receiving ORS among those who reported having diarrhea in the last two
weeks. The ORS is usually defined as either fluid from an Oralyte packet or
prepackaged Oralyte fluid, but the MoH also promotes the use of a government-
recommended homemade solution as another anti-diarrheic option. Two outcome
variables were developed: one coded for typical ORS usage, and another that includes
the MoH option. The outcome takes the value 1 if children are treated through these
measures during diarrhea, and zero otherwise. Because the CBN component could
have direct effects on reducing the incidence of diarrhea itself through teaching
techniques for improved sanitation and hand washing education, the incidence of
diarrhea within the last two weeks was also examined. Similarly, for those who
contracted diarrhea in the last two weeks, the report assesses the likelihood of seeking
advice and/or treatment from skilled health personnel for the diarrhea or not (not
limited to taking ORS measures).
Anthropometry: Anthropometric measures, such as low birth weight, stunting,
underweight, and wasting, are not explicitly included in the PDOs, though they are
used to frame the project in the appraisal document. These measures could be
improved if CNP had positive influences on maternal and child health-seeking
behaviors. Mother and child weights were measured to the nearest 0.1 kilogram. The
child’s height was measured to the nearest 0.1 centimeter. The height-for-age (HAZ),
weight-for-age (WAZ), weight-for-height (WHZ) standardized z-scores were
normalized by month and computed in accordance with the WHO 2006 child growth
73
standards (WHO 2006).
42
Stunting is dichotomous and defined as HAZ less than two
standard deviations below the median. Similarly, underweight is defined as WAZ
scores two standard deviations or more below the median, and wasting is defined as
WHZ scores two standard deviations or more below median.
43
2.4 Methods
2.4.1 DIFFERENCE-IN-DIFFERENCES
This report uses a difference-in-differences (DD) evaluation method to identify the
plausible causal relationships between project outcomes and the combined demand-
side intervention of conditional cash transfer and community-based nutrition. For
child 𝑖 between 0–23 months of age, the basic difference-in-differences estimating
strategy is described as a linear specification for outcome 𝑌 𝑗 through
𝒀 𝒊𝒋
= 𝜷 𝟎 + 𝜷 𝟏 𝑫 𝒊 + 𝜷 𝟐 𝑻 𝒊 + 𝜸 (𝑫 𝒊 × 𝑻 𝒊 ) + 𝑿 𝒊 ′
𝜷 + 𝜺 𝒊𝒋
where 𝛽 0
is a constant term, a binary variable 𝐷 captures the fixed effects summarizing
the time invariant unobserved effects between the Community Nutrition Project
(CNP) intervention and comparison areas (unity for treatment and zero for
comparison)), 𝑇 is a binary variable representing baseline and endline time variations
(unity for endline and zero for baseline), (𝐷 × 𝑇 ) is an interaction of the CNP
intervention area and time, 𝑋 is a set of observed characteristics for child i that is used
consistently across all outcomes j, and 𝘀 is an individual specific error term. The
evaluation team was interested in estimating 𝛾 , the coefficient of (𝐷 × 𝑇 )
demonstrating the overall effect of the CNP intervention under the ordinary least
42
Biologically implausible outliers are recoded to missing. These include WAZ < 6, WAZ > 5, HAZ < 6, HAZ > 6, WHZ
< 5 and WHZ > 5 as described in the Readme file of the Stata igrowup package prepared by WHO.
43
Low birth weight, defined as less than 2,500 grams, could also be a potentially important outcome, but birth weight
information is not credible because most of the available responses (n=1,079) for final impact evaluation analysis are based
on recall. Half of the samples are placed disproportionately to either exactly 2 kg, 2.5 kg, 3 kg, 3.5 kg or 4 kg.
74
squares. The resulting program estimates are the intention to treat effects at the
household level.
𝑌 represents project development objective (PDOs) indicators, which are predefined
by the project documents. Most of the project outcomes are measured by a binary
indicator that takes the value 1 if maternal and child health (MCH) service–seeking
behavior is made for the child 𝑖 under the study, and zero otherwise (for example,
whether or not a woman had an antenatal visit attended by skilled health personnel).
This report presents linear probability models (LPM) of the DD—and matched
difference-in-differences (MDD) discussed later in this chapter—specifications
including province-level fixed effects and robust standard errors clustered at the
village level.
As a robustness check, the report also estimates a nonlinear logit model. With a DD
estimation strategy on binary outcomes, the interpretation of γ under a logit (or
probit) specification does not directly indicate a causal relationship because all
expected outcomes are bounded by zero and 1, which could violate the parallel trend
assumption (Puhani 2012). To this end, the report calculates the marginal effect of the
dummy variable of interaction term for causal interpretation as the difference in cross-
differences between the conditional expectations of the observed outcome and the
counterfactual outcome.
44
Continuous outcomes, including most anthropometrics,
such as standardized height-for-age, are estimated through a linear DD model; results
with and without covariates are reported. The impact evaluation estimates for 𝛾 are
reported both with and without controlling for covariates (under the heading DD) for
the linear and nonlinear models for dichotomous outcomes. Regression tables report
only the coefficients for the outcomes of interest, and report the baseline and endline
44
This is done through the “margin” post-estimation command in Stata 12.1 (Stata Corporation, College Station, Texas,
USA). Since the difference-in-differences estimates the average treatment effect on the treated at the endline, to run the
“margin” command, this report sets both the time and intervention variables as 1, and the rest of the variables are set to the
sample mean.
75
mean values of the comparison and treatment groups. The analysis of dichotomous
dependent variables also includes nonlinear logit estimates.
The set of covariates in vector 𝑋 are chosen to explicitly control for variables that could
potentially confound the relationship between treatment and the MCH outcomes of
interest. The set of covariates consists of four different levels of individual, household,
village, and health center attributes. Individual, child-level characteristics include the
child’s age, gender, and whether the child is the mother’s firstborn. The household-
level characteristics captured in the regression are the mother’s age, weight,
education, presence of the child’s grandmother and grandfather, whether any
children passed away in the household, ethnicity, the number of children in the
household, the total number of births the mother delivered, socioeconomic status
(SES), and external shocks. The SES variable is produced by a principal component
analysis (PCA) of the household asset indices (Filmer and Pritchett 2001)—through
the examination of scree plots of eigenvalues and factor loadings, and the SES is
divided into one short-term (consumables) asset index
45
and two long-term (durables)
asset indexes
46
which use consumption-type assets and durable assets (further details
are found in appendix C). A shock index was also created using PCA.
47
Considering
that the intervention was born out of the global food crisis and because it does not fit
well in psychometric analysis of the other shock items, price shock dummies—
increases in consumption food prices and decreases in agricultural selling prices—
enter into the regressions separately. The presence of the child’s grandmother and
grandfather is included to control for the intra-household decision-making dynamics
in determining MCH practices as conveying intergenerational knowledge on delivery
options and experience at the time when the mother was born. Village-level variables
45
The short-term asset index includes ownership of a motorcycle, bicycle, refrigerator, electric rice cooker, electric fan, two-
wheel tractor (tuk-tuk), boat, fishing net, radio, telephone, mobile phone, and satellite dish.
46
The long-term asset index includes availability of a toilet, main source of electricity (electricity or fuel), main source of
floor (high class, wood, low class), and main source of wall (wood or bamboo).
47
The shock index includes dichotomous variables for whether the household experienced drought, fire, floods, crop
disease, illness or death of the household head, illness or death of other household members, resettlement, and robbery.
76
include travel time to the closest health center from the village; urban or rural status;
and ethnic congruence between the village majority and the household head. Finally,
since gender and ethnicity of the health staff at the public health center could be
important considerations for access to delivery in the context of Lao PDR (Sychareun
et al. 2012), those variables were explicitly controlled. For the consistency of the
analysis, the same set of covariates 𝑋 is applied in analyzing the various outcomes of
interest unless otherwise stated. Further details on the definitions of each covariate
are provided in appendix D.
The identifying assumption of the DD method is parallel trending after controlling for
covariates—that is, the observed change in the outcome for the comparison group
would be the same as the counterfactual change in the treatment group had there been
no intervention. This is a strong assumption which cannot be directly tested. To
bolster this assumption in practice, many empirical studies use multiple rounds of
pre-treatment data for both the treatment and comparison group. If the parallel trend
can be observed between periods before intervention, it can be argued that this is also
likely to be true for the later period between baseline and endline. In this report,
despite having only one pre-treatment dataset, time trajectories of outcomes can be
constructed for both the treatment and comparison group, taking advantage of the
child’s age in months (as shown in appendix E). This would mimic an actual “pre-
trend” for antenatal and delivery outcomes that occurred at or just around child birth.
The similarities of such trajectories between the treatment and comparison area
provide useful—if still somewhat indirect—evidence on the plausibility of the
identifying assumption. As shown in appendix E, the parallel trend for the pre-
treatment period (different age in month under the baseline dataset) seems to hold for
PDO1, PDO2, PDO4 (12–23 months) and the tested anthropometric outcomes.
However, PDO3, PDO5 and PDO6 follow a similar but less exact trend during the
pre-treatment period, meriting caution in the interpretation of the outcomes for these
measures.
77
2.4.2 MATCHED DIFFERENCE-IN-DIFFERENCES
As noted above, the comparators for the 20 treatment health centers were selected
before initiation of the baseline survey. This pre-matching exercise identified
treatment and comparison health center pairs (appendix A) using administrative data
and followed the selection criteria ostensibly employed by the Ministry of Health. As
a robustness check, this report also uses a propensity score matching in conjunction
with the difference-in-differences estimation procedure (a matched difference-in-
differences or MDD model).
This second round of matching was done using the survey data collected for this
evaluation through calculating the propensity score by again following the treatment
assignment decision of the MoH using characteristics of the health centers and their
service areas. Given the relatively small number of health centers (n = 38), k nearest
neighbors with k=4 propensity score matching was specified. The health center level
pre-treatment baseline variables for matching included the number of proper staff at
the public health center, mean and range of elevation of the public health center, mean
access time to the health center from the villages in the service area, population of the
health center service area, ethnicity, and basic infrastructure of the health center. The
proxy of the pre-treatment outcomes, such as the number of deliveries at the public
health center and the postnatal visit at baseline, was also included in the matching
algorithm to account for unobserved characteristics that might be correlated with the
intended outcome. As a result, the risk of bias was reduced for all of the above pre-
treatment baseline variables except electrification of the public health center
(appendix F). In particular, the bias reduction in the proxies of the pre-treatment
outcomes is 99.5 percent for public health center delivery and 13.6 percent for
postnatal visits. However, because 11 of the 38 health centers were off-support after
applying this second matching algorithm, the analyzed sample is reduced to 27 health
centers (12 intervention and 15 comparison). Most of the off-support health centers
are located in northern regions of the project area. Furthermore, though the baseline
78
balance was improved for delivery outcome, mother’s educational background, and
socioeconomic status of the household, the balance of other variables, such as
ethnicity, became somewhat worse (appendix G). These imbalances were, however,
explicitly controlled for in the set of covariates, and time invariant unobservable
confounders on the village- and health center–level are ameliorated through the
difference-in-differences specification. After this matching exercise, the difference-in-
differences analysis was repeated on the weighted on-support sample, as previously
described.
2.5 Results
2.5.1 MAIN RESULTS
The only statistically significant effects of the project on improving the project
development objective (PDOs) indicators is for PDO3 for children between one and
two years old having received the full complement of diphtheria, pertussis, and
tetanus (DPT) vaccinations (at least three shots).
48
In the unconditioned linear models,
there is also a marginally significant result on the probability of attending at least one
routine growth checkup for the full sample (10.7 percentage points, PDO4), though
the effect washes out when controlling for covariates and even have a marginally
significant negative effect in matched sample with logit. Contrary to original
expectations and a recent comprehensive systematic review of the literature (IEG
2013), although the Community Nutrition Project (CNP) has marginally significant
effect on matched sample with logit, the conditional cash transfer (CCT) in Lao PDR
does not lead to a statistically significant difference in the rate of institutional
deliveries (PDO2). Neither did the project demonstrate an effect on the PDO-defined
outcomes of antenatal visits (PDO1), breastfeeding within one hour of birth (PDO5),
or of administering oral rehydration solutions (ORS) to a diarrheic child (PDO6).
48
The official project development objective designates that the DPT3 series should have been completed before the child
turns one year old, but since vaccine administration dates are unavailable for a large share of the children, this requirement
(vaccination before the first birthday) is not included in the definition used for this evaluation.
79
Table 2.3
Difference-in-Differences Results on the Full and Matched Sample for PDOs
Notes: The robust standard error is reported in parentheses, which is clustered at the village level. The coefficient and standard error of the
time and intervention interaction term are reported through linear probability model (LPM) and marginal effect of logit regression. Sample
size for LPM and logit is not always the same, but sample size for LPM is only reported here given the minor difference between the two
specifications. comp = comparison area; DD = difference-in-differences; LPM = linear probability model; treat = treatment area.
a. DD means simple difference-in-differences without covariates, and DD is equal to [(Endline Treat) (Baseline Treat)] [(Endline
Comparison) (Baseline Comparison)].
Significance level: * = 10 percent, ** = 5 percent, *** = 1 percent.
2.5.2 ASSOCIATED RESULTS
Even though the project can take credit for improvements in only one of the six PDOs,
that is largely because of a mismatch between the definition of the PDOs and what
the project was designed to do or what the data can reliably answer. By examining the
domains of these objectives in more detail, this evaluation shows that the project had
a significant effect on a meaningful aspect of nearly all of the objectives.
The results in the tables in this section come in pairs by age group. Most PDOs are
defined as being relevant for children under 1 year (or analogously, births within the
last year). Each of the alternative views of the PDOs gives estimated results for this
age group and the complete age group that was sampled: children under 2 years. The
exception is PDO3, which can only be defined for children between 1 and 2 years of
age, and PDO6 which is defined for children under 2 (the table presents alternative
outcomes for children under 1). The relevant PDO from table 2.3 is reproduced at the
top of tables 2.4–2.8.
PDO Indicator Comp Treat Comp Treat n LPM Logit (margin) n LPM Logit (margin)
0.28 0.43 0.55 0.73 0.035 3274 0.015 0.008 2320 0.087 0.079
(0.060) (0.051) (0.057) (0.053) (0.065)
0.13 0.21 0.25 0.35 0.014 3274 -0.018 -0.018 2320 0.047 0.070 *
(0.045) (0.040) (0.053) (0.048) (0.039)
0.41 0.45 0.46 0.66 0.164 *** 2277 0.149 *** 0.193 *** 1643 0.119 * 0.162 **
(0.058) (0.050) (0.064) (0.062) (0.089)
0.04 0.10 0.42 0.59 0.107 * 3257 0.046 -0.104 2307 0.004 -0.222 *
(0.056) (0.054) (0.100) (0.082) (0.121)
0.41 0.40 0.36 0.42 0.074 3257 0.070 0.076 2308 0.042 0.070
(0.062) (0.060) (0.062) (0.074) (0.072)
0.64 0.64 0.62 0.70 0.077 933 0.042 0.048 731 0.010 0.020
(0.082) (0.085) (0.088) (0.114) (0.113)
PDO6
PDO1
PDO3
PDO5
PDO2
PDO4
Full Sample Matched Sample
Baseline Endline DD
a
With covariates
DPT is at least three times (last birth, 12-23
months)
Any growth checkup vs. None (last birth, 0-11
months)
Breastfeeding within one hour of birth vs. None
(last birth, 0-11 months)
Received ORS (1, 2 or 3) during diarrhea vs.
Not received (last two births, 0-23 months)
With covariates
Attended by health staff vs. Attended by other
persons + None (all births, 0-11 months)
Institutional delivery vs. Non-institutional delivery
(all births, 0-11 months)
(Et-Bt)-(Ec-Bc)
80
Antenatal Care: Do the PDO1 results mean that CNP has no impact at all on antenatal
care? Some important nuances are found when other associated antenatal care
behavioral outcomes are examined in table 2.5.
The impact evaluation results show that the project cannot claim attribution for the
improvements seen in PDO 1 (table 2.3, row 1; table 2.5, row 1a), defined as antenatal
care visits by skilled health personnel for mothers giving birth in the last year. Despite
the significant results for the more complete sample of beneficiaries (births within the
last two years) in the better identified specifications (the matched sample), results are
empirically similar for the representative (full) sample, measuring by whether care
was given by personnel or by institutions (table 2.3, rows 1a and 1b, 2a and 2b). This
is not surprising given the 0.97 correlation rate between antenatal care visits measured
by staff and institutional attendance. The matched difference-in-differences results do
suggest that a small share of mothers in the intervention area who would have had
visits by health staff outside of the health center (often at home or in the village center)
instead have those visits at health centers where they would be paid for their time;
49
however, these results are not generalizable to the intervention population as a whole
because the matching process drops health centers from the representative sample.
49
The margins inducing this change are quite small: The share of mothers that had a health staff visit outside of a health
institution dropped from 3.6 percent in the baseline to 0.75 percent in the endline for the treatment area, but increased from
2.6 percent to 2.9 percent in the comparison service area.
81
Table 2.4
Antenatal Care Behavioral Outcomes
Notes: The robust standard error is reported in parentheses, which is clustered at the village level. The coefficient and standard error of the
time and intervention interaction term are reported through linear probability model (LPM) and marginal effect of logit regression. Sample
size for LPM and logit is not always the same, but sample size for LPM is only reported here given the minor difference between the two
specifications. comp = comparison area; DD = difference-in-differences; LPM = linear probability model; treat = treatment area.
a. DD means simple difference-in-differences without covariates, and DD is equal to (Endline Treat) (Baseline Treat) ((Endline
Comparison) (Baseline Comparison)).
Significance level: * = 10 percent, ** = 5 percent, *** = 1 percent.
Lines 3 to 5 of table 2.3 examine the effects of the program according to what actually
was incentivized by the CCT. Only antenatal care visits at health centers were eligible
for the transfer, as opposed to a broader category of health institutions that can
include province and district hospitals and private clinics.
50
Across both the 0–11
month age group consistent with the PDO accounting and the 0–2 year age group
included in the survey work and eligible for benefits from the project, there is an
increase of between 10.5 percentage points and 21.5 percentage points in the share of
pregnant mothers who sought care specifically at public health centers versus all other
options in linear probability specifications. This result holds for nearly all
specifications in lines 3a and 3b (the logit specifications in the full sample being the
50
Visits to those other institutions would be eligible only in cases of referral from the health center.
PDO Indicator 1: Antenatal Care Comp Treat Comp Treat n LPM Logit (margin) n LPM Logit (margin)
0.28 0.43 0.55 0.73 0.035 3274 0.015 0.008 2320 0.087 0.079
(0.060) (0.051) (0.057) (0.053) (0.065)
0.26 0.43 0.51 0.74 0.055 5759 0.057 0.048 4129 0.126 ** 0.111 *
(0.049) (0.045) (0.050) (0.052) (0.065)
0.27 0.41 0.54 0.72 0.048 3258 0.031 0.021 2308 0.097 * 0.090
(0.059) (0.050) (0.058) (0.053) (0.066)
0.25 0.41 0.50 0.74 0.075 5565 0.076 * 0.064 3983 0.143 *** 0.131 **
(0.051) (0.045) (0.051) (0.050) (0.064)
0.13 0.24 0.42 0.64 0.117 ** 3258 0.105 * 0.053 2308 0.197 *** 0.166 *
(0.056) (0.054) (0.082) (0.065) (0.092)
0.12 0.24 0.38 0.64 0.132 *** 5565 0.142 *** 0.094 3983 0.215 *** 0.173 **
(0.049) (0.049) (0.074) (0.062) (0.087)
0.48 0.58 0.77 0.88 0.012 1558 0.017 0.039 1034 0.028 0.053
(0.102) (0.074) (0.038) (0.089) (0.035)
0.48 0.58 0.77 0.86 -0.008 2607 0.035 0.053 1713 0.030 0.050
(0.084) (0.068) (0.046) (0.086) (0.044)
0.15 0.29 0.47 0.70 0.090 2897 0.057 0.008 2061 0.123 * 0.094
(0.061) (0.057) (0.076) (0.067) (0.087)
0.14 0.29 0.44 0.71 0.123 ** 4950 0.110 ** 0.059 3555 0.173 *** 0.134
(0.053) (0.052) (0.071) (0.063) (0.084)
With covariates
Matched Sample Full Sample
With covariates
1a
Baseline Endline DD
a
PDO1: Attended by health staff vs. Attended by other
persons + None (all births, 0-11 months)
1b
4a
5a
5b
2a
2b
3a
3b
4b
PDO1 - older: Attended by health staff vs. Attended by other
persons + None (all births, 0-23 months)
(Et-Bt)-(Ec-Bc)
Institutional ANC visit vs. Non-institutional + No ANC visit
(last birth, 0-11 months)
Health center visit vs. Non-institutional + No ANC visit (last
birth, 0-11 months)
Health center visit vs. Non-institutional + No ANC visit (last
birth, 0-23 months)
Institutional ANC visit vs. Non-institutional + No ANC visit
(last birth, 0-23 months)
Health center visit vs. Other institutional + Non-institutional
+ No ANC visit (last birth, 0-11 months)
Health center visit vs. Other institutional + Non-institutional
+ No ANC visit (last birth, 0-23 months)
Health center visit vs. Other institutional visit (last birth, 0-11
months)
Health center visit vs. Other institutional visit (last birth, 0-23
months)
82
only exception). So although it cannot be said that the PDO was met for increasing the
rate of antenatal care overall, the project did what it was designed to do: increase
antenatal care visits to public health centers.
To understand why there would be an effect on health center–specific antenatal care,
but not on general antenatal care, the analysis investigated whether the increase in
health center antenatal care came from a substitution effect (an inframarginal effect)
that simply induced women who likely would have sought antenatal care anyway to
switch from other institutional care into health center care, or whether there was an
extramarginal effect, in which mothers who would not have sought institutional care
came to the health center. Although no evidence was found for a substitution effect,
there is more robust evidence of an extramarginal effect for the complete age group
in the linear models.
51
This highlights a mismatch between the PDO indicators and
the project design: The extramarginal effect for health center-specific visits among the
complete age group is swamped by the non-differentiated null results, which include
visits for institutions not incentivized by the project.
Delivery: The primary indicator for assessing the progress of the fifth Millennium
Development Goal—a reduction in maternal mortality of 75 percent—is measured by
the improvements in the proportion of births attended by skilled health personnel.
The Independent Evaluation Group recently completed a systematic review of all
interventions with an impact evaluation that estimates improvements in skilled birth
attendance (SBA) (IEG 2013). That review concluded that vouchers and CCTs were
the interventions most likely to result in improvements in SBA. By contrast, this CNP
intervention in Lao PDR found positive effects for the matched sample, but not for the
representative full sample (rows 5a and 5b of table 2.6). In the face of the
implementation challenges previously described, the instruments in this intervention
51
As will be seen in the rest of the report, the full age group models often report larger point estimates that are more likely
to be statistically significant. This may be because the backlog in CCT payments to early participants (those 12–23 months)
undermined participation of later beneficiaries (those age 0–11 months).
83
were insufficient to improve health-seeking behavior in most specifications,
particularly in the full sample estimates. Most frequently in the specifications in table
2.6, the point estimate for program effectiveness is near zero, with relatively small
standard errors.
Table 2.5
Delivery Behavioral Outcomes
Notes: The robust standard error is reported in parentheses, which is clustered at the village level. The coefficient and standard error of the
time and intervention interaction term are reported through linear probability model (LPM) and marginal effect of logit regression. Sample
size for LPM and logit is not always the same, but sample size for LPM is only reported here given the minor difference between the two
specifications. comp = comparison area; DD = difference-in-differences; LPM = linear probability model; treat = treatment area.
a. DD means simple difference-in-differences without covariates, and DD is equal to [(Endline Treat) (Baseline Treat)] [(Endline
Comparison) (Baseline Comparison)].
Significance level: * = 10 percent, ** = 5 percent, *** = 1 percent.
As with the PDO1 outcomes, rows 3a through 5b examine whether there was an effect
on the incentivized behavior of delivering specifically at a health center. Surprisingly,
evidence indicates the opposite may be true. The full sample regression estimates
show that among those who recently delivered at a health institution, the institution
was some 20 percentage points less likely to be a health center for mothers living in
intervention catchment areas compared with mothers living in comparison areas. In
other words, the proportion of pregnant mothers delivering at public health centers
PDO Indicator 2: Delivery Comp Treat Comp Treat n LPM Logit (margin) n LPM Logit (margin)
0.13 0.21 0.25 0.35 0.014 3274 -0.018 -0.018 2320 0.047 0.070 *
(0.047) (0.040) (0.053) (0.048) (0.039)
0.11 0.20 0.23 0.33 0.013 5759 -0.010 -0.020 4129 0.056 * 0.071 ***
(0.039) (0.035) (0.045) (0.034) (0.027)
0.04 0.08 0.14 0.17 -0.013 3274 -0.022 -0.043 2320 0.050 0.029 **
(0.043) (0.037) (0.042) (0.041) (0.012)
0.03 0.08 0.12 0.16 -0.004 5759 -0.009 -0.038 4129 0.054 0.028 ***
(0.035) (0.032) (0.032) (0.034) (0.012)
0.29 0.39 0.54 0.49 -0.161 713 -0.219 *** -0.287 ** 381 0.016 -0.033
(0.121) (0.083) (0.117) (0.095) (0.142)
0.28 0.40 0.50 0.49 -0.140 1156 -0.132 * -0.158 621 0.046 0.039
(0.105) (0.078) (0.117) (0.082) (0.113)
0.04 0.10 0.15 0.21 -0.002 2927 -0.025 -0.056 2102 0.044 0.029 **
(0.048) (0.041) (0.055) (0.041) (0.014)
0.03 0.09 0.13 0.19 0.005 5183 -0.011 -0.051 3769 0.053 0.030 **
(0.040) (0.035) (0.042) (0.034) (0.013)
0.17 0.23 0.29 0.38 0.034 3274 0.008 0.031 2320 0.085 * 0.121 ***
(0.048) (0.040) (0.047) (0.047) (0.039)
0.15 0.22 0.27 0.36 0.025 5759 0.010 0.022 4129 0.077 ** 0.104 ***
(0.040) (0.035) (0.040) (0.034) (0.028)
3a
Baseline Endline
Health center delivery vs. Other institutional delivery + Non-
institutional (all births, 0-11 months)
Health center delivery vs. Other institutional delivery + Non-
institutional (all births, 0-23 months)
Health center delivery vs. Other institutional delivery (all
births, 0-11 months)
DD
a
(Et-Bt)-(Ec-Bc)
Full Sample Matched Sample
With covariates With covariates
Delivery assisted by skilled health staff (all births, 0-11
months)
Delivery assisted by skilled health staff (all births, 0-23
months)
5a
5b
1a
1b
PDO2: Institutional delivery vs. Non-institutional delivery (all
births, 0-11 months)
PDO2 - older: Institutional delivery vs. Non-institutional
delivery (all births, 0-23 months)
4b
3b
Health center delivery vs. Other institutional delivery (all
births, 0-23 months)
Health center delivery vs. Non-institutional delivery (all
births, 0-11 months)
Health center delivery vs. Non-institutional delivery (all
births, 0-23 months)
4a
2a
2b
84
versus other health facilities increased for both the treatment and comparison groups,
but the rate of increase is faster in comparison areas than in the treatment areas—
despite the potential of a sizeable cash incentive for mothers in intervention areas.
One potential explanation for this counterintuitive result of impaired growth may be
the possible reputational damage caused by the bumpy implementation of the project.
More than 65 percent of women who delivered at the public health center did not
receive the minimum amount of cash transfer in the treatment area. Such a breach of
trust among the early adopters may have been shared with other expectant mothers,
which caused later cohorts to be disaffected by the CCT scheme and induced fewer to
deliver at health centers than would have if the project had not happened.
DPT3: There is a statistically strong positive effect on DPT3 vaccination. Even if the
definition of DPT3 was changed slightly to be “exactly three vaccinations” instead of
“three or more DPT vaccinations,” the results still hold with the same magnitude
(table 2.6).
52
Table 2.6
DPT3 Behavioral Outcomes
Notes: The robust standard error is reported in parentheses, which is clustered at the village level. The coefficient and standard error of the
time and intervention interaction term are reported through linear probability model (LPM) and marginal effect of logit regression. Sample
size for LPM and logit is not always the same, but sample size for LPM is only reported here given the minor difference between the two
specifications. comp = comparison area; DD = difference-in-differences; LPM = linear probability model; treat = treatment area.
a. DD means simple difference-in-differences without covariates, and DD is equal to [(Endline Treat) (Baseline Treat)] [(Endline
Comparison) (Baseline Comparison)].
Significance level: * = 10 percent, ** = 5 percent, *** = 1 percent.
52
Payments for well-child visits were available every month for the first 12 months and six times for children between the
ages of 13 and 24 months.
PDO Indicator 3: DPT3 Comp Treat Comp Treat n LPM Logit (margin) n LPM Logit (margin)
0.41 0.45 0.46 0.66 0.164 *** 2277 0.149 *** 0.193 *** 1643 0.119 * 0.162 *
(0.058) (0.050) (0.064) (0.062) (0.089)
0.81 0.65 0.69 0.77 0.236 *** 924 0.256 *** 0.326 *** 612 0.155 0.170
(0.081) (0.073) (0.109) (0.102) (0.134)
0.41 0.43 0.45 0.65 0.188 *** 2277 0.167 *** 0.217 *** 1643 0.140 ** 0.190 **
(0.059) (0.050) (0.063) (0.062) (0.088)
0.28 0.24 0.38 0.75 0.413 *** 2307 0.362 *** 0.467 *** 1675 0.475 *** 0.623 ***
(0.057) (0.055) (0.063) (0.065) (0.073)
2
4
1
DPT is at least three times (last birth, 12-23 months)
within vaccination card holder
Proportion of mothers/guardians who could show
vaccination card (last birth, 12-23 months)
PDO3: DPT is at least three times (last birth, 12-23
months)
3 DPT equals three times (last birth, 12-23 months)
DD
a
With covariates
(Et-Bt)-(Ec-Bc)
Full Sample Matched Sample
With covariates Baseline Endline
85
In a somewhat related outcome, the rate of improvement of biological mothers or
guardians being able to show the immunization card to the enumerator was much
larger for the intervention group—increasing by 51 percentage points from the
baseline proportion of 24 percent. Although the comparison groups also increased,
the gains in the intervention group were still 41 percentage points higher. All of these
results strongly suggest that the CNP induced positive behavioral changes regarding
immunization. The latter also implies that parents in intervention areas may be taking
their children’s health more seriously since they are more fastidious about caring for
the child’s card. Or perhaps the clinics are better at providing cards and keeping
immunization records. Regardless of whether the improvements come from changes
in provision or utilization, being able to produce a card is potentially a proxy for more
careful attention to child health, which may well result in better care and child health
in many small ways not observable by the relatively blunt instrument of this survey.
Checkup: The rate of checkups saw phenomenal growth between the baseline and
endline, and the simple difference-in-differences estimates indicate that the simple
growth rate in treatment areas was 10–14 percentage points higher than in baseline
areas. Even so, the more robust multivariate models in rows 1a and 1b of table 2.8 do
not support such claims (a marginally significant result in one specification of the 0-
to 23-month-old group notwithstanding).
It is notable that the intensive margin improved for both the treatment and
comparison groups. For both groups, the share visiting at all or multiple times moved
from less than 10 percent to the 30 to 50 percent range. Notwithstanding the fact that
parents could have received a payment nearly every month for taking their child to a
monthly well-child visit,
53
the robust models do not support the assertion that it was
the intervention that induced parents to bring their children for checkups at a higher
rate than parents who did not have the incentive (table 2.7, rows 2a and 2b). This
53
Payments for well-child visits were available every month for the first 12 months and six times for children between the
age of 13 and 24 months.
86
suggests that the cash transfer amount may not be enough to induce mothers to bring
in a child who appears healthy more than once. Short-term public financial support
might not induce long-term adoption of health services if there is no learning effect or
a negative learning effect in using those services (Dupas 2014c); this dynamic may be
occurring in the case of well-child checkups in this context.
Table 2.7
Routine Checkup Behavioral Outcomes
Notes: The robust standard error is reported in parentheses, which is clustered at the village level. The coefficient and standard error of the
time and intervention interaction term are reported through linear probability model (LPM) and marginal effect of logit regression. Sample
size for LPM and logit is not always the same, but sample size for LPM is only reported here given the minor difference between the two
specifications. comp = comparison area; DD = difference-in-differences; LPM = linear probability model; treat = treatment area.
a. DD means simple difference-in-differences without covariates, and DD is equal to [(Endline Treat) (Baseline Treat)] [(Endline
Comparison) (Baseline Comparison)].
Significance level: * = 10 percent, ** = 5 percent, *** = 1 percent.
Breastfeeding: Although the project did not cause a statistically significant effect on
breastfeeding within one hour of birth (PDO5), another variant of this outcome does
demonstrate a robust effect. As previously described, there is potentially an issue of
comparability in the construction of the questionnaire between the baseline and the
endline for breastfeeding within one hour. However, that challenge is mitigated when
defining the initiation of breastfeeding within three hours of birth (the minimum time
explicitly comparable between the baseline and endline instruments). Using this
definition of breastfeeding, the project successfully caused a 6 to 14 percentage point
increase in mothers nursing their babies; these results are robust across most
specifications.
PDO Indicator 4: Checkup Comp Treat Comp Treat n LPM Logit (margin) n LPM Logit (margin)
0.04 0.10 0.42 0.58 0.107 * 3257 0.046 -0.104 2307 0.004 -0.222 *
(0.056) (0.054) (0.100) (0.082) (0.121)
0.05 0.12 0.45 0.66 0.144 *** 5562 0.097 * -0.026 3980 0.075 -0.122
(0.051) (0.049) (0.075) (0.073) (0.093)
0.03 0.07 0.30 0.41 0.067 3257 0.010 -0.138 2307 -0.009 -0.274
(0.050) (0.047) (0.115) (0.078) (0.168)
0.04 0.10 0.35 0.50 0.095 ** 5562 0.052 -0.079 3980 0.042 -0.190
(0.047) (0.044) (0.085) (0.072) (0.119)
Full Sample Matched Sample
DD
a
With covariates
(Et-Bt)-(Ec-Bc)
PDO4: Any growth checkup vs. None (last birth,
0-11 months)
1b
2a
2b
With covariates Baseline Endline
PDO4 - older: Any growth checkup vs. None
(last birth, 0-23 months)
2+ checkups vs. 1 or 0 checkup (last birth, 0-11
months)
2+ checkups vs. 1 or 0 checkup (last birth, 0-23
months)
1a
87
Table 2.8
Breastfeeding Behavioral Outcomes
Notes: The robust standard error is reported in parentheses, which is clustered at the village level. The coefficient and standard error of the
time and intervention interaction term are reported through linear probability model (LPM) and marginal effect of logit regression. Sample
size for LPM and logit is not always the same, but sample size for LPM is only reported here given the minor difference between the two
specifications. comp = comparison area; DD = difference-in-differences; LPM = linear probability model; treat = treatment area.
a. DD means simple difference-in-differences without covariates, and DD is equal to [(Endline Treat) (Baseline Treat)] [(Endline
Comparison) (Baseline Comparison)].
Significance level: * = 10 percent, ** = 5 percent, *** = 1 percent.
Receive ORS with Diarrhea: There was no detectable difference in trends for PDO6
for children age 0–23 months (table 2.9, rows 1a and 2b) or 0–11 months (table 2.9,
rows 1b and 2a), regardless of whether or not its definition including the government-
approved homemade recipe. The CNP did not reduce the incidence of diarrhea (table
2.9, rows 3a and 3b). However, the project demonstrated a positive impact on seeking
treatment and/or advice for children who had diarrhea in the last two weeks for both
children under age one and children under age two (table 2.9, rows 4a and 4b). The
results imply that mothers from intervention areas sought treatment at a higher rate,
even if they were no more likely to receive it than mothers in comparison areas.
PDO Indicator 5: Breastfeeding Comp Treat Comp Treat n LPM Logit (margin) n LPM Logit (margin)
0.41 0.40 0.35 0.42 0.074 3257 0.070 0.076 2308 0.036 0.042
(0.062) (0.060) (0.062) (0.067) (0.074)
0.42 0.41 0.37 0.46 0.099 5564 0.085 0.091 3983 0.057 0.063
(0.063) (0.062) (0.065) (0.069) (0.075)
0.51 0.51 0.74 0.81 0.075 3257 0.062 0.064 * 2308 0.129 ** 0.113 **
(0.045) (0.044) (0.036) (0.055) (0.047)
0.52 0.51 0.73 0.82 0.091 ** 5564 0.069 0.071 ** 3983 0.137 *** 0.119 ***
(0.043) (0.043) (0.035) (0.049) (0.043)
With covariates
Full Sample Matched Sample
DD
a
With covariates
(Et-Bt)-(Ec-Bc)
2a
2b
Baseline Endline
Breastfeeding within 3h of birth vs. None (last
birth, 0-11 mo)
Breastfeeding within 3h of birth vs. None (last
birth, 0-23 mo)
1a
1b
PDO5: Breastfeeding within 1h of birth vs.
None (last birth, 0-11 mo)
PDO5 - older: Breastfeeding within 1h of birth
vs. None (last birth, 0-23 mo)
88
Table 2.9
Oral Rehydration Solutions with Diarrhea Behavioral Outcomes
Notes: The robust standard error is reported in parentheses, which is clustered at the village level. The coefficient and standard error of the
time and intervention interaction term are reported through linear probability model (LPM) and marginal effect of logit regression. Sample
size for LPM and logit is not always the same, but sample size for LPM is only reported here given the minor difference between the two
specifications. comp = comparison area; DD = difference-in-differences; LPM = linear probability model; treat = treatment area.
a. DD means simple difference-in-differences without covariates, and DD is equal to ((Endline Treat) (Baseline Treat) ((Endline
Comparison) (Baseline Comparison))).
Significance level: * = 10 percent, ** = 5 percent, *** = 1 percent.
Because the incidence of diarrhea may be non-random even after controlling for
observable characteristics, the evaluation team members applied a Heckman selection
model to the diarrhea treatment indicator as a robustness check, using village-level
incidence of diarrhea as the selecting variable. This covariate is highly significant in
the first stage on incidence of diarrhea for the child. Wald test results do not reject the
null hypothesis of no correlation between the error term and unobserved
determinants of diarrhea incidence, which implies that the incidence of diarrhea after
controlling for covariates is indeed random. The magnitude of the coefficient and
standard error are remarkably stable between the selected and non-selected models.
The slight reduction in significance levels is likely due to the decrease in power in the
selection models. Consequently, the non-selected results in table 2.8 are maintained
as the main findings for this PDO.
PDO Indicator 6: Diarrhea with ORS Comp Treat Comp Treat n LPM Logit (margin) n LPM Logit (margin)
1a 0.64 0.64 0.62 0.70 0.077 933 0.042 0.048 731 0.01 0.02
(0.082) (0.085) (0.088) (0.114) (0.113)
1b 0.51 0.59 0.56 0.63 -0.006 465 -0.011 -0.006 354 -0.025 -0.020
(0.110) (0.117) (0.118) (0.146) (0.156)
2a 0.43 0.47 0.45 0.60 0.112 465 0.101 0.118 354 0.079 0.126
(0.101) (0.104) (0.114) (0.126) (0.159)
2b 0.55 0.53 0.52 0.68 0.173 ** 933 0.115 0.131 731 0.121 0.140
(0.0757) (0.0785) (0.0840) (0.111) (0.119)
3a 0.13 0.14 0.15 0.13 -0.036 3271 -0.027 -0.024 2313 -0.028 -0.020
(0.038) (0.033) (0.031) (0.038) (0.035)
3b 0.16 0.16 0.16 0.14 -0.019 5753 -0.008 -0.005 4122 0.006 0.004
(0.032) (0.029) (0.027) (0.038) (0.032)
0.58 0.53 0.60 0.76 0.206 * 465 0.239 * 0.257 * 354 0.318 ** 0.358 **
(0.117) (0.122) (0.135) (0.141) (0.151)
0.62 0.56 0.62 0.77 0.212 ** 933 0.196 ** 0.206 ** 731 0.191 * 0.202 *
(0.084) (0.086) (0.090) (0.110) (0.115)
Full Sample
With covariates
(Et-Bt)-(Ec-Bc)
PDO6 - younger: Received ORSplus (1, 2 or 3) during
diarrhea vs. Not received (last two births, 0-11 mo)
With covariates
Matched Sample
Baseline Endline
PDO6: Received ORSplus (1, 2 or 3) during diarrhea vs.
Not received (last two births, 0-23 mo)
DD
a
Treated for diarrhea conditional on having diarrhea (last
two births, 0-11 mo)
Treated for diarrhea conditional on having diarrhea (last
two births, 0-23 mo)
4a
4b
Received ORS (1 or 2) during diarrhea vs. Not received
(last two births, 0-11 mo)
Received ORS (1 or 2) during diarrhea vs. Not received
(last two births, 0-23 mo)
Had diarrhea in last two weeks or not (last two births, 0-
23 mo)
Had diarrhea in last two weeks or not (last two births, 0-
11 mo)
89
2.6 Heterogeneity
Lao PDR has a vast geographical, sociocultural, and economic diversity, and care-
seeking behavior varies by these factors. This chapter explores potential
heterogeneous effects of the Community Nutrition Project (CNP) through subgroup
analyses on the established project development objective (PDO) indicators along
these dimensions: gender, ethnicity, whether or not the child’s mother had a previous
child die, the mother’s educational background, being the first-born child or not, the
household being in the bottom 40 percent of the short-term and long-term asset
indexes or not, experience price increase or decrease shocks (self-reported), and
geographic aspects, including province, rural-urban classification of the village, and
distance from the village to the health center. All subgroup analyses compare those
within same subgroup across the intervention versus comparison areas as opposed to
comparing treatment effects across subgroups within the same treatment status
(intervention or comparison).
All the covariates used for the main results in table 2.3 are controlled for in these
analyses except for the variable being investigated through subgroup analysis. The
effect estimates for the program on PDO indicators for each subgroup are
summarized in appendix I; each of the tables in appendix I corresponds with a PDO,
and the top row of each of those tables reproduces the overall estimates for each PDO
indicator from table 2.3. Although estimates do not explicitly correct for multiple
hypothesis testing, the only results discussed below are those that are fairly consistent
across specifications and with at least one result significant at the 1 percent level.
A substantial challenge for subgroup analysis is the reduction in sample size. This
results in a decrease in statistical power, often increasing the minimum detectable
effect size beyond what the survey was originally designed to detect. Consequently,
this section focuses on those dimensions showing significant differences instead of
null results. Furthermore, some subsamples are too small to produce reliable
90
estimates—these are reported in the appendix I tables with a dash ( - ). All estimates
are for children (or their mothers) born within the year preceding the survey, except
for those in PDO3, which estimated effects for children between one and two years
old, and PDO6, which pooled children under two years old.
PDO1: Although there was no significant effect for the general results for PDO1 (table
5.1), the project seems to have significantly increased the likelihood of seeing a
professional health worker for an antenatal visit in the Khommaun province by 19.3–
23.3 percentage points. A large, significant, and fairly robust positive effect is
observed for those who experienced a price increase shock in purchased staples. There
is weaker evidence that mothers from intervention areas were more likely to seek
antenatal care than observably identical mothers in matched comparison areas.
PDO2: Delivery at the health facility for births within the preceding year varies by
province. As with PDO1, there is evidence that CNP-area mothers in Khammaun (and
more weakly, the Saravan province) were more likely to give birth at a health
institution. The overall null result for this PDO indicator, however, may be explained
by the result of these positive effects being offset by a large, highly significant negative
result in Bolikhamxay. Socioeconomic status effects are somewhat mixed. Although
wealthy intervention residents were more likely to deliver in health institutions than
wealthy residents of comparison areas,
54
the intervention was effective for the Mon-
Khmer ethnic minority and the uneducated. The intervention caused an increase in
institutional delivery among those who lived near a health center.
PDO3: The CNP caused a significant increase in PDO3 for the general population
estimates. Therefore, many of the subgroups examined also show positive effects on
children between one and two years old receiving at least three diphtheria, pertussis,
and tetanus immunizations—in particular, mothers with some primary but no
54
As measured by the second component of the long-term asset index, which loads more heavily on high-quality building
materials in floors and walls.
91
secondary schooling, those living between 3 kilometers and 6 kilometers from the
nearest public health center, children without a deceased sibling, and households at
the bottom 40 percent of the second principal component of the short-term asset
(consumables) index. Also, families immunizes their children as a result of the project,
both among those who did and did not report a price increase in their consumables,
and from both the top and bottom of the wealth distribution.
55
PDO4: Subgroup analysis reveals an intriguing mix of positive and negative effects
for different segments of the beneficiary population for PDO4, giving insight into the
null overall effect that alternates between small positive point estimates for the linear
probability models and negative point estimates for the nonlinear models (table 2.3).
Unfortunately, motivations for this mix of effects by motherhood experience, wealth,
ethnic background, and locality are not always easily understood or explained.
Experience in motherhood seems to have led to mixed interactions with the project.
First-born children were less likely to have a growth checkup if they lived in a project
area than if they did not—perhaps, again, because the negative reputational effects of
the project’s delay in disbursing the transfer was especially influential for new parents
who had weaker priors. By contrast, there was weaker evidence that mothers who
had previously lost a child were more likely to take advantage of the project and bring
their baby in for a growth checkup. Wealth was also associated with variation in the
likelihood of seeking a growth checkup because of the intervention. Better-off families
(those in the top two quintiles for the long- or short-term asset indexes) were less
likely to seek a checkup if living in a CNP area than similar families living in a control
area. Conversely, there are mixed results for those in the bottom two quintiles:
Estimates indicate a positive interaction for the poor as measured by the first principal
component of the long-term (durables) asset index, but a negative interaction effect
for those in the bottom of the second principal component. Ethnic background had a
55
As measured by the second component of the long-term (durable) asset index, which loads more heavily on high-quality
building materials in floors and walls.
92
polarizing influence on the project’s ability to induce demand for growth checkups.
Those in the Lao Tai majority ethnic group were less likely to bring their children for
visits, but the project was utilized among the Mon-Khmer. Finally, location seems to
have played a role in project effectiveness. Like the institutional delivery result of
PDO2, the Saravan province saw large benefits from the program with regard to child
growth checkups—perhaps because almost all sampled families in the province were
ethnic Mon-Khmer. However, the project may have made parents who lived more
than 6 kilometers away from a health center less likely to bring their child to a checkup
than similar parents from comparison areas.
PDO5: The project was particularly effective among the poor for inducing an increase
in the likelihood that mothers breastfed their babies within the first hour of birth.
Mothers from households from the bottom two quintiles of the durables index and
those who experienced a price increase shock in consumed goods were significantly
more likely to immediately breastfeed their children if they lived in an intervention
area. But so, too, did mothers from the top two quintiles of the consumables index.
Province and ethnicity again played a moderating role. Households from Khammaun
were negatively influenced by the project, but residents of Savanhnakhet and
Champasak, along with ethnic Mon-Khmer, were favorably affected.
PDO6: As shown in table 2.3, the project did not affect the likelihood of receiving an
oral rehydration solution (ORS) for children with diarrhea. Compared with the other
PDOs, none of the subgroups demonstrated strong, consistent effects. Positive,
consistent but marginal effects were observed among more educated mothers. A
significant interaction is observed for first children in the logit models, and a single
highly significant result appears for those who experienced a decrease in the price of
products sold by the household. Beyond that single result, and in contrast to the other
PDOs, there is no evidence that the poor benefitted from this project with regard to
receiving ORS for diarrheic children. However, this could be partially due to the
93
smaller sample size of these estimates imposed by the conditionality of having
diarrhea.
2.7 Conclusion
This impact evaluation evaluated the efficacy of the first World Bank program
administered by the government of Lao PDR and the first project incorporating a
conditional cash transfer (CCT) scheme. The transfer program, in conjunction with
community-based nutrition education, was able to induce health-seeking behavior
and awareness of maternal and child health (MCH) and nutrition outcomes for
pregnant and lactating women and children younger than two years old. Through
quasi-experimental methods, this evaluation finds that the Community Nutrition
Project (CNP) cannot claim to have affected most of the project development objective
(PDO) indicators, but it did influence closely-related indicators that are more
appropriately aligned with the project’s design. It also benefitted the poor and
vulnerable.
More specifically, the CNP had a positive influence on child caring practices, such as
diphtheria, pertussis, and tetanus (DPT) 3 vaccination, and breastfeeding after birth.
The effects on increasing MCH utilization, however, were subtle and nuanced.
Although there is some evidence pointing to improvements in the rate of antenatal
and well-child visits, specifically to the health centers incentivized through the CCT
for children 0–23 months, the evidence is not convincing that the project was able to
inspire higher rates of institutional delivery generally, despite the cash incentive
subject to antenatal care, delivery, and routine growth checkups. In light of the global
literature showing the potential for success through CCTs in improving skilled birth
attendance (IEG 2013), there were some positive effects on the matched sample. But
somewhat surprisingly, robust positive effects were not witnessed in the full sample.
The subgroup analysis shows that there are positive influences on the use of public
health center services, particularly for those in the bottom 40 percent of the durables
94
asset index and those who experienced price increase shock. Furthermore, there are
heterogeneous program effects across different socioeconomic statuses and provinces.
Even though the overall effect is modest, the project caused an improvement for the
bottom 40 percent of the wealth distribution for nearly every outcome, including
institutional delivery. This suggests that the project contributed to MCH coverage and
behavior for the bottom 40 percent, and could reduce the inequality of MCH
interventions within Lao PDR. These findings could also be informative for better
targeting in case scaling up of the program is considered as a policy option under the
limited fiscal space. CNP also worked as a social protection measure against shocks
from rising prices.
These achievements are encouraging, but the modest effects are almost certainly
lower than they could be. Because essential design elements were not sufficiently
completed at the time of appraisal in 2009, the design had to be fully realized during
the implementation period. Moreover, this was the first World Bank project fully
executed by the in-house capacity of a line ministry in Lao PDR, and not surprisingly
there were growing pains that led to additional implementation delays. Moreover,
these delays allowed other donors to enter into the comparison areas with potentially
similar interventions, which may have contributed to the lower effect sizes than other
CCT programs in the literature which in general have counterfactuals that do not
include competing programs. The null effect on anthropometric measures might be
due to the short duration of the project as much as these implementation limitations.
Also, the relatively low uptake rates imply that significant improvements can still be
made to the program’s design.
95
3. Corruption, Governance Institutions and the Extent to
Which Governance Institutions Can Reduce the
Pervasiveness of Corruption: Evidence from Firms
Perceptions and Behavior
96
3.1. Introduction
Few topics in public policy, public economics or development economics have
received as much attention as corruption. In this literature, many different aspects of
corruption have been studied. Among these are the various different forms and levels
that corruption can take, e.g. bribes to public officials by individuals and firms,
falsification of accounts so as to avoid or lesson the tax liability of an individual or
firm, payments to officials to obtain better information on the direction of future
policy or detailed rules to be followed in government programs, and smuggling
operations. It is not uncommon that favorable decisions by government officials can
be an extremely important means by which individuals and firms can obtain favors,
and become rich and successful. Yet, the process is also often thought to be one that
locks whole groups and even countries into slow growth and poverty traps. Some
important examples of public policy decisions that may be distorted by corruption
include privileged contracts with government, regulations that would protect their
interests, the issuance of licenses of various types, zoning and campaign contributions
legal or otherwise.
The topic is unusual and controversial in many ways. First, there is considerable
disagreement as to whether it is good or bad. Second, there is considerable
disagreement on which forms of corruption are most common and most important.
Third, there is considerable disagreement on both its main determinants and main
effects. Fourth, the ability to settle any of these disputes is rendered difficult due to
the difficulty of being able to identify and measure it, given that it is illegal. Take the
case of a bribe. Because bribery is in most countries illegal, it is unlikely that either the
payer of the bribe or the recipient of the bribe will be willing to report it. Even if
questions are put to individuals or firms about it, they are quite unlikely to own up to
it for fear of possible prosecution. Finally, even if discovered, there is often great
disagreement on how best to deal with it. Indeed, this has led to a rapidly rising tide
97
of field experiments of virtually any imaginable type to examine the effectiveness of
a given approach.
The purpose of this paper is to take advantage of firm survey data in which, after
years of experience, survey designers have identified quite satisfactory ways in
getting people to answer questions about both their perceptions of corruption and
their experience in it, i.e., in actually paying bribes (phrased as “gifts to officials” to
elicit responses) and to tie those gifts to particular types of agents and purposes (but
not their names). This is the data from the World Bank’s Enterprise Surveys.
Whereas much of the literature is merely either macro-level cross-sectional based on
differences in perceptions of corruption across countries and perhaps also over time,
or single country studies examining differences across states or over time, this study
focuses on differences across countries (and indeed across an unusually large sample
of countries, in some cases at different points in time, and across firms of different
sizes, age, industries, technology, ownership form. Because much of the macro-level
cross-sectional literature has identified various kinds of desirable governance
characteristics that seem to play a significant role in explaining cross country
differences in perceptions of corruption, considerable emphasis is given to each of six
different but interrelated governance indicators, namely the World Bank’s World
Governance Indicators. As will be explained below, we extend the existing literature
on the role of these different governance institutions, by examining the extent to
which their effects differ by type of firm and certain other country characteristics.
Whereas much of the literature has treated perceptions of corruption as if it was the
same as actual corruption, our results show many cases in which a certain variable
such as a governance indicator or favorable firm characteristic or a term representing
the interaction between two of these has a strong negative effect on perceptions of
corruption, but at the same time little or no effect on our actual measures of
corruption. All these results underscore the point that actual corruption is harder to
eliminate than the perception of corruption. There are also some cases, such as the
98
effects of high tax rates, medium size of firm and most importantly regulatory
enforcement, in which even the direction of the effects may differ between the two
types of measures.
3.2. Relevant Literature
Earlier studies such as Leff (1964) viewed corruption as a natural result of very
inappropriate regulations and public policies such that corruption should be seen as
good thing for a country’s economic development since gifts are paid to regulators
and enforcers to get around the inefficient and inappropriate regulations. However,
since the most obvious and despised corruption typically involved scandals at the
highest levels, with top leaders taking bribes from some of the richest individuals and
powerful firms to help protect their vested interests, the pendulum soon shifted to a
more general belief that corruption was bad, not only for development but for society
in general, e.g., Rose-Ackerman (1978), Caiden and Caiden (1977), Mauro (1995).
Indeed, the paper of Mauro (1995) was particularly persuasive. It showed that both
investment as a share of GDP and the overall growth rate of GDP per capita over the
1960-1985 period were substantially and significantly raised by several freedom from
corruption indexes, and bureaucratic quality indexes. Indeed, by showing that these
effects were almost identical between subsamples of countries in which indexes of red
tape were alternatively low or high, the results tended to go against the idea used by
Leff and others that corruption was beneficial to get around regulatory red tape. One
shortcoming in this study, however, was that the indexes of freedom from corruption
and of bureaucratic efficiency that it used were based exclusively on the most recent
years of growth and investment.
56
Hence, the results could be interpreted as
representing reverse causality going from growth to greater freedom from corruption
and higher bureaucratic quality rather than the other way around.
56
The reason for this of course was that it was only in the mid-1980s that data collection agencies such as in that case the
Economist’s Economic Intelligence Unit first started to obtain the reports of businessmen and other relevant experts needed
to construct such indices.
99
As a result of this turnaround in views about whether corruption was good or bad,
considerable efforts have been taken to take legal action against corrupt individuals
such as those detected as having taken bribes. Often this has proven difficult,
especially when it was so rampant. Indeed, it tended to require rather drastic action.
For example, not long after gaining its independence, corruption in Indonesia was
deemed so serious and pervasive among customs officials of that country that the
Indonesian government decided to replace all these officials by a highly skilled and
honest set of customs officers from the Netherlands. Clearly, an operation such as this
is expensive and can be blocked by entrenched interests. Hence, successful actions of
this sort are not common. This is especially difficult in the case of corruption at the
top level, where the military and mafia like organizations might succeed in heading
off any efforts to get after the powerful leaders. Yet, it should be noted that, at the time
of writing this paper, one country, Guatemala, is voting in a democratic election in
which a replacement is being sought to a president who has been charged and
convicted of bribery. That president, moreover, is currently in jail. For this reason,
identifying programs to identify existing corruption, factors critical to it, and means
of overcoming it are generally very popular, but just how to do so?
As scholars and governments have delved into this in a serious way, it has become
increasingly clear that simply passing a highly publicized anti-corruption Law would
not be enough. Indeed, this has led to an enormous field of how best to discourage it.
Numerous such models of how to do that have been developed, such as by Becker,
Bardhan (1997) Murphy et al (1996). An increasingly wide variety of factors has been
identified as lying behind corruption, such as cultural (Caiden, 2013), political
(Campos and Giovannini, 2007; Fisman and Gatti, 2002), economic (Bardhan, 1997;
Ades and Di Tella. 1999), administrative (Caiden and Caiden 1977), and conflict (Dix
and Jayawickrama 2010).
All through the 1980s and especially in developing countries there were numerous
reports on factors that seemed to be responsible for various failures to reduce poverty
100
and to sustain or even raise overall rates of development, and to achieve respectable
scores on various development indicators such as the United Nations Development
Program’s Human Development Indexes. Corruption of different types and levels
was frequently commonly identified as a major contributing factor, although
satisfactory proof has often been lacking, in large part because of the lack of
satisfactory measures of vulnerability to corruption.
In 1993 and almost simultaneous to the aforementioned study of Mauro (1995) a
group of concerned officials and advocates of development launched an international
NGO known as Transparency International which dedicated itself to raise public
awareness of corruption and to lead international efforts to reduce it wherever it could
be found throughout the world. These efforts included constructing an index known
as the Corruption Perceptions Index (CPI) measuring the extent to which various
agents in society perceive corruption to be serious and pervasive, by country and year.
One of the sources used was identical to that used by Mauro (1995), namely the
Economist’s Economic Intelligence Unit. The creation of Transparency International
has also fostered the growth of country- level chapters of Transparency International.
At present, Transparency International has chapters in more than one hundred
countries and has an International Secretariat based in Berlin.
Through its country-level chapters, international conferences, publications and
calling attention of the world of the quality and availability of its country-specific CPI
scores, Transparency International and other like-minded institutions have fostered
international efforts to deal with the problem. Especially prominent among these
efforts have been studies trying to identify the best anti-corruption strategies and
policies. Indeed, the latter quest has led to the formation of international agreements
among OECD countries, and subsequently members of the United Nations to adopt
the United Nations Compact for Business in 2000 and the United Nations National
Council against Corruption (UNCAC) in 2003. Transparency International has also
focused on the role of corruption in newer and more specific topic areas such as
101
Education, Defense and Security, approaches to limit climate change, international
trade and even to sports. These are all areas in which various organizations and
governments are spending large amounts of money to lower poverty, but raise health
and education and the efficiency of these programs. Corruption has been seen as a
factor greatly reducing the efficiency of these programs, again justifying the fight
against corruption. Transparency International has also been demonstrating the
quality of its CPI scores and publishing them for almost all countries in the world
annually since 1995.
At roughly the same time, there began another major international effort to
characterize the Governance institutions which scholars, policy makers and
businesses were increasingly identifying as linked to both the determinants and
effects of corruption. Recognizing the breadth, depth and degree of the
interconnectedness among these different characteristics of governance, Daniel
Kauffman and Arthur Kraay led a World Bank effort to develop a multidimensional
measures of governance which were commonly believed to both limit the incidence
of corruption (i.e., affect the determinants of corruption) and the effects of corruption.
These authors wrote a series of widely read papers under the banner “Corruption
Matters” (Kaufmann and Kraay, Kaufmann, D., A. Kraay and P. Zoido-Lobaton,
Kaufmann, D., A. Kraay and M. Mastruzzi (2004, 2006 and 2010). Governance was
defined so as to consist “of the traditions and institutions by which authority in a
country is exercised”. This multidimensional measurement of governance was
designed to integrate all the different facets that would be needed to effectively limit
the incidence and effects of corruption. These were divided into three areas: (A) the
process by which governments are selected, monitored and replaced, (B) the capacity
of the government to effectively formulate and implement sound policies, and (C) the
respect of citizens and the state for the institutions that govern economic and social
interactions among them. Beginning in 1996 the World Bank has constructed country-
102
specific indexes of two different dimensions for each of these three broad areas, giving
rise to six separate indexes identified as follows:
(1) Voice and Accountability (perceptions of the extent to which a country’s citizens
are able to participate in selecting their government, as well as freedom of expression,
freedom of association and a free media)
(2) Political Stability and the Absence of Violence/Terrorism (perceptions of the
likelihood that the government will be destabilized or overthrown by
unconstitutional or violent means, including politically-motivated violence and
terrorism)
(3) Government Effectiveness (perceptions of the quality of public services, the civil
service and the degree of its independence from political pressures, the quality of
policy formulation and the credibility of the government’s commitment to such
policies
(4) Regulatory Quality (perceptions of the extent to which a country’s citizens are able
to participate in selecting their government, as well as freedom of expression, freedom
of association, and a free media.)
(5) Rule of Law (perceptions of the extent to which agents have confidence in and
abide by the rules of society, and in particular the quality of contract enforcement,
property rights, the police, and the courts, as well as the likelihood of crime and
violence)
(6) Control of Corruption (perceptions of the extent to which public power is exercised
for private gain, including both petty and grand forms of corruption, as well as
“capture” of the state by elites and private interests)
(The first for of these indexes were designed to represent Area A, the third and fourth
to represent Area B and the last two Area C. All six indexes were based on perceptions
of citizens constructed from detailed surveys, in no case constructed from
government–provided self-assessments.)
103
Not surprisingly, the availability of these indexes has given rise to ever-increasing
numbers of studies comparing countries with one another at any one point in time
and tracking progress or retardation over time. Indeed, Kaufmann and his co-authors
have been doing just that. For example, Kaufmann (2015) is but the most recent of
these updates, in this particular case focusing on Latin American countries,
comparing this region with others, and within this region one country with another
at different points in time. As a result, he has managed to distinguish countries that
have made considerable progress on these dimensions such as Chile from countries
that experienced regress on many of these dimensions since 1996, such as Venezuela.
More generally, OECD countries are usually shown to score considerably higher on
these indexes than poor countries. Global progress has been witnessed with reduced
income inequality and improvement in both political stability and economic growth.
Rarely, however, have these studies managed to reach down to the firm or individual
level whose behavior would be affected both by corrupt agents and by the various
indexes of good governance believed to help reduce both the perception of corruption
and actual corruption and thereby affect development performance. A notable and
very commendable exception is Clarke and Xu (2004) who used firm-level data from
the World Bank’s Business Environment Survey which was a predecessor survey to
that used in this study, the Enterprise Surveys (also done by the World Bank). Clarke
and Xu limit their analysis to one region, namely Eastern Europe and Central Asia
(based on 21 country surveys), and to one area of regulation/corruption, namely, gifts
made by sample firms to electric and telecommunications utilities for improving and
accelerating access to these utilities. Interestingly, they relate the magnitudes of these
gifts (as shares of firm sales) to actual conditions and characteristics of both the firms
themselves and the utilities supplying these services in each specific country. They
show that the gifts tend to be larger when the utilities face more severe capacity
constraints, are subject to less competition and more likely to be publicly instead of
privately owned and when the firms are more profitable and newer.
104
Quite naturally, much of the literature based on these indexes has been directed at
identifying those policies and other factors most responsible for improvements in
governance. On the other hand, much of it has also been directed at relating such
changes to specific kinds of behavior such as the inflow of foreign direct investment
(FDI) or to the extent of under invoicing of imports, or to objective measures of market
competition.
The aim of this study is two-fold. First, it is to extend the analysis with these
impressively well researched governance indexes to examine the extent to which firm
perceptions of corruption as an obstacle to their businesses are affected by each of
these six governance indicators and more importantly to distinguish the
characteristics of those firms which seem to be more affected than others by these
individual governance indexes. Second, complementary to the first objective, is to
examine the extent to which the governance indexes relate to actual corrupt behavior
of firms, such as giving gifts to various different agents for access to a certain type of
permit or service, and then to determine the extent to which the effects of these
governance indicators would vary between firms of different types and
characteristics. This should go some way to identifying the kinds of firms and
countries for which the governance indicators seem to be less effective in mitigating
corruption. From that it should be possible to think of other kinds of policies which
might be more effective in such circumstances.
3.3. Data, Sample, Variable Definitions and Estimation Procedure
3.3.1. MEASUREMENT OF THE OUTCOMES OF INTEREST
World Bank’s Enterprise Surveys (ES) provide the data for both sets of outcome
variables examined in this study, i.e. firm perceptions of corruption as an obstacle and
their actual corruption behaviors. In all, the behavior of six different types of gifts for
access or service are examined (1) gift for access to electricity, (2) gift for access to
telephone, (3) gift for obtaining a license to import, (4) gift to a tax inspector, (5) gift
105
for an operating license and (6) gift for a construction permit. Our primary measure
of actual corrupt behavior by firms is the total number of different types of gifts that
the firm provides (conditional on having been subject to at least one type of constraint
in which a gift might have been expected). In addition, we also examine the
determinants of each of the six types of “gifts” identified in the surveys separately,
again conditional on it being identified as relevant.
Similar to the way the perception of labor regulations as an obstacle to the firm’s
business was constructed in chapter 1, the firm’s perception of corruption as an
obstacle to its business is also derived from the firm’s response to the following
question—“Please tell us if any of the following issues are a problem for the operation
and growth of your business. If an issue poses a problem, please judge its severity as
an obstacle on a four-point scale.” It is treated as an ordered response variable in the
following categories “no obstacle”, “minor obstacle”, “moderate obstacle”, “major
obstacle” and “very serious obstacle”, and is rated on a 0–4 point scale.
On the other hand, the various corruption behavior variables are derived from the
question “In reference to that [name of activity in the last two years], was an informal
gift or payment expected or requested?” All the specific corruption behaviors are
coded as dummy variables, 1 for having engaged in such activities in the past two
years and 0 otherwise. Table 3.1.1 (panel A) shows the summary statistics of these
outcomes. Notice that the responses are obtained only for a relatively moderate
proportion of the whole sample, given that understandably only a fraction of firms
engage in each of the activities in which a gift payment might be expected. In addition,
given that this set of gift questions are incorporated in ES questionnaires 2006 and
onwards, the sample in this analysis is comprised of firms surveyed considerably
more recently compared to those in chapter 1.
The overall sample mean of ObstCorruption is not especially high and not
surprisingly is somewhat below the corresponding averages of certain other obstacles
106
to the firms’ business, such as access to finance, tax administration and political
uncertainty. Yet there is considerable variation in it across firms and countries.
Likewise, on the reports by the firms of gifts asked for and paid by and to various
agents (by type), the overall incidence of such gifts across the sample is not high, with
considerably less than one percent of the firms reporting having made such gifts.
The full sample consists of over 110,000 firms taken from 135 developing or transition
countries where corruption is generally deemed to be more serious than in OECD
countries. The entire list of countries is given in Appendix Table A1. Along with each
country name is given the number of firms surveyed in that country and their share
in the total sample. In some of these countries, the survey has been carried out in two
different years, though generally for different samples of firms. As can be easily seen,
the samples were larger for countries with larger populations and manufacturing
output. Note for example that India has the largest sample, with China, Russia,
Nigeria, not far behind and Argentina, Bangladesh, Brazil, Colombia, Egypt, Mexico,
Pakistan, Peru, Turkey, and Ukraine each with over 1,500 firms in their samples. On
the other end of the scale, many countries come with only about 150 firms in their
samples. Same as in chapter 1, firms from all five continents are represented in the full
sample.
107
Table 3.1
3.3.2. DATA FOR MEASURES OF THE EXPLANATORY VARIABLES
The source of the six governance indicators used as key explanatory variables in the
analysis is World Bank World Governance Indicators. Kaufmann et al. (2010) provides
a detailed summary on the methodology in constructing the indexes. In short, for each
of the following three areas were identified (A) the process by which governments are
selected, monitored and replaced, (B) the capacity of the government to effectively
formulate and implement sound policies, and (C) the respect of citizens and the state
Table 1 Summary Statistics
Panel A: Dependent Variables
N Mean Std. Dev. Min 10th
percentile
90th
percentile
Max
ObstCorruption 113428 1.76 1.49 0 0 4 4
Gift for various purposes(dummies)
Electricity connection 17122 0.16 0.36 0 - - 1
Telephone connection 15894 0.067 0.25 0 - - 1
Construction permit 15818 0.19 0.39 0 - - 1
Tax inspection 69634 0.14 0.35 0 - - 1
Import license 14598 0.12 0.33 0 - - 1
Operation license 28337 0.16 0.36 0 - - 1
Total occasions of paying gifts 117105 0.19 0.56 0 0 1 6
Panel B: Selected Explanatory
Variables
N Mean Std. Dev. Min 10th
percentile
90th
percentile
Max
Years of operation (6–20 years) 114391 0.57 0.49 0 - - 1
Years of operation (>20 years) 114391 0.28 0.45 0 - - 1
Manager's experience (in years) 113801 16.9 10.8 0 5 32 50
Owner/top manager is female 117098 0.29 0.45 0 - - 1
Foreign owned 117105 0.080 0.27 0 - - 1
Government owned 117105 0.005 0.074 0 - - 1
meanLog_ManTimeReg 117098 1.54 0.65 0 0.72 2.38 3.68
Sole proprietorship 117101 0.34 0.48 0 - - 1
Partnership 117101 0.08 0.27 0 - - 1
Medium size 117105 0.34 0.47 0 - - 1
Large size 117105 0.19 0.39 0 - - 1
Agroindustry and food 117105 0.11 0.31 0 - - 1
Textiles and garments 117105 0.12 0.32 0 - - 1
Manufacturing 117105 0.35 0.48 0 - - 1
Service 117105 0.14 0.35 0 - - 1
Retail and wholesale 117105 0.23 0.42 0 - - 1
Having international quality certificate 115378 0.23 0.42 0 - - 1
City level, country, year, unemployment rate not shown
Data Source: World Bank Enterprise Surveys and Doing Business Surveys
108
for the institutions that govern economic and social interactions, the World Bank,
beginning in 1996, has constructed country-specific indexes of two different
dimensions of each of these broad areas, constituting the following six components.
These indexes have been constructed from four different types of sources:
1. Household and firm level surveys, of which there were nine, including
Afrobarometer, Gallup World Poll, and the Global Competiveness Report Survey
2. Commercial business information providers of which there were four, including
the Economist’s Economic Intelligence Unit, Global Insight, and Political Risk
Services,
3. Non-governmental organizations (NGOs) of which there were eleven, including
Global Integrity, Freedom House, and Reporters without Borders
4. Public sector organizations including the CPIA assessments of the World Bank,
regional development banks, the European Bank for Reconstruction and
Development’s Transition Report, the French Ministry of Financial Institutions
Profiles
Therefore, altogether 32 different major sources are used to construct the six different
indicators in a step-by-step manner, as outlined on the World Government Indicators
webpage. Many of the 32 individual sources do not cover the full set of countries for
any given index. Therefore, the overall evaluations have had to be constructed by the
World Bank’s Governance Indicator team by converting the original scores on
different scales into a common 0-1 scale and then aggregating them into an overall
ranking with a 0-100 scale for each of the six different indicators.
Descriptive statistics on all six of these Governance Indicators are given in Panel B of
Table 1.1. Not surprisingly, the indexes receiving the lowest scores overall are Political
Stability, Control of Corruption and Rule of Law. Comparatively at least, the sample
countries received somewhat better ratings for Government Effectiveness and
Regulatory Quality.
109
Apart from the six country-level governance indexes used to explain the outcomes,
other control variables are largely a standard list of firm characteristics comparable to
that in chapter 1 (e.g. firm’s years operation in the country, ownership type, manager’s
experience legal status, the time spent on dealing with regulations by the management
as a measure of labor enforcing, industry and year fixed effects, and an overall
national-level tax rate obtained from World Bank Doing Business Database). In
particular, an extra dummy variable is included, indicating whether or not the owner
or top manager of a firm is female as demonstrated to be a significant determinant in
Clark (2015) with a somewhat smaller sample of surveys. Note that the “any related
activity” is a dummy for firms which conducted in at least one of the six business
activities for which the Enterprise Surveys asked whether a gift was given.
Descriptive statistics on these additional explanatory variables are given in panel C of
Table 3.1. Note especially the wide variation across the sample in terms of firm age
(years of operation), manager’s experience, ownership type, size and industry and
technological sophistication. This is also true for the proxy measure of regulatory
enforcement, namely, percentage of managers’ time spent on regulations
(meanLog_ManTimeReg).
3.3.3. ESTIMATION PROCEDURE
Since that the dependent variables are all defined in terms of integers ranging from 0
to either 4 for perceptions concerning the seriousness of Corruption (ObstCorruption)
or 6 for the total number of different kinds of “gifts” given by the firms to officials for
the six different kinds of access identified in the Enterprise Surveys, the main
relationships are estimated by ordered logit. Yet, for robustness purposes many of
these relationships have been estimated by ordinary least squares and when dummy
variables were used as dependent variables by probit.
110
3.4. Empirical Results
3.4.1. BASELINE RESULTS: HOW DO THE GOVERNANCE INDEXES AFFECT FIRMS ’
PERCEPTIONS OF CORRUPTION AND LIKELIHOOD OF ENGAGING IN IT?
As noted above, the main focus of this study is to assess the extent to which the
relationship between the country-level World Governance Indexes and the two sets
of outcomes (1) the individual firm’s perception of corruption as an obstacle
(ObstCorrupt) to its business and (2) the firm’s likelihood of engaging in giving out
“gift” to different types of agents to help gain access of their business to alternative
permits and services may vary between different types of firms. We begin however
with what might be considered as the baseline results, the direct relationships without
taking into consideration the effects on these relationships of different firm
characteristics.
We begin our presentation of the results with Table 3.2 providing estimates obtained
from the ordered logistic regressions for ObstCorrupt. From the top row of the table
it can be seen that the firm’s perception of ObstCorrupt as an obstacle to business is
positively related to the dummy variable “Any Related Activity” representing the
firm saying that it needed at least one of the six permits or services identified in the
Enterprise Surveys as possible subjects for bribery (gifts asked for or suppled) by or
to relevant officials. Given that the overall mean score of ObstCorrupt from Table 3.1
is 1.76 and that in all columns of the table, the coefficient of Any Related Activity is
statistically significant and ranging from a little below 0.2 to a little above 0.3, this
means that the magnitude of this increase in ObstCorrupt is quite considerable, well
over 10 percent in percentile terms.
In the next six rows of the table are the estimated effects on ObstCorrupt of each of
the six different Governance Indexes beginning with Control of Corruption and
ending with Voice and Accountability. In all six cases, the estimated coefficient of this
index is negative but is statistically significant at the five percent level or less only in
the first five. This means that the effect of the Voice and Accountability index which
111
relates to the perceptions of the extent to which a country’s citizens are able to
participate in selecting their government, as well as freedom of expression, freedom
of association and a free media on ObstCorrupt is not statistically significant. Note
also that, in that case the direct effect of any related activity is largest (0.305). In all the
other cases, however these Governance Indicators have highly significant negative
influences, confirming the presumption in much of the literature that higher values of
each of these five Governance Indexes should have a significant negative effect on the
firm’s overall perception of ObstCorrupt. The three largest of these negative effects
on ObstCorrupt are those of Control of Corruption (CC), Government Effectiveness
(GE) and Rule of Law (RL).
Table 3.2
(1) (2) (3) (4) (5) (6)
Any related activity 0.238** 0.186** 0.286*** 0.258*** 0.244** 0.305***
(0.099) (0.095) (0.095) (0.096) (0.097) (0.094)
Gov Index - CC -0.806***
(0.088)
Gov Index - GE -1.090***
(0.099)
Gov Index - PS -0.406***
(0.101)
Gov Index - RQ -0.632***
(0.111)
Gov Index - RL -0.799***
(0.100)
Gov Index - VA -0.167
(0.111)
Firm characteristics Yes Yes Yes Yes Yes Yes
Industry fixed effects Yes Yes Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes Yes Yes
N 100458 100458 100458 100458 100458 100458
*** Significant at the 1 percent level.
** Significant at the 5 percent level.
* Significant at the 10 percent level.
Table 2 Effects of Governmance Indexes on Firm's Perceptions of Corruption (ordered logit)
Key explanatory
variables
Dependent variable: ObstCorrupt (0 to 4; estimated using ordered logit)
Notes: Standard errors are clustered at each level of country × industry × year. Firm characteristics include
all variables in Panel B of Table 1 except country-level indexes, plus an average of the ratings on other
obstacles to its business by a firm. CC = control of corruption; GE = government effectiveness; PS =
political stability (and absense of violence); RQ = regulatory quality; RL = rule of law
112
Table 3.3, on the other hand, presents the corresponding ordered probit estimates for
exactly the same specifications of variables as in Table 3.2, but in this case for the
dependent variable total number of gift-requested occasions in the last two years.
Once again, the any related activity measure has positive and significant effects and
the governance indexes negative and significant effects. Note that, the estimates of the
effect of any related activity are remarkably stable across columns. While for both
regressors, the magnitudes of the effects on this measure of actual corruption are
smaller than they were in the case of ObstCorrupt, in this case all six of the negative
effects of the Governance Indicators are statistically significant (at the 1% level).
Hence, not only does having to engage in any one of the six surveyed activities raise
a sampled firm’s perception of corruption but also it increases the incidence of actual
gift giving to relevant officials. However, if the firm is in a country for which any of
the six governance indexes is higher, that should lead not only to lower perception of
corruption but also to lower practice of corruption.
Is it possible that one of these corruption-prone activities could account for all these
positive effects on perceptions of corruption or corruption itself? To help answer this
question, we investigate this by estimating the same relationships, but for each of the
six corruption-related activities separately (Table A2 of Appendix 3). Since in this case
the dependent variables are simply dummy variables for gifts given for Electricity
Connection in Panel 1, Telephone Connection in Panel 2, Tax Inspection in Panel 3,
Import License in Panel 4, Operating License in Panel 5 and Construction Permit in
Panel 6, these results are Probit estimates. Notably, in every case the need for this
specific activity has a positive and highly significant effect on the likelihood of the
firm’s providing a gift to the relevant agent. The coefficients are smallest for
Telephone Connection and largest for Operating License and Construction Permit
where one might well suspect that timely supply is likely to be most important to
success of the business. Although the magnitudes of the negative effects of the
individual Governance Indexes are smaller than they were in Table 3.3 for the total
113
gift giving occasions, they are again statistically significant in all cases except column
(3) of Panel 2 for Telephone Connection. Therefore in general these results
demonstrate quite clearly that the overall likelihood of gift giving by firms to agents
is by no means driven by the effects of one single corruption-prone activity. Rather
each of the six such activities has a significant positive effect on such a gift and each
of the Governance Indexes has an offsetting negative influence on the likelihood of
gift-giving.
Table 3.3
(1) (2) (3) (4) (5) (6)
Any related activity 0.145*** 0.143*** 0.152*** 0.147*** 0.147*** 0.146***
(0.013) (0.013) (0.014) (0.013) (0.014) (0.013)
Gov Index - CC -0.116***
(0.020)
Gov Index - GE -0.108***
(0.014)
Gov Index - PS -0.053***
(0.010)
Gov Index - RQ -0.102***
(0.014)
Gov Index - RL -0.108***
(0.013)
Gov Index - VA -0.087***
(0.013)
Firm characteristics Yes Yes Yes Yes Yes Yes
Industry fixed effects Yes Yes Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes Yes Yes
N 103547 103547 103547 103547 103547 103547
adj. R-sq 0.071 0.067 0.060 0.068 0.068 0.064
*** Significant at the 1 percent level.
** Significant at the 5 percent level.
* Significant at the 10 percent level.
Table 3 Effects of Governmance Indexes on Firm's Total Occasions of Giving Gifts
Key explanatory
variables
Dependent variable: Number of gift-requested occasions (from 0 to 6)
Notes: Standard errors are clustered at each level of country × industry × year. Firm characteristics include
all variables in Panel B of Table 1 except country-level indexes. CC = control of corruption; GE =
government effectiveness; PS = political stability (and absense of violence); RQ = regulatory quality; RL =
rule of law
114
3.4.2. RESULTS OF HETEROGENEOUS SPECIFICATIONS: WHAT ROLE DO FIRM
CHARACTERISTICS PLAY IN THE CORRUPTION- GOVERNANCE
RELATIONSHIP?
Given that firms in the higher governance countries perceive corruption to be a less
serious problem and to be less inclined to provide corruption-like gifts in exchange
for certain permits or access to services, the natural next step in our analysis is to
explore the extent to which the seemingly universal negative and significant effect of
the governance indexes on both the perception of corruption and actual “gift-giving”
activities might vary with specific firm characteristics. For example, we might want
to identify those form types which might be less benefitted by such institutional
reforms, at least so far as vulnerability to corruption is concerned.
In line with what was done for the heterogeneous effects of labor regulations and
enforcement in Chapter 1, we explore the extent to which effects of the different
Governance Indexes might differ along differ by several important dimensions of firm
characteristics, namely firm’s age (years of operation), size, industry/sector, export
intensity, technology and access to credit. If the effect does vary across different types
of firms, it will be captured by the magnitudes and significance of terms representing
interactions between firm type dummies and one or another of the Governance
Indexes. We also examine whether or not any of these heterogeneous effects differ
between the two different sets of outcomes, perception of corruption and apparent
bribery behavior. In Tables 3.2 and 3.3 above, we found that, after controlling for
relevant firm and country characteristics, both the perceptions of corruption
(ObstCorrupt) and actual corruption reflected in gifts by firms to various agents
involved in the corruption-prone permits and services are in almost all cases
significantly reduced by national Governance Indexes of higher rank. Yet, this does
not imply that the effects of the different Governance Indexes on some of these
outcome variables might not also vary with the firm and country characteristics.
115
In Table 3.4-1 we examine the extent to which the effects of each of the six different
Governance Indexes would vary according to firm age (represented by dummy
variables for mid-age firms (“age group 2”, 6–20 years of operation) and old firms
(more than 20 years of operation. The Ordered Logit estimates for each of the relevant
interaction terms show that, without exception, the effects of the interaction terms are
estimated to be negative for both age groups. Yet, whereas for the intermediate age
group, the negative effects are never statistically significant, also without exception
for the older age group the negative effects are larger and always statistically
significant. In all cases, except again the Voice and Accountability, the direct effect of
each individual governance index is negative and significant but generally somewhat
smaller in magnitude than the corresponding estimates of Table 3.2.
Hence it would seem that the negative effects of these individual governance indexes
are largely limited to the oldest firms for which these effects may have been in effect
for a considerable period of time. Once again, the direct effects of the Governance
Indexes are largest for Control of Corruption, Government Effectiveness and Rule of
Law. The same is true of the magnitudes of the negative interaction terms with Age
Group 3, except that in the latter case this negative effect is also quantitatively large
in the case of Regulatory Quality. For all six Governance Indexes the negative effects
of the Governance Indexes on ObstCorrupt are significantly larger for older firms than
for young ones (with less than six years in operation). Put another way, the results
suggest that the existence of high quality governance indexes are negligible for firms
with no more than six years in operation.
Table 3.4-2 reports the estimated effects on ObstCorrupt of corresponding interactions
between each of the six Governance Indexes and dummy variables for firms of
medium size (20 to 99 employees) and large size (100+ employees). With the exception
of the Political Stability and to a lesser extent the Regulatory Quality Indexes, in these
cases the effects of the Governance Indexes do not vary by firm size. In the case of the
Political Stability Index, however, while by itself this index has only a fairly small
116
negative effect on ObstCorrupt, for large firms the magnitude of this negative effect
is almost twice as large as that for small firms.
Table 3.4-1
(1) (2) (3) (4) (5) (6)
Gov Index - CC -0.448***
(0.097)
CC × AgeGrp2 -0.113
(0.121)
CC × AgeGrp3 -0.340***
(0.122)
Gov Index - GE -0.500***
(0.124)
GE × AgeGrp2 -0.069
(0.120)
GE × AgeGrp3 -0.364***
(0.136)
Gov Index - PS -0.323***
(0.083)
PS × AgeGrp2 0.013
(0.084)
PS × AgeGrp3 -0.202**
(0.102)
Gov Index - RQ -0.303***
(0.098)
RQ × AgeGrp2 -0.056
(0.103)
RQ × AgeGrp3 -0.326**
(0.129)
Gov Index - RL -0.424***
(0.101)
RL × AgeGrp2 -0.040
(0.112)
RL × AgeGrp3 -0.311**
(0.132)
Gov Index - VA -0.137
(0.086)
VA × AgeGrp2 -0.010
(0.085)
VA × AgeGrp3 -0.216*
(0.119)
Firm characteristics Yes Yes Yes Yes Yes Yes
Industry fixed effects Yes Yes Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes Yes Yes
N 100457 100457 100457 100457 100457 100457
Table 4-1 Effects of Governmance Indexes on Firm's Perceptions of Corruption (ordered logit),
by Firm's Years of Operation in the Country
Key explanatory
variables
Dependent variable: ObstCorrupt (0 to 4; estimated using ordered logit)
Notes: Standard errors are clustered at each level of country × industry × year. Firm characteristics include
all variables in Panel B of Table 1 except country-level indexes, plus an average of the ratings on other
obstacles to its business by a firm. CC = control of corruption; GE = government effectiveness; PS =
political stability (and absense of violence); RQ = regulatory quality; RL = rule of law; AgeGrp2 = dummy
for 10-19 years; AgeGrp3 = dummy for 20 years and above
117
Table 3.5-2
Table 3.4-3 in appendix 3 presents the corresponding estimates of the interaction
effects between each of the Governance Indexes and the industry or sector to which
(1) (2) (3) (4) (5) (6)
Gov Index - CC -0.575***
(0.084)
CC × Medium Size -0.093
(0.098)
CC × Large Size -0.042
(0.149)
Gov Index - GE -0.589***
(0.094)
GE × Medium Size -0.140
(0.092)
GE × Large Size -0.055
(0.108)
Gov Index - PS -0.296***
(0.066)
PS × Medium Size -0.166**
(0.070)
PS × Large Size -0.216**
(0.085)
Gov Index - RQ -0.346***
(0.092)
RQ × Medium Size -0.191**
(0.086)
RQ × Large Size -0.135
(0.100)
Gov Index - RL -0.509***
(0.091)
RL × Medium Size -0.069
(0.091)
RL × Large Size 0.001
(0.106)
Gov Index - VA -0.199**
(0.081)
VA × Medium Size 0.021
(0.087)
VA × Large Size 0.032
(0.082)
Firm characteristics Yes Yes Yes Yes Yes Yes
Industry fixed effects Yes Yes Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes Yes Yes
N 100458 100458 100458 100458 100458 100458
Table 4-2 Effects of Governmance Indexes on Firm's Perceptions of Corruption (ordered logit),
by Size of Firm
Key explanatory
variables
Dependent variable: ObstCorrupt (0 to 4; estimated using ordered logit)
Notes: Standard errors are clustered at each level of country × industry × year. Firm characteristics include
all variables in Panel B of Table 1 except country-level indexes, plus an average of the ratings on other
obstacles to its business by a firm. CC = control of corruption; GE = government effectiveness; PS =
political stability (and absense of violence); RQ = regulatory quality; RL = rule of law
118
the firm belongs. Once again, these results show large direct negative effects of each
of the six Governance Indexes on ObstCorrupt. In most cases, however, the
interactions between these indexes and the Sector dummy variables are not
statistically significant. In five of the six Governance Indexes, i.e., those for Control of
Corruption, Government Effectiveness, Regulatory Quality, Rule of Law and Voice
and Accountability, those negative effects on ObstCorrupt are very substantially
offset by positive effects of interactions with the Textiles and Garments dummy.
Hence, for firms in Textiles and Garments, the negative and significant effects of the
Governance Indicators are limited largely to the Political Stability Index alone.
Appendix Tables 5-1, 5-2 and 5-3 present the ordered logit estimates corresponding
to those of Tables 4-1 through Table 4-3 for the case in which the dependent variable
is again the Total Number of Types of Gifts Given instead of ObstCorrupt. In every
column of each of these tables the direct effect of Any Related Activity is again
positive and significant as it was in Table 3. Indeed, even the magnitudes of these
positive direct effects of the Number of Types of Gifts Given are almost identical in
these new tables to what they were in Table 3. We also find in Tables 5-1 through 5-3
estimates of the direct effects of the Governance Indexes that are very similar in sign
(negative) and magnitude to those in Table 3.3-3.
In Appendix Table 5-1, however, we can see some age group–governance Indicator
interaction terms that are statistically significant. In particular, the interactions
between Large Size and the Control of Corruption, Government Effectiveness and
Rule of Law Indexes all have relatively small positive effects, indicating partial offsets
to the otherwise negative effects of these Governance Indexes on the number of gift
types paid by sample firms. By contrast, in appendix Table 5-2 each of the six
interactions between large size and the individual governance indexes is negative and
significant, indicating that the effect of each of these governance indexes on gift giving
is larger for firms of large size than for firms of small size.
119
In the case of appendix Table 5-3, however, in only one case, is an interaction between
one of the governance indicators and a sectoral dummy variable statistically
significant. This is the small negative coefficient for the interaction between Voice and
Accountability and the Retail dummy, implying that the negative effect of Voice and
Accountability on this measure of corruption is distinctive (in particular, larger) only
for firms in Retail activities. While the estimates of the interactions between the
dummy for firms in Textiles and Garments and each of the first five of the governance
indicators (Control of Corruption, Government Effectiveness, Political Stability,
Regulatory Quality and Rule of Law) are all negative and relatively large in
magnitude, none of these coefficients is statistically significant.
By comparing Tables 4-3 and 5-3 in appendix, one can see at least some indication of
an interesting difference in the effects of interactions between the Governance
Indicators and the textiles/garments and manufacturing sector dummy variables
between those on the Perceptions of Corruption (ObstCorrupt in Table 4-3) and those
on the practice of corruption (types of gift giving in appendix Table 5-3). In particular,
while in Table 4-3, the interactions between this sectoral dummy and the Control of
Corruption, Government Effectiveness, Regulatory Quality, Rule of Law and Voice
and Accountability reveal positive and significant effects on the perceptions of
corruption index (ObstCorrupt), from Table 5-3 these same interactions have no
statistically significant effects on the measure of actual gift giving (number of types of
gift giving).
Other heterogeneous analyses that we have examined are those involving the firm’s
export intensity (the firms exports as a percent of total sales, technological capabilities
(measured by the firm’s having a quality certificate or alternatively use of its own
website) and the firm’s access to credit from a private bank , by firm’s source of
capital). But, by no means do all of these heterogeneous effects turn out to be
significant or consistent. For example, in the case in the case of Export Intensity in
Table 4-4, not a single one of the Governance-Export Intensity interactions has a
120
statistically significant effect on the Corruption Perceptions measure (ObstCorrupt).
Yet, in the case of the effects of these interactions on the total number of gift types
(Table 5-4), in most cases there is at least a small negative and statistically significant
effect. Hence, the Governance Indexes seem to have slightly more significant negative
effects on actual corruption for Firms with high shares of exports in total sales than
they do on firm with lower or zero export shares.
In the case of the firms with higher technology (measured by possession of a Quality
Certificate), however, the results in appendix Tables 4-5 and 5-5 are quite different. In
this case the effects of the interactions between the Governance Indicators and this
measure of technology on the Perceptions of Corruption (ObstCorrupt) are large and
statistically significant in all but two of the columns (column (3 with Regulatory
Quality, and column (6) with Voice and Accountability. In most of these cases, the
magnitudes of the direct and interaction coefficients are such that the effects of these
Governance Indexes in reducing the perceptions of Corruption are from 30 percent to
80 percent larger for firms with Quality Certificates than for other firms. On the other
hand, higher Governance Indicators have only very slightly larger effects in reducing
actual Gift Giving on firms with Quality Certificates.
The question about credit access was much less frequently included in the different
Enterprise Survey questionnaires. This explains why the sample size is so very
significantly reduces in appendix Tables 4-6 and 5-6. From these tables, however, one
can see some particularly interesting evidence of heterogeneous effects of some of the
different Governance Indicators in the case of firms with access to private credit. In
the case in which ObstCorrupt is the dependent variable, the negative effects of
Control of Corruption, Regulatory Quality and Voice and Accountability are all
significantly larger for firms with access to Credit from a Private Bank. In the latter
two cases, these effects are more than double the direct effect and are much more
statistically significant. The same is true for the effects on actual corruption (Number
of types of gift giving by firms. Hence, although in neither of these tables are the
121
interactions of credit access with either Political Stability or Rule of Law statistically
significant, for all other Governance Indicators their effects are considerably larger in
reducing both perceptions of corruption and actual corruption for firms with access
to credit from private banks. These firms are likely to be more formal in general than
firms without such access to credit from private banks. Based on this interpretation,
this result underscores the point that the Governance Indicators, especially those of
Control of Corruption, Regulatory Quality, and Voice and Accountability, are likely
to be much more effective in reducing corruption (both perceptions and actual) for
formal than for informal firms.
3.4.3. EFFECTS OF CONTROL VARIABLES
While the emphasis in this study has been to extend the analysis of the effects of
Governance indicators to include how these differ across different firm characteristics
and types, it is also quite relevant and important to examine the direct effects of these
different characteristics. Estimates of the effects of these obtained from the same
framework as in Tables 3.2 and 3.3 are presented in Appendix 3 Tables A2 and A3.
Some highlights are as follows:
(1) The effects of the variable meanLog_ManTimeReg, representing the average time
spent by managers with officials in dealing regulations may be of considerable interest
since it is constructed in such a way as to reflect that average of managers in the same
region, sector and size group but excluding the firm’s own response . It may be
considered a proxy for the complexity and enforcement of regulations. Notice that it
has very different effects on the two measures of corruption, i.e., large, positive effects
on the perception of corruption measure ObstCorrupt but small negative effects on
actual corruption (Number of Types of Gifts Given by the Firm). In the latter case,
however, the magnitudes and significance of the effects varies across the columns,
being small and insignificant when either Regulatory Quality or Voice and
Accountability serve as the Governance Indicator.
122
(2) With respect to the sector dummy variables, the effects are rarely statistically
significant, though there is a tendency for those of Textiles and Garments and
Manufacturing to be somewhat positive on ObstCorrupt but negative on number of
types of gifts given by the firm.
(3) Although again not statistically significant, there is a tendency for the effects of
Age Group 2 (mid-age firms) and especially Age Group 3 (old firms) on perceptions
of corruption to be positive but those on actual gift giving to be negative.
(4) Not surprisingly, Government Ownership tends to be associated with lower scores
on both perceptions of corruption and actual corruption, though in part because of
the small numbers of firms with government ownership in these samples, these effects
are not statistically significant.
(5) Solely owned firms seem to have significantly lower scores on perceptions of
corruption but no significant differences in actual corruption practices (number of
types of gifts given by the firm).
(6) Medium and Large Sized firms tend to have both higher perceptions of Corruption
and actual experience with corruption. Indeed, the positive effect of the Medium Size
dummy on Number of Types of Gifts Given by the Firm are the among the most
consistent and statistically significant of all the results in Table A3.
(7) Female Owned firms seem to have significantly lower scores on the perception of
corruption measure ObstCorrupt in Table A2 but not on actual corruption in Table
A3. The latter result conflicts with the finding of Clarke (2015) which did reveal
significant negative effects of female ownership on actual corruption, but in a smaller
sample confined to transition economies.
(8) Not surprisingly, managers with longer experience have slightly higher
perceptions of corruption, but not higher gift-giving.
(9) Firms with larger Shares of Exports in Total Sales tend to have higher scores on
both ObstCorrupt and Total Number of Types of Gift Giving by Firms though only in
the latter case are they statistically significant.
123
(10) Firms with their own Websites tend to have higher values of ObstCorrupt and
actual corruption than those without websites but these differences are significant
only in the case of ObstCorrupt.
(11) In countries with higher income and corporate tax rates, perceptions of corruption
are higher, though significantly so only when Political Stability is the Governance
Index. On the other hand, when Regulatory Quality is the Governance Index, firms in
countries with higher tax rates have significantly lower incidence of actual corruption.
3.5. Conclusion
By combining data from the governance indicators of the World Bank over time with
the various Enterprise Surveys, we have constructed a data set of over 110,000
individual firms in 132 developing countries
57
capable of identifying the effects of
both firm characteristics and six different country-level governance indicators and
their interactions on both perceptions of corruption (ObstCorrupt) and actual
corruption (measured by the number of different types of gifts given by firms). The
validity of the results is strengthened by (1) the careful way in which the questions
concerning corruption were posed in the Enterprise Surveys and (2) the numerous
robustness checks we have performed, showing the results to be rather insensitive to
different estimation procedures, and specifications.
Of special importance as contributions are the extension of the analysis on corruption
to (1) examining differences in effects of the governance indicators and other
measures between the perceptions of corruption (ObstCorrupt) and Actual
Corruption captured by the number of different types of gifts Given to relevant
officials, and (2) heterogeneity in the effects of the different governance indicators
across different types of firm characteristics.
With respect to differences in effects between the perceptions of corruption
(ObstCorrupt) and our proxy for actual corruption the number of different types of
57
These firms are from 135 countries in total, including three OECD member countries (Israel, Poland and
Sweden).
124
Gifts that firms give, the results above and in particular Tables A2 and A3 have
identified many differences. Some of these differences are limited to magnitudes of
the effects, the effects of the governance indicators, government ownership, sole
ownership, medium/large size, female ownership dummy, and website all being
considerably larger and more significant on ObstCorrupt than on the number of
different types of gifts that firms give. In some cases, like exports as a percent of sales,
medium size the size and significance of the effects are larger on actual corruption
than on perceptions of corruption.
With respect to the heterogeneity in the effects of the governance indicators, we have
found numerous examples of fairly strong heterogeneity across firm characteristics,
both in the case of ObstCorrupt as in Tables 4-1 through 4-6 and the number of
different types of gifts that firms give in Appendix Tables 5-1 through 5-6. For
example, from Appendix Tables 4-6 and 5-6 we find several of the governance
indicators to have much greater success in reducing corruption among formal firms
(with access to credit from private banks) than among informal or less formal firms.
We also found this to be the case for older firms (age group 3) for perceptions of
corruption in Appendix Table 4-1 but interestingly enough, quite the opposite is true
for actual corruption in Appendix Tables 5-1, where the effects of these interactions
are frequently positive. In the case of Size of Firm the effect of governance interactions
contributes significantly to reducing perceptions of corruption (ObstCorrupt) in only
one column (column (3) of Appendix Table 4-2) but in all columns of Appendix Table
5-2 for actual corruption. With respect to sectors of industry, however, there is little
variation in the effects of the governance indicators across sectors in either perceptions
of, or actual corruption.
125
Bibliography
Ackerman, R.S. 1999. Corruption and Government: Causes, Consequences and Reform.
Cambridge University Press.
Addison, J., and P. Teixeira. 2003. “The Economics of Employment Protection.”
Journal of Labor Research, 24 (1): 85–129.
Addison, J., and J. Grosso. 1996. “Job Security Provisions and Employment: Revised
Estimates.” Industrial Relations, 35 (4): 585–603.
Ades, A., and R. Di Tella. 1999. “Rents, Competition and Corruption.” American
Economic Review, 89 (4): 982-993.
Adhvaryu, A., A.V. Chari, and S. Sharma. 2013. “Firing Costs and Flexibility:
Evidence from Firms' Labor Adjustments to Shocks in India.” Review of
Economics and Statistic, 95 (3): 725–740.
Agenor, P.R., and K. El-Aynaoui. 2003. “Labor Market Policies and Unemployment
in Morocco: A Quantitative Analysis.” World Bank Policy Research Working
Paper 3091.
Aleksynska, M., and M. Schindler. 2011. “Labor Market Regulations in Low-,
Middle- and High-Income Countries: A New Panel Database.” IMF Working
Paper No. 11/154.
Allard, G. 2005. “Measuring Job Security Over Time: In Search of a Historical
Indicator for EPL (Employment protection legislation).” Madrid: Instituto de
Empresa, Working Paper 05–17.
Almeida, R.K., and R. Aterido. 2008. “The Incentives to Invest in Job Training: Do
Strict Labor Codes Influence this Decision?” World Bank SP Discussion Paper
No. 0832.
Almeida, R., and P. Carneiro. 2011. “Enforcement of Regulation, Informal
Employment, Firm Size and Firm Performance.” Journal of Comparative
Economics, 37 (1): 28–46.
Angel-Urdinola, D.F., and Leon-Solano, R. 2013. Overview in D.F. Angel-Urdinola,
A. Kuddo, and A. Semlali. Building effective employment programs for
unemployed youth in the Middle East and North Africa. Washington DC: World
Bank.
Arpaia, A., P. Braila, and F. Pierini. 2007. “Tracking Labor Market Reforms in the EU
using the LABREF Database.” Bonn: presented at the IZA-fRDB Workshop:
Measurement of Labor Market Institutions.
Ayyagari, M., A. Demirgüç-Kunt, and V. Maksimovic. 2008. “How Important Are
Financing Constraints? The Role of Finance in the Business Environment.”
World Bank Economic Review, 22 (3): 483–516.
126
Bardhan, P. 1997. “Corruption and Development: A Review of Issues.” Journal of
Economic Literature, 35 (3): 1320–1346.
Barros, A. J., C. Ronsmans, H. Axelson, E. Loaiza, A. D. Bertoldi, G. França, J. Bryce,
J. T. Boerma, and C. G. Victora. 2012. “Equity in Maternal, Newborn, and
Child Health Interventions in Countdown to 2015: A Retrospective Review of
Survey Data from 54 Countries.” The Lancet, 379 (9822): 1225–1233.
Bastagli, F. 2011. “Conditional Cash Transfers as a Tool of Social Policy.” Economic
and Political Weekly, 46 (21): 61–66.
Becker, G.S., and G.J. Stigler. 1974. “Law Enforcement, Malfeasance, and the
Compensation of Enforcers” Journal of Legal Studies, 3 (1): 1–18.
Belot, M., J. Boone, and J. Ours. 2007. “Welfare Improving Employment Protection.”
Economica, 74 (295): 381–396.
Besley, T., and R. Burgess. 2004. “Can Labor Regulation Hinder Economic
Performance? Evidence from India.” Quarterly Journal of Economics, 119 (1):
91–134.
Bhattacharjea, A. 2006. “Labour Market Regulation and Industrial Performance in
India: A Critical Review of the Empirical Evidence.” Indian Journal of Labour
Economics, 49 (2), 211–232.
Bhaumik, S.K., R. Dimova, S.C. Kumbhakar, and K. Sun. 2012. “Does Institutional
Quality Affect Firm Performance? Insights from a Semiparametric
Approach.” IZA Discussion Paper No. 6351.
Blanchard, O., and J. Wolfers. 2000. “The Role of Shocks and Institutions in the Rise
of European Unemployment: The Aggregate Evidence.” Economic Journal, 110
(462): C1–C33.
Blanchard, O., and P. Portugal. 2001. “What Hides Behind an Unemployment Rate:
Comparing Portuguese and U.S. Labor Markets.” American Economic Review,
91 (1): 187–207.
Botero, J., S. Djankov, R. La Porta, F. Lopez-de-Silanes, and A. Shleifer. 2004. “The
Regulation of Labor.” Quarterly Journal of Economics, 119 (4): 1339–1382.
Caiden, G.E. 2013. “Accounting for Success in Combatting Corruption” in J. Quah,
ed. Different Paths to Curbing Corruption, Emerald Publishing. Bringley 189–
2014.
Caiden, G.E., and N.J. Caiden 1977. “Administrative Corruption.” Public
Administration Review, 37 (3): 301–309.
Campos, N., and F. Giovannoni. 2007. “Lobbying, Corruption and Political
Influence.” Public Choice, 131 (1) 1–21.
127
Campos, N.F., C. Hsiao., and J.B. Nugent. 2010. “Crises, What Crises? New Evidence
on the Relative Roles of Political and Economic Crises in Begetting Reforms.”
The Journal of Development Studies, 46 (10): 1670–1691.
Carlin, W., and P. Seabright. 2009. “Bring Me Sunshine: Which Parts of the Business
Climate Should Public Policy Try to Fix?” Proceedings of the Annual Bank
Conference on Development Economics 2008. Washington D.C.: World Bank.
Carlin, W., M.E. Schaffer, and P. Seabright. 2006. “Where are the Real Bottlenecks? A
Lagrangian Approach to Identifying Constraints on Growth from Subjective
Survey Data.” CEPR Discussion Paper No. 5719.
Chaudhury, N., J. Hammer, M. Kremer, K. Muralidharan, and F. H. Rogers. 2006.
“Missing in Action: Teacher and Health Worker Absence in Developing
Countries.” The Journal of Economic Perspectives, 20 (1): 91–116.
Clarke, G.R.G. 2011. “How Petty is Petty Corruption? Evidence from Firm Level
Surveys in Africa.” World Development, 39 (7): 1122–1232.
―――. 2015. “More Evidence on Corruption and Gender: Evidence from Firm Level
Surveys.” Mimeo
Clarke, G.R.G, and L.C. Xu. 2004. “Privatization, Competition and Corruption: How
Characteristics of Bribe Takers and Payers Affect Bribes to Utilities.” Journal of
Public Economics, 88 (9–10): 2067 –2097.
Cunat, A., and M.J. Melitz. 2009. “Volatility, Labor Market Flexibility and the
Patterms of Comparative Advantage.” Journal of the European Economics
Association, 10 (2): 225–254.
Deakin, S., P. Lele, and M. Siems. 2007. “The Evolution of Labor Law: Calibrating
and Comparing Regulatory Regimes.” International Labour Review, 146 (1):
133–162.
Dix, S., and N. Jayawickrama 2010. Fighting Corruption in a Post-Conflict and Recovery
Situation: Lessons from the Past. New York: United Nations Development
Programme.
Dollar, D., M. Hallward-Driemeier, and T. Mengistae. 2005. “Investment Climate
and Firm Performance in Developing Economies.” Economic Development and
Cultural Change, 54 (1): 1–31.
Dupas, P. 2014a. “Getting Essential Health Products to Their End Users: Subsidize,
But How Much?” Science, 1256973 (1279): 345.
―――. 2014b. “Global Health Systems: Pricing and User Fees.” In Encyclopedia of
Health Economics, Volume 2, edited by A. J. Culyer, 136–141. San Diego:
Elsevier.
―――. 2014c. “Short‐Run Subsidies and Long‐Run Adoption of New Health
Products: Evidence from a Field Experiment.” Econometrica, 82 (1): 197–228.
128
Elbadawi, I., and N. Loayza. 2008. “Informality, Employment and Economic
Development in the Arab World.” Journal of Development and Economic Policies,
10 (2): 25–75.
Filmer, D., and L. H. Pritchett. 2001. “Estimating Wealth Effects Without
Expenditure Data—Or Tears: An Application to Educational Enrollments in
States of India.” Demography, 38 (1): 115–132.
Fiori, G., G. Nicoletti, S. Scarpetta, and F. Schiantarelli. 2007. “Employment
Outcomes and the Interaction between Product and Labor Market
Deregulation: Are They Substitutes or Complements?” IZA Discussion Paper
No. 2770.
Fisman, R., and R. Gatti. 2002. “Decentralization and Corruption: Evidence across
Countries.” Journal of Public Economics, 83 (3): 325–345.
Forteza, A., and M. Rama. 2006. “Labor Market ‘Rigidity’ and the Success of
Economic Reforms across More than 100 Countries.” Journal of Policy Reform, 9
(1): 75–106.
Fredriksson, P.G., and J. Svensson. 2003. “Political Instability, Corruption and Policy
Formation: The Case of Environmental Policy.” Journal of Public Economics, 87
(7–8): 1383–1405.
Freeman, R.B. 2004. “Labor Market Institutions without Blinders: The Debate over
Flexibility and Labor Market Performance.” NBER Working Paper 11286.
Glassman, A., D. Duran, L. Fleisher, D. Singer, R. Sturke, G. Angeles, J. Charles, B.
Emrey, J. Gleason, W. Mwebsa, K. Saldana, K. Yarrow, and M. Koblinsky.
2014. “Impact of Conditional Cash Transfers on Maternal and Newborn
Health.” Journal of Health, Population and Nutrition (JHPN), 31 (4): S48–S66.
Greenhill, B., L. Mosley and A. Prakash. 2009. “Trade-Based Diffusion of Labor
Rights: A Panel Study, 1986–2002.” American Political Science Review, 103 (4):
669–690.
Haltiwanger, J., L. Foster, and C. Syverson. 2008. “Reallocation, Firm Turnover and
Efficiency: Selection on Productivity or Profitability.” American Economic
Review, 98 (1): 394–425.
Heckman, J., and Pages, C. 2000. “The Cost of Job Security Regulation: Evidence
from Latin American Labor Markets.” NBER Working Paper 7773.
―――. eds. 2004. Law and employment: Lessons from Latin America and the Caribbean.
New York: University of Chicago Press.
Helpman, E., and O. Itskhoki. 2010. “Labor Market Rigidities, Trade and
Unemployment.” Review of Economic Studies, 77 (3): 1100–1137.
Holmes, W., D. Hoy, A. Lockley, K. Thammavongxay, S. Bounnaphol, A.
Xeuatvongsa, and M. Toole. 2007. “Influences on Maternal and Child
129
Nutrition in the Highlands of the Northern Lao PDR.” Asia Pacific Journal of
Clinical Nutrition, 16 (3): 537.
Hopenhayn, H.A., and R. Rogerson. 1993. “Job Turnover and Policy Evaluation: A
General Equilibrium Analysis.” Journal of Political Economy, 101 (5): 915–938.
Independent Evaluation Group (IEG). 2013. Delivering the Millennium Development
Goals to Reduce Maternal and Child Mortality: A Systematic Review of Impact
Evaluation Evidence. Washington, DC: World Bank.
Jain, A.K. 2001. “Corruption: A Review.” Journal of Economic Surveys, 15 (1): 71–121.
Kaplan, D.S. 2009. “Job Creation and Labor Reform in Latin America.” Journal of
Comparative Economics, 37 (1): 91–105.
Kaplan, D.S., and V. Pathania. 2010. “What influences firms' perceptions?” Journal of
Comparative Economic, 38 (4): 419–431.
Kamiya, Y. 2011. “Socioeconomic Determinants of Nutritional Status of Children in
Lao PDR: Effects of Household and Community Factors.” Journal of Health,
Population, and Nutrition, 29 (4): 339.
Kaufmann, D., A. Kraay, and P. Zoido-Lobaton. 1999. “Governance Matters.” World
Bank Policy Research Working Paper 2196.
Kaufmann, D., A. Kraay, and M. Mastruzzi. 2004 “Governance Matters III:
Governance Indicators for 1996 –2002. World Bank Policy Research Working
Paper 3106.
―――. 2007. “Governance Matters V: Governance Indicators for 1996–2005.” World
Bank Policy Research Working Paper 4280.
―――. 2010. “The Worldwide Governance Indicators: A Summary of Methodology,
Data and Analytical Issues.” World Bank Policy Research working Paper
5430.
Kinda, T., P. Plane, and M. Veganzones-Varoudakis. 2011. “Firm Productivity and
Investment Climate in Developing Countries: How Does Middle East and
North Africa Manufacturing Perform?” The Developing Economies, 49 (4): 429–
462.
Knack, S., and P. Keefer. 1997. “Does Social Capital Have an Economic Payoff? A
Cross-Country Investigation.” Quarterly Journal of Economics, 112 (4): 1251 –
1288.
Kounnavong, S., T. Sunahara, C. Mascie-Taylor, M. Hashizume, J. Okumura, K.
Moji, B. Boupha, and T. Yamamoto. 2011. “Effect of Daily Versus Weekly
Home Fortification with Multiple Micronutrient Powder on Haemoglobin
Concentration of Young Children in a Rural Area, Lao People’s Democratic
Republic: A Randomised Trial.” Nutrition Journal, 10 (129).
130
Kpodar, K. 2007. “Why Has Unemployment in Algeria Been Higher than in MENA
and Transition Countries?” IMF Working Paper 07/210.
Kucera, D. 2002. “Core Labor Standards and Foreign Direct Investment.”
International Labor Review, 141 (1–2): 31–69.
Kuntchev, V., R. Ramalho, J. Rodriguez-Meza, and J. Yang. 2013. “What Have We
Learned from the Enterprise Surveys Regarding Access to Finance by SMEs?”
World Bank Policy Research Working Paper 6670.
Lagos, R. 2006. “A Model of TFP.” Review of Economic Studies, 73 (4): 983–1007.
Lambsdorff, J.G. 2006. “Causes and Consequences of Corruption: What Do We
Know from a Cross-section of Countries” in S. Rose-Ackerman, ed.
International Handbook on the Economics of Corruption. Cheltenham:
Edward Elgar, 3 –51.
Lederman, D., N. Loayza, and R.S. Rodrigo. 2005. “Accountability and Corruption:
Political Institutions Matter.” Economics and Politics, 17 (1): 1–35.
Leff, N. 1964. “Economic Development through Bureaucratic Corruption” American
Behavioral Scientist, 8 (3): 8–14.
Leite, C., and J. Weidemann. 1999. “Does Mother Nature Corrupt? Natural
Resources, Corruption and Economic Growth.” IMF Working Paper 85.
Malomo, F. 2013. “Factors Influencing the Propensity to Bribe and Size of Bribe
Payments: Evidence from Formal Manufacturing Firms in West Africa.”
Brighton: University of Sussex.
Masuno, K., D. Xaysomphoo, A. Phengsavanh, S. Douangmala, and C. Kuroiwa.
2009. “Scaling Up Interventions to Eliminate Neonatal Tetanus: Factors
Associated with the Coverage of Tetanus Toxoid and Clean Deliveries
Among Women in Vientiane, Lao PDR.” Vaccine, 27 (32): 4284–4288.
Mauro, P. 1995. “Corruption and Growth.” Quarterly Journal of Economics, 60 (3): 681–
712.
Mauro, P. 1998. “Corruption and the Composition of Government Expenditure.”
Journal of Public Economics, 69 (2): 263–279.
Melitz , M. 2003. “The Impact of Trade on Intra-industry Reallocation of Aggregate
Industry Productivity.” Econometrica, 71 (6): 694–726.
Muravyev, A. 2014. “Evolution of Employment Protection Legislation in the USSR,
CIS and Baltic States, 1985–2009.” Europe-Asia Studies, 66 (8): 1270–1294.
Nicoletti, G., R.C.G. Haffner, S. Nickell, S. Scarpetta, and G. Zoega. 2000. “European
integration, liberalization and labor market reform.” in G. Bertola, T. Boeri,
and G. Nicoletta, eds. Welfare and Employment in a United Europe. Cambridge:
MIT Press.
131
Nickell, S. 1997. “Unemployment and Labor Market Rigidities: Europe versus North
America.” Journal of Economic Perspectives, 11 (3): 55–74.
Nugent, J.B. 2012. “Detecting Corruption and Evaluating Programs to Control It:
Some Lessons for MENA.” ERF Working Paper 738.
Park, H. 2003. “Determinants of Corruption: A Cross-National Analysis.”
Multinational Business Review, 13 (2): 29 –48.
Pellegrini, L., and R. Gerlagh. 2004. “Corruption’s Effect on Growth and Its
Transmission Channels.” Kyklos, 57 (3): 429 –456.
Persson, T., G. Tabellini, and F. Trebbi. 2003. “Electoral Rules and Corruption.”
Journal of the European Economic Association, 1 (4): 958–989.
Phimmasone, K., I. Douangpoutha, V. Fauveau, and P. Pholsena. 1996. “Nutritional
Status of Children in the Lao PDR.” Journal of Tropical Pediatrics, 42 (1): 5–11.
Pierre, G., and S. Scarpetta. 2006. “Employment Protection: Do Firms' Perceptions
Match with Legislation?” Economics Letters, 90 (3): 328–334.
Pimhidzai, O., N. Fenton, P. Souksavath, and V. Sisoulath. 2014. Poverty Profile in Lao
PDR Poverty Report for the Lao Consumption and Expenditure Survey, 2012–2013.
Washington, DC: World Bank.
Puhani, P. A. 2012. “The Treatment Effect, the Cross Difference, and the Interaction
Term in Nonlinear “Difference-In-Differences” Models.” Economics Letters,
115 (1): 85–87.
Rama, M., and R. Artecona. 2002. “A Database of Labor Market Indicators across
Countries.” Washington, D.C.: World Bank, mimeo.
Rivas, M.F. 2013. “An Experiment on Corruption and Gender.” Bulletin of Economic
Research, 65 (1): 10–42.
Rose-Ackerman, S. 1978. Corruption: A Study of Political Economy. New York:
Academic Press.
Seker, M. 2012. “Rigidities in Employment Protection and Exporting.” World
Development, 40 (2): 238–250.
Serra, D. 2006. “Empirical Determinants of Corruption: A Sensitivity Analysis.”
Public Choice, 126 (1): 225 –256.
Serra, D., and L. Wantchekon, eds. 2012. New Advances in Experimental Research on
Corruption. Bingley: Emerald Group Publishing.
Shleifer, A., and R. Vishny. 1993. “Corruption.” Quarterly Journal of Economics, 108
(3): 599 –617.
Spector, B. 2012. Detecting Corruption in Developing Countries: Identifying
Causes/Strategies for Action. Sterling: Kumarian Press.
132
Svensson, J. 2003. “Who Must Pay Bribes and How Much? Evidence from a Cross
Section of Firms.” Quarterly Journal of Economics, 118 (1): 207–230.
Swamy, A., S. Knack, Y.Lee, and O. Azfar. 2001. “Gender and Corruption.” Journal of
Development Economics, 64 (1): 25 –55.
Sychareun, V., V. Hansana, V. Somphet, S. Xayavong, A. Phengsavanh, and R.
Popenoe. 2012. “Reasons Rural Laotians Choose Home Deliveries over
Delivery at Health Facilities: A Qualitative Study.” BMC Pregnancy and
Childbirth, 12 (1): 86.
Tanzi, V. 1998. “Corruption around the World: Causes, Consequences, Scope and
Cures.” IMF Staff Papers, 45 (4): 559 –594.
Treisman, D. 2000. “The Causes of Corruption: A Cross-National Study.” Journal of
Public Economics, 76 (3): 399 –457.
Viengsakhone, L., Y. Yoshida, and J. Sakamoto. 2010. “Factors Affecting Low Birth
Weight at Four Central Hospitals in Vientiane, Lao PDR.” Nagoya Journal of
Medical Science, 72 (1–2): 51–58.
Wagstaff, A., C. Bredenkamp, and L. R. Buisman. 2014. “Progress Toward the Health
MDGs: Are the Poor Being Left Behind?” World Bank Policy Research
Working Paper 6894.
World Health Organization (WHO). 2006. WHO Child Growth Standards Based on
length/height, weight and age. Acta Paediatrica Supplement, 450: 76–85.
―――. 2010. Indicators for Assessing Infant and Young Child Feeding Practices. Part 3:
Country Profiles. Geneva: WHO.
World Bank. 2009. Emergency Project Paper on a Proposed Grant in the Amount of US$ 2
Million to the Lao People’s Democratic Republic for a Community Nutrition Project.
Washington DC: World Bank.
―――. 2013. Maternal Health Out-of-Pocket Expenditure and Service Readiness in Lao
PDR: Evidence for the National Free Maternal and Child Health Policy from a
Household and Health Center Survey. Washington, DC: World Bank.
―――. 2015. Laos; Laos - Community Nutrition Project; Community Nutrition Project.
Washington, DC: World Bank Group.
https://hubs.worldbank.org/docs/imagebank/Pages/docProfile.aspx?nodeid=24112474
Yamada, H., Y. Sawada, and X. Luo. 2013. “Why is Absenteeism Low among Public
Health Workers in Lao PDR?” The Journal of Development Studies, 49 (1): 125–
133.
133
Appendix 1
Table A1: List of Countries with Employment Change Questions
Country/year #Firms Percent Country/year #Firms Percent
Asia Latin America and Caribbean
Cambodia (2003) 497 0.9% Brazil (2003) 1,638 3.1%
China (2002, 2003) 2,156 4.0% Chile (2004, 2006) 919 1.7%
India (2002, 2006) 3,906 7.3% Costa Rica (2005) 338 0.6%
Indonesia (2003) 709 1.3% Ecuador (2003, 2006) 448 0.8%
Kazakhstan (2002, 2005) 828 1.5% El Salvador (2003, 2006) 463 0.9%
Kyrgyzstan (2002, 2003, 2005) 476 0.9% Guatemala (2003, 2006) 454 0.8%
Laos (2006) 92 0.2% Honduras (2003, 2006) 442 0.8%
Malaysia (2002) 895 1.7% Guyana (2004) 154 0.3%
Mongolia (2004) 192 0.4% Jamaica (2005) 7 0.0%
Pakistan (2002) 886 1.6% Nicaragua (2003, 2006) 446 0.8%
Philippines (2003) 680 1.3% Peru (2002, 2006) 570 1.1%
South Korea (2005) 515 1.0% Subtotal 5,879 10.9%
Sri Lanka (2004) 448 0.8%
Tajikistan (2002, 2003, 2005) 481 0.9% Middle East and North Africa
Thailand (2004) 1,384 2.6% Algeria (2002, 2007) 857 1.6%
Uzbekistan (2002, 2003, 2005) 658 1.2% Egypt (2004, 2007, 2008) 2963 5.5%
Vietnam (2005) 1,621 3.0% Jordan (2006) 483 0.9%
Subtotal 16,424 30.6% Lebanon (2006, 2009) 736 1.4%
Morocco (2004, 2007) 1384 2.6%
Europe Oman (2003) 337 0.6%
Albania (2002, 2005) 352 0.7% Saudi Arabia (2005) 628 1.2%
Armenia (2002, 2005) 522 1.0% Syria (2003) 408 0.8%
Azerbaijan (2002, 2005) 517 1.0% Turkey (2002, 2004, 2005) 2,318 4.3%
Belarus (2002, 2005) 571 1.1% West Bank and Gaza (2006) 399 0.7%
Bosnia and Herz (2002, 2005) 358 0.7% Subtotal 10,513 19.6%
Bulgaria (2002, 2004, 2005) 1,056 2.0%
Croatia (2002, 2005) 383 0.7% Sub-Saharan Africa
Czech (2002, 2005) 571 1.1% Angola (2006) 7 0.0%
Estonia (2002, 2005) 359 0.7% Benin (2004) 143 0.3%
Macedonia (2002, 2005) 366 0.7% Botswana (2006) 45 0.1%
Hungary (2002, 2005) 837 1.6% Burkina Faso (2006) 11 0.0%
Georgia (2002, 2005) 374 0.7% Burundi (2006) 13 0.0%
Germany (2005) 1,191 2.2% Cameroon (2006) 12 0.0%
Greece (2005) 530 1.0% Cape Verde (2006) 3 0.0%
Ireland (2005) 498 0.9% DR Congo (2006) 15 0.0%
Latvia (2002, 2006) 370 0.7% Ethiopia (2002) 422 0.8%
Lithuania (2002, 2004, 2005) 636 1.2% Kenya (2003) 231 0.4%
Moldova (2002, 2003, 2005) 613 1.1% Lesotho (2003) 70 0.1%
Montenegro (2003) 73 0.1% Madagascar (2005) 281 0.5%
Poland (2002, 2003, 2005) 1,550 2.9% Malawi (2005) 146 0.3%
Portugal (2005) 501 0.9% Mali (2003) 145 0.3%
Romania (2002, 2005) 849 1.6% Mauritania (2006) 22 0.0%
Russia (2002, 2005) 1,085 2.0% Mauritius (2005) 177 0.3%
Serbia & Montenegro (2002, 2005) 898 1.7% Namibia (2006) 30 0.1%
Slovakia (2002, 2005) 385 0.7% Senegal (2003) 219 0.4%
Slovenia (2002, 2005) 411 0.8% South Africa (2003) 602 1.1%
Spain (2005) 601 1.1% Swaziland (2006) 31 0.1%
Ukraine (2002, 2005) 1,053 2.0% Tanzania (2003) 262 0.5%
Subtotal 17,510 32.6% Uganda (2006) 280 0.5%
Zambia (2002) 205 0.4%
Subtotal 3,372 6.3%
Total 53,698 100.0%
134
Table A2: Definition of Variables
Variables Definition
Dependent Variables
AnyJobChange Absolute value of NetJobChange
NetJobCreation max(0, NetJobChange)
Log_NetJobCreation Natural logarithm of (NetJobCreation + 1)
NetJobDestruction max(0, -NetJobChange)
Log_NetJobDestruction Natural logarithm of (NetJobDestruction + 1)
Explanatory Variables
TaxDaysInsp See Q2
Log_TaxDatsInsp Natural logarithm of (TaxDaysInsp + 1)
PctTimeReg See Q3
Log_PctTimeReg Natural logarithm of (PctTimeReg + 1)
ObstLabor See Q4
AgeFirm Number of years since the firm been in operation
AgeFirm_10 Number of years since the firm been in operation (divided by 10)
AgeFirm2 Square of AgeFirm
AgeFirm2_100 Square of AgeFirm (divided by 100)
SmlSize dummy: 1 for small-size firms with 0–19 (permanent and temporary) employees
MedSize dummy: 1 for medium-size firms with 20–99 employees
LrgSize dummy: 1 for large-size firms with at least 100 employees
PriOwned dummy: 1 for firms with private ownership >50%
FgnOwned dummy: 1 for firms with foreign ownership >50%
GovOwned dummy: 1 for firms with government ownership >50%
Sole dummy: 1 for sole proprietorship
Partner dummy: 1 for partnership
Ind_AgriFood dummy: Agriculture and Food
Ind_TxlGmt dummy: Textiles and Garments
Ind_Manu dummy: Manufacturing
Ind_Service dummy: Services other than Retail
Ind_Retail dummy: Retail
Capital cities dummy: 1 if establishment and headquarter located in a capital city
Other large cities dummy: 1 if located in a non-capital city over 1 million population
QualCert 1 for firms that have ISO certification
MultiPlant 1 if the firm has more than one establishments operating in the country
Email 1 if the firm regularly uses email to interact with clients and suppliers
Website 1 if the website regularly uses website to interact with clients and suppliers
135
(1) (2) (3) (4) (5) (6) (7) (8) (9)
3.650*** 3.782*** 3.309*** 3.187*** 3.342*** 2.767*** -0.463** -0.440* -0.542**
(1.037) (1.134) (1.021) (0.943) (1.031) (0.923) (0.210) (0.231) (0.210)
2.779*** 3.227*** 2.754*** 2.508*** 2.898*** 2.482*** -0.271 -0.329* -0.272
(0.711) (0.749) (0.710) (0.651) (0.689) (0.648) (0.184) (0.180) (0.184)
1.310*** 1.239*** 0.961*** 1.342*** 1.275*** 0.918*** 0.032 0.035 -0.043
(0.288) (0.310) (0.295) (0.270) (0.291) (0.277) (0.069) (0.073) (0.067)
2.676*** 3.220*** 0.544***
(0.443) (0.414) (0.109)
Covariates in Table 3 Yes Yes Yes Yes Yes Yes Yes Yes Yes
Additional covariates No Yes No No Yes No No Yes No
Country fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes
Industry fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes
City-level fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes
Observations 41763 36066 41679 41763 36066 41679 41763 36066 41679
Adjusted R-sq 0.081 0.078 0.083 0.079 0.077 0.082 0.060 0.065 0.061
*** Significant at the 1 percent level.
** Significant at the 5 percent level.
* Significant at the 10 percent level.
Average Obstacle
Notes: meanObstLabor, meanLog_ManTimeReg, meanObstInfml_Low and Log_DaysPowerOut are standardized in the
same way as indicated in Table 1. Standard errors are clustered at each level of country × industry × city-level × year. Firm
characteristics include all variables in Table XX. Additional covariates include the percentage of capital financed internally, as
well as two dummy variables for a firm 1) being part of a franchise or with multiple locations in the country 2) having its own
Table A3 Robustness to Additional Covariates
meanObstLabor
meanLog_ManTimeReg
Log_DaysPowerOut
Net Job Change Net Job Creation Net Job Destruction
136
Table A4 Robustness to Alternatives Forms of Net Job Change
Ordered Probit
Standard OLS
Excluding zero
values
Winsorizing 5% at
top and bottom tail
Trimming 5% at
top and bottom tail
In quantiles
(1) (2) (3) (4) (5)
3.650*** 6.546*** 1.791*** 0.854*** 0.082***
(1.037) (1.845) (0.455) (0.279) (0.027)
2.779*** 5.382*** 0.984*** 0.369 0.060***
(0.711) (1.370) (0.338) (0.246) (0.021)
1.310*** 1.377** 0.850*** 0.562*** 0.048***
(0.288) (0.560) (0.128) (0.098) (0.008)
Firm characteristics Yes Yes Yes Yes Yes
Country fixed effects Yes Yes Yes Yes Yes
Industry fixed effects Yes Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes Yes
City-level fixed effects Yes Yes Yes Yes Yes
Observations 41763 21183 41763 38669 41763
Adjusted R-sq 0.081 0.151 0.101 0.071 -
*** Significant at the 1 percent level.
** Significant at the 5 percent level.
* Significant at the 10 percent level.
Ordinary Least Square
meanObstLabor
meanLog_PctTimeReg
Log_DaysPowerOut
Notes : meanObstLabor, meanLog_PctTimeReg and Log_DaysPowerOut are standardized in the same
way as indicated in Table 1. Standard errors are clustered at each level of country × industry × city-level ×
year. Firm characteristics include all variables in Table XX.
137
(1) (2) (3) (4) (5)
-0.474** -0.463** -0.460** -0.431** -0.377*
(0.221) (0.210) (0.210) (0.216) (0.203)
-0.271 -0.277
(0.184) (0.189)
-0.026
(0.112)
0.123 0.163
(0.160) (0.165)
0.198
(0.130)
0.020 0.032 0.032 0.020 0.027
(0.069) (0.069) (0.069) (0.069) (0.069)
Panel B: Firm
Characteristics (1) (2) (3) (4) (5)
0.352** 0.332** 0.331** 0.353** 0.351**
(0.142) (0.143) (0.142) (0.142) (0.142)
1.151*** 1.111*** 1.110*** 1.152*** 1.153***
(0.192) (0.191) (0.191) (0.192) (0.192)
-0.506*** -0.497*** -0.498*** -0.513*** -0.518***
(0.170) (0.170) (0.170) (0.170) (0.170)
1.830*** 1.873*** 1.873*** 1.823*** 1.821***
(0.288) (0.289) (0.289) (0.289) (0.289)
-0.143 -0.120 -0.119 -0.145 -0.152
(0.191) (0.191) (0.191) (0.191) (0.191)
-0.537*** -0.520*** -0.519*** -0.539*** -0.550***
(0.182) (0.183) (0.182) (0.182) (0.182)
0.359** 0.345** 0.344** 0.360** 0.357**
(0.147) (0.146) (0.146) (0.147) (0.147)
0.757*** 0.775*** 0.774*** 0.754*** 0.746***
(0.167) (0.167) (0.167) (0.167) (0.168)
0.013** 0.015** 0.015** 0.013** 0.012**
(0.006) (0.006) (0.006) (0.006) (0.006)
Panel C: Industry
Dummies (1) (2) (3) (4) (5)
-0.124 -0.192 -0.193 -0.119 -0.091
(0.315) (0.323) (0.323) (0.315) (0.315)
0.161 0.079 0.081 0.155 0.225
(0.323) (0.335) (0.336) (0.323) (0.329)
-0.035 -0.086 -0.089 -0.045 -0.030
(0.263) (0.271) (0.272) (0.265) (0.265)
-0.315 -0.366 -0.365 -0.345 -0.312
(0.266) (0.269) (0.269) (0.272) (0.273)
-0.317 -0.449 -0.442 -0.302 -0.293
(0.282) (0.290) (0.292) (0.283) (0.283)
Country fixed effects Yes Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes Yes
City-level fixed effects Yes Yes Yes Yes Yes
Observations 41968 41763 41763 41959 41959
Adjusted R-sq 0.060 0.060 0.060 0.060 0.060
Effective tax rate
Agroindustry and food
ObstLabor_InfmlLow
Log_DaysPowerOut
Years of operation in
country (6 - 20 years)
Years of operation in
country (> 20 years)
Foreign owned
Government owned
Table A5 OLS Estimates - Determinants of Net Job Destruction
Sole proprietorship
Partnership
Medium size
Large size
Panel A: Labor Law
Rigdity and
Net job destruction
meanObstLabor
meanLog_ManTimeReg
ObstLabor_TimeReg
meanObstInfml_Low
Textiles and garments
Manufacturing
Service
Retail and wholesale
138
(1) (2) (3) (4) (5) (6)
0.301*** 0.299*** 0.297*** 0.319*** 0.306*** 0.246***
(0.041) (0.034) (0.041) (0.039) (0.044) (0.036)
-0.016 -0.049
(0.037) (0.033)
-0.294***
(0.032)
-0.037 -0.065
(0.039) (0.040)
-0.159***
(0.036)
-0.054 -0.034
(0.040) (0.034)
-0.232***
(0.037)
Panel B: Selected
other covariates (1) (2) (3) (4) (5) (6)
0.148*** 0.149*** 0.148*** 0.151*** 0.144*** 0.131***
(0.021) (0.021) (0.021) (0.022) (0.021) (0.020)
0.115*** 0.097*** 0.117*** 0.106*** 0.112*** 0.135***
(0.025) (0.024) (0.026) (0.025) (0.023) (0.023)
0.214*** 0.173*** 0.216*** 0.200*** 0.212*** 0.234***
(0.035) (0.032) (0.035) (0.036) (0.033) (0.033)
-0.186*** -0.124*** -0.198*** -0.167*** -0.179*** -0.224***
(0.044) (0.043) (0.041) (0.039) (0.042) (0.040)
-0.099** -0.064 -0.113*** -0.110*** -0.099** -0.119***
(0.047) (0.044) (0.042) (0.041) (0.049) (0.046)
0.121** 0.073* 0.120** 0.085* 0.120** 0.073*
(0.048) (0.043) (0.049) (0.046) (0.049) (0.043)
-0.005*** -0.006*** -0.005*** -0.006*** -0.005*** -0.005***
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
0.460*** 0.491*** 0.452*** 0.442*** 0.478*** 0.539***
(0.127) (0.119) (0.127) (0.125) (0.129) (0.126)
Country fixed effects No No No No No No
Year fixed effects Yes Yes Yes Yes Yes Yes
City-level fixed effects Yes Yes Yes Yes Yes Yes
Observations 47624 47624 47624 47624 47624 47624
Adjusted R-sq 0.320 0.393 0.321 0.345 0.323 0.365
Sole proprietorship
Government owned
Large size
Effective tax rate
Textiles and garments
Years of operation in
country (> 20 years)
Table A6-1 Relating Mean Obstacle Labor Back to the Rigidity of Labor Regulation
Indexes, Enforcement and Firm & Industry Characteristics
Panel A:Main
covariates
meanObstLabor
Average obstacle
Years of operation in
country (6 - 20 years)
meanLog_ManTimeReg
std1_indexo
std1_o_PctTimeReg
std1_indexh
std1_h_PctTimeReg
std1_indexf
std1_f_PctTimeReg
139
(1) (2) (3) (4) (5) (6)
0.034** 0.035** 0.034** 0.035** 0.033** 0.034**
(0.016) (0.016) (0.016) (0.016) (0.016) (0.016)
-0.020 -0.026*
(0.016) (0.016)
0.022*
(0.013)
-0.037** -0.040***
(0.016) (0.015)
0.031**
(0.016)
0.014 0.012
(0.016) (0.015)
0.008
(0.012)
Panel B: Selected
other covariates (1) (2) (3) (4) (5) (6)
0.516*** 0.515*** 0.516*** 0.517*** 0.516*** 0.516***
(0.009) (0.009) (0.009) (0.009) (0.009) (0.009)
0.079*** 0.078*** 0.081*** 0.078*** 0.080*** 0.080***
(0.013) (0.013) (0.013) (0.013) (0.013) (0.013)
0.126*** 0.125*** 0.128*** 0.123*** 0.125*** 0.126***
(0.018) (0.018) (0.018) (0.018) (0.018) (0.018)
0.038** 0.038** 0.037** 0.038** 0.042** 0.043**
(0.016) (0.016) (0.016) (0.016) (0.017) (0.017)
-0.094*** -0.095*** -0.106*** -0.107*** -0.098*** -0.097***
(0.024) (0.024) (0.024) (0.024) (0.024) (0.024)
-0.037* -0.037* -0.049*** -0.057*** -0.034* -0.032
(0.020) (0.020) (0.019) (0.019) (0.021) (0.020)
0.187*** 0.186*** 0.186*** 0.186*** 0.191*** 0.192***
(0.021) (0.021) (0.021) (0.021) (0.021) (0.021)
-0.002*** -0.002*** -0.002*** -0.002*** -0.002*** -0.002***
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
0.113** 0.122*** 0.105** 0.113** 0.111** 0.113**
(0.047) (0.046) (0.046) (0.045) (0.047) (0.047)
Country fixed effects No No No No No No
Year fixed effects Yes Yes Yes Yes Yes Yes
City-level fixed effects Yes Yes Yes Yes Yes Yes
Observations 47624 47624 47624 47624 47624 47624
Adjusted R-sq 0.320 0.393 0.321 0.345 0.323 0.365
Government owned
Sole proprietorship
Large size
Effective tax rate
Textiles and garments
Foreign owned
Years of operation in
country (6 - 20 years)
Table A6-2 Relating Obstacle Labor Back to the Rigidity of Labor Regulation Indexes,
Enforcement and Firm & Industry Characteristics
Panel A:Main
covariates
ObstLabor
meanObstInfml_Low
std1_indexo
std1_o_ObstInfml_Low
std1_indexh
std1_h_ObstInfml_Low
std1_indexf
std1_f_ObstInfml_Low
Average obstacle
Years of operation in
country (> 20 years)
140
(1) (2) (3) (4) (5) (6)
-0.356*** -0.353*** -0.353*** -0.352*** -0.352*** -0.348***
(0.039) (0.039) (0.039) (0.039) (0.041) (0.040)
-0.071* -0.086**
(0.042) (0.042)
0.056
(0.035)
-0.082** -0.085**
(0.040) (0.040)
0.037
(0.040)
0.005 -0.002
(0.040) (0.038)
0.026
(0.029)
Panel B: Selected
other covariates (1) (2) (3) (4) (5) (6)
0.077*** 0.077*** 0.079*** 0.080*** 0.079*** 0.078***
(0.015) (0.015) (0.015) (0.015) (0.015) (0.016)
0.098*** 0.097*** 0.103*** 0.099*** 0.099*** 0.100***
(0.019) (0.019) (0.020) (0.021) (0.019) (0.019)
0.187*** 0.186*** 0.190*** 0.185*** 0.183*** 0.187***
(0.028) (0.029) (0.029) (0.030) (0.029) (0.028)
-0.250*** -0.251*** -0.277*** -0.278*** -0.254*** -0.251***
(0.039) (0.039) (0.039) (0.040) (0.038) (0.039)
-0.181*** -0.181*** -0.205*** -0.215*** -0.172*** -0.164***
(0.047) (0.048) (0.042) (0.042) (0.050) (0.048)
-0.146*** -0.140*** -0.189*** -0.201*** -0.158*** -0.145***
(0.046) (0.046) (0.043) (0.043) (0.046) (0.044)
0.084** 0.084** 0.087** 0.088** 0.094*** 0.095***
(0.033) (0.033) (0.035) (0.035) (0.036) (0.035)
0.144*** 0.141*** 0.147*** 0.147*** 0.156*** 0.158***
(0.043) (0.043) (0.045) (0.045) (0.045) (0.045)
-0.003*** -0.003*** -0.004*** -0.004*** -0.003*** -0.003***
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
0.492*** 0.516*** 0.478*** 0.487*** 0.499*** 0.503***
(0.115) (0.114) (0.113) (0.112) (0.116) (0.115)
0.218** 0.243** 0.212** 0.221** 0.232** 0.237**
(0.103) (0.102) (0.103) (0.103) (0.108) (0.107)
-0.302*** -0.300*** -0.301*** -0.301*** -0.309*** -0.306***
(0.092) (0.091) (0.091) (0.089) (0.090) (0.090)
Country fixed effects No No No No No No
Year fixed effects Yes Yes Yes Yes Yes Yes
City-level fixed effects Yes Yes Yes Yes Yes Yes
Observations 47624 47624 47624 47624 47624 47624
Adjusted R-sq 0.320 0.393 0.321 0.345 0.323 0.365
Textiles and garments
Manufacturing
Wholesale and retail
Large size
Effective tax rate
Partnership
MedSize
Years of operation in
country (6 - 20 years)
Table A6-3 Relating Mean Obstacle Labor Back to the Rigidity of Labor Regulation
Indexes, Enforcement and Firm & Industry Characteristics
Panel A:Main
covariates
meanObstLabor
meanObstInfml_Low
std1_indexo
std1_o_ObstInfml_Low
std1_indexh
std1_h_ObstInfml_Low
std1_indexf
std1_f_ObstInfml_Low
Average obstacle
Years of operation in
country (> 20 years)
Government owned
Sole proprietorship
141
Appendix 2
A. List of Paired Health Centers
The following matched pair of health center between treatment and comparison area is the
result of the matching by the country team at project design for survey.
Table A.1. Initial Matched Pairs of Health Centers
Source: IEG, The World Bank.
Note: The catchment area of health centers (Nongboua health center and Sob One health center) which received relocation benefit
because of the Nam Theun 2 Hydropower Project are excluded from the list.
No Health Center District Province Health Center District Province
1 Denvilai Nong Savannakhet Amine Samouay Salavane
2 Xe Keu Thapangthong Savannakhet Asok Samouay Salavane
3 Ban Bo Bolikhanh Boulikhamxay Kengkia Bachieng Champassak
4 Kengchone Xaibouathong Khammouane Sock Boulapha Khammouane
5 Kengmakkheua Saysettha Attapeu Nakong Nong Savanakhet
6 Kuangsy Bachieng Champassak Nasai Phalanxai Savanakhet
7 Ladhor Xepon Savannakhet Kimea Samouay Salavane
8 Manh chi Xepon Savannakhet Dongsavanh Xepon Savanakhet
9 Nakoun Bolikhanh Boulikhamxay Nadou Toumlan Salavane
10 Phameuang Khamkeud Boulikhamxay Nam One Xayabouathong Boulikhamxay
11 Namphao Xaibouathong Khammouane Hai Nyommalad Khammouane
12 Naseuark Phouvong Attapeu Phabang Xepon Savanakhet
13 Natane Nakai Khammouane Nanoi thong Xaibouathong Khammouane
14 Nayom Vilabouly Savannakhet Snod Pathumphone Champassak
15 Nongdeng Soukhouma Champassak Ban That Soukhouma Champassak
16 Panam Mahaxai Khammouane Sobpeng Boulapha Khammouane
17 Phortang Taoi Salavane Phid Nyommalad Khammouane
18 Lak 24 Pathumphone Champassak Khammuane Khamkeud Boulikhamxay
19 Tahouark Taoi Salavane Kokbok Taoi Salavane
Control Intervention
142
B. Data Collection Sampling Framework
Figure B.1. Sampling Procedure
Source: IEG, The World Bank
Notes: The baseline survey covers 2,979 households, 207 villages, and 41 health centers after the baseline survey, including those health
centers that were eventually flooded or received relocation benefit as a result of the NT2 hydropower project. Health centers include the
health center that was flooded or received relocation benefit of hydropower project. ADB = Asian Development Bank; HC = health center;
Vil = village.
The same local survey company collected both baseline and endline datasets. The
enumerators were trained through a field pilot test before the full survey and were blind to
the treatment allocation information. The enumerators visited each household with eligible
children and obtained verbal consent from the parents or guardians at the beginning of the
interview. Data cleaning was executed by the survey firm, the Independent Evaluation
Group, and the World Bank country office staff. The village household roster used to select
the 15 households was compiled by information provided by the health center and the
interview with the village head and the Lao Women’s Union. Despite these efforts to
construct a complete sampling frame, children between 18 and 23 months of age are
somewhat undersampled in both waves.
143
C. Principal Component Analysis
Durable asset variables are used to develop two long-term asset indexes. More specifically,
nine different durable asset variables—availability of toilet, main source of electricity
(electricity or fuel), main source of floor (high class, wood, low class) and main source of
wall (wood, bamboo)—are applied for principal component analysis (PCA). The first two
principal components, which are more than two eigenvalues, are selected as long-term asset
indexes. The first principal component largely consists of (i) access to electricity (rather than
fuel), (ii) main source of floor is wood, (iii) main source of wall is wood. The first principal
component also has large, negative eigenvectors on electricity source from fuel, and low
class source of floor and wall. The second principal component represents high-class material
of floor and wall.
Table C.1. Principal Components (Eigenvectors) for Long-term Asset Index
Variable Component 1 Component 2 Component 3 Unexplained
Toilet 0.256 0.182 -0.047 0.723
Main source of energy: electricity 0.331 0.285 -0.528 0.082
Main source of energy: fuel -0.326 -0.274 0.536 0.092
Main source of floor: high class 0.063 0.562 0.345 0.138
Main source of floor: wood 0.391 -0.369 0.025 0.237
Main source of floor: lower class -0.437 0.144 -0.176 0.323
Main source of wall: high class 0.049 0.551 0.347 0.169
Main source of wall: wood 0.426 -0.184 0.211 0.303
Main source of wall: bamboo -0.432 0.017 -0.340 0.256
Figure C.1. Scree Plot of PCA Eigenvalues for Long-term Asset Index
Source: Authors’ calculation
0
1
2
3
Eigenvalues
0 2 4 6 8 10
Number
144
Short-term (Consumables) Asset Index
Consumption asset variables are used to develop one short-term asset index. There
are 13 variables, which are motorcycle, bicycle, refrigerator, electric rice cooker,
electric fan, two-wheel tractor, boat, fishing net, radio, telephone, mobile phone, and
satellite dish. The first principal component of these nine items loads predominantly
on short-term assets and luxury items such as owning a TV, satellite dish, electric fan,
refrigerator, radio, and mobile phone.
Table C.2. Principal Components (Eigenvectors) for Short-term Asset Index
Variable Component 1 Component 2 Unexplained
Motorcycle 0.277 0.212 0.608
Bicycle 0.157 0.291 0.780
Refrigerator 0.341 -0.232 0.423
Electric rice cooker 0.252 -0.385 0.528
Electric fan 0.369 -0.120 0.392
Two-wheel tractor 0.154 0.480 0.588
Boat 0.152 -0.373 0.713
Fishing net 0.103 0.411 0.728
TV 0.409 0.025 0.274
Radio 0.313 0.129 0.555
Telephone 0.048 -0.305 0.865
Mobile phone 0.305 0.067 0.592
Satellite dish 0.407 0.000 0.282
Figure C.2. Scree Plot of PCA Eigenvalues for Short-term (Consumables) Asset Index
Source: Authors’ calculation
0
1
2
3
4
Eigenvalues
0 5 10 15
Number
145
Household Welfare Shock Index
Similarly, PCA is applied for eight items representing different external shocks
(drought, fire, floods, crop disease, sickness or death of household head, sickness or
death of other household member, resettlement, and robbery) to develop one shock
index. Since the Community Nutrition Project (CNP) is prepared to respond to the
global food crisis, food price increase and decreases are controlled for explicitly as
independent covariates in the regression analysis and are not included in the PCA.
The first principal component of the shock index loads most prominently on items of
drought, floods, crop disease, and sickness or death of household head.
Table C.3. Scree Plot of PCA Eigenvalues for Shock Index
Variable Component 1 Unexplained
Drought 0.458 0.666
Fire 0.267 0.887
Floods 0.422 0.718
Crop disease 0.475 0.642
Sickness or death of household head 0.412 0.731
Sickness or death of other household members 0.352 0.803
Resettlement 0.093 0.986
Robbery 0.118 0.978
Figure C.3. Scree Plot of PCA Eigenvalues for Welfare Shocks
Source: Authors’ Calculations
.8
1
1.2
1.4
1.6
Eigenvalues
0 2 4 6 8
Number
146
D. Definition of Covariates
There are 41 covariates. Each covariate is defined below.
Table D.1. Covariate Definitions
No. Variable Definition
1 Constant Constant term
2 Time Binary variable (1 for endline, 0 for baseline)
3 Intervention Binary variable (1 for treatment area, 0 for comparison area)
4 Interaction Time * Intervention
5 Mother age 1 Mother’s age spline 1 (mother age <= 20)
6 Mother age 2 Mother’s age spline 2 (20 < mother age <= 40)
7 Mother age 3 Mother’s age spline 3 (40 < mother age)
8 Mother weight Mother’s weight (kg)
9 Mother weight sq Mother’s weight square
10
13
Mother educ Mother’s educational background (four dummy variables)
a
14 Grandfather Grandfather living within the household
15 Grandmother Grandmother living within the household
16 Dead child Mother had a child who passed away
17 Child age in months Child age in months
18 Child girl Gender (binary variable: 1 for girl, 0 for boy)
19 Short asset index Short-term asset index: created through PCA
b
20 Long asset index 1 Long-term asset index 1: created through PCA
c
21 Long asset index 2 Long-term asset index 2: created through PCA
d
22-
24
Ethnicity Ethnicity gender: three dummy variables consistent with LSIS
e
25 Time to HC Time to health center from village: logged form
26 First child First child (binary variable: 1 for first child, 0 for others)
27 HH size under 5 Number of children under age 5 within the same household
28 Total birth Number of total births from the same natural mother
29 Shock index Shock index created through PCA
30 Price decrease shock Price decrease shock
31 Price increase shock Price increase shock
32 Urban/rural Village is in urban or rural area
33 Ethnic congruence 1 Ethnic congruence between village and household head
34 Ethnic congruence 2 Ethnic congruence between health center staff and household head
35 HC male propor Proportion of males among nearest health center staff
36 Province Five dummy variables for provinces
f
Note: HC = health center; HH = household; kg = kilograms; LSIS = Lao Social Indicator Survey 2011–12; PCA = principal component
analysis.
a. No education, primary, lower secondary, upper secondary, post-secondary and higher.
b. Consumption type assets: motorcycle, bicycle, refrigerator, electric rice cooker, electric fan, 2 wheel tractor, boat, fishing net, radio,
telephone, mobile phone, satellite dish
c. Durable assets: toilet, main source of electricity (electricity or fuel), main source of floor (high class, wood, low class), main source of wall
(wood, bamboo)
d. External shocks: drought, fire, floods, crop disease, ill or death of household head, ill or death of other household member, resettlement,
robbery
e. Lao-Tai, Mon-Khmer, Hmong-Mien, Others (Tibetan included in others due to negligible sample size)
f. Bolikhamxay, Khammaun, Savanhnakhet, Saravan, Champasak, Attapue
147
E. Parallel Trending at Pre-treatment Period
Figure E.1. Utilization of Health Services by Age (0-23 mo)
Panel a. PDO1: ANC assisted by health staff Panel b. PDO2: Delivery at health facility
Source: IEG, The World Bank.
Note: ANC = antenatal care.
Panel c. PDO3: Receive DPT at least three times
Source: IEG, The World Bank.
Panel d. PDO4: At least one routine checkup
Note: DPT = diphtheria, pertussis, and tetanus.
Panel e. PDO5: Breastfeeding within one hour of birth
Source: IEG, The World Bank.
Panel f. PDO6: ORS with diarrhea
Source: IEG, The World Bank.
Note: ORS = oral rehydration solutions.
.2 .3 .4 .5 .6
DPT at least 3 times
10 15 20 25
age in month at baseline
Treatment Control
0
.05
.1
.15
.2
At least one routine check-up
0 5 10 15 20 25
age in month at baseline
Treatment Control
148
Panel g. Height-for-age z-score
Panel h. Stunting
Source: IEG, The World Bank. Source: IEG, The World Bank.
-2.5
-2
-1.5
-1 -.5
Height-for-age z-score
0 5 10 15 20 25
age in month at baseline
Treatment Control
.2 .3 .4 .5 .6 .7
Height-for-age z-score < -2
0 5 10 15 20 25
age in month at baseline
Treatment Control
149
F. Propensity Score Matching
Figure F.1. Map of Treatment and Comparison Villages and Health Centers
Source: IEG, The World Bank.
Note: comparison = comparison area; HC = health center; treat = treatment area; Vil = village.
150
Figure F.2. Common Support
Source: IEG, The World Bank.
Figure F.3. Bias reduction
Source: IEG, The World Bank.
151
Note: hc_cat_pop_hc = population in health center catchment area; hc_delivery_inst = proportion of institutional deliveries in health center
catchment area; hc_grid = health center having access to electrical grid; hc_poor_cond = health center buidling in poor condition;
hc_postnatal_1wk = proportion of receiving postnatal visits with 1 week in health center catchment area; hc_staff_proper = number of
health center staff excling volunteers; pct_ECLao_Tai = percentage of Lao/Tai population in health center catchmen area;
pct_ECMon_Khmer = percentage of Mon/Khmer population in health center catchmen area; meanaccs_hcs = mean travel time to nearest
health center; meanelev_hccatch = mean elevation in health center catchment area; rangeelev_hccatch = range of elevation in health
center catchment area.
Table F.1. Bias Reduction
Mean
Bias
reduction
Variable Treat Comp % bias t-stat p > |t|
Ethnicity (Lao Thai)
Unmatched 38.7 43.6 -13.2 -0.41 0.69
Matched 32.6 32.1 1.2 90.8 0.03 0.98
Ethnicity (Mon Khmer)
Unmatched 55.9 52.8 7.9 0.24 0.81
Matched 67.4 66.8 1.5 80.8 0.04 0.97
Health center catchment
population
Unmatched 3,625 3,710 -4.7 -0.14 0.89
Matched 3,399 3,345 3.0 36.6 0.07 0.94
Range of elevation in health
center catchment area
Unmatched 799.1 673.7 30.8 0.95 0.35
Matched 668.7 741.7 -17.9 41.8 -0.43 0.68
Mean elevation of health center
Unmatched 400.3 405.4 -2.5 -0.08 0.94
Matched 369.7 410.7 -20.1 -699.6 -0.50 0.62
Mean access time to health
center
Unmatched 111.7 98.7 19.1 0.59 0.56
Matched 100.8 103.2 -3.4 82 -0.08 0.94
Health center connected to grid
Unmatched 0.47 0.53 -10.3 -0.32 0.75
Matched 0.33 0.31 4.1 60.4 0.10 0.92
Number of proper health staff at
health center
Unmatched 2.3 2.1 32.9 1.01 0.32
Matched 2.0 2.0 0.0 100 0.00 1.00
Proportion of delivery at health
facility (health center level)
Unmatched 0.19 0.11 57.9 1.78 0.08
Matched 0.10 0.10 -0.3 99.5 -0.01 0.99
Proportion of post-natal visit
after 1 week of birth
Unmatched 0.04 0.03 22.0 0.68 0.50
Matched 0.05 0.04 19.0 13.6 0.40 0.69
Proportion of health center in
poor condition
Unmatched 0.21 0.26 -12.1 -0.37 0.71
Matched 0.25 0.25 0.0 100 0.00 1.00
Mean Bias Unmatched - - 19.4 - - -
Matched - - 6.4 - - -
Median Bias Unmatched - - 13.2 - - -
Matched - - 3.0 - - -
Source: IEG, The World Bank.
Notes: comp = comparison area; treat = treatment area. The t-statistics and p-values are derived from balancing t-test for unmatched and
matched samples.
152
G. Baseline Balance Check
Table G.1. Baseline Balance on Pretreatment Variables
Full Sample Matched Sample
Variable
Treatment
(n=1,428)
Comparison
(n=1,417)
Differ
ence
p-
value
Treatment
(n=890)
Comparison
(n=1,107) Difference
p-
value
Child
Age in months 9.80 10.00 -0.20 0.53 10.03 10.29 -0.26 0.53
Gender (girl = 1) 0.49 0.50 -0.01 0.70 0.49 0.51 -0.02 0.57
First child 0.26 0.27 -0.01 0.79 0.22 0.28 -0.06 0.04 **
Mother's education
No education 0.48 0.54 -0.06 0.02 ** 0.55 0.60 -0.05 0.11
Primary 0.30 0.34 -0.04 0.06 * 0.29 0.31 -0.01 0.69
Lower secondary 0.10 0.04 0.06 0.00 *** 0.05 0.03 0.02 0.09 *
Upper secondary 0.07 0.04 0.03 0.01 *** 0.05 0.04 0.01 0.64
Post secondary or higher 0.04 0.03 0.01 0.28 0.05 0.02 0.03 0.01 ***
Mother's age 26.5 26.0 0.50 0.08 * 26.8 26.2 0.62 0.09 *
Household
Long asset index 1 0.02 0.01 0.00 0.98 -0.16 -0.19 0.04 0.72
Long asset index 2 0.26 -0.02 0.28 0.00 *** -0.25 -0.17 -0.08 0.36
Short asset index -0.14 0.12 -0.26 0.02 ** 0.46 0.52 -0.06 0.58
Shock index -0.20 0.05 -0.25 0.00 *** -0.24 -0.12 -0.12 0.17
Price increase 0.01 0.01 0.00 0.71 0.02 0.01 0.00 0.84
Price decrease 0.01 0.04 -0.03 0.00 *** 0.02 0.03 -0.01 0.13
Grandfather 0.30 0.27 0.02 0.25 0.23 0.28 -0.05 0.07 *
Grandmother 0.37 0.36 0.01 0.70 0.31 0.36 -0.06 0.04 **
Household size under five years old 1.62 1.57 0.05 0.11 1.58 1.55 0.03 0.45
Total births 3.42 3.24 0.19 0.09 * 3.52 3.27 0.26 0.06 *
Ethnicity
Lao Tai 0.51 0.43 0.08 0.00 *** 0.58 0.36 0.23 0.00 ***
Monkhmer 0.46 0.50 -0.04 0.10 * 0.41 0.61 -0.20 0.00 ***
Hmon Mien 0.03 0.02 0.00 0.78 0.00 0.01 -0.01 0.00 ***
Other 0.01 0.04 -0.04 0.00 *** 0.00 0.01 -0.01 0.00 ***
Village
Urban/rural (urban = 1) 0.18 0.17 0.01 0.49 0.12 0.12 0.01 0.75
Time to health center (log) -0.56 -0.49 -0.07 0.21 -0.49 -0.30 -0.19 0.01 **
PDO
ANC attended by health staff 0.43 0.26 0.17 0.00 *** 0.38 0.25 0.14 0.00 **
Institutional delivery 0.20 0.11 0.09 0.00 *** 0.11 0.10 0.01 0.44
DPT at least three times 0.45 0.41 0.03 0.36 0.50 0.38 0.12 0.01 **
Any growth checkup 0.12 0.05 0.07 0.00 *** 0.15 0.05 0.11 0.00 ***
Breastfeeding within one hour 0.41 0.42 -0.01 0.66 0.38 0.46 -0.08 0.01 ***
Received ORS during diarrhea 0.64 0.64 0.00 1.00 0.65 0.65 0.00 1.00
Source: IEG, The World Bank.
Notes: All datasets restrict the sample to the newborns less than 24 months who are children of mothers aged between 15 and 49 years
old. The sample sizes are for child age in months; depending on the missing values of the other variables, sample sizes could change.
DPT = diphtheria, pertussis, and tetanus; ORS = oral rehydration solutions.
Significance level: * = 10 percent, ** = 5 percent, *** = 1 percent.
153
H. Heckman Selection Model
Table H.1. Heckman and Non-Heckman Selection Model Results on ORS
Full Sample Matched Sample
0-11 mo (N=465) 0-23 mo (N=933) 0-11 mo (N=354) 0-23 mo (N=731)
First Stage Heckman LPM First Stage Heckman LPM First Stage Heckman LPM First Stage Heckman LPM
Coefficient 4.123 0.092 0.101 4.014 0.113 0.115 3.916 0.077 0.079 3.975 0.118 0.121
Standard Error (0.264) *** (0.101) (0.104) (0.128) *** (0.077) (0.079) (0.315) *** (0.118) (0.126) (0.177) *** (0.106) (0.111)
Wald Test - 0.186 - - 0.130 - - 0.848 - - 0.381 -
First
Stage
Heckman
logit ME
Logit
ME First Stage
Heckman
logit ME
Logit
ME First Stage
Heckman
logit ME
Logit
ME First Stage
Heckman
logit ME Logit ME
Coefficient 4.128 0.105 0.118 4.015 0.127 0.131 3.921 0.122 0.126 3.977 0.138 0.140
Standard Error (0.265) *** (0.115) (0.114) (0.129) *** (0.085) (0.084) (0.316) *** (0.159) (0.159) (0.177) *** (0.119) (0.119)
First
Stage Heckprob Probit First Stage Heckprob Probit First Stage Heckprob Probit First Stage Heckprob Probit
Coefficient 4.123 0.258 0.300 4.013 0.318 0.335 3.916 0.273 0.283 3.975 0.370 0.380
Standard Error (0.264) *** (0.272) (0.276) (0.129) *** (0.213) (0.215) (0.315) *** (0.362) (0.361) (0.177) *** (0.301) (0.307)
Wald Test - 0.200 - - 0.161 - - 0.860 - - 0.479 -
Source: IEG, The World Bank.
Notes: The column "First stage" reports the coefficient of the village level incidence of diarrhea on the individual incidence of diarrhea. The Heckman/LPM/Logit marginal effect
(ME)/Probit column reports the results of the interaction term between time and intervention. The Wald test row reports the probability greater than chi square value under the
null hypothesis that correlation is equal to zero. Heckman logit with marginal effect is estimated separately in first stage with probit, and second stage with logit with marginal
effect. Heckman = Heckman’s sample selection model; Heckprob = Heckman model using probit in the second step; LPM = linear probability model; ME = marginal effect; ORS
= oral rehydration solutions.
Significance level: * = 10 percent, ** = 5 percent, *** = 1 percent.
Table H.2. Heckman and Non-Heckman Selection Model Results on ORS (including
government recommended fluid)
Full Sample Matched Sample
0-11 mo (N=465) 0-23 mo (N=933) 0-11 mo (N=354) 0-23 mo (N=731)
First Stage Heckman LPM First Stage Heckman LPM First Stage Heckman LPM First Stage Heckman LPM
Coefficient 4.120 -0.020 -0.011 4.013 0.040 0.042 3.915 -0.027 -0.025 3.975 0.008 0.010
Standard Error (0.264) *** (0.114) (0.117) (0.128) *** (0.083) (0.085) (0.316) *** (0.138) (0.146) (0.177) *** (0.110) (0.114)
Wald Test - 0.176 - - 0.165 - - 0.874 - - 0.460 -
First
Stage
Heckman
logit ME
Logit
ME First Stage
Heckman
logit ME
Logit
ME First Stage
Heckman
logit ME
Logit
ME First Stage
Heckman
logit ME Logit ME
Coefficient 4.128 -0.019 -0.006 4.015 0.044 0.048 3.921 -0.023 -0.020 3.977 0.019 0.020
Standard Error (0.265) *** (0.119) (0.118) (0.129) *** (0.089) (0.088) (0.316) *** (0.158) (0.156) (0.177) *** (0.113) (0.113)
First
Stage Heckprob Probit First Stage Heckprob Probit First Stage Heckprob Probit First Stage Heckprob Probit
Coefficient 4.119 -0.047 -0.011 4.012 0.118 0.131 3.915 -0.053 -0.045 3.975 0.057 0.063
Standard Error (0.265) *** (0.306) (0.308) (0.129) *** (0.243) (0.245) (0.316) *** (0.412) (0.409) (0.177) *** (0.329) (0.332)
Wald Test - 0.188 - - 0.189 - - 0.870 - - 0.631 -
Source: IEG, The World Bank.
Notes: The column "First stage" reports the coefficient of the village level incidence of diarrhea on the individual incidence of diarrhea. The Heckman/LPM/Logit marginal effect
(ME)/Probit column reports the results of the interaction term between time and intervention. The Wald test row reports the probability greater than chi square value under the
null hypothesis that correlation is equal to zero. Heckman logit with marginal effect is estimated separately in first stage with probit, and second stage with logit with marginal
effect. Heckman = Heckman’s sample selection model; Heckprob = Heckman model using probit in the second step; LPM = linear probability model; ME = marginal effect; ORS
= oral rehydration solutions.
Significance level: * = 10 percent, ** = 5 percent, *** = 1 percent.
154
Appendix 3
(1) (2) (3) (4) (5) (6)
Gov Index - CC -0.871***
(0.154)
CC_Food 0.418
(0.260)
CC_TxlGmt 0.829***
(0.228)
CC_Manu 0.357*
(0.202)
CC_Service 0.153
(0.189)
CC_Retail 0.162
(0.210)
Gov Index - GE -0.771***
(0.162)
GE_Food 0.059
(0.318)
GE_TxlGmt 0.552*
(0.331)
GE_Manu 0.131
(0.213)
GE_Service 0.153
(0.214)
GE_Retail 0.045
(0.243)
Gov Index - PS -0.391***
(0.125)
PS_Food -0.013
(0.204)
PS_TxlGmt 0.093
(0.180)
PS_Manu -0.015
(0.174)
PS_Service 0.017
(0.160)
PS_Retail 0.051
(0.180)
Table 4-3 Effects of Governmance Indexes on Firm's Perceptions of Corruption (ordered logit),
by Firm's Industry
Key explanatory
variables
Dependent variable: ObstCorrupt (0 to 4; estimated using ordered logit)
155
(1) (2) (3) (4) (5) (6)
Gov Index - RQ -0.548***
(0.164)
RQ_Food 0.083
(0.259)
RQ_TxlGmt 0.570**
(0.254)
RQ_Manu 0.121
(0.213)
RQ_Service 0.028
(0.207)
RQ_Retail 0.120
(0.235)
Gov Index - RL -0.734***
(0.152)
RL_Food 0.507**
(0.239)
RL_TxlGmt 0.812***
(0.261)
RL_Manu 0.317
(0.203)
RL_Service 0.079
(0.192)
RL_Retail 0.113
(0.228)
Gov Index - VA -0.413***
(0.139)
VA_Food 0.205
(0.236)
VA_TxlGmt 0.601***
(0.161)
VA_Manu 0.386**
(0.163)
VA_Service 0.139
(0.160)
VA_Retail 0.096
(0.181)
Table 4-3 (Cont'd)
Dependent variable: ObstCorrupt (0 to 4; estimated using ordered logit)
156
(1) (2) (3) (4) (5) (6)
Gov Index - CC -0.597***
(0.083)
CC × % Sale Exported -0.002
(0.002)
Gov Index - GE -0.633***
(0.096)
GE × % Sale Exported -0.000
(0.002)
Gov Index - PS -0.352***
(0.067)
PS × % Sale Exported -0.001
(0.001)
Gov Index - RQ -0.401***
(0.094)
RQ × % Sale Exported -0.002
(0.001)
Gov Index - RL -0.515***
(0.094)
RL × % Sale Exported -0.002
(0.001)
Gov Index - VA -0.176**
(0.075)
VA × % Sale Exported -0.002
(0.001)
Firm characteristics Yes Yes Yes Yes Yes Yes
Industry fixed effects Yes Yes Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes Yes Yes
N 100457 100457 100457 100457 100457 100457
Table 4-4 Effects of Governmance Indexes on Firm's Perceptions of Corruption (ordered logit),
by Export Share
Key explanatory
variables
Dependent variable: ObstCorrupt (0 to 4; estimated using ordered logit)
Notes: Standard errors are clustered at each level of country × industry × year. Firm characteristics include
all variables in Panel B of Table 1 except country-level indexes, plus an average of the ratings on other
obstacles to its business by a firm. CC = control of corruption; GE = government effectiveness; PS =
political stability (and absense of violence); RQ = regulatory quality; RL = rule of law
157
(1) (2) (3) (4) (5) (6)
Gov Index - CC -0.565***
(0.077)
CC × QualCert -0.263**
(0.109)
Gov Index - GE -0.599***
(0.093)
GE × QualCert -0.219**
(0.097)
Gov Index - PS -0.357***
(0.062)
PS × QualCert -0.062
(0.080)
Gov Index - RQ -0.361***
(0.088)
RQ × QualCert -0.308***
(0.087)
Gov Index - RL -0.481***
(0.090)
RL × QualCert -0.287***
(0.096)
Gov Index - VA -0.177**
(0.074)
VA × QualCert -0.105
(0.100)
Firm characteristics Yes Yes Yes Yes Yes Yes
Industry fixed effects Yes Yes Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes Yes Yes
N 99253 99253 99253 99253 99253 99253
Table 4-5 Effects of Governmance Indexes on Firm's Perceptions of Corruption (ordered logit),
by by Having Quality Certificate or Not
Key explanatory
variables
Dependent variable: ObstCorrupt (0 to 4; estimated using ordered logit)
Notes: Standard errors are clustered at each level of country × industry × year. Firm characteristics include
all variables in Panel B of Table 1 except country-level indexes, plus an average of the ratings on other
obstacles to its business by a firm. CC = control of corruption; GE = government effectiveness; PS =
political stability (and absense of violence); RQ = regulatory quality; RL = rule of law
158
(1) (2) (3) (4) (5) (6)
Gov Index - CC -0.406***
(0.129)
CC × Private Credit -0.272*
(0.145)
Gov Index - GE -0.370***
(0.139)
GE × Private Credit -0.205
(0.133)
Gov Index - PS -0.280***
(0.102)
PS × Private Credit -0.008
(0.092)
Gov Index - RQ -0.087
(0.120)
RQ × Private Credit -0.249**
(0.121)
Gov Index - RL -0.389***
(0.130)
RL × Private Credit -0.119
(0.127)
Gov Index - VA -0.118
(0.103)
VA × Private Credit -0.278**
(0.122)
Firm characteristics Yes Yes Yes Yes Yes Yes
Industry fixed effects Yes Yes Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes Yes Yes
N 26269 26269 26269 26269 26269 26269
Table 4-6 Effects of Governmance Indexes on Firm's Perceptions of Corruption (ordered logit),
by Source of Credit
Key explanatory
variables
Dependent variable: ObstCorrupt (0 to 4; estimated using ordered logit)
Notes: Standard errors are clustered at each level of country × industry × year. Firm characteristics include
all variables in Panel B of Table 1 except country-level indexes, plus an average of the ratings on other
obstacles to its business by a firm. CC = control of corruption; GE = government effectiveness; PS =
political stability (and absense of violence); RQ = regulatory quality; RL = rule of law; Private credit =
dummy for having secured a line of credit from private banks during the last two years
159
(1) (2) (3) (4) (5) (6)
Any related activity 0.145*** 0.144*** 0.152*** 0.147*** 0.147*** 0.146***
(0.013) (0.013) (0.014) (0.013) (0.014) (0.013)
Gov Index - CC -0.129***
(0.032)
CC × AgeGrp2 -0.003
(0.025)
CC × AgeGrp3 0.046*
(0.026)
Gov Index - GE -0.125***
(0.024)
GE × AgeGrp2 0.004
(0.021)
GE × AgeGrp3 0.051**
(0.022)
Gov Index - PS -0.057***
(0.016)
PS × AgeGrp2 0.007
(0.015)
PS × AgeGrp3 -0.002
(0.017)
Gov Index - RQ -0.110***
(0.023)
RQ × AgeGrp2 -0.000
(0.019)
RQ × AgeGrp3 0.031
(0.019)
Gov Index - RL -0.120***
(0.026)
RL × AgeGrp2 0.002
(0.022)
RL × AgeGrp3 0.042*
(0.024)
Gov Index - VA -0.080***
(0.016)
VA × AgeGrp2 -0.009
(0.012)
VA × AgeGrp3 -0.006
(0.012)
Firm characteristics Yes Yes Yes Yes Yes Yes
Industry fixed effects Yes Yes Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes Yes Yes
N 103547 103547 103547 103547 103547 103547
Table 5-1 Effects of Governmance Indexes on Firm's Total Occasions of Giving Gifts, by Firm's
Years of Operation in the Country
Key explanatory
variables
Dependent variable: Number of gift-requested occasions (from 0 to 6)
Notes: Standard errors are clustered at each level of country × industry × year. Firm characteristics include
all variables in Panel B of Table 1 except country-level indexes. CC = control of corruption; GE =
government effectiveness; PS = political stability (and absense of violence); RQ = regulatory quality; RL =
rule of law; AgeGrp2 = dummy for 10-19 years; AgeGrp3 = dummy for 20 years and above
160
(1) (2) (3) (4) (5) (6)
Any related activity 0.146*** 0.144*** 0.153*** 0.147*** 0.146*** 0.146***
(0.013) (0.013) (0.014) (0.014) (0.013) (0.013)
Gov Index - CC -0.097***
(0.015)
CC × Medium Size -0.011
(0.017)
CC × Large Size -0.146***
(0.044)
Gov Index - GE -0.090***
(0.011)
GE × Medium Size -0.033**
(0.014)
GE × Large Size -0.098**
(0.049)
Gov Index - PS -0.044***
(0.009)
PS × Medium Size -0.010
(0.011)
PS × Large Size -0.073**
(0.034)
Gov Index - RQ -0.085***
(0.011)
RQ × Medium Size -0.023*
(0.012)
RQ × Large Size -0.112***
(0.039)
Gov Index - RL -0.086***
(0.011)
RL × Medium Size -0.040***
(0.015)
RL × Large Size -0.112***
(0.037)
Gov Index - VA -0.084***
(0.012)
VA × Medium Size 0.006
(0.010)
VA × Large Size -0.047***
(0.015)
Firm characteristics Yes Yes Yes Yes Yes Yes
Industry fixed effects Yes Yes Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes Yes Yes
N 103547 103547 103547 103547 103547 103547
Table 5-2 Effects of Governmance Indexes on Firm's Total Occasions of Giving Gifts, by Size of
Firm
Key explanatory
variables
Dependent variable: Number of gift-requested occasions (from 0 to 6)
Notes: Standard errors are clustered at each level of country × industry × year. Firm characteristics include
all variables in Panel B of Table 1 except country-level indexes. CC = control of corruption; GE =
government effectiveness; PS = political stability (and absense of violence); RQ = regulatory quality; RL =
rule of law
161
(1) (2) (3) (4) (5) (6)
Any related activity 0.146*** 0.144*** 0.153*** 0.147*** 0.148*** 0.146***
(0.013) (0.013) (0.014) (0.013) (0.014) (0.013)
Gov Index - CC -0.114***
(0.021)
CC_Food -0.032
(0.047)
CC_TxlGmt -0.123
(0.090)
CC_Manu -0.025
(0.035)
CC_Service 0.032
(0.031)
CC_Retail 0.012
(0.036)
Gov Index - GE -0.120***
(0.023)
GE_Food -0.014
(0.048)
GE_TxlGmt -0.075
(0.089)
GE_Manu 0.000
(0.035)
GE_Service 0.016
(0.033)
GE_Retail 0.039
(0.029)
Gov Index - PS -0.034*
(0.019)
PS_Food -0.020
(0.033)
PS_TxlGmt -0.065
(0.053)
PS_Manu -0.028
(0.027)
PS_Service -0.004
(0.024)
PS_Retail -0.018
(0.022)
Table 5-3 Effects of Governmance Indexes on Firm's Total Occasions of Giving Gifts, by Firm's
Industry
Key explanatory
variables
Dependent variable: Number of gift-requested occasions (from 0 to 6)
162
(1) (2) (3) (4) (5) (6)
Gov Index - RQ -0.103***
(0.018)
RQ_Food -0.006
(0.036)
RQ_TxlGmt -0.072
(0.076)
RQ_Manu -0.024
(0.030)
RQ_Service 0.024
(0.024)
RQ_Retail 0.019
(0.025)
Gov Index - RL -0.107***
(0.016)
RL_Food -0.040
(0.037)
RL_TxlGmt -0.092
(0.085)
RL_Manu -0.007
(0.029)
RL_Service 0.019
(0.025)
RL_Retail 0.008
(0.026)
Gov Index - VA -0.070***
(0.013)
VA_Food -0.020
(0.032)
VA_TxlGmt 0.001
(0.021)
VA_Manu -0.023
(0.016)
VA_Service -0.007
(0.015)
VA_Retail -0.028*
(0.016)
Table 5-3 (Cont'd)
Dependent variable: ObstCorrupt (0 to 4; estimated using ordered logit)
163
(1) (2) (3) (4) (5) (6)
Any related activity 0.145*** 0.143*** 0.152*** 0.146*** 0.147*** 0.146***
(0.013) (0.013) (0.014) (0.013) (0.014) (0.013)
Gov Index - CC -0.104***
(0.017)
CC × % Sale Exported -0.001
(0.001)
Gov Index - GE -0.101***
(0.012)
GE × % Sale Exported -0.001
(0.001)
Gov Index - PS -0.047***
(0.009)
PS × % Sale Exported -0.001
(0.001)
Gov Index - RQ -0.091***
(0.012)
RQ × % Sale Exported -0.001*
(0.001)
Gov Index - RL -0.098***
(0.011)
RL × % Sale Exported -0.001*
(0.001)
Gov Index - VA -0.083***
(0.013)
VA × % Sale Exported -0.000***
(0.000)
Firm characteristics Yes Yes Yes Yes Yes Yes
Industry fixed effects Yes Yes Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes Yes Yes
N 103547 103547 103547 103547 103547 103547
Notes: Standard errors are clustered at each level of country × industry × year. Firm characteristics include
all variables in Panel B of Table 1 except country-level indexes, plus an average of the ratings on other
obstacles to its business by a firm. CC = control of corruption; GE = government effectiveness; PS =
political stability (and absense of violence); RQ = regulatory quality; RL = rule of law
Table 5-4 Effects of Governmance Indexes on Firm's Total Occasions of Giving Gifts, by Export
Share
Key explanatory
variables
Dependent variable: Number of gift-requested occasions (from 0 to 6)
164
(1) (2) (3) (4) (5) (6)
Any related activity 0.145*** 0.143*** 0.152*** 0.146*** 0.147*** 0.146***
(0.013) (0.013) (0.014) (0.013) (0.014) (0.013)
Gov Index - CC -0.104***
(0.017)
CC × % Sale Exported -0.001
(0.001)
Gov Index - GE -0.101***
(0.012)
GE × % Sale Exported -0.001
(0.001)
Gov Index - PS -0.047***
(0.009)
PS × % Sale Exported -0.001
(0.001)
Gov Index - RQ -0.091***
(0.012)
RQ × % Sale Exported -0.001*
(0.001)
Gov Index - RL -0.098***
(0.011)
RL × % Sale Exported -0.001*
(0.001)
Gov Index - VA -0.083***
(0.013)
VA × % Sale Exported -0.000***
(0.000)
Firm characteristics Yes Yes Yes Yes Yes Yes
Industry fixed effects Yes Yes Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes Yes Yes
N 103547 103547 103547 103547 103547 103547
Table 5-5 Effects of Governmance Indexes on Firm's Total Occasions of Giving Gifts, by Having
Quality Certificate or Not
Key explanatory
variables
Dependent variable: Number of gift-requested occasions (from 0 to 6)
Notes: Standard errors are clustered at each level of country × industry × year. Firm characteristics include
all variables in Panel B of Table 1 except country-level indexes, plus an average of the ratings on other
obstacles to its business by a firm. CC = control of corruption; GE = government effectiveness; PS =
political stability (and absense of violence); RQ = regulatory quality; RL = rule of law
165
(1) (2) (3) (4) (5) (6)
Any related activity 0.124*** 0.122*** 0.127*** 0.123*** 0.124*** 0.126***
(0.012) (0.012) (0.012) (0.012) (0.012) (0.012)
Gov Index - CC -0.068***
(0.019)
CC × Private Credit -0.064***
(0.023)
Gov Index - GE -0.079***
(0.017)
GE × Private Credit -0.042*
(0.025)
Gov Index - PS -0.011
(0.012)
PS × Private Credit 0.003
(0.012)
Gov Index - RQ -0.078***
(0.018)
RQ × Private Credit -0.043**
(0.021)
Gov Index - RL -0.087***
(0.020)
RL × Private Credit -0.020
(0.026)
Gov Index - VA -0.059***
(0.019)
VA × Private Credit -0.075***
(0.025)
Firm characteristics Yes Yes Yes Yes Yes Yes
Industry fixed effects Yes Yes Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes Yes Yes
N 27020 27020 27020 27020 27020 27020
Table 5-6 Effects of Governmance Indexes on Firm's Total Occasions of Giving Gifts, by Source
of Credit
Key explanatory
variables
Dependent variable: Number of gift-requested occasions (from 0 to 6)
Notes: Standard errors are clustered at each level of country × industry × year. Firm characteristics include
all variables in Panel B of Table 1 except country-level indexes, plus an average of the ratings on other
obstacles to its business by a firm. CC = control of corruption; GE = government effectiveness; PS =
political stability (and absense of violence); RQ = regulatory quality; RL = rule of law; Private credit =
dummy for having secured a line of credit from private banks during the last two years
166
Table A1: List of Countries
Country (region) # Firms Percent Country (region) # Firms Percent Country (region) # Firms Percent
Afghanistan 945 0.8% Gabon 179 0.2% Nigeria 4567 3.9%
Albania 664 0.6% Gambia 174 0.1% Pakistan 2182 1.9%
Angola 785 0.7% Georgia 733 0.6% Panama 969 0.8%
Antigua and Barbuda 151 0.1% Ghana 1,214 1.0% Paraguay 974 0.8%
Argentina 2,117 1.8% Grenada 153 0.1% Peru 1632 1.4%
Armenia 734 0.6% Guatemala 1,112 0.9% Philippines 1326 1.1%
Azerbaijan 770 0.7% Guinea 223 0.2% Poland 997 0.9%
Bahamas 150 0.1% Guinea Bissau 159 0.1% Romania 1081 0.9%
Bangladesh 2,946 2.5% Guyana 165 0.1% Russia 5224 4.5%
Barbados 150 0.1% Honduras 796 0.7% Rwanda 453 0.4%
Belarus 633 0.5% Hungary 601 0.5% Samoa 109 0.1%
Belize 150 0.1% India 9,281 7.9% Senegal 1107 0.9%
Benin 150 0.1% Indonesia 1,444 1.2% Serbia 748 0.6%
Bhutan 250 0.2% Iraq 756 0.6% Sierra Leone 150 0.1%
Bolivia 975 0.8% Israel 483 0.4% Slovakia 543 0.5%
Bosnia and Herzegovina 721 0.6% Jamaica 376 0.3% Slovenia 546 0.5%
Botswana 610 0.5% Jordan 573 0.5% South Africa 937 0.8%
Brazil 1,802 1.5% Kazakhstan 1,144 1.0% South Sudan 738 0.6%
Bulgaria 1,596 1.4% Kenya 1,438 1.2% Sri Lanka 610 0.5%
Burkina Faso 394 0.3% Kosovo 472 0.4% St Kitts and Nevis 150 0.1%
Burundi 427 0.4% Kyrgyz Rep 505 0.4% St Lucia 150 0.1%
Cambodia 472 0.4% Lao PDR 630 0.5% St Vincent and the Grenadines 154 0.1%
Cameroon 363 0.3% Latvia 607 0.5% Sudan 662 0.6%
Cape Verde 156 0.1% Lebanon 561 0.5% Suriname 152 0.1%
Central Africa 150 0.1% Lesotho 151 0.1% Swaziland 307 0.3%
Chad 150 0.1% Liberia 150 0.1% Sweden 600 0.5%
Chile 2,050 1.8% Lithuania 546 0.5% Tajikistan 719 0.6%
China 2,700 2.3% Macedonia 726 0.6% Tanzania 1232 1.1%
Colombia 1,942 1.7% Madagascar 977 0.8% Timor Leste 150 0.1%
Congo DR 1,228 1.0% Malawi 673 0.6% Togo 155 0.1%
Congo Rep 151 0.1% Mali 850 0.7% Tonga 150 0.1%
Costa Rica 538 0.5% Mauritania 387 0.3% Trinidad and Tobago 370 0.3%
Côte d'Ivoire 526 0.4% Mauritius 398 0.3% Tunisia 592 0.5%
Croatia 993 0.8% Mexico 2,960 2.5% Turkey 2496 2.1%
Czech Rep 504 0.4% Micronesia 68 0.1% Uganda 1325 1.1%
Djibouti 266 0.2% Moldova 723 0.6% Ukraine 1853 1.6%
Dominica 150 0.1% Mongolia 722 0.6% Uruguay 1228 1.0%
Dominican Rep 360 0.3% Montenegro 266 0.2% Uzbekistan 756 0.6%
Ecuador 1,024 0.9% Morocco 407 0.3% Vanuatu 128 0.1%
Egypt 2,897 2.5% Mozambique 479 0.4% Venezuela 820 0.7%
El Salvador 1,053 0.9% Myanmar 632 0.5% West Bank and Gaza 434 0.4%
Eritrea 179 0.2% Namibia 909 0.8% Vietnam 1053 0.9%
Estonia 546 0.5% Nepal 850 0.7% Yemen 830 0.7%
Ethiopia 644 0.5% Nicaragua 814 0.7% Zambia 1204 1.0%
Fiji 164 0.1% Niger 150 0.1% Zimbabwe 599 0.5%
Total 117,105 100%
167
(1) (2) (3) (4) (5) (6)
Electricity connection 0.098*** 0.097*** 0.098*** 0.098*** 0.097*** 0.098***
(0.014) (0.014) (0.014) (0.014) (0.014) (0.014)
Gov Index - CC -0.012***
(0.002)
Gov Index - GE -0.012***
(0.002)
Gov Index - PS -0.006***
(0.001)
Gov Index - RQ -0.011***
(0.001)
Gov Index - RL -0.013***
(0.002)
Gov Index - VA -0.008***
(0.002)
(1) (2) (3) (4) (5) (6)
Telephone connection 0.033*** 0.033*** 0.033*** 0.033*** 0.033*** 0.033***
(0.006) (0.006) (0.006) (0.006) (0.006) (0.006)
Gov Index - CC -0.004**
(0.002)
Gov Index - GE -0.003**
(0.001)
Gov Index - PS -0.001
(0.001)
Gov Index - RQ -0.004***
(0.001)
Gov Index - RL -0.003**
(0.001)
Gov Index - VA -0.002**
(0.001)
Table A2 Effects of Governmance Indexes on Firm's Likelihood of Giving Gifts
Panel 1
Dependent variable: Gift given for electricity connection (1 if yes, 0 otherwise)
Panel 2
Dependent variable: Gift given for telephone connection (1 if yes, 0 otherwise)
168
(1) (2) (3) (4) (5) (6)
Tax inspection 0.089*** 0.089*** 0.091*** 0.090*** 0.090*** 0.089***
(0.008) (0.008) (0.008) (0.008) (0.008) (0.008)
Gov Index - CC -0.044***
(0.008)
Gov Index - GE -0.042***
(0.006)
Gov Index - PS -0.023***
(0.005)
Gov Index - RQ -0.040***
(0.006)
Gov Index - RL -0.044***
(0.006)
Gov Index - VA -0.036***
(0.007)
(1) (2) (3) (4) (5) (6)
Import license 0.075*** 0.075*** 0.075*** 0.075*** 0.075*** 0.075***
(0.016) (0.016) (0.016) (0.016) (0.016) (0.016)
Gov Index - CC -0.011***
(0.004)
Gov Index - GE -0.009***
(0.003)
Gov Index - PS -0.004**
(0.002)
Gov Index - RQ -0.009***
(0.003)
Gov Index - RL -0.008***
(0.002)
Gov Index - VA -0.007***
(0.002)
Table A2 (Continued)
Panel 3
Dependent variable: Gift given to a tax inspector (1 if yes, 0 otherwise)
Panel 4
Dependent variable: Gift given for import license (1 if yes, 0 otherwise)
169
(1) (2) (3) (4) (5) (6)
Operation license 0.120*** 0.121*** 0.122*** 0.121*** 0.121*** 0.122***
(0.012) (0.012) (0.012) (0.012) (0.012) (0.012)
Gov Index - CC -0.019***
(0.005)
Gov Index - GE -0.015***
(0.004)
Gov Index - PS -0.010***
(0.002)
Gov Index - RQ -0.013***
(0.004)
Gov Index - RL -0.015***
(0.003)
Gov Index - VA -0.011***
(0.004)
(1) (2) (3) (4) (5) (6)
Construction permit 0.158*** 0.158*** 0.158*** 0.158*** 0.158*** 0.158***
(0.014) (0.014) (0.015) (0.014) (0.014) (0.014)
Gov Index - CC -0.017***
(0.002)
Gov Index - GE -0.015***
(0.002)
Gov Index - PS -0.004*
(0.002)
Gov Index - RQ -0.013***
(0.002)
Gov Index - RL -0.014***
(0.002)
Gov Index - VA -0.013***
(0.003)
Table A2 (Continued)
Panel 6
Dependent variable: Gift given for construction permit (1 if yes, 0 otherwise)
Panel 5
Dependent variable: Gift given for operation license (1 if yes, 0 otherwise)
Abstract (if available)
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Three essays on the identification and estimation of structural economic models
PDF
The impact of agglomeration policy on CO₂ emissions: an empirical study using China’s manufacturing data
PDF
Three essays on supply chain networks and R&D investments
PDF
Internal prosperity and external alienation: compounding effects of village elections in rural China
PDF
Essays on development and health economics
PDF
A study of Chinese environmental NGOs: policy advocacy, managerial networking, and leadership succession
PDF
Three essays on the evaluation of long-term care insurance policies
PDF
Three essays on health economics
PDF
Three essays on economics of early life health in developing countries
PDF
Panel data forecasting and application to epidemic disease
PDF
An empirical characterisation of the Bombay Stock Exchange
PDF
Essays on the estimation and inference of heterogeneous treatment effects
PDF
Essays on education and institutions in developing countries
PDF
Catastrophe or adaptation? Explaining the impacts of resource scarcity and adaptability on political instability
PDF
The social making of authoritarian environmentalism: protest-litigation nexus and policy changes in China
PDF
Essays on political economy and corruption
PDF
Essays on effects of reputation and judicial transparency
PDF
Essays on health and aging with focus on the spillover of human capital
PDF
Selection and impacts of early life events on later life outcomes
PDF
Why go green? Cities' adoption of local renewable energy policies and urban sustainability certifications
Asset Metadata
Creator
Li, Yunsun
(author)
Core Title
Essays on policies to create jobs, improve health, and reduce corruption
School
College of Letters, Arts and Sciences
Degree
Doctor of Philosophy
Degree Program
Economics
Publication Date
05/10/2016
Defense Date
11/10/2015
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
conditional cash transfer,Corruption,impact evaluation,job creation,labor rigidity,OAI-PMH Harvest
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Nugent, Jefffrey (
committee chair
), Hsiao, Cheng (
committee member
), Hsieh, Yu-Wei (
committee member
), Tang, Shui-Yan (
committee member
)
Creator Email
yunsunli@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c40-198407
Unique identifier
UC11278477
Identifier
etd-LiYunsun-4028.pdf (filename),usctheses-c40-198407 (legacy record id)
Legacy Identifier
etd-LiYunsun-4028.pdf
Dmrecord
198407
Document Type
Dissertation
Format
application/pdf (imt)
Rights
Li, Yunsun
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Tags
conditional cash transfer
impact evaluation
job creation
labor rigidity