University of Southern California Dissertations and Theses

Spatial convergence patterns in the presence of information technology

(USC Thesis Other)

Spatial convergence patterns in the presence of information technology

PDF

Download

Open document

Flip pages

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content SPATIAL CONVERGENCE PATTERNS IN THE PRESENCE OF INFORMATION TECHNOLOGY Copyright 2003 by Gang Yu A Dissertation Presented to the FACULTY OF THE GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements of the Degree DOCTOR OF PHILOSOPHY (PLANNING) December 2003 Gang Yu Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. UMI Number: 3133358 Copyright 2003 by Yu, Gang All rights reserved. INFORMATION TO USERS The quality of this reproduction is dependent upon the quality of the copy submitted. Broken or indistinct print, colored or poor quality illustrations and photographs, print bleed-through, substandard margins, and improper alignment can adversely affect reproduction. In the unlikely event that the author did not send a complete manuscript and there are missing pages, these will be noted. Also, if unauthorized copyright material had to be removed, a note will indicate the deletion. ® UMI UMI Microform 3133358 Copyright 2004 by ProQuest Information and Learning Company. All rights reserved. This microform edition is protected against unauthorized copying under Title 17, United States Code. ProQuest Information and Learning Company 300 North Zeeb Road P.O. Box 1346 Ann Arbor, Ml 48106-1346 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. UNIVERSITY OF SOUTHERN CALIFORNIA THE GRADUATE SCHOOL UNIVERSITY PARK LOS ANGELES, CALIFORNIA 90089-1695 This dissertation, written by Gang Yu under the direction o f h±.s.. dissertation committee, and approved by all its members, has been presented to and accepted by the Director of Graduate and Professional Programs, in partial fulfillment o f requirements for the degree of DOCTOR OF PHILOSOPHY Director Date December 17. 2003 Dissertation Co\ \ttee Chair Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 11 ACKNOWLEDGEMENTS This dissertation could not have been completed without some very noteworthy contributions. I am deeply grateful for the guidance and support Professor Peter Gordon provided as my advisor and dissertation chairman. For my dissertation, he spent considerable time evaluating the research design, reviewing drafts, offering insightful suggestions, answering questions, and correcting some errors. I benefited a lot from his valuable advice and comments, some of which were strategic and some in detail. Professor Gordon’s assistance also went beyond academics, such as helping me seek financial aid that is important to this dissertation research, as well. In addition, I am appreciative of the constructive contributions made by the other dissertation committee members—Professors Harry Richardson and Jennifer Wolch. Both o f them made insightful comments and suggestions for the improvement of my dissertation. What should be mentioned especially, is that both professors and Professor Gordon helped me on my dissertation, even when they were on sabbaticals or vacations. At USC, I am also indebted to Professors Richard Peiser, James Elliott Moore, II, Genevieve Giuliano, and Robert Kalaba for their instruction, advice, and help. In particular, Professors Peiser, Moore, and Guiliano taught me much about research. Also, I appreciate the nice help from Ms. June Muranaka, Senior Advisor of the School of Policy, Planning, and Development. My special thanks are given to Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. iii Dr. Rena Sivitanidou, who nicely helped me, but left all of us and our world too early. Furthermore, I would like to express my gratitude for the financial support of USC’s Graduate School Haynes Dissertation Fellowship. For several years earlier, the USC School of Policy, Planning, and Development, in particular some of Professors Gordon’s and Peiser’s research funds, had provided me graduate assistant positions. I am grateful for the positions, as well. Many friends and fellow students also offered valuable help. Sherrie Love and the Chitwoods family (Derek and Amy) provided me home-like hosting, when I started my Ph.D. study in a difficult financial situation. The others that I should thank are Ningsheng Zhou, Doug and Nettie Diller, Yue Zhao, Xueming Chen, Rong Xu, Devajyoti Deka, Qisheng Pan, Liang Wei, Xudong An, and Gang Chen. Finally, I could not have gotten through this dissertation without the love, support, understanding, and patience of my family, parents, and sister. I dedicate this dissertation to all who loved, supported, or helped me on my journey toward this destination. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. iv TABLE OF CONTENTS ACKNOWLEDGEMENTS........................................................................................ ii LIST OF TABLES........................................................................................................ vii ABSTRACT................................................................................................................... ix Chapter Page 1. INTRODUCTION....................................................................................... 1 1.1 Statement of Problem s................................................................... 1 1.2 Information Technology (IT), a Possible Explanation............... 5 1.3 Research Questions......................................................................... 10 1.4 Significance of This Research....................................................... 15 1.5 Research M ethods........................................................................... 17 1.6 Key Empirical Findings................................................................. 18 1.7 Theoretical Conclusions................................................................. 19 1.8 A Guide to the Chapters................................................................. 21 2. LITERATURE REVIEW.......................................................................... 28 2.1 Introduction..................................................................................... 28 2.2 Theoretical Framework on Convergence..................................... 29 2.3 Empirical Findings and Challenges............................................. 34 3. RESEARCH QUESTIONS AND HYPOTHESIS................................. 37 3.1 Introduction..................................................................................... 37 3.2 Analysis............................................................................................ 39 3.2.1 What is Information Technology, and What Can Be Its Impact on Convergence?........................................ 39 3.2.1.1. Broad Definition of I T ................................ 39 3.2.1.2 Exogenous or Endogenous Development o f a Technology..................................... 40 3.2.1.3 Special Impacts of IT—A Technology with Spatiality—on Convergence 44 3.2.2 Is IT a Significant Economic Shock in Terms of Productivity Growth?.................................................. 49 3.2.3 How May Spatial Scales Play a Role in Income Convergence?............................................................... 55 3.2.4 How May Trickling-Down Convergence Differ from Spreading-Out Convergence?.......................... 63 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. V Chapter Page 3.3 Research Hypothesis ................................................................. 64 4. RESEARCH APPROACH...................................................................... 6 6 4.1 Introduction...................................................................................... 6 6 4.2 Data Issues....................................................................................... 6 8 4.2.1. Measurement o f the IT Sector.......................................... 6 8 4.2.2 Major Datum Sources........................................................ 72 4.2.2.1 Data from the Regional Economic Information Systems (REIS)...................... 72 4.2.2.2 Data from the Bureau of Labor Statistics (BLS)............................................................. 72 4.2.3 Long-time Period and Short-term.................................... 73 4.3 Unit of Analysis............................................................................... 74 4.3.1 Temporal Units................................................................... 74 4.3.2 Spatial U nits....................................................................... 80 4.3.2.1 States, Counties, and Central-Suburban A reas............................................................. 80 4.3.2.2 IT Intensified Areas (ITIA) vs. Non-IT Intensified...................................................... 82 4.4 Models.............................................................................................. 84 4.4.1 panda M odels.................................................................. 84 4.4.2 T-tests.................................................................................. 87 4.4.3 Linear Regression Tests.................................................... 8 8 4.5 Expected Findings.......................................................................... 8 8 4.5.1 Tests on Income Convergence......................................... 8 8 4.5.2 Tests on Statistical Relationships..................................... 90 5. ESTIMATION AND DISCUSSION...................................................... 91 5.1 Introduction..................................................................................... 91 5.2 P Convergence................................................................................ 94 5.2.1 States: Initial Period Decelerating, Later Period Accelerating, and Long-time Convergence C ases... 94 5.2.2 Counties.............................................................................. 95 5.2.2.1 Counties Throughout the U. S.: Initial Divergence, Later Convergence, and Extremely Slow Long-time Divergence C ases............................................ 95 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. vi Chapter Page 5.2.2.2 Counties Within Sate: Fluctuating Initial and Later Period Convergence, but Almost All Long-time Convergence C ases............................................................. 96 5.2.3 A Comparison of State and County p Convergences 104 5.3 c t Convergence.................................................................................. 105 5.3.1 States: Long-time Convergence, but Short-term Fluctuations................................................................... 105 5.3.2 Counties: Long-time convergence, but Short-term Fluctuations................................................................... 107 5.3.3 A Comparison of State and County c t Convergences.... 107 5.4 T-tests and Other Observations for Central-Suburban Areas I l l 5.4.1 Testing Results: Neither Convergence Nor Divergence.................................................................... I l l 5.4.2 Additional Findings: Initially Diverging, Later Converging, and Long-time Converging Evidence........................................................................ 119 5.5 Tests on the Link of Convergence to ITIA/NITIA...................... 126 5.6 Tests on the Relevancy of Convergence to Productivity 133 6 . CONCLUSIONS....................................................................................... 136 6.1 Introduction.......................................................................... 136 6.2 Summary o f Empirical Findings....................................... 140 6.2.1 Temporal and Relationship Testing R esults 140 6.2.2 Findings for Spatial Scales.................................... 144 6.2.3 Center-Suburb Testing Results and Observations.................................................... 145 6.3 Theoretical Conclusions..................................................... 146 6.3.1 Income Convergence Patterns, IT Impacts, and a New M odel............................................ 146 6.3.2 Exogenous or Endogenous Assumption 149 6.3.3 Spatial Scales and Convergence............................ 152 6.3.4 Spreading-out Type of Income Convergence.... 153 6.4 Further Research Issues...................................................... 155 6.5 Implications for Social Income Inequality....................... 156 6 .6 Policy Implications.............................................................. 157 REFERENCES.............................................................................................................. 159 APPENDIX.................................................................................................................... 172 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. LIST OF TABLES Table Page 3.1 Changes o f Non-farm Productivity in the U. S........................................ 51 3.2 Average Annual Growth Rates of Nonfarm Productivity in the U. S.. 52 4.1 Information Technology Industries........................................................... 70 4.2 Expected Findings for Income Convergence and Its Changes............... 89 4.3 Expected Findings in the Relationship T ests........................................... 90 5.1 P Convergence of States in the U. S.......................................................... 95 5.2 p Convergence o f Counties in the U. S..................................................... 96 5.3 P Convergence of Counties within State 9 7 5.4 Per Capita Income a Values of States in the U. S................................... 106 5.5 Per Capita Income cr Values of Counties in the U. S.............................. 108 5.6 Per Capita Income Ratios of Central Counties to Ring I Counties in the Largest 50 Places in the U.S............................................................... 112 5.7 Per Capita Income Ratios of Central Counties to Ring II Counties in the Largest 50 Places in the U.S......................................................... 115 5.8 T-Tests on Changes of Per Capita Income Ratios o f Central Counties to Ring I/II Counties................................................................................. 118 5.9 Center-Periphery Convergence/Divergence Analysis for the 50 Largest Places..................................................................................................... 120 5.10 Center-Periphery Convergence/Divergence Analysis for the 50 Largest Places in Total...................................................................................... 124 5.11 State IT Indices in the U. S......................................................................... 128 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. viii Table Page 5.12 Regression of Convergence Proxies on IT Indices for States in theU . S................................................................................................... 132 5.13 Regression of State a on Non-farm Productivity in the U.S................. 134 5.14 Regression of County a on Nonfarm Productivity in the U. S 134 6 .1 Summary of Findings in the Convergence Testing................................ 141 6.2 Summary of Findings in the Statistical Relationship T ests.................. 144 LIST OF FIGURES Figure 3.1 Nonfarm Productivity Changes in the U. S.............................................. 52 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ix ABSTRACT SPATIAL CONVERGENCE PATTERNS IN THE PRESENCE OF INFORMATION TECHNOLOGY Two issues drive this research. The first is the lack of understanding on how differences of spatial per capita incomes evolve during rapid technological changes, such as an emergence of information technology (IT). The previous long-term convergence pattern appears to change in the entitled “Information Age,” because studies report nonconvergence or divergence evidence recently. The second is the limited understanding o f how spatial income distributions evolve for substate units, especially in the presence o f IT. The convergence literature predominately investigates large geographical aggregations. This research investigates whether the convergence change is relevant to IT, and whether convergence patterns vary at different spatial scales. The assumptions are that, during the “Information Age,” convergence speeds become slower or even negative initially, but get faster later. For a relatively long span, convergence trend prevails. Convergence is also faster if it is examined with larger geographical units. Moreover, long-time convergence occurs in both the rich-poor region and the center periphery directions. Regarding the rich-poor-region direction, p convergence tests for states and counties within states support these assumptions; p tests for counties throughout the U. S. do not. All c t convergence tests support the assumptions. Tests demonstrate faster convergence at the state level than at the county level. Probably as a result, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. X spatial income inequalities at the county level are more severe. Observations of data indicate long-time convergence between centers and the suburbs, but two-tail t-tests do not show any convergence or divergence. Also, IT impacts seem to be stronger on the pair of center and the outer suburbs, and appear to lag behind in smaller places for the center-periphery convergence. The empirical evidence supports the long-range convergence argument by the neoclassical theory assuming exogenous development of technology and declining returns to capital. No long-time evidence has supported the new growth theory assuming endogenous development o f technology. However, transitory non convergence or divergence, can occur when there are increasing returns initially during significant technological changes. An explanation for faster convergence at a higher spatial level can be across large distance technological diffusions brought by IT firms’ to other state or to other nation (global or off-shoring) relocation patterns. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1 CHAPTER 1 INTRODUCTION th In the last decade of the 20 century, the world witnessed a series of technological innovations that have been labeled the Information Revolution. It started with mainframe and personal computers, continued through communication networks, and subsequently brought about the Internet, World Wide Web, and e- commerce. Nowadays, documentation indicates that more Americans make computers than cars, and more Americans build semiconductors than construction machinery (U. S. Government Working Group on Electronic Commerce 1998). 1.1 Statement of Problems During the development and application of information technology (IT), some strange economic growth patterns have brought issues to growth theories. One of the great issues, concerns spatial income convergence. Although prior to the 1970s, there had existed spatial income convergence, a pattern that could be explained by the “neoclassical growth” or exogenous growth theory, nonconvergence and even divergence observations, have been reported in many countries since then. If we try to resolve the issue with the “new growth” or endogenous theory, it is hard to explain the long-term convergence pattern in the past. Interestingly, recent deviations from an income convergence path seem to coincide temporally with the development and application of IT. Is this a coincidence or is there any association? This question needs research efforts. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 2 However, there is a surprising shortage of research, particularly on empirical work, investigating whether the emergence of IT plays a role in the newly observed income distribution pattern. The dissertation, therefore, addresses this topic. Income convergence is an important issue in both academic and policy communities (Dluhosch, 1997; Sala-i-Martin, 1996). First, the issue has theoretical significance. Whether there exists an automatic tendency for income convergence across spatial units, has been a key issue in economics and urban and regional sciences for some time (Barro & Sala-i-Martin, 1992; Richardson, 1973). The answer involves some fundamental questions dividing several schools in economics, such as whether there are diminishing or increasing returns in capital. In the urban and regional studies, the convergence issue implies whether distributions of economic activities become centralized or decentralized. Second, a better understanding of this issue provides insight for firm and residential location behaviors, in that spatial income distributions are influenced by economic distribution activities and location patterns. To some extent, convergence findings disclose how investments are allocated spatially. For business administration, a better understanding of this issue can provide input for firm location and investment decisions. Third, another current event—unionization of countries—brings the income convergence discussion to the central stage again. Among the discussions on the NAFTA (North America Free Trade Agreement) and the EU (European Union), a Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 3 major concern is whether poor countries in a union can catch up with rich ones in the same union (convergence) or fall more and more behind (divergence) (Dluhosch, 1997; Sala-i-Martin, 1996). From the very beginning, the reduction of cross-national income gaps were regarded as an essential part of EU integration and its institutional design (Commission of the European Communities, 1993). Consequently, income convergence— a possible effect of global region-unionization—is one of the focuses in the discussion. Finally, but not least importantly, the issue has strong policy implications. There are arguments about policy intervention. Some arguments may be favorable to policy intervention. Bauer (1992) reported that unbalanced growth could lead to an overall productivity slowdown. Alesina and Rodrik (1994), Persson and Tabellini (1994), Clarke (1995) and Sylwester (2000), found that greater income inequality lowers subsequent growth. Also, unbalanced growth, such as per capita income gaps between the poorest and richest regions, can be too much for social cohesion and stability (Richardson, 1973). Barro (2000) and Hibbs (1973) argued that inequality of spatial incomes motivate the poor to engage in crime, riots, and other disruptive activities in those regions. If invisible hands fail to lead to income convergence, Maxwell and Hite (1992) argued that governmental agencies should use taxation, investment, and transfer of payment to redistribute incomes and financial resources at regional levels. In the U. S., some federal and local funds have already targeted spatial income inequities (Gyourko, 1998; Gong & Gyourko, 1998) in light of an Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 4 assumption that the inequities are growing rather than diminishing. According to Dluhosch (1997), EU policy, is in effect, channeling transfers between poorer and richer EU countries. However, if spatial incomes can converge on their own, policy decisions should adjust accordingly. In Alonso’s (1968) view, measures of equalization may slow down the growth of total economy. Maudos, Pastor, and Seranno (1999/2000) argued that spatial income inequality may contribute to faster productivity gains. Accordingly, policy intervention is not justified. Since spatial income convergence addresses important academic and policy issues, it has become a hot area of economic debates. The debates started from the “neoclassical growth theory.” The theory predicts that an economy is initially unevenly distributed and later approaches balanced distribution (Alonso, 1980). The most likely reason is that there may be an automatic trend (Barro & Sala-i-Martin, 1990) or that a competitive market mechanism driving capitals to lag-behind areas, leads to spatial income equalization. If we measure with per capita income, some regions are rich and some poor initially; but gradually, poor regions catch up with rich ones and spatial incomes converge. For many years before the 1970s, empirical evidence had provided support for this hypothesis (Barro & Sala-i-Martin, 1990; Carlino & Mills, 1996). Estimates of convergence were recurrent in the literature across three levels of aggregation: (a) nations, (b) regions of a continent, and (c) states in a nation (Austin & Schmidt, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 5 1998). However, since the 1970s, numerous empirical findings have not agreed with this hypothesis. Those findings have puzzled researchers. To solve this puzzle, two major types of responses have been seen in the literature. The first type of response claims that the issue is a measurement or modeling problem (Quah, 1996; Carlino & Mills, 1996; Xie, 2002). Those researchers seek different measurement or modeling approaches to achieve the testing results that they think are relatively accurate. This type of response can be questionable, because if recent observations had been caused by a methodological problem, empirical findings for the years before the 1970s should have presented the same “symptoms.” The second type of response challenges the “neoclassical theory” directly. These responses argue that spatial incomes may diverge rather than converge (Drennan, Tobier, & Lewis, 1996; Romer, 2000). In other words, the notion of income convergence is untenable. In the theoretical discussions, the augment has mainly come from the “new growth” assumption. A difficulty in the second type of response is its explanation for the long-term convergence in history. 1.2 Information Technology (IT): A Possible Explanation There can be a third possible explanation for the puzzle. That is, the emergence of information technology may play a role. This dissertation is a research effort from this perspective. Because, in the literature, there has been no long-range evidence for the “new growth theory,” but for the “neoclassical growth theory.” This research still starts from the “neoclassical” framework to examine the features of Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 6 income convergence in the U. S. in the presence o f IT. The objective is to find out whether an exogenous shock like IT, can disturb the convergence predicted by the “neoclassical model.” The development and use of IT should be a significant shock to the American economy in terms of productivity change. Recent data indicated that, following the emergence of IT, private business productivity in the U.S. accelerated to a speed not seen since the 1960s (Whelan, 2002). Also, substantial studies reported that IT contributes to the acceleration of productivity (Baily, 2002; Baily & Lawrence, 2001; Brynjolfsson & Hitt, 2000; Oliner & Sichel, 2000; Smolny, 2000; Whelan, 2002). If the technological disturbance only causes short-term fluctuations of convergence, especially in the initial phase of the IT presence while convergence is a long-time trend, the conflict between the “neoclassical growth theory” and recent empirical evidence, can be explained by this disturbance. This way, we can solve the puzzle, while reconciling with the long-range convergence evidence in history. In fact, albeit little of the income convergence, literature directly investigates the possible association between IT and income convergence or divergence, where some papers have addressed that exogenous shocks may disturb the income convergence trend. Barro and Sala-i-Martin (1991), Coulombe (2000), and Webber (2001), discussed that the energy crisis causes income divergence in the post gasoline crisis period, and several mentioned that technological changes might affect Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 7 income convergence. In theory, an exogenous shock can disturb income convergence. Just take technological revolutions as an example. Fundamentally, for two reasons, we can assume that a revolutionary technology can impinge on income convergence. First, when it emerges to the world, a technology as a scarce resource offers competitive advantages to a small portion of people in some poles. Those people can gain higher incomes by taking advantage of this scarce resource. Thus, initially, technological innovation tends to raise inequality and disturbs the income convergence trend. This point can also be found in Barro’s (2000) argument. Second, because a great and especially revolutionary technology always spills over and diffuses to other people and places, this progress can generate a new wave of income convergence in the later phase of the technology. Since the people in some poles who initially possess the new technology can earn higher incomes, people who do not possess it or who are in other places, must be eager to learn in order to earn higher incomes. Subsequently, the new technology becomes less and less scarce, and the people who initially have it gradually, lose competitive advantages. Besides, firms that possess the technology may invest in other places to seek the benefit of lower costs. The diffusion of a new technology in its later phase, can generate new dynamics for spreading out high incomes from their initial poles to other areas. As a result, this progress might accelerate income convergence. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 8 Therefore, development of a great technology may influence convergence negatively in its beginning phase, but positively as it matures and diffuses. Although great technological innovations often leave footprints in economies and society, it is always difficult to evaluate their impacts. Currently, there are two opposite views on the impacts of information technology. One of them argues for a revolutionary impact. In their view, “electronic commerce is a revolution that many industry and academic observers believe will transform the conduct and structure of business as we know it” (Kauffman & Walden, 2001). “In just a few years, the Internet has had a visible impact on the daily lives of many Americans— at work, at home, and how they communicate with one another” (Litan & Rivlin, 2001). The technology “has brought substantial changes in living styles, modes of communi cation and the ways of conducting business” (Lau, Wong, Chan, & Law, 2001). Actually, most of them only discussed e-commerce or the Internet, parts of IT. If their discussions had covered more aspects of IT, they might have claimed a stronger impact. The opposite argument cautions that the optimism on the technological revolution is another variant o f “Technological Determinism” (Thrift, 1996) or “Technological Utopianism” (Graham, 1998). Thrifts (1996) argues that the current writing on the effects of “new” electronic communication technologies, has an error like old errors in looking at electronic communication technologies o f the 19th and Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 9 early 20th centuries. He gives an example of the impacts of telegraphy and telephone on London, especially on the financial markets of London. It is too early to reach a conclusion on the total impacts of IT. Though IT has demonstrated some seemingly dramatic impacts on economy and society, these impacts may be transitory, while some long-run, great impacts may not have occurred or been noticed. Moreover, it is always difficult to single out the effects of a technology, even if it is revolutionary, because the effects are integrated and melted into other factors. For instance, if we review Thrift’s case on telegraphy and telephone, we may wonder whether the emergence o f the New York stock market was affected by the development of those technologies. Integrated with other infrastructures, both telegraphy and telephone must have helped the New York stock market to emerge to compete with European markets. Furthermore, the booming of the New York stock market, as well as international trade that was also facilitated with telegraphy and telephone, might have contributed to the growth o f American economy and society. Therefore, we can understand that, if we only evaluate London financial markets or even London for a conclusion on the impacts of telegraphy and telephone, the scope is too limited. In an evaluation of impacts of a technology, it is better to examine the technology from various perspectives. This dissertation research is an effort from one of these perspectives— a possible impact on spatial income distribution trends to improve our understanding on technological impacts. Moreover, for two additional Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 10 reasons concerning IT, the researcher chose this topic. One reason was that, among great technologies, IT has a spatial feature. That is, IT is a technology to tackle spatial constraints for human activities. This technology is like those for airplanes, trains and automobiles, all of which have contributed to spatial redistributions of economic activities and demographic locations; while some other technologies, such as a light bulb or a biological product, may not have similarly strong impacts in the spatial dimension. The other reason is that, in the current literature, there are more and more reports attributing recent years’ productivity acceleration to IT. This is a critical indication that IT has a great impact on economy. Therefore, it is reasonable to consider that there may be a discernible impact of IT on income convergence. 1.3 Research Questions The objective of this research is to investigate the patterns o f spatial personal income convergence during the “Information Age,” and to answer the question of whether the changes of convergence are relevant to the emergence o f IT. In particular, the research hypothesis is that speeds of spatial income convergence are relevant to the development and application o f information technology. That is, there is a relationship between the presence of IT and the change of income convergence speeds. The assumption is that, during the “Information Age,” income convergence speeds become slower or even negative, initially, but gets faster later. This hypothesis suggests that, at the beginning of the “Information Age,” we may see Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 11 a divergence pattern; but for a relatively long time span, the convergence trend prevails. The “neoclassical model” in growth economics is a starting point for the research hypothesis, but the hypothesis contains an important revision. In the researcher’s view, there is a shortcoming in the conventional income convergence assumption. Namely, the assumption implies that there exists only one significant technological shock that has dominated convergence forever, since the early stage of the economy. It predicts an all-time convergence pattern and suggests a single curved income convergence path. This assumption omits that there can be multiple significant technological shocks that can bring about overlapping income convergence curves corresponding to those shocks temporally. The hypothesis acknowledges multiple significant technological shocks and corresponding multiple curves. Based on the revision, recent nonconvergence findings are assumed to be caused by the disturbances of a technological shock and probably a demonstration of the overlapping curves. To get a better understanding of whether IT has impact on convergence and how IT can influence if it has impact, this investigation furthers the inquiry into two research dimensions— spatial scales and spatial directions. In the first dimension, the question is how the convergence story in the “Information Age” may vary on different spatial scales. For several reasons, spatial scales are an indispensable factor in evaluating income convergence in the presence of IT. The first reason is derived Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 12 from the “neoclassical” logic concerning the dynamics leading to convergence. According to this logic, market forces and spatial constraints for competition and mobility cause or impede convergence. Because, at different spatial scales, market forces and spatial constraints differ in terms of their intensity and form, we may expect various convergence patterns. Second, as discussed in the literature, a geographical scale should be recognized as one crucially important dimension of geographical differentiation (Brenner, 2001), and spatial income inequity is a manifestation of the geographical differentiation. Moreover, the scale “is not simply an external fact awaiting discovery, but a way of framing conceptions of reality” (Delaney & Leitner, 1997). It “is not simply a ‘hierarchically ordered system’ placed over pre-existing space” (Marston & Smith, 2001). Rather, it is “socially constructed” (Marston, 2000). Income convergence patterns on different geographic levels, such as among states, counties, and between central-suburban areas, should be influenced by the social construction process at those levels. As early as the 1960s, economist Alonso (1968) called for discussion on the scale of regions in the investigation on unbalanced growth and income inequity. The third reason concerns technologies. The “neoclassical theory” argues that diffusion of a technology results in convergence. However, the diffusion can be distance constrained. In other words, the diffusion of IT can vary in terms of spatial scales (Keller, 2002), and there can be spatially mediated spillovers (Audretsch & Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 13 Feldman, 1996). The distribution of IT and its consequences on income distribution can vary on different spatial scales. The final reason for an examination of the role of spatial scales in the convergence pattern, is regarding a methodological or datum processing issues in convergence research. In the literature, all of the convergence evidence has been reported at high spatial levels. These findings can be subject to an artifact of aggregating, and averaging data that the researcher will entitle “aggregation and averaging effects.” The aggregation of geographic units and averaging o f incomes could underestimate the differences. For this reason, the researcher suspects that the investigations on high spatial levels might have distorted the findings to some extent. The other research dimension explored in this research is the spatial movement direction of income convergence. This project defines two types of spatial directions—trickling-down and spreading-out. The trickling-down type of convergence is that high incomes in rich areas trickle down to poor ones. In Barro and Sala-i-Martin’s (1992) words, poor regions catch up with rich ones. Almost all of the existing income convergence investigations fall into this category. The other type, spreading-out, is that income disparity between a central place and its peripheral areas gradually diminish. The discussion on the center-periphery spreading out pattern o f income convergence is much less common in the literature. In the U. S., based on an assumption that income inequities between central cities and suburban areas are growing or diverging, some of the federal (Gong & Gyourko, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 14 1998) and local (Gyourko, 1998) policies, have aimed at ameliorating their income inequities with both federal and local funds. However, assumption-based policy making is inappropriate. These policies require investigation on the spreading-out type of convergence or divergence. Moreover, the investigation into this dimension also shows how IT can affect convergence. These two types of convergence suggest different spillover patterns of the IT. Spreading-out means that capital decentralizes from centers to surrounding areas. This type o f convergence implies a center-suburb pattern of technological diffusion. Trickling-down convergence indicates that capital moves from rich to poor regions and follows a developed-underdeveloped pattern of technological diffusion. This diffusion may cross a big distance, since developed and underdeveloped regions are usually not adjacent to one another. In fact, in this research, a particular reason for an investigation into these two research dimensions— spatial scales and spatial directions—lies in an initial concern of the researcher’s. The concern is that impact o f IT on convergence may not occur on all spatial scales and in all directions. If the issue is evaluated on one scale and in only one direction, the impact may be missed, if IT happens to have the weakest impact on the scale and direction that is selected. As a result, the research may mistakenly conclude that there is no IT impact, even if the impact occurs intensively in another direction or on other scales. To evaluate some more important scales and directions can help this research catch the impact, if it exists. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 15 1.4 Significance of This Research The findings of this research can contribute to the literature from several perspectives. In the income convergence literature, the impact of IT is inappropriately ignored. Although some papers mention that exogenous shocks and technological changes should affect income convergence, not much serious research, especially no empirical work, has been conducted concerning the impact. This research is an effort to bridge the gap. In addition, this dissertation extends the inquiry of convergence features into spatial scales. The investigation on convergence at the levels of U. S. counties and other spatial units used in this research is rarely seen in the literature. It is necessary to examine the income convergence story at the county level to understand whether income convergence also occurs among smaller spatial units. Counties are also an important economic and political unit in policy-making. Considerable federal and local funds are allocated in this level. State level convergence does not guarantee county-level convergence. Even if it does, lower-level convergence patterns can differ from the higher-level ones. Regions, states, and cities are far from interchangeable concepts (Cheshire & Magrini, 2000). Using higher-level convergence evidence for county-level policy-making can potentially mislead the policy. Thus, the investigation on county convergence is useful for income distribution policy-making at the county level. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 16 Furthermore, in terms of spatial directions, there is little understanding on a center to peripheral area spreading-out type of income convergence, which this research also examines. Unlike the trickling-down convergence tests that do not address spatial vectors, spreading-out convergence tests can show whether there is a radiating type of convergence following a decentralization trend from urban centers. These issues are concerns of policy-makers as well. For example, Gong and Gyourko (1998) mention that assumed growing disparities in the economic fortunes of central cities and their suburbs, have led the federal government’s decision to encourage the banking system and even government-sponsored enterprises (GSEs), such as Fannie Mae and Freddie Mac to increase lending in order to foster more home ownership in cities. Also, Gyourko (1998) argues that inner-city areas with high rates of poverty are seriously disadvantaged, when compared to suburban areas in terms of expenditures on poverty-related costs. He suggests that intergovernmental transfer payments to combat poverty be directed to areas of concentrated poverty, or in his words, place-based aid. Many of those policy arguments are based on the assumption that income inequities between central cities and the suburbs are not diminishing but growing. However, there have been no investigations on that assumption in the literature. The lack of research and evidence does not provide concrete input for the policymakers’ decisions. Thus, it is also meaningful to examine the spreading-out type of convergence patterns. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 17 1.5 Research Methods To evaluate the trickling-down income convergence, the researcher applied beta (P) and sigma (a) models— two widely used approaches in the convergence literature. The P Model tests whether per capita income in poor regions grows faster. The a model examines variations of regional per capita income. The reason for using both models is that the p test can be subject to Gabon’s fallacy of regression (Bliss, 1999; Quah, 1993,1996), and the a model can expose a “switching” problem. A combination of the two models can help to correct possible bias caused by each approach. For the spreading-out convergence, the researcher examined the income ratios of the central counties and their suburban areas in the 50 largest American places with t-tests. Two-tail tests were applied, because they can disclose not only if the income ratios change, but also whether a change found in the test is convergence or divergence from a positive or negative t-score. To seek further information on associations between income convergence and IT, the researcher employed ordinary least square (OLS) regression tests. Just like all other statistical analyses of these kinds, the p, a , t-tests, and regression tests used in this research, cannot provide cause-and-effect conclusions, even though the relationships found or rejected with these techniques are important for improvement of our understandings on growth patterns of spatial incomes. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 18 1.6 Key Empirical Findings The trickling-down convergence tests, such as P tests for states, P tests for counties within states, and all of the a tests for both states and counties, support the research hypothesis. All of the tests demonstrate statistically significant convergence for over 3 decades, the longest time for which data are available. However, the testing on counties throughout the U. S. does not. During the “Information Age,” there was spatial income divergence or decelerating convergence initially; but in later years, the divergence switched to convergence, or the slower convergence accelerated. These findings support the hypothesis, as well. The convergence patterns at two levels— states and counties— are different. For over 3 decades, both p and c j tests have shown faster convergence speeds at the state level. Moreover, as indicated in the both tests, the short-term swings of convergence during the development and application o f IT were stronger at the state level as well. Spatial income inequalities at the state level are less severe than those at county level, probably as a result of the long-time faster convergence. For spreading-out convergence, observations of data indicate long-time convergence between centers and the suburbs. Some data show that, in the early phase of IT presence, divergence became more intensive, but convergence took the momentum in the later period. However, two-tail /-tests do not show any convergence or divergence for either the long-time period or any of the short term. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 19 Also, IT impact appears to be stronger on the pair o f center and the outer suburbs, and lags behind in smaller places for the center-periphery convergence. Furthermore, for additional information on a possible link of convergence to IT, OLS regression tests show that, during the “Information Age,” there were significant relationships between a state convergence pattern and a state total IT index, and between the state convergence pattern and the state hardware and IT equipment index, but no relationship between the state convergence pattern and the state software and IT services index. Also, regression tests show relationships between state and county a convergence and productivity growth that is reported to link to IT. The regression tests provide some supportive information for the research hypothesis. 1.7 Theoretical Conclusions The first conclusion is that all of the long-time empirical evidence indicates that there has been no long-range divergence, but rather, long-range convergence. The trickling-down convergence tests, such as the p test for states, the p test for counties within states, all a tests, and nonstatistical observations for spreading-out convergence, support this conclusion. However, the control of temporal periods to IT shock indicates that short-term convergence fluctuations appear to be relevant to the IT shock. Further regression tests provide additional evidence for linking convergence changes to IT development and application. Thus, a model of overlapping multiple Bell-shaped curves or multiple inverted-U-shaped curves, can Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 20 offer a better explanation for the convergence patterns than the conventional single inverted-U-shaped curve. The second conclusion concerns exogenous or endogenous changes of technologies. The findings using states and counties as units of analysis in this research, support the neoclassical exogenous assumption. Two empirical observations challenge the endogenous model for explaining a long-run income evolution pattern. The first one, as already mentioned, is that there has never been any finding for long run or even relatively long-time divergence. This second observation is the slower convergence in the faster technological diffusion years— the “Information Age”—because the endogenous assumption would suggest faster convergence in a faster diffusion age. Even though the short-term divergence finding could support the endogenous assumption on increasing returns to capital, the evidence o f initial divergence can suggest that, when a technology is in its early phase, there are increasing returns to capital; however, for the long run, the investment of return feature appears to switch from increasing returns to diminishing returns. This feature will be called “development-phase-based increasing or diminishing returns to capitals,” which stands for whether there is increasing or diminishing returns depends on the development phase of a technology. The third theoretical conclusion is that convergence patterns vary on different spatial scales. During the longer time, convergence is faster at a higher level Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 21 states— and fluctuation of convergence is more intensive at the state level, as well, than at lower level counties. In addition, spatial income inequalities at the state level are less severe than those at the county level. This can be owing to faster convergence at the state level for the long term. The final conclusion is regarding core periphery, or spreading-out convergence. Observations of data indicate that there are signs of long-time-period convergence; however, those observations are not supported with statistical testing. One safe theoretical point is that spreading-out convergence is evidently weaker than trickling-down, or rich-poor region convergence. In addition, an interesting observation is that the IT influence the convergence between the central places and the outer suburbs, seemed to be stronger than that between the central places and the inner suburbs. However, this exogenous disturbance to convergence might not hold long; so, to the contrary, the convergence for the pair of the central place and the outer suburbs was rather stronger for a longer periods. Furthermore, the larger places appeared to be affected earlier than smaller places. In other words, there was a time lag between the larger and smaller places being influenced by IT. 1.8 A Guide to the Chapters The remainder of this dissertation is divided into five chapters. Chapter 2 is a survey of the major literature on spatial income convergence or divergence. The survey results are organized into two sections—theoretical discussions and empirical findings. In the second section following the section the Introduction, two major Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 22 schools—the “neoclassical growth theory” that argues for convergence and the “new growth theory” that predicts divergence— as well as some discussion on factors that may contribute to spatial income evolution are reviewed. Then, in the third section, there is a review of empirical studies that document convergence evidence for decades before the 1970s, and nonconvergence or even divergence evidence recently. Chapter 3 begins a discussion on four questions critical to developing the research hypothesis. The first question mainly addresses what the impact can be of IT on income convergence. The second is whether IT constitutes a significant economic shock in terms of productivity growth. This question is important, as well, because only if IT is a significant economic shock is an investigation of its impacts on convergence meaningful. The third concern is how spatial scales play a role in income convergence. In light of the analysis of the four questions, the research hypothesis about what is expected for the relationship between changes of the income convergence speeds and the presence of IT is formulated. The chapter closes with elaborations of this hypothesis on how the emergence of IT can affect income convergence; how income convergence patterns vary on different spatial scales; and what is expected for center to periphery income convergence in the presence of IT. Chapter 4 discusses the research design, including data issues, temporal and spatial units of analysis, research models, and expected results with those models. First, datum issues pose a challenge to the research, just as they do to almost all other Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 23 studies on the convergence issue in the U. S. Some feasible treatments are introduced on the measurement of the IT sector in the literature, and adopt one used by the Taskforce of Economics and Statistics Administration o f the U. S. Department of Commerce. The following section presents what temporal and spatial units this research uses and why. For the temporal units, the “Information Age” is defined, and a division of the time span into subtime units based on three criteria. The first criterion is the social adoption rate of IT, which is a consequence of demand for IT. The second is development phases, or the product development and production phases of IT. The third is contribution of this technology to the economy, i.e., contribution to the GDP (the GDP percentage generated in the industries closely related to this technology) and to productivity growth. In terms of spatial units, states and counties for the trickling-down convergence analyses are used. Some reasons are discussed. For the investigation on spreading-out convergence, the pairs of central counties and their suburban counties in the 50 largest metropolitan areas in the U. S. are used. For comparing convergence in those areas with larger IT employment shares with that in those areas with smaller shares, it is explained how IT Intensified Areas (ITIA) vs. Non-IT Intensified Areas (NITIA) or other partial IT oriented areas with total or partial IT indices, derived from Hoover's index o f regional specialization or localization is defined. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 24 Then, major models— p and a models— for the trickling-down convergence analyses are discussed. For the spreading-out convergence testing, two-tail /-tests of income ratios of a central county to its peripheries, which can disclose whether there is convergence or divergence are employed. Furthermore, the linear regression tests on relationships between convergence patterns and IT, via IT indices and productivity growth rates are explained. This chapter closes with summary tables of expected findings with those models. Chapter 5 reports and discusses estimation results. The second section, following the chapter Introduction section, examines results from the p testing, first for states and then for counties. Then, comparisons o f state and county p convergence follow. The third section demonstrates results from the a convergence testing for states and counties, as well as comparisons of cr convergence between these two units. Also, a finding regarding more severe state level spatial income inequalities than county ones is reported. The fourth section shows spreading-out convergence findings. Results of the two-tail /-tests are reported first, followed by a discussion of nonstatistically-tested observations on convergence or divergence between central counties and their Ring I's and between central counties and their Ring II's. Also, presented are additional observations beyond the research hypothesis, such as stronger fluctuations between core and the outer suburbs than between center and the inner suburbs, and lagging- behind impacts of IT on smaller places. The final two sections include results from Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 25 linear regression testing— one section for a possible relationship between convergence proxy and total or partial IT index, and the other for a possible relationship between state or county a convergence and productivity growth. In the last chapter, the researcher seeks theoretical conclusions, sketches some further research issues, and discusses possible implications in social income convergence research and policy-making. The first conclusion concerns whether there is spatial income convergence or divergence for the long range and how the evolution trend may change during the “Information Age.” The second is regarding how assumptions on exogenous or endogenous changes of technologies explain spatial income evolution patterns. The third discusses the role of different spatial scales in spatial income convergence. The final theoretical conclusion is on core periphery or spreading-out convergence. The following section suggests three dimensions into which the research should be extended. The first is temporal dimension. Investigation into long-range data is needed. Second, more work should address why convergence is faster at higher spatial scales such as states, than at lower scales, such as counties. Explanations may be sought from relocation patterns of firms and capital. Across- large distance decentralization of firms and technologies may be a reason. This needs further research on whether there is significant across-bigger distance decentralization and whether this type of decentralization brings stronger convergence on higher spatial scales. The third is about sub-county level Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 26 investigations, i.e., we should evaluate the income convergence trends in comparatively micro spatial units or on lower spatial scales. Then, it is attempted to extend the discussion scope beyond the topic of this dissertation to raise interest in social income-convergence research. It is appears interesting whether the findings in the spatial income inequality have some possible implications on social income inequality. The assumptions include the theory that social income convergence behavior may share some similar features with spatial income convergence. Therefore, it is likely that exogenous shocks may influence social income convergence, too. Similarly, significant technological shocks may initially drive a short-term divergence or slower convergence; but in later phases, the shocks should contribute to long-run convergence of social inequalities. In addition, social convergence trends on relatively macro scales or larger aggregation units may appear stronger than those on micro scopes or smaller aggregation units may. Also, social inequality issues might be more severe, if we review them on comparatively microscopes or by smaller groups. The dissertation will conclude with a discussion of policy implications from the findings. First, long-term policies may be less necessary than short-term ones, if at all. This is because, for a long run or a relatively long time, spatial incomes converge; and divergence appears to be transitory. Second, whenever policy measures are considered, those on lower levels may be more necessary than those at higher levels, since convergence is slower at a lower spatial level. Finally, due to Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 27 long-time convergence, income redistribution policies in later stages of an economy may be less necessary than in earlier stages. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 28 CHAPTER2 LITERATURE REVIEW 2.1 Introduction This chapter is a survey of the major literature on spatial income convergence. The survey results are organized into two parts—theoretical discussions and empirical findings. For a long time, the “neoclassical growth theory” has argued for convergence. Two important assumptions are a foundation for the convergence argument—diminishing returns to capital and exogenous technological change. To the contrary, assuming increasing returns and endogenous technological change, the “new growth theory” predicts divergence. In addition to the two factors—technology and returns to capital—there is research on other factors that may explain spatial income growth and evolution, such as demographic migration, payment transfer, etc. In terms of empirical studies, there has been a large volume of literature on this issue. Earlier studies document strong convergence evidence for many decades prior to the 1970s. However, there are substantial nonconvergence or even divergence reports recently. Furthermore, our understanding on the issue is limited in two aspects. First, there is limited understanding on how income differences evolve when there are rapid technological changes, such as the emergence of IT. Second, there is little understanding of how income differences evolve for substate spatial units. This dissertation research intends to contribute to the convergence Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 29 literature on those two aspects by way of trying to explain recent spatial income evolution patterns. 2.2 Theoretical Framework on Convergence Convergence is a key feature of the “neoclassical growth” framework (Coulombe, 2000), mainly because of a possibly automatic trend (Barro & Sala-i- Martin, 1992) or a market mechanism, which tends to equilibrate economic growth among economies (Nissan & Carter, 1999). Hicks’ “The Theory o f Wages (1932)” was the first formal statement on income convergence across regions within an economy (Drennan & Lobo, 1999). Later, in his seminal paper, Kuznet (1955) described the growth path as an inverted-U curve, which has been entitled “Kuznet’s curve.” He claimed that regional income disparities increase during the early development stages and decline in the late stages of an economy. Similarly, Hirschman’s (1958) and Myrdal’s (1957) models suggest that regional inequalities are normal in the early stage, due to initial relative advantages in some regions, but will disappear as economies reach maturity. In his "Presidential Address" to the Regional Science Association, Alonso (1980) summarized this feature as one o f the five widely referred to “Bell Shapes” in development; that is, an early stages’ divergence and late stages’ convergence o f regional incomes. He argued that, finally, the “regional inequality diminishes.” Many convergence discussions, similar to those mentioned above, are based on a notion about stages of growth. In his book, The Stages o f Economic Growth, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 30 Rostow (1960) discussed the stages of growth theory to dispel ambiguity of defining growth (Alonso, 1980). He divided economic growth into five stages. The first stage is “the traditional society,” whose structure is developed within limited production functions based on pre-Newtonian science and technology. The second stage is the preconditions for takeoff, when a society is in transformation in the ways necessary for it to exploit the fruits of modem science. The third stage is a takeoff phase. Growth becomes a normal condition. The forces for making economic progress, which yielded limited bursts and enclaves of modem activity, expand and come to dominate the society. Frequently, the first three stages are referred to as early stages o f growth. After the takeoff, there follows a long interval of sustained, if fluctuating, progress, as the regularly growing economy drives to extend modem technology over the entire front of its economic activity. This is the drive to maturity stage. In this stage, the makeup of the economy changes unceasingly as technique improves. The economy finds its place in the international economy. Some 60 years after takeoff begins, what may be called maturity is generally attained in Rostow’s view. The fifth stage is the age of high mass-consumption, in which the leading sectors shift towards durable consumers’ goods and services. Rostow (1960) claimed that the U. S. was beginning to emerge from the fifth stage, when he wrote his book in 1960. After 1960, papers (Alonso, 1968, 1980; Amos, 1988; Barro, 2000; Chen & Fleisher, 1996; Oshima, 1992; Pompili, 1994) often cited or used the terms Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 31 “development stages,” “stages of growth,” “an early stage,” and “a late stage of an economy,” in discussing unbalanced growth or income convergence. Notwithstanding the strong argument for convergence in the late stages of growth, in the literature, economists have debated on whether different regions and countries are converging (Jones, 2002). The “neoclassical model” is entitled “exogenous growth model,” because the model assumes that technologies improve for reasons outside the model and that same technological opportunities are available in all countries of the world (Romer, 1994). The “new growth” or the endogenous growth theory (Jones, 2002, Hoffer & Worgotter, 1997; Romer, 1986; Lucas, 1988; Grossman & Helpman, 1994) argued that growth rates of per capita products are independent of starting levels. The logic of the endogenous growth hypothesis underpinning the argument is that growth is driven by technological changes arising from firms’ desires for profit-maximizing and from the investments from those firms (Romer, 1990). The distinguishing feature of the technology as an input, is that it is neither a conventional good nor a public good; it is a nonrival, partially excludable good. This school emphasizes the private sector activities that contribute to technological advancement rather than public sector funding for research (Romer 1994). Accordingly, technological changes are endogenous rather than exogenous. That is, the “new growth model” withdraws the fundamental assumption— exogenous technological changes— from the convergence hypothesis. This withdrawal is critical. If technological changes that drive growth are endogenous, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 32 the rich or developed regions where the technologies are usually developed should grow faster. Then, the spatial income convergence is not expected. That is why Romer (1986,1994) and other new growth advocates predict income divergence. Another fundamental difference between the “neoclassical” and the “new growth” models is concerning the returns on capital. In the “neoclassical” logic, the market force leading to convergence is driven by diminishing returns to capital. This is because, if there are diminishing returns, monetary capital will flow to poor regions where investment to labor ratios are low. As a result, incomes in poor regions grow faster and convergence is expected. On the contrary, increasing returns could lead to divergence. The competing “new growth theory” argues for increasing returns, and consequently, predicts spatial income divergence. Some other efforts that also seek explanation on spatial income convergence from factors such as population migration, payment transfer and so on. Barro and Sala-i-Martin (1992), Cardenas and Porton (1995) and Mallick (1993), argued that migration of population does not offer a satisfactory explanation for convergence or divergence. In their opinion, migrating labor is likely to embody a significant degree of human capital (Barro & Sala-i-Martin, 1992). Second, mobility o f labor may be impeded by the presence of consumption externalities (Mallick, 1993). Income differentials are not able to explain all of migration (Morrison, 1997; Lande & Gordon, 1977). Also, Mallick (1993) and Drennan, Tobier and Lewis (1996), found that the unemployed stay where they are rather than moving, except they have some Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 33 reliable relatives in the promising regions. At any rate, even when migration does occur in response to wage differentials, it does not contribute to convergence (Barro & Sala-i-Martin, 1992). Moreover, their findings received support in some other countries. Cardenas and Porton (1995) indicated that in Colombia, although there is a very successful story of regional convergence, labor migrations do not play a seemingly important role in the process o f convergence. Sala-i-Martin (1996) concluded that migration is ruled out in the explanation for convergence. However, in the literature, the argument that population migration affects income convergence J ! is also active (Austin & Schmidt, 1998; Cuadrado-Roura, 2001; Livemier, Rickman, ! & Partridge, 1995). j In addition, Livemier, Rickman and Partridge (1995) have examined whether transfer payment plays a role in income inequity changes. They found that this variable has a mixed effect on income inequality. Some evidence showed that in earlier decades the transfer payment was significant at an a = 0.05 level, but the significance has been decreasing over time. They concluded that their finding implies a diminished effect o f transfer payment in reducing income inequality. Nonetheless, the belief that transfer can contribute to convergence is still strong, especially in the movement toward the European Union (Dluhosch, 1997). Many empirical studies, including this one, use the data that do not consist o f transfer payments. The findings here will not be contaminated by the transfer payments. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 34 2.3 Empirical Findings and Challenges Early empirical studies strongly supported the “neoclassical convergence hypothesis.” Findings for the 19th century and the first half of the 20th century, documented a convergence o f per capita income in the U. S. (Borts, 1960; Garnick & Friedenberg, 1982; Perloff, 1963). Kuznet’s curve was accepted through the 1970s as a strong empirical regularity (Barro, 2000). Barro and Sala-i-Martin (1992) observed income convergence in the U. S. in 7 decades of the 20th century at a rate around 2%. Numerous research papers reported similar results of convergence. Furthermore, Drennan and Lobo (1999) used a simple test to show that there was convergence o f per capita personal incomes across metropolitan areas in the U. S. Income convergence had also been observed in the years before the 1980s in other countries and among global regions. Actually, evidence existed in both developed and developing economies. Barro and Sala-i-Martin (1992) showed convergence in Western Europe and among Japanese prefectures. Cuadrado-Roura (2001) observed long period regional convergence among EU until the mid-1970s. Oxley and Greasley (1999) demonstrated that convergence could be identified for a small “Club” of Nordic countries. Kangasharju (1998) found income convergence of about 2% per year among Finnish regions in a long run. For Canada, Coulombe and Lee (1995) and Coulombe (2000) identified strong convergence across provinces. In Africa, Jones (2002) reported convergence evidence among the Economic Community of West African States (ECOWAS). Nagaraj, Varoudakis and Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 35 Vaganzones (2000) saw per capita income convergence in India. In Latin America, Ferreira (2000) demonstrated income convergence in Brazil between 1970 and 1986. Estimates of convergence were recurrent in the literature across three levels of aggregation: (a) nations, (b) regions of a continent, and (c) states in a nation (Austin & Schmidt, 1998). Nonetheless, Lande and Gordon (1977) found an absence of convergence a universal phenomenon. In particular, recent empirical studies showed an elusive picture for this issue. Much evidence of a halt of convergence or even divergence in the decades of the 1980s and 1990s was found (Cuadrado-Roura, 2001; Drennan & Lobo, 1999). Barro and Sala-i-Martin (1992) observed state income divergence in the 1980s in the U. S. Browne (1989), Carlino (1992), Gamick (1990), also discovered that following many decades of income convergence, the U. S. in 1980s, documented divergence of per capita incomes among regions. In some other countries, researchers reported recent divergence evidence as well. Maxwell and Hite’s (1992) findings disclosed income divergence in Australia between the late 1970s and the mid-1980s. Cuadrado-Roura (2001) observed that regional convergence among EU almost completely ended during the 1980s and 1990s. Ferreira (2000) found that the process of convergence seemed to have slowed down almost to a halt after 1986 in Brazil. The controversial findings challenge the “neoclassical” income convergence model. In pursuit of explanations, two types of responses come to the stage. As Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 36 mentioned already, the first response questions, the methodological legitimacy (Carlino & Mills, 1996; Quah, 1996; Xie, 2002), and the other challenges the validity of the convergence theory (Drennan, Tobier, & Lewis, 1996; Romer, 1994). Besides these two major arguments, several papers discussed that exogenous shocks might disturb the income convergence. Barro (2000) discussed that the energy crises affected income convergence. Coulombe (2000) also argued that oil shock in 1970, disturbed the income convergence in Canada. However, there has been no dedicated study, and in particular, no empirical research on whether information technology has an impact on income convergence. Consequently, little is known about whether an exogenous shock like the emergence of IT is a real significant factor affecting convergence. If it is true, nothing is known on how it affects convergence. In reality, one probable reason that there has been no such research is owing to datum limitations. However, this dissertation research will attempt to investigate this issue, even with such datum limitations. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 37 CHAPTER 3 RESEARCH QUESTIONS AND HYPOTHESIS 3.1 Introduction To build a foundation for the research hypothesis that will be presented at the end o f this chapter, four questions critical to the development of the hypothesis will be discussed. The first question mainly addresses what impact can IT have on income convergence. Using a broad definition o f IT, the researcher assumes that the impact contributes to long-run convergence, but leads to a short-term reversal of convergence. For this assumption, the exogenous (neoclassical) growth approach was adopted, an historical trend toward exogenous development and diffusion of technologies were analyzed, and spatial units such as states and counties, rather than firm units in this research were used. The second question is whether IT constitutes a significant economic shock in terms of productivity growth. This is a critical question as well, because only if information technology is a significant economic shock, is an investigation of its impact on meaningful convergence. Data have shown accelerating productivity growth in the U. S. Almost all recent studies report a significant relationship between IT and productivity growth. IT appears to be a significant economic shock. The third question concerns how spatial scales play a role in income convergence. The researcher wonders whether, for a long run or a long period of time, convergence speeds may be faster at higher spatial levels. This assumption is Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 38 based on three considerations. First, the “neoclassical” logic of competitive market mechanism for convergence, suggests faster convergence at higher spatial scales. In the essence of this logic, the dynamics of leading to convergence, is just the beneficial or exploitable spatial differences that drive capital and other production factors to the poor or peripheral regions. Because the exploitable differences are likely greater at higher spatial scales, as elaborated in detail in the discussion, convergence can be stronger at higher scales. Second, the analytical work on construction of scales— in particular, the exploration of uneven development—also has the same implication. Finally, recent across large distance IT diffusion patterns, intensively reported by economic analysis firms and in mass media, suggests stronger convergence at higher geographical levels as well. For the short-term, exogenous impacts on convergence may be stronger on higher scales. This point was derived from relevant empirical evidence in the literature. The final question is on how trickling-down convergence differs from spreading-out convergence. Based on some of the same reasons for the assumption for spatial scales, the researcher assumes that trickling-down convergence is stronger than spreading-out convergence. Indeed, the assumptions for both of the third and the fourth questions are rather bold. This is because lack o f relevant theoretical and empirical discussions in the convergence literature, cannot provide a sufficient basis for developing assumptions on these issues. Those assumptions for spatial scales are taken in a rather bold way as starting points for empirical testing. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 39 In light of the analysis of the four questions, the research hypothesis is defined as: Speeds of spatial income convergence are relevant to presence of information technology. That is, there is a relationship between the emergence of IT and the change of the income convergence speeds. An elaboration of this hypothesis is that the emergence of IT can reverse income convergence at the beginning, but contributes to convergence in later phases. Also, income convergence patterns vary at different spatial scales. During a long period, convergence is expected faster at higher spatial levels. Impacts of IT are stronger at higher levels than at lower levels, and lead to more intensive swings of convergence at higher levels as well. Furthermore, impact can be observed in both rich to poor trickling-down and center- to-periphery spreading-out income convergences, but the trickling-down convergence is stronger than the spreading-out one. 3.2 Analysis 3.2.1. What is Information Technology, and What Can Be Its Impact on Convergence? 3.2.1.1 Broad definition of IT. This research uses a broad definition for the term “information technology.” One reason for a broad definition o f IT, is that a technological revolution can be composed of several nonseparable components. The Internet requires computers and communication networks. The creation of personal computers is a foundation of the innovation o f the World Wide Web. Also, the Internet is just a “cooked” or “soft” version of communication networks. These components are integrated together to contribute to society and economy. Actually, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 40 this technological shock to the economy consists of the innovations o f PCs, software, communication networks, the Internet, and E-commerce. Thus, this dissertation does not use a narrow concept, but a broad definition covering these essential components. The similar broad definition of IT can be found in the literature (Carlson, 2002; Jorgenson, 2001; Lau, Wong, Chan, & Law, 2001). Moreover, some governmental statistics use almost the same definition (Taskforce of Economics and Statistics Administration, 2000). 3.2.1.2 Exogenous or endogenous development of a technology. In an evaluation of the possible impact of IT on income convergence, there could be two opposite approaches— assuming the development of IT endogenous or exogenous. This dissertation assumes exogenous development of technologies primarily owing to a better explanation o f this approach for the long-term convergence before the emergence o f IT. Besides, there could be additional reasons for applying this approach in an evaluation of IT impact. The first reason is that, with the passage of time, technologies become increasingly exogenous. The opposite assumption—the endogenous model—presumes that technologies are mainly generated from private agents that seek profits. This presumption holds more likely in the past than in the present. We can understand the reason by reviewing not only discovery and invention, but also transfer and diffusion of technologies. In ancient times, or the agricultural era, there were few universities and research institutions. National governments invested little in technological research. In addition, if a technology Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 41 was invented, it was limited to a small group of people and, more important to this topic, confined to a geographic place. There were few media or channels for transfer and diffusion of technologies. We can look at an example. In the agricultural society, textile businessmen in a particular geographic location in Asia invented the production technology o f silk. For centuries, whoever wore silk clothes in Europe, thought that the silk was produced in a particular kind of tree. Only after Marco Polo smuggled some silk worms to Europe, did European textile craftsmen learn how to make silk. Later during the industrial revolution, many universities and research institutions in many regions and countries contributed to the machinery technology. Billions of dollars of government funds went to technological research. However, even for Japanese automakers, they learned how to produce cars many decades after the automobile had been invented. For the information technology, there has been even broader involvement for its invention in terms of spatial and regional entities, educational and research institutions, and nonprivate research funds. Its development is well beyond only the involvement of private firms and funds in a limited place. The first computer was invented at the Massachusetts Institute o f Technology. The Internet was bom in a research project o f the U. S. Department of Defense with three universities, including MIT, Carnegie Mellon University, and the University of California at Los Angeles. The markup language HTML that is critical for the Internet and the World Wide Web, was invented at the University o f Illinois at Urbana-Champaign. Moreover, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 42 this technology has spilled over quickly to both developing and other developed countries, let alone to other regions in the U. S. Thus, technologies have become from more endogenous in the silk era to more exogenous in the “Information Age.” The second reason is that IT possesses some attributes, such as a relatively strong public goods property, favorable to the exogenous assumption. For instance, computer languages are free and available to all users all over the world. There are no patents for BASIC, FORTRAN, C, C++, Java, HTML, SQL, XML, and almost all other widely used computer languages. Even in developing countries, computer programmers catch on to computer languages quickly. Furthermore, many IT products are free and available to all over the world. A quick glance can list Netscape and Internet Explorer (browser), Yahoo and Microsoft Hotmail (e-mail), all Internet search engines, such as Google and AltaVista, many free ISPs (Internet service providers), and free electronic datum storage. Also, hundreds and thousands of free software programs can be downloaded from the Internet or installed from free CDs. This can hamstring the profit-maximizing assumption of the endogenous growth model. Moreover, some source codes, e.g., the source code for Sun Microsystem Inc.’s Java 2 Enterprise platform that is widely used among software engineers, are available and free. Even earlier in the 1980s, IBM opened its PC design protocols and architecture. There can be a question as to why so many IT firms make their technologies public and offer free products. Is this because the computer industry is more Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 43 philanthropic than many other industries? The answer is “no.” The real reason is owing to the nature of this technology—interdependency. For instance, Merk’s medicines do not depend on John & John’s medicines; however, Netscape, Oracle or Sun Microsystem Java rely on operating systems, such as Linux and Windows. In order to use Netscape, Oracle, Word, or other application programs, people install Linux or Windows. Thanks to its nature of interdependency, IT products possess a stronger public good’s attribute than many other products do such as automobiles, medicines, and machines. This nature offers relatively strong explanation power for the exogenous assumption. However, the exogenous assumption can be unsatisfactory in explaining some of the issues of IT development. Still, endogenous advocates may challenge the exogenous assumption from those perspectives. For instance, small start-up companies driven by profit seeking also contribute to the development of IT. For a certain kind of microscope analysis, such as an investigation of firm level impacts, the endogenous approach may have an explanation power in that public contributions may be ignored. Nevertheless, the endogenous approach may be less powerful, if we evaluate income distribution trends using spatial units, such as national, state/ provincial or other geographical units; because contributions of start-ups are less significant in those units than in the firm units, but public contributions are also important. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 44 In fact, the debate on exogenous or endogenous has not been intensive in the literature, in that the importance of a debate is questioned. As stated by Romer (1994), one of the initiators of the endogenous growth theory, it is not a very useful debate on endogenous or exogenous models, because there are circumstances in which each model can be a useful expositional device for highlighting different aspects of the growth process. Due to the long-term evidence, a historical trend toward exogenous development and diffusions of technologies, and use of spatial units rather than firm units for an analysis, this research starts by assuming the exogenous development of technologies. 3.2.1.3 Special impacts of IT—a technology with spatiality—on convergence. The analysis so far has only discussed the generic technological impact o f information technology without viewing its particular spatial feature. Here, spatiality means the feature of technologies with a goal o f dealing with spatial constraints. Technologies help human beings overcome some kind of constraints, some from a spatial perspective, but some others from nonspatial ones. For example, a light bulb extends human usable time; a steam engine extends human physical strength; and movies enrich human lives. Those technologies may have much weaker impacts on convergence than trains, automobiles, and airplanes, which have the feature of spatiality. IT is like the latter, so it is more likely to have an additional impact on spatial income convergence. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 45 Essentially, IT can have a special impact on convergence from two perspectives— one from communications and the other from information acquisitions. How IT affects the distributions of economic activities from these two perspectives is in debate. On the one hand, some argue for a centripetal impact leading to a centralization distribution. For instance, Gaspar and Glaeser (1998) argued that e-mail will increase the possibility of face-to-face contacts, because after e-mailing in their words, people are more likely to meet. Comford, Gillespie, and Richardson (2000) challenged the notion that information and communication technologies (ICTs) mean the death of distance by shrinking time and space. They argued that the ICTs clearly affect spatial relations, but do not dispense with regional disparities and may actually accentuate them. Furthermore, on this side o f the argument, there can be an additional reason, such as some on-line retailing with bricks-and-mortar stores, may be doing better than those without physical stores. One example is that Toys R Us has beaten eToys.com. Thus, physical locations close to centers may still count. On the other hand, there are arguments for a centrifugal dynamic bringing about decentralization patterns of distributions. The Taskforce of Economics and Statistics Administration (1998) asserts that the improvement of communications relaxes the spatial constraints. For this argument, there can be several reasons. First, IT can help firms overcome some spatial and temporal constraints for business activities and operations with brand new means and channels of fast and convenient Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 46 communications within and among firms. Second, IT improves telecommunications. Third, e-mail, Web conferencing, the Internet, Instant Messaging, WebEx and other new technologies, make remote business transactions, partnerships, demonstrations, collaborations, and negotiations much easier than before. In terms of information acquisition, the development and application of IT helps the “neoclassical” argument, because one of the big criticisms o f the “neoclassical model,” is its fundamental assumption on perfect or complete information. This assumption is challenged intensively (Stiglitz, 2002). However, information technology, just as being entitled, is a technology for information. Ray (1996) argued that this new technology is changing the way information is distributed. Lau, Wong, Chan, and Law (2001) claimed that IT is regarded as the information superhighway. It provides many additional and often convenient means for acquiring information, and thus, improves availability of information in a relatively ubiquitous sense. The Taskforce of Economics and Statistics Administration (1998) indicated that news from around the world is now available on the Internet, usually free of charge. Nearly 90% of Web users go on-line to get news and information. With respect to cost savings, the Taskforce of Economics and Statistics Administration indicated that the Internet lowers distribution costs for medium companies, reduces transaction and search costs in ticketing businesses, and in the bank industry, a transaction in a branch can cost $1.07, but a similar one only incurs Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 47 1 cent for Internet banking. Borenstein and Saloner (2001) argued that the Internet and related technologies have caused the costs of many kinds of market interactions to plummet. Bakos (2001) also claimed that the savings on search costs for buyers and sellers are likely to be substantial. One example for the centrifugal argument is regarding Wall Street. The Internet offers convenient venues for people, even in remote places or poor regions in the U. S., to get Wall Street quotes almost immediately. The new technological infrastructure contributes improvement not only for some spatial constraints but also for some time constraints, because the speed of dissemination can be quick. Lots of the information that was formerly constrained spatially on Wall Street—a rich and core area—now can be available immediately in remote or poor regions. People in those regions can get New York stock quotes as quickly as people can on Wall Street. What is more, the cost of getting information via the Internet or other electronic channels can be even cheaper than that from newspapers. Wall Street is losing its previously strong competitive advantage. That is why many other new stock brokerage companies are not located in financial centers, such as E*Trade and Datek mainly conducting transactions via the Internet, can overrun brokerages on Wall Street quickly. The changes of information availability and acquisition move in a direction favorable for neoclassical assumption, consequently, providing additional support for the convergence argument. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 48 In view of the centrifugal argument, firms’ and residential location patterns, can be affected due to improvement in infrastructures for communications and for acquisitions of information. The Taskforce of Economics and Statistics Administration (1998) argues that the improvement of communications contributes to a decentralized pattern of economic activities. Barkley and Keith (1991) found that hi-tech branch plants tend to locate in communities with a highly educated work force and at greater distance from metro areas. They reported that the location behavior o f unit plants, rather than headquarters, better fits the current perception that hi-tech plants select high amenity locations with abundant skilled labor. Similarly, Easterlin (1994) claimed that firms' location patterns are more than ever determined by consumers’ preferences rather than, as before, dictated by the firms' location decisions. Information and Communication Technology (ICT) should make firms more “foot-loose.” When the constraints become much weaker, firms should seek cost cutting by going to wherever land, labor, and other operation costs are lower. This can lead to decentralized economic distributions and spatial income convergence. Still, for both the centrifugal and the centripetal arguments, empirical proof is not strong, owing to either lack of such a proof in the real world or a shortage of empirical studies. We need to examine these points empirically. The investigation on income convergence can contribute to the debate, because convergence stands for decentralized distribution trends driven by centrifugal forces, while divergence Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 49 indicates centralized trends led by centripetal dynamics. For this reason, the statistical tests in this research can be considered as indirect empirical examinations of those arguments. Statistical acceptance of a hypothesis of convergence evidence will support the argument for a centrifugal dynamic and decentralized behavior. Rejection of the hypothesis or a divergence finding will provide evidence for a centripetal force and centralized pattern. 3.2.2. Is IT a Significant Economic Shock in Terms of Productivity Growth? This is a critical question, because only if information technology is a significant economic shock, is an investigation o f its impact on convergence meaningful. In the recent economics literature, there are arguments about whether the development o f IT is a great economic shock. For instance, Jorgenson (2001) views the IT development as one of a series of positive, but temporary, shocks. The competing perspective from Alan Greenspan’s address to the Board of Governors of the Federal Reserve System (2000) is that IT has produced a fundamental change in the U. S. economy, leading to a permanent improvement in growth prospects. The Taskforce o f Economics and Statistics Administration (1998) also argued that the impact of IT is significant in a report based on some governmental statistics. It may be vague to discuss whether information technology is a significant economic shock owing to difficulties of concepts, data, and measurements. A relatively reasonable and measurable discussion can be regarding the productivity growth caused by IT, because one of the primary driving forces for growth of labor productivity is Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 50 assumed to be technology shocks (Baily, Bartelsman, & Haltiwanger, 2001). Namely, we can evaluate whether IT drives productivity growth. This way, we can identify whether IT has significant impact on the economy, too. The contribution of IT to productivity used to be a puzzle or a so-called “IT productivity paradox” (Griliches, 1994; Haynes & Thompson, 2000). In 1987, the Nobel Prize winning economist Robert Solow spoke about this issue. In his words, “You can see the computer age everywhere but in the productivity statistics” (Carlson, 2002; Morrison, 1997). However, the puzzle has been self-resolved in the real world (Hicks & Nivin, 2000). Increasingly recent studies do not support the “IT productivity paradox” view (Brynjolfsson & Hitt, 1996; Haynes & Thompson, 2000; Lichtenberg, 1995). The productivity literature reports accelerating productivity in the late 1990s and the beginning of the 21st century (Baily, 2002). As shown in Table 3.1 or graphically in Figure 3.1, even since the 1980s, the nonfarm productivity reported by the Bureau o f Labor Statistics has been accelerating. The productivity growth before 1973 may be owing to cheaper production inputs, such as energy and some other technological improvements, for instance, mass production in manufacturing. Later, the average annual growth rates of nonfarm productivity accelerated from 0.84% from 1973 to 1982 to 1.81% from 1982 to 1993,1.96% from 1993 to 1999, and then to 2.02% from 1999 to 2001, as indicated in Table 3.2. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 51 Table 3.1 Changes in Non-Farm Productivity in the U.S. (Index—1992 + 100) Year 1969 1970 Productivity 67.9 68.9 Year 1977 1978 Productivity 81.5 82.6 Year 1985 1986 Productivity 89.3 92.0 Year 1993 1994 Productivity 100.5 101.8 Year 2001 Productivity 117.5 1971 1972 1973 1974 1975 1976 71.8 74.2 76.5 75.3 77.4 80.3 1979 1980 1981 1982 1983 1984 82.2 82.0 83.0 82.5 86.2 88.1 1987 1988 1989 1990 1991 1992 92.3 93.5 94.2 95.3 96.4 100.0 1995 1996 1997 1998 1999 2000 102.8 105.4 107.5 110.3 112.9 116.2 Source: Bureau of Labor Statistics, Washington, DC. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 52 Non-Farm Productivity Changes in the U.S. 1 4 0 120 100 o o CN O s O s s c • 1 3 Annual Productivity - o 2 O h OS VO O s r - * OO o s o s o o o s O s O s O s O s O s o o O s o o O S o Year Source: Bureau of Labor Statistics Figure 3.1. Nonfarm productivity changes in the U. S. Table 3.2 Average Annual Growth Rates o f Nonfarm Productivity in the U.S. Time Periods 1969-73 1973-82 1982-93 1993-99 1999-2001 Annual Productivity Growth Rate 3.03% 0.84% 1.81% 1.96% 2.02% Source: Bureau of Labor Statistics, Washington, DC. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 53 Then, why was there a puzzle in the early phase? Some papers have discussed the reasons. The first reason can be that there was a measurement difficulty. It is clear that IT investment was overwhelmingly concentrated in what Griliches (1999,1994) has termed the “immeasurable” sectors of the economy, where output determination is difficult and where productivity effects are correspondingly harder to find (Haynes & Thompson, 2000). Whelan (2002) stated that some o f the early studies reporting IT paradox, did not capture the effect of a unit of computer capital on productivity due to technological obsolescence. That is, those studies examined the machines that were no longer near the technological frontier. Using a model that incorporated obsolescence, Whelan’s research implied a larger computer-usage effect. Also, the failure of measuring the accelerating productivity growth in the early phase, was due to the decline o f productivity growth elsewhere in the economy (Jorgenson, 2001). For instance, some o f the positive impact of IT was offset by the negative impact of the traditional manufacturing technology. The second reason is that the output measurement problem is intensified, since most IT applications are both process and product innovations. That is, they have impact on both the quantity and quality of any good or service being produced (Haynes & Thompson, 2000). The third reason can be a time lag. It has been suggested that radically new technologies require an unusually long lag for full implementation, perhaps requiring complementary investments in human and Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 54 physical capital and even organizational and contractual changes (David, 1990; Haynes & Thompson, 2000). Baily (2002) used an analogy to describe this productivity lag. In his words, the economy, like a large oil tanker, does not change direction or transform itself quickly. Probably the best support for the time lag explanation is the later period accelerating productivity growth. Then, the question is whether IT contributes to accelerating productivity growth. Technological development and improvement has been considered an important part in productivity growth. Smolny (2000) cited that the differences of productivity growth in the industrial countries since the 1950s could be explained with growth models relying on knowledge spillovers, technological diffusion, and convergence toward “best practice” technology. Since 1994, there have been few reports denying an association between productivity and IT. The researcher’s survey of the recent literature on the IT productivity relationship has showed only one denial case, in which a finding of McKinsey Global Institute mentioned by Baily (2002) undermined IT and productivity link in a case study of the Institute. However, more and more findings report a link between the accelerating productivity growth and information technology. Stiroh (2002) summarized that a consensus is now emerging from aggregate growth accounting studies that both the production and the use of IT contributed substantially to the U. S. aggregate productivity revival in the late 1990s. Baily (2002) showed that IT capital accumulation became more important, and all other types of capital less important in Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 55 explaining productivity growth. Having reviewed the studies in the literature, Baily concluded that “there is a rather strong case for saying an important connection exists between information technology and the productivity acceleration.” Also, Whelan’s model (2002) showed that the substantive contribution was to document the role that computers played in the improved productivity performance o f the U.S. economy in the later 1990s. His data indicated that more than half of the labor productivity growth in this period could be explained by computer-related factors. There has been substantial evidence presented suggesting that the productivity acceleration is linked to IT (Baily & Lawrence, 2001; Brynjolfsson & Hitt, 2000; Oliner & Sichel, 2000; Stiroh, 2002). The findings on productivity acceleration and its link to IT support the argument that IT is a significant economic shock. 3.2.3 How May Spatial Scales Play a Role in Income Convergence? This is a difficult question, due to a shortage o f existing work either theoretically or empirically addressing this issue in the income convergence literature. Assumptions have to be developed from this weak foundation. Based on sparsely existing research and some suppositions, it is surprising to note that for long periods o f convergence, speeds appear to be faster at higher scales. Also, for the short-term exogenous on convergence, it may be stronger at higher scales. Here, the word “suppositions” is used for the reason that the foundation in the convergence literature for these assumptions has been weak, but a starting point needs to be set up to analyze the important issue from the weak foundation. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 56 The first assumption is whether the “neoclassical” logic of the competitive market mechanism for convergence suggests faster convergence at higher spatial scales. In the essence of this logic, the dynamic leading to convergence is just the spatial differences that drive capital and other production factors to the poor or peripheral regions. Surely, the differences imply beneficial or exploitable ones, because some differences are below an investment threshold and cannot be exploited for business profits. The latter deters, rather than attracts, capital movement. For example, even if the investment differences between the U. S. and Cuba or between Los Angeles and a Rocky Mountain area are very big capital movement, may not occur because of a political barrier for the former and an infrastructure constraint for the latter. These differences may be called nonexploitable differences, since they are not beneficial or below an investment threshold. The exploitable differences drive movement of capital and then contribute to convergence, so the greater the exploitable differences are, the stronger the movement momentum and convergence could be. Furthermore, the exploitable differences are likely greater at higher spatial scales. This is because, in reality, the difference o f labor costs, land rents, tax rates, as well as laws and regulations, can be greater at higher geographic levels or across bigger distances. For example, if a firm in the Silicon Valley goes to another county in California, it still has to pay state income tax; but if it goes to Washington State, the firm will pay no state income tax. Also, the regions in California suitable for hi-tech companies are suffering with Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 57 soaring salaries, land rents and housing prices, while those costs in Coloradoan or Texan hi-tech areas are still much lower. Actually, if a firm moves across country lines, such as moving to a developing nation, those costs can be even lower. Since the exploitable differences are possibly bigger at higher spatial scales, it is likely that the forces driving the capital movement tend to be stronger at higher scales. Consequently, there might be stronger convergence at higher spatial scales. The second assumption on faster convergence at higher spatial scales, derives from some of the analytical work on construction of scales, in particular, the discussion on uneven development in geography. A fundamental point is that, to some extent, scales are socially constructed (Marston, 2000) in complicated processes, though they can be crystallized into some kind of spatial fixes (Brenner, 2001). The construction of scales or the scaling process plays an important role in production, consumption, geographical differentiation and redifferentiation, and uneven development of an economy (Brenner, 2001; Marston, 2000; Marston & Smith, 2001; Smith, 1984). Thus, as one of the manifestations of social and economic interactions, the social construction process of scales can affect spatial income convergence. Nevertheless, the effect of social and political processes on spatial income convergence has been ignored in the convergence literature. The reason is, as noted by geographer Morrill (2000) on income inequality discussions in the United States, “existing studies of the geographic variability of inequality among regions or the states have all been carried out by economists.” Those studies Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 58 predominately discuss market mechanisms or economic reasons such as competition, capital movement, and other market forces forming and changing income distribution patterns spatially. In a comment on Barro and Sala-i-Martin’s (1991) income convergence paper, Blanchard (1991) even claimed that “Macroeconomists have rediscovered regional economics,” later being criticized by geographers for ignorance o f spatiality. The current economist driven situation in the convergence literature leads to emphasis on economic factors, while noneconomics such as geographical, social, cultural, and political dynamics for convergence, have not been studied to the extent that they deserve. Actually, this situation does not indicate a denial of those noneconomic dynamics, but a consequence o f researchers’ academic backgrounds. Even so, some discussions on the social construction of scales in geography, such as Talor’s (1982,1984, and 1987), Smith’s (1984), and Marston’s (2000) work on uneven development, may help us understand the convergence differentiation with a consideration o f scaling process. Although driven by its researchers’ political views, those discussions can provide some hints for examining the role o f spatial scales in convergence patterns. In his book, Uneven Development: Nature, Capital and the Production o f Space, Smith provided a sketch on how scale is central to uneven development. His sketch, drawn upon Taylor and Wallerstein's (1975) theoretical work, divides world space into three realms: (a) world-economy scale, (b) nation-state scale, and (c) urban scale. Smith argued that the capitalist production Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 59 of the urban scale is largely the result of differentiation. In his view, the global scale is a product of the tendency of the capitalist system toward equalization. In attending to the production of the nation-state scale, he maintained that “the scale of the nation-state is less a direct product o f this contradiction” that the urban scale represents toward differentiation and the global scale represents toward equalization. In terms of the cause, Smith argued that this process is driven by the wage rate differentials that drive the process of the concentration of capital. Less important, though not insignificant, is the existing pattern of labor skills. Marston (2000) furthered Smith's work on this theoretical framework. “In short, the organization of capital into different sectoral divisions— research and development, manufacturing, corporate administration and management—will result in geographical separation.” Interestingly, for this argument, Marston gave an example that relates to this dissertation research. “For example, the research and development activities of the computer industry can be concentrated in locations where technically-trained laborers abound, whereas manufacturing o f chips and assembly of computers can occur in regions with an abundance of unskilled labor.” The spatial differentiation can result in uneven development and income distribution. Smith and Marston’s analyses of equalization at higher geographic levels can reconcile with the researcher’s assumption that stronger convergence trends may occur at higher spatial scales. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 60 Furthermore, the different convergence results on various spatial scales may be expected from an addition perspective of considering the scaling process. In this process, political institutions at higher scales usually have stronger political and legislative powers and many more resources. We may see the differences in the powers and controlled resources by viewing them at the federal, state, county, urban, and neighborhood levels. If equality is a goal o f many policies, there could be an assumption on a stronger dynamic for equalization driven by those powers and resources on higher scales, as well. The third assumption is regarding spillovers of technologies that are relevant to distance. Actually, this one is relevant to the first assumption. There is a phenomenon frequently reported in mass media, but not investigated in the literature. For instance, The Economist (2003), U. S. News & World Report (Benjamin & Perry, 2003), Reuters, CNET and Yahoo! News (Auchard, 2003), CNN.com (2003), MSNBC (Schoen, 2003; Weaver, 2003), Datamation and IntemetNews.com (Pastore, 2003) all report that, currently, many American IT companies in some poles, for instance in the Silicon Valley, relocate their branches to other states or even foreign nations rather than stay where they are, or move to the suburbs or other counties in the same state. The Economist summarizes two features o f the new geo- technological movements. One is that companies are “moving away from the (Silicon) Valley to places such as Redmond, Austin, Armonk and Walldorf (in Germany).” The other is that IT companies are “migrating offshore, mainly to India, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 61 but also to such places as China, Russia and Vietnam. This is already being likened to what happened to manufacturing.” According to Reuters, CNET, MSNBC, and U. S. News & World Report cited above, a study in Gartner Inc. indicates that 1 in 10 U. S. tech jobs may move overseas by the end of 2004. Several decades ago, moving to the suburbs seemed to be a frequently observed behavior, as suburbanization was intensive. But now, the trend appears to have changed. This kind of behavior can be understandable from the analysis of exploitable differences. We may attribute this relocation behavior change to a change of exploitable differences. That is, due to technological improvements in IT, transportation, and some other technologies, formerly nonexploitable differences become exploitable ones for moving to faraway lagging-behind regions. As a result of the across higher spatial scale relocation behavior, spillovers of technology following the movement o f those firms can occur in relatively greater distances and at higher spatial scales. This can lead to an assumption that this form of technological spillovers also brings stronger convergence at higher spatial scales. The three assumptions noted above are for the long-run. For the short-term, there have been some relevant empirical findings. Although not intending to examine differences of convergence at different spatial scales, Cashin and Strappazzon’s work (1998) happened to disclose that, in the time period corresponding to the initial IT age, stronger divergence was found at a higher spatial level states a lower level, such as substate statistical divisions (SDs) in Australia. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 62 That is, a stronger reversal of convergence, which is assumed to be caused by the exogenous shock in this dissertation, was observed at a higher scale. One possible explanation for this different result might be found in Hicks and Nivin’s (2000) so- called “scale-biased” IT capital investment. According to them, IT-induced income gains are eroded if being surveyed at higher levels, though their discussion used two nonspatial levels—the firm level and the aggregated industry level. As a result, at higher or aggregated levels, income differentiation could be stronger. Keller’s study (2002) on technology diffusion has a similar implication. He reported that IT initially is to a substantial degree local, not global, but over time, technological knowledge has become considerably more global. The initial nonglobal technological advantages may lead to more uneven development, and consequently, more intensive income inequality at higher spatial scales, i.e., a stronger divergence or reversal of convergence showing up at higher geographical levels. However, the overtime global diffusion reported by Keller (2002), Hedley (1999), Frantzen (2002), and the mass media mentioned earlier can suggest that, in later periods, stronger convergence may occur at higher spatial scales. In other words, there could be an assumption on a more intensive impact of IT, initially stronger negative impact and later stronger positive impact, at higher spatial scales based on these current findings. Accordingly, we may observe stronger swings of convergence at higher scales than at lower ones. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 63 The long-time period faster convergence and the short-term stronger impact at higher spatial scales, are provided here as a starting point for empirical testing. This research will compare convergence patterns in states, counties, and some central suburban areas to seek clues on the role of spatial scales. The testing results can help us understand whether these assumptions presented here, may gain some empirical ground. Since the spatial scale issue is a weak side in the convergence literature, the tests on these assumptions, can be a first step for an empirical evaluation of the role of spatial scales in the convergence story. 3.2.4 How May Trickling-Down Convergence Differ from Spreading-Out Convergence? The previous discussion on spatial scales already mentioned or suggested differences between the trickling-down and the spreading-out convergences. The exploitable differences among compared areas discussed in the first assumption are applicable to the comparison here as well. Those exploitable differences are usually bigger between poor and rich regions than between central places and their suburbs; because the differences of land, housing, salary and tax rates, are usually greater between the former pairs than between the latter ones. Otherwise, there would have been no “rich and poor” talk provided that the differences between the former pairs were not big. Thus, the corresponding exploitable differences driving the trickling- down convergence may be bigger than those leading to the spreading-out one. Based on the logic that greater exploitable differences could lead to stronger convergence, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 64 there can be an assumption that the trickling-down convergence may be stronger than the spreading-out one. The third assumption in the analysis of the role of spatial scales in convergence is also related to the comparison here. The assumption indicated that across large distance relocation behavior might assist fast spillovers of technologies, and therefore, lead to stronger convergence across big spatial distances and at higher spatial scales. In particular, suburbanization seems not as strong as several decades ago; however, crossing state or country movement of capital seems to become stronger. This may suggest stronger trickling-down convergence rather than spreading-out convergence, as well. 3.3 Research Hypothesis In the previous “3.2 Analysis” section of the research hypothesis, several basic issues— whether IT may have an impact and how the impact might be— have been discussed. As discussed, the research hypothesis was developed primarily from the exogenous framework with a revision. Fundamentally, for a long run or at least a long time span, convergence should be expected. However, technological development could disturb such a trend. Initially, information technology provides some people in particular areas with competitive advantages. This initial disturbance may cause a short-term divergence or decelerating convergence. Gradually, the diffusions and spillovers of IT help noncore or poor areas to compete. Therefore, this should generate a new dynamic of spatial income equalization and, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 65 consequently, accelerate convergence in the later phase of the presence of IT. The assumption of initial decelerating convergence or divergence for each significant technological shock is a difference between this dissertation and the conventional “neoclassical theory.” Also, to make convergence measurable, as in most papers in the convergence literature, this dissertation measures income convergence or divergence with speeds that can be not only fast and slow (convergence) but also negative (divergence). Hypothesis. Speeds of spatial income convergence are relevant to the presence of information technology. That is, there is a relationship between the emergence of IT and the change of the income convergence speeds. A further elaboration o f this hypothesis is that the emergence of IT can reverse the income convergence in the short-term at the beginning, but contributes to convergence positively in the later phase and for a longer time period. Also, income convergence patterns vary on different spatial scales. During a longer time period, convergence is expected to be faster on higher spatial scales. The impact of IT should be stronger on higher scales than that on lower scales and leads to more intensive swings of convergence at higher scales. Furthermore, this impact should be observed in both rich to poor trickling-down and center-to-periphery spreading- out income convergences, but the trickling-down convergence may be stronger than the spreading-out one. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 66 CHAPTER IV RESEARCH APPROACH 4.1 Introduction This chapter discusses the research design, such as treatments to data issues, temporal and spatial units of analysis, research models, and expected results with those models. The major challenge to this research, as well as to all research on income convergence in the United States, comes from the datum side. With the current census or other major economic datum collection and categorization channels, it is impossible to draw hard-and-fast boundaries for the IT sector and single out the data. A feasible treatment is to group the sector with 4-digit SIC data, e.g., a treatment used by the Taskforce of Economics and Statistics Administration of the U. S. Department of Commerce. This categorization is adopted from this Taskforce. The major data are from the Regional Economic Information Systems (REIS) and the Bureau of Labor Statistics (BLS). The “Information Age” is defined and the time span 1969-1999, for which data are available, is divided into subtime units based on three criteria. The first criterion is the social adoption rate of IT, which is a consequence of demand for IT. The second is the development phase, or the product development and production phases o f IT. The third is the contribution of this technology to the economy, i.e., contribution to the GDP (the GDP percentage generated in the industries closely related to this technology) and to the productivity growth. Based on the three Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 67 criteria, the whole time span is divided into three subperiods: (a) 1969-1982, “pre information Age” Years (PIAY); (b) 1982-1993, Information Infrastructure Construction Years (IICY); and (c) 1993-1999, Information Boom Period (IBP). There could be other temporal divisions based on other rationales. For example, economic cycles can be a control variable. Actually, the progress of IT accompanies the ups and downs of the American economy. For example, the two boom periods in the U. S. economy in the 1980s and 1990s were associated with IICY and IBP. Thus, economic cycles have been considered indirectly in the sample periods through the three criteria. However, a weakness in the existing convergence literature is that most o f the studies divide the temporal span by decades or even casually without mentioning why they divided sample periods their ways. This is inappropriate for the income convergence discussion. For the trickling-down convergence testing, states and counties were used as spatial units of analysis. Both are important economic and political entities. In terms of investment, taxation, payment transfer, income redistribution, and many other public policies, they play indispensable roles. Also, county level convergence investigation is limited, and the finding for it is new to the literature. Furthermore, a comparison of these two units can disclose whether income convergence patterns differ at different spatial levels, and if the patterns differ, how the difference can be. For the investigation on spreading-out convergence, the pairs of central counties and their suburban counties in the 50 American largest metropolitan areas were used. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 68 For comparing convergence in those areas with larger IT shares with that in those areas with smaller shares, this research defines IT Intensified Areas (ITIA) vs. Non- IT Intensified Areas (NITIA) with total and partial IT indices derived from Hoover's index o f regional specialization or localization. The major models for trickling-down convergence analysis are p and a models, both used widely in the income convergence literature. The P model tests whether per capita income in poor regions grows faster. The a model examines variations of regional per capita income. For the spreading-out convergence testing, two-tail /-tests of income ratios o f a central county to its peripheries, which can disclose whether there is convergence or divergence, were employed. Furthermore, this research applies linear regression tests on variables representing income convergence to variables representing IT development and application to see whether relationships between IT and convergence could be found. At the end o f this chapter is a summary o f expected findings. 4.2 Data Issues 4.2.1 Measurement of the IT Sector How to categorize the IT sector is a challenging issue. Amazon.com and eToys.com are Internet companies. However, the Barnes & Noble bookstore and Toys R Us are also using the Web to sell books, CDs, and toys. The revenues and jobs from the former companies are counted in the IT sector, but the equivalent in the latter companies, are calculated into the conventional industries such as retail. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 69 This is an issue that is not easy to overcome with current the U. S. statistics. The major reason is that the U. S. statistics use the Standard Industrial Classification (SIC) code that was developed during Franklin Roosevelt’s presidency. SIC uses broad sector categories, but those categories have grown most rapidly over the post- World War II period (Lum, Moyer, & Yuskavage, 2000). The Taskforce of Economics and Statistics Administration (1998) claimed that, as IT goods and services are increasingly incorporated into non-IT good and services, it is difficult to draw hard-and-fast boundaries with SIC classification. As a practical matter, it does not seem feasible to separate e-commerce activities from other activities (Fraumeni, 2001). The issues using current SIC codes and the future North American Industry Classification System (NAICS) for the U. S. statistics are reviewed in the Appendix. However, we have to seek practical treatments to tackle those datum issues. One treatment used in the literature and also in this research, is to aggregate 4 digit SIC data in subsections into a category for the IT industries. There have been some data aggregations for a category that may represent major parts of the IT sector. For instance, the classification by the Department of Commerce is one of acceptable choices. In the report, "The Emerging Digital Economy," prepared by the Taskforce of Economics and Statistics Administration, U. S. Department of Commerce (1998), IT industries are defined as shown in Table 4.1. This aggregation covers the major IT sector, though some o f the above sub sections include more than only IT-related sub-sectors. Since it is infeasible to Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 70 Table 4.1 Information Technology Industries SIC Industry Hardware 3571,2,5,7 Computers and equipment 5045 Wholesale trade of computers and equipment 5734 Retail trade of computers and equipment 3578,9 Calculating and office machines, nec 3659 Magnetic and optical recording media 3671 Electron tubes 3672 Printed circuit boards 3674 Semiconductors 3675-9 Passive electronic components 3823 Industrial instruments for measurement 3825 Instruments for measuring electricity 3826 Laboratory analytical instruments Communications Equipment 3651 Household audio and video equipment 3661 Telephone and telegraph equipment 3663 Radio and TV and communications equipment Software and Services 7371 Computer programming services 7372 Prepackaged software 5734 Retail trade of software 7373 Computer integrated systems design Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 71 Table 4.1 (continued). SIC Industry 7374 Computer processing, data preparation 7375 Information retrieval services 7376 Computer services management 7377 Computer rental and leasing 7378 Computer maintenance and repair 7379 Computer related services, nec Communications Services All 481,4822,4899 Telephone and telegraph Communications 4832 Radio broadcasting 4833 Television broadcasting 4841 Cable and other pay TV services Source: Taskforce of Economics and Statistics Administration, U.S. Department of Commerce (1998). The emerging digital economy. Washington, DC: U.S. Government Printing Office. precisely single out the IT industries with current census channels and data using SIC codes, the classification in Table 4.1 provides a reasonable categorization for the total IT sector. Following the Taskforce o f Economics and Statistics Administration in the U. S. Department o f Commerce, this research uses this mix as the total IT sector. Furthermore, to examine how two IT subsectors—hardware and IT equipment subsector and software and IT services subsector—have impacts on convergence, the total IT sector is grouped into two subsectors—those with SIC Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 72 codes starting with 3 and 5, as the hardware and IT equipment subsector, and those with SIC codes starting with 4 and 7, as the software and IT services subsector. 4.2.2 Major Datum Sources 4.2.2.1 Data from the Regional Economic Information Systems (REIS) REIS is the major income datum source for this research. The Bureau of Economic Analysis of the Department of Commerce collects and prepares the REIS data. Income and employment data are reported by states, counties, Metropolitan Statistical Areas (MSAs), and the entire U. S. The REIS data have some features and limitations. First, since states and counties are relatively stable geographic units, the data are consistent across a long time span. However, the counties are not as small as cities, so the data have the limitation of being disaggregated into smaller units. Second, the REIS data cover several decades and economic cycles, though they are only available at the 2-digit SIC level. Some disaggregated data from other sources— for instance, the Bureau o f Labor Statistics data—only cover shorter time periods. 4.2.2.2 Data from the Bureau of Labor Statistics (BLS) BLS data can overcome some of the shortcomings of the REIS data. The advantage o f the BLS data is that the employment data of the BLS are provided at a 4-digit level. At the 4-digit level, these data are more easily classified into the IT industries and non-IT industries, as defined in Table 4.1 than REIS data. Thus, BLS Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 73 employment data is used for the IT sector. However, the employment data are unavailable for some levels of spatial units. Furthermore, the data are only available for several recent years. This prevents us from doing a longer time period analysis. In addition, the productivity data from the BLS was used, as well. 4.2.3 Long-time Period and Short term Economic analysis focuses on long-run outcomes (Cheshire & Magrini, 2000). Even so, understandings o f the short-term impacts are meaningful too, because short-term impacts are not totally irrelevant to long-term ones. Moreover, analyses o f short-term patterns help us differentiate long-term and short-term features. For a long-run analysis, the constraint is datum availability. Since the IT is relatively new, data cannot be available for a long rim. This problem is, therefore, inevitable. In a convergence study, Cheshire and Magrini encountered a similar problem and claimed that the 16 years of available data for their research covered “a medium run period.” The REIS data used in this research are available for 31 years. Though some papers claim “long range” findings if their data are as long as 3 decades, it can be questionable to define the period o f 31 years to be long term. However, 31 years should cover more than a short-term period and can be considered a quasi-long-term period. For convenience of discussion and to avoid confusion, this research uses the term “long time period” to describe the 31 years in question to contrast it with those unquestionable short-term periods. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 74 4.3 Unit of Analysis 4.3.1 Temporal Units This dissertation defines the “Information Age” and divides the time span into subtime units based on three criteria. The first criterion is social or demographic acceptance of information technology, which can be measured with the adoption rate of IT. The adoption rate is also a consequence of the demand, thus, an indirect measurement of the demand for IT. The second is the development phases of the technology, which is measured with the product development or production phases of IT. The development or production phases reflect the supply o f this technology. The third is contribution of this technology to economy. Two measurements are used. One is the contribution of IT to the portion of GDP. The indicator can be the GDP percentage generated in the industries closely related to this technology. For the other measurement, productivity growth rate is picked up, just because the recent literature reported a strong link of the productivity growth in information technology. Income data for this research are mainly from the REIS. The starting year of REIS data was 1969. Based on three criteria, this research divided the whole time span into three subperiods: 1969-1982,1982-1993, and 1993-1999. Just as mentioned, the first criterion for the division is the adoption rate. The subperiod, 1982-1993, is IT infrastructure build-up years. The first commercial PC kit came out in 1982. Since then, the PC adoption rate has changed quickly each year and reached approximately 50 million in about 10 years (Taskforce of Economics and Statistics Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 75 Administration, 1998). Also, according to the adoption rate, the subperiod starting from 1993 as the information boom period with the feature of rapid development of the Internet and e-commerce is defined. Once it was opened to the general public in 1993, the Internet reached an adoption rate of 5 million in 4 years (Taskforce of Economics and Statistics Administration, 1998). In 1999, the population accessing the Internet was expected to be over 100 million. A research project conducted by Varian and Lyman (2000) at UC Berkeley, reported that between 1992 and 2000, the percentage of hours per year of time spent on the Internet in U. S. households increased by 2,050%. In terms of the three subperiods, the first one was an age of preadoption of PCs and the Internet. The second represented the quick adoption period for PCs. The third period was the fast adoption years for the Internet and e- commerce. The second criterion is the development and production phases of the technology, which actually accompanied with the adoption rates. Before 1982, computers had not been mature enough for widespread personal use. Between 1982 and 1993, millions of PCs were produced for over 5 million people to possess and use. Meanwhile, models, quality, and speeds of PCs were improved dramatically. During 1982 and 1993, both PCs and networks established the physical infrastructure for the “Information Age.” This was the information hardware developmental age. Like the year 1982, as an almost starting point of the development of PCs, the year 1993 was about the beginning point o f the Internet’s development. The Taskforce of Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 76 Economics and Statistics Administration (1998) reported that the number of domain names registered in the Domain Name System rose 5 times from 26,000 in July 1993 to 1,310,000 in July 1997; during the same period, the Internet hosts (i.e., the unique Internet Protocol addresses) jumped from 1.78 million to 19.54 million— almost an increase of 11 times. In light o f these figures, we can say that from 1993 to 1999 the U. S. realized the Internet revolution. The year 1999 was a peak. Starting from 2000, the Internet and related industries experienced an evident slow-down; and subsequently, the U. S. economy declined (Baily, 2002). The third criterion is the contribution of the technology to the economy. Although the data on computer and other IT industries’ contribution to American GDP from 1982 to 1993 were unavailable due to the census classification, during that period, the computer and other IT industries must have contributed to the economic growth significantly. Some documentation shows that, during that period, computer and other IT industries were one of the major powerhouses for economic growth (Taskforce o f Economics and Statistics Administration, 1998). Thus, the period 1982-1993 should have been supported by GDP data, if such a categorization in GDP had been available. For the later period, data are available. According to the census data, the IT development was the most important source contributing to economic growth among all growth factors between 1993 and 1997, based the data available up to 1998 (Taskforce, based on Bureau o f Economic Analysis and Census data, 1998). A narrow definition of IT industries that did not include many e- Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 77 commerce industries (such as Internet retailing, financing, wholesaling, and B-to-B) was responsible for more than one-quarter of real economic growth (Taskforce of Economics and Statistics Administration, 1998). If those excluded relevant IT industries had been included, the information technology and related industries could have accounted for much more than 25% of the real economic growth in the U. S. during that period. The Taskforce of Economics and Statistics Administration also marked 1993 as a pivotal point. In their words, the “next spurt” o f rapid economic growth after PC development “started in 1993, with the burst of commercial activity driven by the Internet.” Furthermore, Figure 1 shows that 1973,1982, and 1993, seemed to be pivotal points separating the productivity growth into segments graphically. As indicated in Table 3.2, during the segment of 1973 to 1983, the average annual growth rate of nonfarm productivity was only 0.84%. Between 1982 and 1993,1993 and 1999 and between 1999 and 2001, the rates accelerated from 1.81 % to 1.96%, and then to 2.02%. Because the interest of this research is Information Technology, the periods before 1982 were combined as the pre information period. The periods of 1982-1993 and 1993-1999 are the most important in this research. The period after 1999 is so short, that not enough REIS data for this period were unavailable. Thus, based on the three criteria, this research divides the “Information Age” into two major periods or phases: (a) 1982-1993, Information Infrastructure Construction Years (IICY), and (b) 1993-1999, Information Boom Period (IBP)— Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 78 and entitles the years 1969-1982 as “pre-information Age” Years (PIAY). Unlike those o f many other great technologies, the development and application of IT seems to involve dual progress or two peaks— one with the information infrastructures and instruments and the other with the Internet and e-commerce. In the IICY, the information economy was driven mainly by the production and construction of information infrastructures, such as computers and networks. In the IBP, the information economy was mainly empowered by the derived usage of the infrastructures, such as e-commerce and Internet communications. This research investigated what the income convergence patterns were like and what impact IT had on income distribution during these two special periods. Furthermore, the income convergence rates among the IBP, IICY, and PIAY periods will be contrasted to identify the features in the “Information Age.” It can always be debatable on how to divide sample time periods in economic research. There could be other temporal divisions based on other rationales. For example, there could be an argument for using economic cycles as a control variable for sample periods. Actually, the progress of the Information Technology accompanies the ups and downs of the American economy. For example, the two boom periods in the U. S. economy in the 1980s and 1990s were associated with IICY and IBP. The three criteria applied here are closely related to economic cycles. The first one is a footprint of the demand. The second reflects the supply. The third Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 79 directly links to GDP and productivity growth. Thus, the economic cycle has been considered indirectly in the sample periods through the three criteria. In the convergence literature, most authors do not discuss how they divide their sample periods, which indicates that they may not consider the division important. This is inappropriate. Some of them picked up each decade such as 1970-1979 and 1980-1989 (McCoskey, 2002; Carlino & Mills, 1996; Drennan, Tobier, & Lewis, 1996; Cardenas & Porton, 1995). Some others (Cuadrado-Roura, 2001; Ferreira, 2000; Rey & Montouri, 1999; Bimie & Hitchens, 1998; Kangasharju, 1998) seemed only to divide time spans casually without mentioning why they divided sample periods their ways. Nevertheless, the three criteria for temporal division in this research— adoption rate, development phase of the technology, and contribution o f this technology to GDP and productivity growth—are important in the research regarding the impact of the information technology. This division is more legitimate and reasonable than at-will temporal division or than just dividing by each decade. In addition, in this research, there is one more safeguard for any possible temporal division bias: a time division is not used in the a convergence test— one o f the two trickling-down convergence tests. The time unit for the a test is each year, so the a test is immune to any possible bias in the temporal division. To further take advantage o f this safeguard, while using the division for the /-tests, this research also examined the income ratios for the spreading-out convergence test for each year. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 80 4.3.2 Spatial Units 4.3.2.1 States, Counties, and Central-Suburban Areas For trickling-down convergence tests, two geographical units— states and counties in the U. S.— are used. In terms of states, considerable research on spatial convergence has used them as the unit of analysis. One of the reasons for using states as a unit of analysis, is that they are important economic and political entities. In terms of investment, taxation, payment transfer, income redistribution, and many other public policies, states play indispensable roles. Besides, data for state level incomes and economic activities are collected for a relatively long period, and the data quality is thought to be reliable in contrast to that for many other spatial units. There is another reason that this research uses states as a unit of analysis. Since a great deal o f research has been done on the income convergence question used states, the state level convergence test in this research can be compared with various previous findings. Counties are a microspatial unit relative to the existing empirical work on convergence. Like states, they are important economic and political entities, too. Also, the county convergence patterns can be compared with those of state convergence so as to identify the difference in convergence at different levels. For county convergence, this research goes further to narrow down the spatial comparison framework, from all of the U. S. to each state to see, if counties in a smaller comparison framework are examined, what convergence pattern can be observed. For two reasons, this test was extended. First, the test can show whether Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 81 county convergence differs if the spatial comparison framework is different. The second reason is that, later, the regression tests will use the output from the within- state tests as an input to test a spatial association between IT and income convergence. For the study of spreading-out convergence, the pairs of central counties and their suburban counties in the 50 American largest metropolitan areas are used. This research examined income differences and their trends for the 50 largest American places ranked by the 1999 population. This research applies the geographic terms in Gordon, Richardson, and Yu’s paper (1998) such as a central place, Ring I, and Ring II. The central place is the central county o f an MSA area. Ring I is defined as an MSA area deducting the central county. Ring II is the adjoining counties of an MSA area. The per capita incomes of the central counties are compared with those of their Ring I ’ s and Ring I I ’ s, respectively. In this research, the 50 largest places are categorized into five groups: (a) Ist-lOth, (b) 1 lth-20th, (c) 21st-30th, (d) 31 st-40th, and the (e) 41st-50th largest places. Per capita income ratios of a central county to its suburban counties in both Ring I ’ s and Ring I I ’ s are tested. These tests can identify the trend of income distribution from cores to their surrounding areas. The 50 largest places cover all major hi-tech regions in the U. S., as shown in Tables 5.6 and 5.7 in chapter 5. Both Silicon Valley (San Jose) and Silicon Alley (Boston), as well as Long Island, Los Angeles with Orange County and Dallas, are in the first largest group. Denver, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 82 Seattle, Phoenix, and Pittsburgh fall into the second group. Portland, Oregon, where Intel and several key IT companies are located, is in the third group. Charlotte and Chapel Hill around or in the North Carolina Research Triangle area are in the fourth and fifth largest groups. With the 50 largest places, it is likely to catch the IT impacts in the central and suburban cases, since almost all leading IT areas in the United States fall into those 50 places. 4.3.2.2 IT Intensified Areas (ITIA) vs. Non-IT Intensified Areas (NITIA) The ITIA and NITIA units are for an investigation on a spatial association between the development and application of IT and income convergence. All of the tests discussed previously will mainly analyze whether there are such associations temporally. The objective of the test here is to understand whether there is such an association spatially. The reason for doing this is that, although a country-like the United States is in the “Information Age,” this does not mean that all geographic areas are entering the Age at the same pace. Some areas move fast, and some others lag behind. This research compares income convergence evidence between the information technology intensive areas (ITIA) such as hi-tech regions, and information technology less intensive areas (NITIA). This way, we can evaluate whether the development and application of IT accounts for convergence patterns spatially. Moreover, an ordinary least square (OLS) regression of convergence proxy to IT index of a state will be employed to examine if there is a spatial correlation between IT presence and convergence. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 83 This research defines IT indices based on Hoover’s (1936) index of regional specialization or localization. The Hoover index is defined as: Lij = (Ejj/Ejus)/(Ej/Eus), (4.1), where Ly is the localization index in industry; I for region j; Ey, employment in industry; I for region j; Ej, total employment in region j; Ejus, employment in industry I; and E us is total employment in the United States. Based on the Hoover index and the IT sector classification in Table 4.1, this research defines a total IT index, Ir.t, to differentiate the ITIA and NITIA areas. Ir.t is defined as Ir-t = (Jr/Jus)/(Tr/Tus), (4.2), where Ir-t is the jobs in the IT industry in region r; Jus, the IT jobs in the U. S., Tr, the total jobs in region r; and Tus is the total jobs in the U. S. A value of Ir-t larger than 1 means that region r has a higher IT job percentage than the U. S. average, and a value of less than 1 stands for a lower IT job percentage than the U. S. average in region r. Then, in light of the IT index, the ITIA and NITIA areas are identified. The partial IT indices— Ir-he for hardware and IT equipment and Ir.S s for software and IT services— are defined as: Ir-he = (Jr-he /Ju s)/(T r-he /T us), (4.3), and Ir-ss = (Jr-ss /Ju s)/(T r-S s /T us), (4.4), where the subscripts r-he and r-ss represent the hardware and IT equipment sub sector and the software and IT services subsector, respectively. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 84 4.4 Models 4.4.1 p and a Models One o f the widely used models to evaluate income convergence in the literature is P convergence. This model applies a regression equation in which income growth is regressed against the initial level of incomes. The rationale behind the model is that, if the incomes in poor regions grow faster than those in rich ones, income differences between the poor and the rich should be diminishing. However, the model is criticized for its possibility o f Gabon’s fallacy of regression.3 In fact, the P convergence examines changing growth rates of regional incomes instead of changing inequalities of regional incomes. Barro and Sala-i-Martin (1991) provided a full economic analysis and illustration of the P model. Here, the p model in a shorter form is illustrated. We start with the Cobb-Douglas production function: YT = A«Ka (4.5), where Yt is the final income state, here, the final regional per capita income; A and a , two constants; and K is the initial capital stocks that are determined by the initial income level. We assume that “ Galton's Fallacy o f Regression is named after Sir Francis Galton (1822-1911). Sir Francis found that the persons who have tall fathers, tend to grow shorter than their fathers and the persons who have short fathers, tend to grow taller than their fathers. Their heights tend to regress to the mean o f the population. Thus, Galton falsely concluded that the dispersion of heights across the population diminished over time. However, in the real world, this is not the fact. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 85 -pT Ka = Yt0 e (4.6), where Yto is an initial state, here, the initial regional per capita income; p, a coefficient; T, a time span; thus, we have: -PT Y t - A »(Y t0 e ). Dividing by Yto, we get -PT YT /Yto = A .l/(Y t01-e ), and Log(YT /Y,o) = Log(A) - (1 - e-p T ).Log (Yt0). We divide the equation by T and replace Log(A)/T with another constant, B. Subsequently, we get (l/T).Log(Y T /Yt0) = B - (1 - e-pT ).Log (Yt0)/T (4.7). This is the widely used p Model in convergence or divergence tests. If there is convergence, we must expect a statistically significant p, so that with a lower initial state Y to, a region must grow faster to catch up with richer regions at the end of the time. The absolute value of p indicates the speed of convergence or divergence. Also, the longer the time span T, the smaller the effect of the initial state Yto on the growth rate of the final state Y j. Barro and Sala-i- Martin (1992) and many others applied this model to their research. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 86 Because the ( 3 Model is criticized for the possibility of Galton’s fallacy of regression, another model— c-convergence— is also widely used to correct a possibly misleading understanding obtained from P-convergence tests. The o- convergence examines if the dispersion in incomes among places diminishes over time. However, a “switching” issue is a concern. By “switching,” it is meant that a poor member switches to a rich one, and a rich member switches to a poor one. The “switching” distorts the convergence story, too. The a model could expose the “switching” issue, while the variations hold constant or even become smaller. This can be a disadvantage relative to the P test. That is, even if there is no real convergence, the a values can falsely show a convergence pattern. In applying the or model, Barro and Sala-i-Martin (1991) used the standard deviation for the log of per capita personal income for 48 U. S. states (excluding Hawaii and Alaska). Similarly, Coughlin and Mandebaum (1988) analyzed income convergence with the coefficient of variation cr, a = V £ (y, - y)2 /N (4.8) where yi is the log of regional per capita income; y, the mean value o f the log o f per capita income of the whole set, and N is the number of regions. This research applies the two models— p and a —in the statistical work on income convergence at both the state and county levels. As discussed previously, the P test can be subject to Galton’s fallacy of regression, and the g model can expose to Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 87 the “switching” problem. A combination o f the two models can help to correct the bias caused by each approach. 4.4.2 T-Tests To examine whether there are changes of per capita income ratios between two compared center and periphery units, this research applies two-tail t-tests for the hypothesis: p0 = P t (4.9), where p is the per capita income ratio between the two compared units, the subscript 0 means the initial year and T, the final year. The p is defined as P = yc/ys (4.10), where yc is the log of per capita income of the central county in place;, and ys is the log of per capita income of surrounding counties, in either Ring I or Ring II, in place;. The null hypothesis is that there is no change of ratios between the initial year and the ending year. That is, there is neither convergence nor divergence, because the per capita income ratios between the compared units— central and suburban areas— hold constant. If we reject this null hypothesis, we can conclude that there is a significant change o f income distribution between the compared center and periphery units. Then, in the next step, we can identify whether there is convergence or divergence based on whether the ratios become larger or smaller. Thus, the Mest examines not only whether there is convergence but also if there is divergence. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 88 4.4.3 Linear Regression Tests For two issues, linear regression tests are employed. The first issue is whether income convergence is associated with IT presence spatially. Regression tests are employed to test relationships between P convergence and IT indices. The second issue is whether income convergence links to productivity growth. In two linear regression tests, the relationships between productivity growth rate and state- county g change, and between productivity growth rate and county a change are examined. 4.5 Expected Findings 4.5.1 Tests on Income Convergence The statistical work consists of many tests. To help to read the statistical work, Table 4.2 summarizes the expected findings for the research hypothesis. For all of the spatial units, there is convergence for the long time o f 31 years. The expected result in the IICY is divergence or decelerating convergence. Then, for the IBP period, there is accelerating convergence. This is a comparison in the temporal framework. In terms of a comparison among spatial units, for the long time, state convergence should be faster than county convergence, as the hypothesis expects faster convergence at higher scales. Furthermore, the convergence swing at the state level is more intensive than that at the county level. However, both state and county convergences, i.e., trickling-down convergence, are expected to be stronger than the centers-suburbs convergence during the 31 years. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. Table 4.2 Expected Findings for Income Convergence and Its Changes 31-year Period Initial Period Recent Period Unit of Analysis Methods Pattern Comparison Pattern Comparison Pattern Comparison States P & a C Faster than county C D or decelerating C Stronger swing than county one Accelerating C Stronger swing than county one Counties P & a C Slower than state C D or decelerating C Weaker swing than state one Accelerating C Weaker swing than state one Centers-Suburbs T-test C Weaker than trickling- down C D or decelerating C Accelerating C C = Convergence, D = Divergence oo VO 90 4.5.2 Tests on Statistical Relationships Table 4.3 indicates the expected findings for the research hypothesis in linear regression tests. For the research hypothesis, statistical associations between ITIA/ITLIA and convergence speeds or between annual productivity growth rate and convergence a value are expected. Table 4.3 Expected Findings in the Relationship Tests Variable Convergence Pattern Method a Level ITIA/ITLIA Areas Significantly Related Linear Regression Test 0.1 Productivity Significantly Related Linear Regression Test 0.1 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 91 CHAPTER 5 ESTIMATION AND DISCUSSION 5.1 Introduction This chapter reports and discusses results from p tests, a tests, two-tail /-tests, and regression tests with the models, data, and units of analysis explained in chapter 4. The p testing shows that, per capita incomes among states converged during the longer time of 1969-1999. However, the convergence speed of 0.05% per year was slower than 2%—the rate Barro and Sala-i-Martin (1992) and other researchers reported for the decades before the 1970s. During the development and application of IT, incomes initially converged at a rate of .0012%. Later, incomes converged at a rate o f .0040%— over three times faster than they did initially. These findings support the research hypothesis. For counties, p convergence in two comparison frameworks was tested, throughout the U. S. and within states, to see whether comparison frameworks matter or not. The testing on counties throughout the U. S. shows that, for the longer period, there was negligible divergence at a rate of almost zero. During the “Information Age,” initially, there was divergence at a rate of -.0017%. Later, incomes converged at a slow rate of .0008%. These findings support part o f the hypothesis, i.e., later period accelerating convergence. For counties within states, county per capita incomes converged significantly within 45 states during the longer period. There was no evidence of divergence. In the “Information Age,” many Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 92 divergence cases appeared. The findings support the hypothesis on the longer time period convergence and on a reversal of convergence in the presence o f IT. The or testing provides stronger support for the research hypothesis than the p testing. For the longer periods, there was a convergence for both states and counties. The a values decreased from 0.1707 to 0.1517 between 1969 and 1999 for states; meanwhile, the a values dropped from 0.2474 to 0.2256 for counties. During the “Information Age,” the a values became larger initially but later dropped somewhat. These results support the research hypothesis on longer time period convergence, and initial divergence but later convergence in the “Information Age.” In terms of various convergence patterns at different spatial scales, comparisons between state and county of all p and a convergence testing show that, for longer periods of time, state level convergence was faster than county-level convergence; and during the development and application of IT, the fluctuation of state level convergence was more intensive than that of county level convergence. These results support the research hypothesis on faster long-term convergence and stronger fluctuations on higher spatial scales. An additional finding beyond the research hypothesis is that per capita incomes are more evenly distributed in larger spatial units such as states, than in smaller spatial units, such as counties. For spreading-out convergence, not all of the two tail /-tests show any statistically significant convergence or divergence between central counties and their Ring I ’ s and between central counties and their Ring I I ’ s. However, nonstatistically Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 93 tested observations indicate initial divergence, later convergence, and longer-term convergence. Also, IT impacts seem to be stronger on the pair of center and the outer suburbs, and lag behind in smaller places for the center-periphery convergence. Furthermore, for additional information on a possible link o f convergence to IT, OLS regression tests show that, during the two periods of IT development and application, there were significant relationships between a state’s convergence pattern and a state’s total IT index, and between the convergence pattern and a state’s hardware and IT equipment index, but no relationship between the convergence pattern and a state’s software and IT services index. Also, regression tests show significant reverse relationships between state and county a values and productivity growth rate that is reported to link to IT. Though it would have been more supportive if a direct relationship between IT shock and convergence could be tested, it is difficult, if ever possible, to quantify the IT shock. Those indirect tests were, therefore, employed via IT indices and productivity growth, to seek further information on the relationship between income convergence and the IT shock. These indirect tests provided further information, much of which is supportive, for the research hypothesis. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 94 5.2 p Convergence 5.2.1 States: Initial Period Decelerating, Later Period Accelerating, and Long-Time Convergence Cases As indicated in Table 5.1, the longer time between 1969 and 1999, per capita incomes among states in the U. S. converged. This convergence is statistically significant with a /(-value smaller than 0.0001. However, the speed o f 0.05% was slower than 2% around which Barro and Sala-i-Martin (1992) and other researchers reported for the decades before the 1970s. For each of the three subperiods, the p coefficient is statistically significant, too, with eachp -value smaller than 0.0001. During 1969 and 1982, per capita incomes diverged at a speed of 0.1531 %. This confirms the divergence findings in the literature, and the cause might be the energy crisis, as argued in some papers. Later in the IICY from 1982 to 1993, per capita incomes converged at a rate of 0.0012%. In the IBP, 1993-1999, incomes continued to converge at a speed of 0.0040%, over 3 times faster than that during the IICY. The findings here support the hypothesis for a longer time of income convergence. Though no divergence was observed in the initial years of the “Information Age,” the convergence speed was slower in the initial years; but in the later years, the convergence speed accelerated. This observation also supports the research hypothesis. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 95 Table 5.1 P Convergence o f States in the U. S. Time span P value p-value Significance (a = 0.01) 1969-1999 0.0500% <.0001 Yes 1969-1982 -0.1531% <.0001 Yes 1982-1993 0.0012% <.0001 Yes 1993-1999 0.0040% <0001 Yes Date Source: REIS 5.2.2 Counties 5.2.2.1 Counties throughout the U. S.: Initial Divergence, Later Convergence, and Extremely Slow Long-time Divergence Cases. As shown in Table 5.2, from 1969 to 1999, there was negligible divergence evidence. The divergence speed was 0.0047%. This observation does not support the research hypothesis concerning long-time convergence; however, it is hard to support a contrary hypothesis, because the divergence speed was extremely small. In the PIAY from 1969 to 1982, the divergence speed was higher at a rate of 0.0108%. During the IICY, there was divergence evidence, but at a rate of 0.0017%, which was a slower rate than that in previous years. In the IBP, the county per capita incomes converged, also at a small rate of 0.0008%. Only in the Information Boom Period did the per capita incomes converge. The finding at the county level supports part of the research hypothesis. That is, the income convergence is slower or even Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 96 Table 5.2 P Convergence of Counties in the U . S . Time span ( 3 value p-value Significance (a = 0.01) 1969-99 -0.0047% <.0001 Yes 1969-82 -0.0108% <.0001 Yes 1982-93 -0.0017% <.0001 Yes 1993-99 0.0008% <.0001 Yes Data Source: REIS negative in an initial phase, but faster in a later phase. The later phase’s convergence finding can imply that, if we extend the time, or if we have longer time period data, a long-range convergence may be observed. 5.2.2.2 Counties within State: Fluctuating Initial and Later Period Convergence, but Almost All Long-time Convergence Cases. The discussion above is a picture regarding county convergence within the whole country, i.e., the United States. The spatial comparison framework was narrowed down for county convergence, from within the whole United States to each state, to see what convergence picture appeared in a smaller comparison framework. Table 5.3 shows interesting statistical findings. For the longer periods of time, in all o f the states, there was no county divergence evidence. County per capita incomes converged significantly within 45 states among the 50 states. Even for the five remaining states, their county p convergence values are positive (possible Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. w ith permission o f th e copyright owner. Further reproduction prohibited without permission. CD T3 “5 o Q . c o C D Q . Table 5.3 P Convergence of Counties within State State 1969-99 p p- value Proxy 1969-82 p p- value Proxy AL 0.000161 1 -0.000295 -1 AK 0.000085 0.0042 AZ 0.000157 0.0015 AR 0.000256 CA 0.000134 CO 0.000089 CT 0.000500 0.0489 -0.000083 0.0293 -1 -0.000044 0.5198 0 -0.000145 0.0012 -1 0.000125 -0.000057 0.0135 -1 0.000145 0.1151 0 DE 0.001784 0.6695 0 0.000285 0.2501 0 1982-93 P p -value Proxy 1993-99 p p -value Proxy -0.000038 0.0208 -1 -0.000042 0.0100 -1 -0.000051 0.0077 -1 -0.000070 0.0045 -1 0.000049 0.0883 1 0.000030 0.3246 0 -0.000006 0.7120 0 -0.000008 0.6112 0 0.000081 1 0.000063 1 0.000026 0.0007 1 0.000015 0.0508 1 0.000045 0.1592 0 0.000030 0.1883 0 0.000156 0.3139 0 0.000197 0.4981 0 T ab le 5 .3 (continued). 98 S' O t" 00 VO © © © VO 3 3 00 <N © © VO os © © r- © © © cn © © © o s o s I C *") o s o s VO © © © © © < * * > © © © © © CN CN © © o o © © © © © © © © r - © © © © © © © x * p < L > 3 > ^ cn o s m s s CN v s © VO © © Os • <N 00 Os CN « —1 8 8 © © © © © © CN T f © © © © © © © © CN VO © © © © m VO 8 © © m vs © © © © S' p r- o o r*S © VO c n Os © © © v > 3 VO r * - CN © © © CN 00 I o s VO O n <N CN 8 S © © CN 00 © © CN CO © © © © © © © © v s © © © © © © X p < D 3 * c 3 > m © r^ <N Os OS i Os VO Os ^ so — CN © © © © © © CN r*S © © © © VS © © © © Os CO © VO © © © © © © © o s © © © © s CO U * < O co U > Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 0.000138 1 -0.000145 - 1 -0.000012 0.1395 0 -0.000032 0.0004 99 £ < L > > © © © © © 1 © © © © © c n v s v s CN © r- Os c n <n Os c n VO © 00 © v s OS Os v s CN VO 00 v s 00 VO © CN Os v s r f © v s VO © c n CN VO c n c n © © © © © © © © © © © © O c a Os Os i c n Os Os O o o o © © CN v s CN © © © © © © © r - * © § © © v s © VS v s © © © © p «* CN VO c n c n c n VS 00 c n v s v s c n 0 0 c n r ^ © © © r - © vo © © © v s © v s CN c n © © © © © © © © 0 0 0 0 c n © r - © c n © © © © © ca c n Os i CN 0 0 Os cn c n © © © © © vs CN © © 8 © T f © © © © © Os c n r - © © © © © © © v o CN 8 © © 00 © © © © © s © © 8 © © © CN © © 00 o s t- © © © © © © © © © © S ' p < L > 3 > r- v s 00 c n c n v s © vo r-* © © © © © © © CN © r- c n r-* © © © © Os T f . CN © Os © © © ca CN 00 I o s SO o s s © o © VO CN *— © © © © © © © © © © VS CN c n © © © 8 8 VS n- © © © r - * v> © © © S 8 S 8 8 8 c n r - © © T f — * 00 © © © © © © © $ 3 c 1 u cn v s u i Os Os i Os VO Os £ Os c n © © © © © © © © c n r - CN © v s CN c n © © © © © © © © © © © © 00 rr r- © © 8 8 00 00 © © © < * > s r^ - CN v s © — s o CN © © © © © © V ) CN 00 CN © © © © © © Os t - * H 2 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission 0.000112 1 -0.000100 0.0161 - 1 0.000015 0.3002 0 0.000007 0.6609 T able 5 .3 (continued). 100 S ' p O N CN © vo m cn V O CN O N s o © 00 C^ r - 00 r - » © © CN © 00 00 00 © s © © m r * - © © © © © © © © © 3 o o © V O i n © i n © r*- c n © 00 © so © cn © © © © © on O N On ON ° § 8 I © £ d © r * * 2 S g © £ 8 S . © 9 o o in o o o o © © C N m © © C N © © C N © © © © in so 8 S © © © © © © © © r - m l-N O © * - l © © © © © © © © S ' p © « 3 1 3 > cn © 00 © oo O N 00 m W - * © r - cn © 00 cn «n © » — cn i n © © © © © c n so © m 00 ON c n in © © © © © © co ON i C N 00 O N 00 C N C N — © © © © © © © © © © © © © I-* © cn © © © m so ON © © © © d © © CN O N 00 CN i n c n © § © © © © © © © © © © © £ p cn © m r - cn in cn in . on © © c n © © © © 3 © CN T j* cn CN 00 © CN r* 00 O N C N vo r- © 00 © © © © © © © ca C N 00 I ON V O On 00 00 © © 8 © © © § 8 S © 2 § o s 00 i n O N i n 1 00 © © © 8 © © 8 8 8 8 8 © 1 © i © © 1 d i © © © © © © C N 00 cn On © © 8 8 V O © © 8 C N © © © 5 ? p O n On i ON vo ON © © © ON SO © © © © © * * CN ON c n CN CN VO © © © r- vo © © © O N © © © © © © © i n CN CN V O cn SO ON O N CN SO m cn © 00 00 © 00 i n 00 © CN © CN © © © © © © © © © © © © © © © © © © © © © © © © © © © © © © © © © © © © © © © © © © © © © © d © © I S o % s c o o 0 6 O ss u C /3 Q C /3 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. w ith permission o f th e copyright owner. Further reproduction prohibited without permission. CD ■ o - 5 O Q . Table 5.3 (continued). State 1969-99 P p-v alue Proxy 1969-82 P /7-value Proxy 1982-93 P p-value Proxy 1993-99 p /7-value Proxy VA 0.000132 1 -0.000031 0.0992 -1 0.000014 0.0421 1 0.000007 0.3437 0 WA 0.000034 0.0606 1 -0.000095 0.0260 -1 0.000045 0.0054 1 0.000030 0.0657 1 WV 0.000149 1 -0.000111 0.0044 -1 -0.000002 0.8713 0 -0.000037 0.0178 -1 WI 0.000168 1 -0.000086 0.0036 -1 -0.000019 0.1196 0 -0.000015 0.1655 0 WY 0.000073 0.0045 1 -0.000029 0.5391 0 0.000055 0.0023 1 -0.000003 0.7701 0 Note: P-value is smaller than 0.0001, if the cell is empty. Date Sources: REIS and BLS Proxy Value = 1 (convergence), -1 (divergence), 0 (neither) 102 convergence), though the values are not statistically significant. The evidence supports the research hypothesis that, for longer time periods, county incomes converge within their own states. During the PI AY, county per capita incomes diverged within most states. There were only five exceptions— California, Massachusetts, New Jersey, New York, and Vermont. All of them are either located in the Pacific West or in New England-Middle Atlantic. Interestingly, they constitute two large geographic regions. One is California, and the other is a region of almost similar size with four adjoining states. Both regions are coastal and major industrial renovation areas (manufacturing in the past and IT currently). Two big IT innovation areas— Silicon Valley and Silicon Alley—are located within these two regions. In the IICY, 24 of the 49 U. S. states showed income convergence among their own counties. Only two states—Alabama and Alaska—showed divergence among their counties. All other states neither converged nor diverged. They stayed in a rather stable state. The picture in the IBP, 1993-1999, is quite complicated. In most states, there was neither convergence nor divergence, even if an a value as big as 0.1 was chosen. There was convergence evidence within fourteen states, while divergence appeared within seven states. The seven divergence states were Alabama, Alaska, Kansas, Kentucky, Mississippi, West Virginia, and Texas. Except Texas, all o f the states are agricultural states. Actually, Texas is also an agriculture heavy state. Moreover, the divergence rate of 0.000016 within Texas was the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 103 slowest among the seven divergence states. The divergence rates within all other divergence states were at least double this rate in Texas. Therefore, the divergence in the IBP might not be due to the IT impact. This observation may imply that those states with the higher percentage of IT sector might have diverged like those agricultural states, if these states had not developed the IT. The strong divergence in agricultural states and the simultaneous convergence in the states with more IT industries in the later Information Age can support the research hypothesis that the IT contributes to convergence eventually. Some questions remain. For the four periods in question, i.e., one long-time period and three short subperiods, four states showed convergence evidence in all of the four periods. They are California, New Jersey, New York and Vermont, which are located in two major coastal and leading industrial regions, as mentioned above. California has Silicon Valley. The other three states are close to Massachusetts, where Silicon Alley was formed. Massachusetts converged in the 31-year period, though it neither converged nor diverged in the IICY and the IBP. An interesting finding regarding Alabama and Alaska is that they showed all three subperiod divergences and were the only states of this kind; however, they showed convergence for the total time span. This observation may be an example of Galton’s fallacy o f regression. This finding implied that, even if we saw divergence evidence for all shorter periods, for the longer times, incomes could converge. The Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 104 short-term divergence may not stand for a long-term divergence due to Galton’s fallacy (Barro & Sala-i-Martin, 1992; Bliss, 1999; Quah, 1993,1996). 5.2.3 A Comparison of State and County P Convergences Comparing the state p convergence with within U. S. county P convergence, we can see that for the longer periods, the state’s p absolute value was over 1 0 times larger than the county one. Besides, incomes across states were converging, but those across counties were not. This result supports the hypothesis on a faster income convergence at the state level than at county level. Between 1969 and 1982, the rate at the state level was almost 15 times that of the county level. This indicates that, when there was reverse convergence, the speed was faster at the state level. From 1982 to 1993, states rolled back to convergence; however, counties still stayed in a divergence situation, showing a slower change. From 1993 to 1999, the state level convergence rate was five-fold of the county’s level convergence rate, i.e., demonstrating a stronger convergence in the state level as well. In terms o f the amplitude of percentages, the state convergence had a greater swing—between 0.05 and -0.15—than the county, within a smaller range from 0.0008 to -0.01. All of these findings support the research hypothesis. In a comparison of the within-state county p values and the state ones, we can see that all of the county long-time P percentage values, falling into a range between 0.000001 and 0.001784, are much smaller than the state value 0.05. That is, like the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 105 one within the U. S., and within the state county ( 3 convergences during the period were slower than the state. During the short-term, some states there were in convergence but in the others, there was divergence. It is hard to compare the county within state P convergence or divergence trends with the state ones. Even so, an important feature can still be seen. All of the county P absolute values were much smaller than the state ones. In other words, the range of county p values was smaller than that of the state ones. This feature indicates that the swings at the county level within the state framework, were weaker than at the state level. Like within the U. S. ones, these observations in the smaller comparison framework also provide support for the research hypothesis. 5.3 a Convergence 5.3.1 States: Long-time Convergence, but Short-term Fluctuations In addition to the immunity to Galton’s fallacy, another strong point of c convergence is that we can observe income convergence or divergence for each year. As indicated in Table 5.4, from 1969 to 1999, the a values dropped from 0.1707 to 0.1517 though this research includes Alaska and Hawaii. Thus, for over 3 decades, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 106 Table 5.4 Per Capita Income a Values o f States in the U. S. Year 1969 1970 1971 1972 1973 1974 1975 Per capita income o value 0.1707 0.1657 0.1597 0.1510 0.1484 0.1489 0.1607 Year 1976 1977 1978 1979 1980 1981 1982 Per capita income o value 0.1579 0.1540 0.1437 0.1412 0.1501 0.1447 0.1509 Year 1983 1984 1985 1986 1987 1988 1989 Per capita income a value 0.1511 0.1464 0.1493 0.1522 0.1577 0.1671 0.1658 Year 1990 1991 1992 1993 1994 1995 1996 Per capita income a value 0.1600 0.1541 0.1517 0.1494 0.1446 0.1447 0.1436 Year 1997 1998 1999 Per capita income o value 0.1470 0.1475 0.1517 Data Source: REIS there was state a convergence. These numbers support the research hypothesis on the long period of time convergence. Between 1969 and 1974, the finding o f this research confirms Barro and Sala-i-Martin’s (1992) finding, that the a values decreased significantly from 0.1707 to 0.1484. After that period, the a values rose, probably due to an exogenous shock— the gasoline crisis— according to Barro and Sala-i-Martin. Then, from 1982 to 1988, the a values gradually became larger. However, after this initial phase of the IICY, the a values gradually dropped from 1988 to 1993, the later years o f the IICY. In the IBP, the a values increased by a small amount; but for the years 1995- Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 107 1996, the values dropped somewhat. The observations support the research hypothesis on the initial divergence, but later convergence in the “Information Age.” 5.3.2 Counties: Long-time Convergence, but Short-term Fluctuations As shown in Table 5.5, from 1960 to 1999, the o value dropped from 0.2474 to 0.2256. For the longer periods, there was evidence of county a convergence. Between 1969 and 1978, the a values dropped; however, after 1978, the values rose. From 1982 to 1988, the beginning years of IICY, like those of the states, the cr values of the counties gradually became larger. Then, in the later phase of IICY, the cr values declined from 1988 to 1993. In the years of IBP, the a values increased by a small amount; but for the years from 1996 to 1997, the values dropped a little. In general, most of the findings here for the longer periods and for the short term support the research hypothesis. 5.3.3 A Comparison of State and County a Convergences Several observations should be stressed here. The first one is that, for the past 3 three decades, cr convergence appeared more strongly at the state level than at the county level, because the state a dropped by 12.52%, from 0.1707 in 1969 to 0.1517 in 1999, while the county a decreased by 9.66%, from 0.2474 to 0.2256. The Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 108 Table 5.5 Per Capita Income a Values o f Counties in the U. S. Year 1969 1970 1971 1972 1973 1974 1975 Per capita income a value 0.2474 0.2346 0.2308 0.2323 0.2547 0.2358 0.2377 Year 1976 1977 1978 1979 1980 1981 1982 Per capita income a value 0.2161 0.2180 0.2145 0.2152 0.2287 0.2146 0.2146 Year 1983 1984 1985 1986 1987 1988 1989 Per capita income a value 0.2191 0.2138 0.2150 0.2176 0.2186 0.2250 0.2194 Year 1990 1991 1992 1993 1994 1995 1996 Per capita income a value 0.2178 0.2098 0.2051 0.2075 0.2071 0.2126 0.2199 Year 1997 1998 1999 Per capita income cr value 0.2181 0.2227 0.2256 Data Source: REIS second is that, during initial and recent IT age, the swings were bigger in state a convergence. Measuring with the standard deviation of a values, we could get values 0.0066 and 0.0071 for initial and later IT periods for the state c values, but 0.0056 and 0.0061 for the county a values for the respective periods. Hence, the state a values swung in a larger range than the county a values. During the 31-years period, the county a change range was 14.97% above its lowest value, but the state one was 20.90% over its lowest value. Consequently, the income distributions were more stable at the county level than at the state level. The differences of the changes Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 109 can reconcile with what we observed during the p convergence analysis for states and counties. Moreover, in terms of when convergence or divergence occurred, there were differences in p and c t testing results. Those differences are under standable, because Sala-i-Martin (1996) and Quah (1993) have pointed out that c t convergence is sufficient, but not necessary for p convergence. However, some debates are still ongoing, and the absence of c t convergence cannot be taken as implying the absence of P convergence (Tsionas, 2000/2001). The third observation is beyond the research hypothesis, but an interesting one. Incomes are more evenly distributed in larger spatial units than in ones, because the state c t values were much smaller than the county’s. We should note that the c t values for counties fell into a range between the lowest value— 0.2051 (1992)— and the highest—0.2547 (1973). The county c t values have never dropped below the 0.2 level. This value of 0.2 seemed to be the lowest boundary for the county a values. Also, all of the county cr values were significantly higher than the highest state c t value. The lowest county c t was 0.2051, but the highest state c t value was 0.1707. For almost every year, the county c t value was at least 0.6 higher than the state c t value in the same year. This difference was huge in terms of their total values. Actually, the county c t value was approximately 30% higher than the state c t value each year. Thus, the empirical evidence shows bigger income differences among counties than those among states. This observation indicates that incomes are Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 110 less evenly distributed in counties, which might be considered as a lower spatial scale, than those in states. This may be because the state convergence was stronger or faster than the county convergence. As a result, the already big difference between the state income inequalities and the county’s may not become smaller but become larger for a long run. In light of the above evidence on different convergence patterns using various spatial units, one more issue should be pointed out here. Those findings also disclose a possible bias, if we use large geographic units for convergence testing due to an involvement of aggregation o f data. This is because, when we aggregate spatial units, we may artificially exaggerate income convergence and dilute income inequity. The researcher calls this bias “diluting aggregation and averaging effects” of inequality. A further example can be given here. If we compare incomes in Beverly Hills with those in downtown Los Angeles, we must see a big difference and a slow convergence. However, if we aggregate Beverly Hills with its surrounding cities and average their incomes, and meanwhile aggregate downtown Los Angeles with many other surrounding cities and average the incomes, we will see much less income inequality and stronger convergence. Therefore, a convergence at higher spatial levels does not necessarily mean a real convergence. The convergence research mainly on nations and states in the existing literature, might underestimate income inequality issues and overestimate income convergence to some extent. This Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. I I ll research is trying to improve the understanding by also evaluating counties. However, further research on smaller spatial units is needed. 5.4 T-tests and Other Observations for Central Suburban Areas The previous ( 3 and a convergence tests analyzed the rich to poor trickling- down type of convergence. Now, for the spreading-out convergence testing, income differences between central counties and their Ring I ’ s and between central counties and their Ring I I ’ s in the 50 largest American locations is examined. 5.4.1 Testing Results: Neither Convergence Nor Divergence With the REIS data, the findings are summarized in Tables 5.6 and 5.7. Table 15.6 shows the ratios of per capita incomes of the central counties to those of their own Ring I counties. During 1969 to 1999, and 3 short-term periods, income differences in some areas became larger, while some got smaller. The picture in the table seems to be mixed without showing a clear pattern of convergence or divergence. A similarly mixed picture for the changes of ratios of the central counties to their Ring II counties is indicated in Table 5.7. The income differences between the central counties and their Ring I ’ s and between the central counties and their Ring I I ’ s did not change significantly in the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 112 Table 5.6 Per capita Income Ratios o f Central Counties to Ring I Counties in the 50 Largest Places in the U. S. Top lst-50th largest places (central county/Ring I) 1969 1982 1993 1999 Rank New York-No. New Jersey-Long Island, NY-NJ-CT-PA 1.06 1.05 1.07 1.08 1 Los Angeles-Riverside-Orange County, CA 1.01 1.00 1.00 1.00 2 Chicago-Gary-Kenosha, IL-IN-WI 1.01 1.00 1.00 1.00 3 Washington-Baltimore, DC-MD-VA-WV 1.00 1.00 1.01 1.01 4 San Francisco-Oakland-San Jose, CA 1.02 1.02 1.02 1.02 5 Philadelphia-Wilmington-Atlantic City, PA-NJ-DE-MD 0.98 0.98 0.97 0.97 6 Boston-Worcester-Lawrence-Lowell-Brockton, MA-NH 1.01 1.01 1.02 1.02 7 Detroit-Ann Arbor-Flint, MI 0.99 0.99 0.98 0.98 8 Dallas-Fort Worth, TX 1.02 1.02 1.02 1.02 9 Houston-Galveston-Brazoria, TX 1.03 1.02 1.02 1.03 10 Atlanta, GA 1.01 1.01 1.03 1.04 11 Miami-Fort Lauderdale, FL 0.99 0.98 0.98 0.98 12 Seattle-Tacoma-Bremerton, WA 1.03 1.03 1.04 1.05 13 Phoenix-Mesa, AZ 1.03 1.05 1.04 1.05 14 Cleveland-Akron, OH 1.02 1.02 1.02 1.01 15 Minneapolis-St. Paul, MN-WI 1.02 1.02 1.03 1.03 16 San Diego, CA N/A N/A N/A N/A 17 St. Louis, MO-IL 1.04 1.03 1.04 1.04 18 Denver-Boulder-Greeley, CO 1.01 1.01 1.01 1.02 19 Pittsburgh, PA 1.02 1.02 1.03 1.03 20 Tampa-St. Petersburg-Clearwater, FL 0.98 0.98 0.99 1.00 21 Portland-Salem, OR-WA 1.01 1.01 1.01 1.01 22 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 Table 5.6 (continued). Top lst-50th largest places (central county/Rm gI) 1969 1982 1993 1999 Cincinnati-Hamilton, OH-KY-IN 1.03 1.02 1.03 1.03 Kansas City, MO-KS 1.00 0.99 0.99 0.99 Sacramento-Yolo, CA 1.00 1.00 0.99 0.99 Miiwaukee-Racine, W I 1.00 1.00 0.99 0.98 San Antonio, TX 1.01 0.99 1.00 1.00 Norfolk-Virginia Beach-Newport News, VA-NC 1.01 1.02 1.02 1.02 Indianapolis, IN 1.01 1.01 1.01 1.00 Orlando, FL 1.01 1.00 1.01 1.01 Columbus, OH 1.02 1.01 1.02 1.01 Charlotte-Gastonia-Rock Hill, NC-SC 1.03 1.03 1.03 1.04 Las Vegas, NV-AZ 1.02 1.04 1.04 1.04 New Orleans, LA 1.00 0.99 1.01 1.00 Salt Lake City-Ogden, UT 1.01 1.01 1.01 1.02 Greensboro-Winston-Salem-High Point, NC 1.01 1.01 1.01 1.01 Nashville, TN 1.02 1.01 1.02 1.02 Austin-San Marcos, TX 1.04 1.02 1.02 1.03 BufFalo-Niagara Falls, NY 1.01 1.01 1.01 1.01 Hartford, CT 1.02 1.01 1.01 1.01 Raleigh-Durham-Chapel Hill, NC 1.02 1.02 1.02 1.02 Memphis, TN-AR-MS 1.05 1.03 1.03 1.03 Rochester, NY 1.03 1.03 1.02 1.03 Jacksonville, FL 1.00 1.00 1.00 0.99 permission of the copyright owner. Further reproduction prohibited without permission. 114 Table 5.6 (continued). Top lst-50th largest places (central county/Ringl) 1969 1982 1993 1999 Rank Grand Rapids-Muskegon-Holland, MI 1.01 1.01 1.01 1.02 45 West Palm Beach-Boca Raton, FL N/A N/A N/A N/A 46 Oklahoma City, OK 1.02 1.02 1.02 1.02 47 Louisville, KY-IN 1.01 1.02 1.02 1.02 48 Richmond-Petersburg, VA 0.95 0.96 0.96 0.95 49 Dayton-Springfield, OH 1.02 1.02 1.01 1.01 50 Data Source: REIS, ranked by 1999 population Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 115 Table 5.7 Per capita Income Ratios o f Central Counties to Ring II Counties in the 50 Largest Places in the U.S. Top l st-50th largest places (centra! county/R/ng II) 1969 1982 1993 1999 Rank New York-No. New Jersey-Long Island, NY-NJ-CT-PA 1.08 1.07 1.09 1.11 1 Los Angeles-Riverside-Orange County, CA 1.03 1.03 1.03 1.04 2 Chicago-Gary-Kenosha, IL-IN-WI 1.03 1.02 1.03 1.04 3 Washington-Baltimore, DC-MD-VA-WV 1.06 1.05 1.06 1.05 4 San Francisco-Oakland-San Jose, CA 1.06 1.06 1.06 1.07 5 Philadelphia-Wilmington-Atlantic City, PA-NJ-DE-MD N/A N/A N/A N/A 6 Boston-Worcester-Lawrence-Lowell-Brockton, MA-NH 1.02 1.02 1.03 1.03 7 Detroit-Ann Arbor-Flint, MI 1.03 1.02 1.02 1.02 8 Dallas-Fort Worth, TX 1.06 1.04 1.05 1.05 9 Houston-Galveston-Brazoria, TX 1.06 1.05 1.05 1.05 10 Atlanta, GA 1.05 1.04 1.06 1.07 1 1 Miami-Fort Lauderdale, FL 1.01 1.01 0.98 0.98 12 Seattle-Tacoma-Bremerton, WA 1.04 1.04 1.04 1.06 13 Phoenix-Mesa, AZ 1.04 1.03 1.03 1.03 14 Cleveland-Akron, OH 1.03 1.03 1.03 1.03 15 Minneapolis-St. Paul, MN-WI 1.06 1.05 1.06 1.06 16 San Diego, CA N/A N/A N/A N/A 17 St. Louis, MO-IL 1.06 1.05 1.06 1.07 18 Denver-Boulder-Greeley, CO 1.03 1.03 1.03 1.05 19 Pittsburgh, PA 1.04 1.03 1.04 1.04 20 Tampa-St. Petersburg-Clearwater, FL 1.03 1.03 1.02 1.03 21 Portland-Salem, OR-WA 1.03 1.02 1.03 1.03 22 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 116 Table 5.7 (continued) Top lst-50th largest places (central county/Ring II) 1969 1982 1993 1999 Rani Cincinnati-Hamilton, OH-KY-IN 1.04 1.04 1.05 1.05 23 Kansas City, MO-KS 1.04 1.03 1.04 1.03 24 Sacramento-Yolo, CA 0.99 1.00 1.00 1.00 25 Milwaukee-Racine, W I 1.02 1.02 1.01 1.01 26 San Antonio, TX 1.02 1.01 1.01 1.02 27 Norfolk-Virginia Beach-Newport News, VA-NC 1.06 1.05 1.03 1.02 28 Indianapolis, IN 1.02 1.02 1.02 1.02 29 Orlando, FL 1.00 1.00 0.98 0.98 30 Columbus, OH 1.03 1.02 1.03 1.04 31 Charlotte-Gastonia-Rock Hill, NC-SC 1.05 1.04 1.05 1.06 32 Las Vegas, NV-AZ 1.04 1.03 1.04 1.04 33 New Orleans, LA 1.06 1.04 1.04 1.04 34 Salt Lake City-Ogden, UT 1.02 1.02 1.01 1.02 35 Greensboro-Winston-Salem-High Point, NC 1.03 1.03 1.02 1.03 36 Nashville, TN 1.04 1.03 1.04 1.05 37 Austin-San Marcos, TX 1.02 1.02 1.03 1.06 38 Buffalo-Niagara Falls, NY 1.02 1.02 1.02 1.03 39 Hartford, CT 1.03 1.03 1.03 1.03 40 Raleigh-Durham-Chapel Hill, NC 1.05 1.05 1.04 1.05 41 Memphis, TN-AR-MS 1.06 1.05 1.05 1.05 42 Rochester, NY 1.04 1.04 1.04 1.03 43 Jacksonville, FL 1.05 1.03 1.04 1.04 44 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 117 Table 5.7 (continued) Top lst-50th largest places (central county/Ring 11) 1969 1982 1993 1999 Rank Grand Rapids-Muskegon-Holland, MI 1.02 1.03 1.04 1.04 45 West Palm Beach-Boca Raton, FL 1.06 1.08 1.08 1.08 46 Oklahoma City, OK 1.05 1.04 1.03 1.03 47 Louisville, KY-IN 1.02 1.03 1.03 1.04 48 Richmond-Petersburg, VA 1.01 1.01 1.01 1.00 49 Dayton-Springfield, OH 1.02 1.03 1.02 1.01 50 Data Source: REIS, ranked by 1999 population short- or long run. T-tests confirmed the observations. Table 5.8 indicates the /-test results. The /-tests examine whether the ratios in the ending years differ significantly from the ratios in the beginning years for each of the five groups of the same size. Not all of the /-tests for all ratios during those periods are statistically significant. This means that there were no statistically significant changes for all o f the income differences during all periods. In other words, the per capita income ratios held constant statistically. There is no statistical support for the research hypothesis, assuming that there is center to the suburb convergence. Nonetheless, all of these findings indicate that the spreading-out convergence must be weaker than the trickling-down convergence. Furthermore, the /-tests exclude any statistically significant divergence. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 118 Table 5.8 T-Tests on Changes o f Per capita Ratios of Central Counties to Ring I/II Counties 1969-99 1969-82 1982-93 1993-99 a level Central countyiRing I (the 1-10 largest places): T-score -0.1315 -1.0998 0.3655 0.2732 Statistical significance no no no no 0.1 Central county/Ring II (the 1-10 largest places): T-score 0.4630 -1.1714 0.9002 0.7747 Statistical significance no no no no 0.1 Central county/Ring I (the 11-20 largest places): T-score 0.7832 -0.0638 0.6342 0.8522 Statistical significance no no no no 0.1 Central county/Ring II (the 11-20 largest places): T-score 0.2322 -0.6204 0.2014 0.6903 Statistical significance no no no no 0.1 Central county//?i'ng I (the 21-30 largest places): T-score -0.3666 -0.4660 0.2018 -0.7983 Statistical significance no no no no 0.1 Central county/ZJi'ng II (the 21-30 largest places): T-score -0.2925 -0.3936 -0.1946 0.2202 Statistical Significance no no no no 0.1 Central county/Ring / (the 31 -40 largest places): T-score 0.3103 -0.3099 0.6554 0.5416 Statistical significance no no no no 0.1 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 119 Table 5.8 (continued). 1969-99 1969-82 1982-93 1993-99 a level Central county/Ring II (the 31-■ 4 0 largest places): T-score 0.3261 -0.6470 0.5583 0.8052 Statistical significance no no no no 0.1 Central county/Mng / (The 41-■ 5 0 Largest Places): T-score -0.3968 -0.1924 -0.4002 -0.1789 Statistical significance no no no no 0.1 Central county/Ring II (The 41-50 Largest Places): T-score -0.0308 -0.0854 -0.1000 0.1703 Statistical significance no no no no 0.1 Data Sources: REIS and BLS, ranked by 1999 population 5.4.2. Additional Findings: Initially Diverging, Later Converging, and Long-time Converging Evidence Even so, some observations that are not statistically significant are also worth reporting. As indicated in Table 5.9, in terms of per capita incomes, the ratios of the central counties to their Ring I ’ s, among three groups, including top 1-10,11-20, and 21-30 largest places, show some evidence supporting the research hypothesis. In the IICY, percentages of divergence cases increased from 0% to 50%, 22% to 56%, and 20% to 40%, respectively. Meanwhile, the convergence cases dropped or remained the same from 40% to 0%, 11% to 11%, and 20% to 20%. However, in the IBP, the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. Table 5.9 Center-Periphery Convergence/Divergence Analysis for the 50 Largest Places Group Case percentage 1969-99 1969-82 1982-93 1993-99 1969-99 1969-82 1982-93 1993-99 Cases Percentages Ring I 1-10 Convergence cases & percentage 2 4 0 0 20% 40% 0% 0% Divergence cases & percentage 5 0 5 1 50% 0% 50% 10% Stable cases & percentage 3 6 5 9 30% 60% 50% 90% 11-20 Convergence cases & percentage 2 1 1 1 22% 11% 11% 11% Divergence cases & percentage 6 2 5 4 67% 22% 56% 44% Stable cases & percentage 1 6 3 4 11% 67% 33% 44% 21-30 Convergence cases & percentage 2 2 2 2 20% 20% 20% 20% Divergence cases & percentage 4 2 4 1 40% 20% 40% 10% Stable cases & percentage 4 6 4 7 40% 60% 40% 70% 31-40 Convergence cases & percentage 3 4 0 2 30% 40% 0% 20% Divergence cases & percentage 3 2 2 3 30% 20% 20% 30% Stable cases & percentage 4 4 8 5 40% 40% 80% 50% N _ > o Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. Table 5.9 (continued). Group Case percentage 1969-99 1969-82 1982-93 1993-99 1969-99 1969-82 1982-93 1993-99 Cases Percentages Ring I 41-50 Convergence cases & percentage 2 2 2 0 22% 22% 22% 0% Divergence cases & percentage 3 1 0 4 33% 11% 0% 44% Stable cases & percentage 4 6 7 5 44% 67% 78% 56% Ring II 1-10 Convergence cases & percentage 4 6 0 1 44% 67% 0% 11% Divergence cases & percentage 5 0 5 4 56% 0% 56% 44% Stable cases & percentage 3 4 4 0% 33% 44% 44% 11-20 Convergence cases & percentage 1 5 0 0 11% 56% 0% 0% Divergence cases & percentage 4 0 5 4 44% 0% 56% 44% Stable cases & percentage 4 4 4 5 44% 44% 44% 56% Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. Table 5.9 (continued). Group Case percentage 1969-99 1969-82 1982-93 1993-99 1969-99 1969-82 1982-93 1993-99 Cases Percentages 21-30 Convergence cases & percentage 4 5 2 2 40% 50% 20% 20% Divergence cases & percentage 2 0 4 2 20% 0% 40% 20% Stable cases & percentage 4 5 4 6 40% 50% 40% 60% 31-40 Convergence cases & percentage 2 5 2 0 20% 50% 20% 0% Divergence cases & percentage 5 0 5 7 50% 0% 50% 70% Stable cases & percentage 3 5 3 3 30% 50% 30% 30% 41-50 Convergence cases & percentage 6 2 3 3 60% 20% 30% 30% Divergence cases & percentage 3 4 2 2 30% 40% 20% 20% Stable cases & percentage 1 3 5 5 10% 30% 50% 50% Data Source: REIS, ranked by 1999 population. 123 respectively. The two groups with the 31 to 50 largest places seemed to show a rise of divergence cases in the IBP, from 20% to 30% and 0% to 44%, but a relatively stable state in the IICY phase. Actually, this may suggest that the development and application of IT affected larger places earlier than smaller places. (This could be a new research topic.) For the ratios of central places in their Ring I I ’ s, the three groups with the 30 largest places demonstrate relatively strong evidence to support the research hypothesis. If we measure with percentage the convergence cases for the three groups dropped from 67% to 0%, 56% to 0%, and 50% to 20%, while the divergence cases rose from 0% to 56%, 0% to 56%, and 0% to 40%, respectively in the IICY. After that period, the convergence cases increased, or at least stayed stable, while all of the divergence cases decreased. Also as seen for the income ratios between the central places and the Ring I ’ s, in the smaller places, the impact seemed to lag one period, i.e., a decrease o f convergence cases and, meanwhile, an increase of divergence in the later period. If we review the 50 places as a whole just as shown in Table 5.10, there is even stronger evidence for the research hypothesis. For the pairs o f central places and their respective Ring I ’ s, the convergence case dropped initially from 27% to 10% in the IICY, but did not continue to dip later in the IBP. The divergence cases rose from 15% to 33%, but decreased to 27% later. For the ratios of central places to Ring IPs, the convergence cases decreased from 50% to 17% in the IICY, while the divergence cases increased from 8% to 44%, initially, but dropped to 40% in the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 124 Table 5.10 Center-Periphery Convergence/Divergence Analysis fo r the 50 Largest Places in Total Geographic pair Cases/percentage 1969-99 1969-82 1982-93 1993-99 Center vs. Ring I Convergence cases 11 13 5 5 Center vs. Ring I Divergence cases 21 7 16 13 Center vs. Ring I Stable cases 16 28 27 30 Center vs. Ring I Convergence percentage 23% 27% 10% 10% Center vs. Ring I Divergence percentage 44% 15% 33% 27% Center vs. Ring I Stable percentage 33% 58% 56% 63% Center vs. Ring II Convergence cases 17 24 8 6 Center vs. Ring II Divergence cases 19 4 21 19 Center vs. Ring II Stable cases 12 20 19 23 Center vs. Ring II Convergence percentage 35% 50% 17% 13% Center vs. Ring II Divergence percentage 40% 8% 44% 40% Center vs. Ring II Stable percentage 25% 42% 40% 48% Data Source: REIS IBP. In addition, the smaller places seemed to be impacted in the later period. In other words, there was a time lag between the larger places being affected and the smaller places being affected by IT. Furthermore, one more interesting observation is that the IT influenced the convergence between the central places and their Ring I I ’ s, might be stronger than between the central places and their Ring I ’ s. At the beginning, there were more divergence cases among the former pairs than among the latter pairs. However, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 125 during the longest time, there were more convergence cases between the central places and their Ring I I ’ s. This clue is shown in Table 5.10. In the longest time period, the percentage for the convergence between central places and their Ring I I ’ s was 35%, but the percentage for that between central places and Ring I ’ s was 23%. This finding means that it is more likely to observe income convergence between the central places and the outer suburbs than between the central places and the inner suburbs. In addition, during the initial year, the divergence percentage for the central places and their Ring ITs jumped from 8% to 44%, but for the central places and their Ring I ’ s, the percentage only rose from 15% to 33%. This could mean that IT enriches the outer suburbs more than enriching the inner suburbs. However, this exogenous disturbance to convergence might not hold for long; so to the contrary, the convergence for the pair of the central place and the outer suburbs was rather stronger for a longer time. The initial period enrichment of outer suburbs may imply that a high income cohort living in the outer suburbs, seemed that residents in central places and even in the inner suburbs, cannot enjoy the fruits of hi-tech very much. The later, stronger convergence between the more intensively impacted pairs, may indicate a spreading of benefits to the formerly disadvantaged cohorts in central places and the inner suburbs. Moreover, this phenomenon suggests that the initial impact leading to divergence is temporary but not long. This is what the research hypothesis also assumes. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 126 Nevertheless, not all of the evidence in favor of the research hypothesis is strong. Though one caveat is that none of the findings is statistically significant, a point should be stressed here. No evidence, either statistically or from observations of data, can support an opposite hypothesis for a long time core-periphery income divergence. Thus, a hypothesis for an opposite income divergence will be rejected. Furthermore, the finding here can have an implication in labor productivity between central and peripheral areas. The above convergence findings imply that, during short-term periods, the IT might contribute to faster growth o f central place productivity, and then later, faster growth of peripheral productivity. However, for a longer time, there was no sign that productivity in central counties grew faster than that in suburban counties among the top 50 largest places in the U. S. No evidence supports the assumption that there was productivity divergence between central places and suburban areas. Rather, the findings suggest that there might be productivity convergence between central places and peripheral areas in the later phase, and in the relatively longer timeframe of the Information Age. 5.5 Tests on the Link of Convergence to ITIA/NITIA All o f the previous analyses mainly sought temporal evidence, that is, the comparisons among the years before and during the years when the IT was developed. This part o f the testing was seeking spatial evidence, i.e., making the comparisons among the spatial units where IT industries were mainly developed and those not. To make the spatial comparisons concerning the IT impacts, this research Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 127 compared the income convergence between the information technology intensive areas (ITIA), such as hi-tech regions, and information technology less intensive areas (NITIA). The total IT index, Ir.t, was used to define these two kinds o f areas. For the Ir-t, 4-digit SIC data from the BLS were used. The state Ir-t's are calculated with the 1999 state job data and the total U.S. job data. Table 5.11 shows the Ir.t indices for the states in the U. S. In terms of the total IT index, Colorado, Massachusetts, New Hampshire, California, Virginia, Texas, Oregon, Washington, Minnesota, Georgia, Utah, New Jersey, Maryland, Arizona, and Idaho had higher IT job percentages than the national average. The states with high Ir-t's were composed of the ITIA. Wyoming, Arkansas, Mississippi, Hawaii, Louisiana, West Virginia, Kentucky, Nevada, and Montana ranked the lowest, with IT job percentages less than half of the national average. The states with low Ir-t’s consisted of the NITIA in the U. S. In terms o f the partial IT index, such as hardware and IT equipment index, Ir. he, and software and IT services index, Ir.S s, the results from the same data source are also listed in Table 5.11. Those states with Ir-h e higher than 1 can be considered “hardware and IT equipment” development states. Also, those states with Ir.ss higher than 1 can be considered “software and IT services” states. Comparing convergence patterns in Table 5.3 with the total IT indices in Table 5.11, we see rather complicated pictures for the long time period and the IICY. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 128 Table 5.11 State IT Indices in the U. S. IT (software & services) Rank State IT (total) index IT (hardware & equipment) index index 1 CO 1.9782 1.8273 2.0248 2 MA. 1.7057 2.3102 1.5190 3 NH 1.5573 3.3910 0.9909 4 CA 1.5491 2.2961 1.3184 5 VA 1.4004 0.5333 1.6683 6 TX 1.2898 1.5681 1.2039 7 OR 1.2565 2.6703 0.8198 8 WA 1.1980 0.8759 1.2975 9 MN 1.1922 1.8253 0.9967 10 GA 1.1915 0.3340 1.4564 1 1 UT 1.1792 0.7569 1.3096 12 NJ 1.1774 0.6544 1.3389 13 MD 1.1464 0.5131 1.3420 14 AZ 1.1395 2.0667 0.8532 15 ID 1.0198 2.2792 0.6308 16 AL 1.0143 0.6904 1.1144 17 CT 0.9867 0.9543 0.9967 18 NE 0.9650 0.5737 1.0858 19 IL 0.9320 1.1025 0.8794 20 NY 0.8780 0.6001 0.9638 21 SD 0.8652 1.9431 0.5322 22 KS 0.8614 0.2051 1.0641 23 MO 0.8474 0.3209 1.0101 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 129 Table 5.11 (continued). IT (software & services) Rank State IT (total) index IT (hardware & equipment) index index 24 FL 0.8386 0.6783 0.8882 25 DC 0.8292 0.0069 1.0831 26 PA 0.8258 0.8643 0.8139 27 NM 0.8005 1.1871 0.6811 28 NC 0.7528 0.6997 0.7692 29 OH 0.7060 0.6291 0.7297 30 OK 0.6434 0.2086 0.7776 31 IA 0.6383 0.2850 0.7474 32 DE 0.6204 0.2683 0.7292 33 MI 0.6057 0.3150 0.6954 34 WI 0.5863 0.5848 0.5868 35 ME 0.5716 0.6600 0.5442 36 ND 0.5437 0.1024 0.6800 37 SC 0.5416 0.4673 0.5646 38 TN 0.5289 0.4235 0.5615 39 AK 0.5227 0.0000 0.6841 40 IN 0.5182 0.6313 0.4833 41 RI 0.5023 0.3466 0.5503 42 VT 0.4978 0.1851 0.5944 43 MT 0.4838 0.0361 0.6221 44 NV 0.4827 0.1320 0.5911 45 KY 0.4699 0.0980 0.5848 46 WV 0.4323 0.1577 0.5171 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 130 Table 5.11 (continued) Rank State IT (total) index IT (hardware & equipment) index IT (software & services) index 47 LA 0.4133 0.1211 0.5035 48 HI 0.3951 0.0000 0.5171 49 MS 0.3478 0.0237 0.4479 50 AR 0.3415 0.0794 0.4225 51 WY 0.2343 0.0000 0.3067 Data Source: BLS, ranked by the total IT Index However, during the IBP, within some of those ITIA states such as Colorado, California, Idaho, Maryland, Oregon, New Jersey, Washington, there were statistically significant convergences among their counties. Among the top ITIA states, Texas was the only exception showing divergence. As discussed earlier, the agricultural states that had the lowest total of IT indices, showed income divergence among their counties. Texan divergence could be due to this reason. Moreover, OLS regression analyses are applied to examine the relationship between convergence and IT development and application. This research regresses a convergence/divergence proxy variable on the total or partial IT index. If a state shows significant convergence, a number 1 is assigned to this proxy variable. If a state indicates significant divergence, the proxy variable gets a value of -1. If neither Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 131 convergence nor divergence is statistically significant, the proxy variable is assigned 0 for that state.b The regression tests with the total IT index support the hypothesis that IT development and application is relevant to income convergence patterns. As demonstrated in Table 5.12, the linear regression shows that convergence was significantly related to the total IT index in both of the IBP and the IICY periods at an a level = 0.05. Furthermore, in PIAY before 1982, the total IT index had no relationship with convergence. Thep -value is 0.35, which shows that convergence in the PIAY had no association with the IT index of a state, even if we select a large a value at 0.1. This finding indicates that, before IT came into existence, income convergence had bome no relationship with later IT development and application. In other words, prior to the IT Revolution, convergence or divergence patterns had not been related to the later presence o f IT industries. The relationship only showed up after the ITIA states had developed their IT industries. Hence, the evidence here provides additional support for the point that the development and application of IT is relevant to the spatial income convergence or divergence patterns. b Using a -1,0,1 dependent proxy variable for convergence possibly compromises OLS results. However, because some convergence values are significant (negative or positive), but some others are insignificant, it is inappropriate to use convergence/divergence values that do not disclose whether the values are significant or not. The employment of the -1,0,1 dependent proxy variable is a way to represent these three situations— significant convergence, significant divergence, and insignificant results. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 132 Table 5.12 Regressions o f Convergence Proxies on IT Indices fo r States in the U. S. Time span IT index Parameter b p-value Statistical significance 1969-82 IT (Total) Index 0.23121 0.35 No (a = 0.1) 1982-93 IT (total) index 0.44600 0.03 Yes (a = 0.05) 1993-99 IT (total) index 0.44845 0.05 Yes (a = 0.05) 1969-82 IT (hardware & equipment) index 0.14021 0.24 No (a = 0.1) 1982-93 IT (hardware & equipment) index 0.20301 0.04 Yes (a = 0.05) 1993-99 IT (hardware & equipment) index 0.23019 0.04 Yes (a = 0.05) 1969-82 IT (software & services) index 0.13213 0.62 No (a = 0.1) 1982-93 IT (software & services) index 0.35885 0.11 No (a = 0.1) 1993-99 IT (software & services) index 0.32066 0.20 No (a = 0.1) Note: Convergence^ = a + b* (IT Index), where Convergencep = 1 if converging, Convergencep = -1 if diverging, Convergencep = 0 if neither converging nor diverging. Data sources: REIS and BLS The regression tests with the hardware and IT equipment index have similar results to the tests with the total IT index. Only the parameter b ’s are smaller than the parameter b obtained from the regression on the total IT index. However, all of the regression tests of convergence proxy values to the software and IT services index are not significant statistically. It appears that the software and IT services subsector has a weaker relationship with convergence patterns than the hardware and IT equipment sub-sector or than the total IT industry. Even so, all o f the tests Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 133 indicate that the parameter b’s for 1969-1982 are larger than those for the two IT periods, and that the p- values are larger for 1969-1982 as well. This discloses that the total or partial IT indices account more for the convergence patterns in the two IT periods than prior to the IT periods. 5.6 Tests on the Relevancy of Convergence to Productivity The previous regression tests analyzed the spatial relationship between convergence and the IT index. In this section, the relationship via productivity growth is examined. In fact, it would have been more supportive if a direct relationship between IT shock and convergence could be tested. However, it is difficult, if ever possible, to quantify the IT shock. An indirect test, therefore, was employed via the productivity growth, to seek the relationship between income convergence and the IT shock. Though the deviation from direct testing constrained by data and measurement may not truly report such a relationship either found or rejected, at least this indirect testing can offer additional information on the relationship between convergence and IT. In a linear regression test indicated in Table 5.13, state a is significantly related to nonfarm productivity during 1969-1999 at an a value o f 0.05. This finding shows that they co-vary simultaneously. As shown in Table 5.14, the county a has an even stronger correlation—more than double the state regression coefficient— Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 134 with the nonfarm productivity at a smaller a value of 0.01. The negative coefficients disclose that decreases in the state and county cr values were associated with growth Table 5.13 Regression o f State cron Nofarm Productivity in the U.S. Dependent variable Parameter b p-value Statistical significance R-square Adj. R-square Model p-value State a -0.00022 0.05 Yes (a = 0.05) 0.1247 0.0945 0.05 Note: State a = a + b* (nonFarm productivity) Data Sources: REIS and BLS Table 5.14 Regression o f County cron Nonfarm Productivity in the U.S. Dependent variable Parameter b p-value Statistical significance R-square Adj. R-square Model p-value County c t -0.00053 0.00 Yes (a = 0.01) 0.3309 0.3078 0.00 Note: County a = a + b* (Non-Farm Productivity) Data Sources: REIS and BLS of productivity during 1969-1999. The findings for both states and counties support the argument that a convergence is significantly linked to productivity acceleration. In the presence of empirical evidence that information technology contributes to the productivity acceleration, the a convergence may be linked to information technology. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 135 This association can also be found if p convergence in Tables 3.2, 5.1, 5.2 and 5.3 are examined, as well. In comparing Tables 3.2, 5.1 and 5.2, it is found that both o f the state and county p values got larger as productivity accelerated. For the U. S. states, the only divergence occurred from 1969 to 1982, during most of the period (1973-1982) when productivity grew so slowly at a rate of approximately one-third to one-half of later productivity growth. For the county p convergence, Table 3.2 and 5.3 together show a similar relationship. Prior to 1982, there were many cases o f income divergence while productivity grew slowly. Later, when productivity caught up, more convergence cases were observed. Hence, via the productivity, both of the o and the p convergences may link to the emergence of IT from another perspective. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 136 CHAPTER 6 CONCLUSIONS 6.1 Introduction In this chapter, theoretical conclusions are intended to be drawn based on previous analyses and empirical work. Some future research issues will then be addressed. The final sections will discuss implications for social income inequality and policy-making. The first theoretical conclusion is that a combination o f the findings and what has been reported in the literature supports the point that there has been no long range divergence, but rather, long-range convergence. The trickling- down convergence tests, such as the P test for states, the P test for counties within- states, all e x tests, and nonstatistical observations for spreading-out convergence support this conclusion. However, the control of temporal periods to the IT shock indicates that short term convergence fluctuations appear to be relevant to the IT shock. Regression tests provide additional evidence for the link of convergence changes to the IT presence. Thus, a model of overlapping multiple Bell-shaped curves or multiple inverted-U- shaped curves can offer a better explanation for the convergence patterns than the conventional model of single inverted-U curve. The second conclusion concerns exogenous or endogenous changes in technologies. The findings using states and counties as units of analysis in this research, support the neoclassical exogenous assumption. Two empirical Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 137 observations challenge the endogenous model for explaining a long-run income evolution pattern. The first one, as already mentioned, is that there has never been any finding for long- run or even relatively long-time divergence, but for long-run convergence. This second observation is the slower convergence in the faster technological diffusion years—the “Information Age”—because the endogenous assumption would suggest faster convergence in a faster diffusion age. Even so, the short-term divergence finding could support the endogenous assumption on increasing returns to capitals. The evidence of initial divergence can suggest that, when technology is in its early phase, there are increasing returns to capital; however, later or for a long run, the investment return feature appears to switch from increasing returns to diminishing returns. The researcher calls this feature “development-phase-based increasing or diminishing returns to capitals,” which stands for whether there are increasing or diminishing returns depends on the development phase of a technology. The third theoretical conclusion is that convergence patterns vary at different spatial scales. During the longer time period, convergence is faster at a higher level— states— and fluctuation of convergence is more intensive at the state level, as well, than at a lower level— counties. In addition, spatial income inequalities at the state level are less severe than those at the county level. This can be owing to faster convergence at the state level for the long term. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 138 The final conclusion is regarding core-periphery or spreading-out convergence. Observations of data indicate that there are signs of long-time-period convergence; however, those observations are not supported with statistical testing. One safe theoretical point is that spreading-out convergence is evidently weaker than trickling-down, or rich-poor region convergence. In addition, an interesting observation is that the IT impact on the convergence between the central places and the outer suburbs seemed to be stronger than that between the central places and the inner suburbs. However, this exogenous disturbance to convergence might not hold long. So to the contrary, the convergence for the pair of the central place and the outer suburbs, was rather stronger for a longer time Furthermore, larger places appeared to be impacted earlier than smaller places. In other words, there was a time lag between smaller places being affected and larger places being affected by IT. The shortcomings and limitations of our understanding of convergence patterns analyzed in this dissertation, suggest that research should extend into three dimensions. The first is temporal dimension. One concern in this research is that the findings here still cannot account for long-term observations. We should examine the long-run patterns in the presence of information technology. This can be conducted when long-range data become available. Second, more work should address why convergence is faster at higher spatial scales such as states, than at lower scales— counties. Across-large-distance decentralization of firms and technologies may be an explanation. This needs further research on whether the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 139 across-bigger distance decentralization is significant and whether this type of decentralization brings faster convergence on higher spatial scales. Third, income convergence trends should be evaluated in comparatively micro spatial units or on lower spatial scales. Investigations at city or zip code zone levels can extend our understanding. Going beyond the scope of this dissertation, the findings in spatial income inequality are extended to some possible implications on social income inequality. The goal of going an extra mile is to raise interest in social income-convergence research. The assumptions include that social income convergence behavior may share some similar features that are found in spatial income convergence. Therefore, it is likely that exogenous shocks may have impacts on social income convergence, too. Similarly, significant technological shocks may initially drive a short-term divergence or slower convergence; but in later phases, the shocks should contribute to long-run convergence of social inequalities. In addition, there might be a possibility that social income convergence occurs faster in the early stages, but slower in the later stages of an economy. Furthermore, social convergence trends on relatively macro scales or larger aggregation units, may appear stronger than those in micro scopes or smaller aggregation units. Also, social inequality issues might be more severe, if we review them in comparatively micro scopes or by smaller groups. That is, if we evaluate income inequality on lower spatial scales or in comparatively Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 140 micro scopes, the inequality may be more severe than being evaluated at higher scales or macro scopes. The final part of this chapter discusses implications of the findings in this research for policymaking. First, long-term policies may be less necessary than short-term ones, if at all. This is because, for the long-run or a relatively long-time period, spatial incomes converge, and divergence appears to be transitory. Second, whenever policy measures are considered, those on lower levels may be more necessary than those at higher levels, since convergence is slower at a lower spatial level. Finally, due to long-time convergence, income redistribution policies in later stages o f an economy, may be less necessary than in earlier stages. 6.2 Summary of Empirical Findings 6.2.1 Temporal and Relationship Testing Results As indicated in Table 6.1, which is corresponding to Table 4.2, substantial empirical findings in the P and a tests support the research hypothesis. In particular, the p tests for states and all of the cr tests provide relatively persuasive evidence for the research hypothesis. The a tests for both states and counties support the hypothesis that spatial incomes diverged in the initial years of the emergence of IT and converged in the later years. The state and county (throughout the U. S.) P tests show that, in the initial years, convergence speeds were slower than in the later years. However, there are glitches in the p tests. One of the glitches is that state personal incomes shifted from pre-information-Age divergence Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. Table 6.1 Summary o f Findings in the Convergence Testing 31-Year Period Initial Period Recent Period Unit of analysis Method Pattern Comparison Pattern Comparison Pattern Comparison States P C(S) Faster C than county C (S) Slow but accelerating C (partial S) Stronger swing than county one (S) Accelerating C(S) Stronger swing than county one (S) c C(S) Faster C than county C (S) Accelerating D or Decelerating C(S) Stronger swing than county one (S) Decelerating D or accelerating C(S) Stronger swing than county one (S) Counties P (counties in throughout the U. S.) Extremely weak D (NS) Slower than state C (S) D (S) Weaker swing than state one (S) Accelerating C(S) Weaker swing than state one (S) P (counties within-states) Almost all C cases (S) N/A More D & fewer C cases than in the 31- year period (S) N/A More D & fewer C cases than in the 31- year period (NS) N/A o (counties in throughout the U.S.) C(S) Slower C than state C (S) Accelerating D or Decelerating C(S) Weaker swing than state one (S) Decelerating D or accelerating C(S) Weaker swing than state one (S) Centers-Suburbs T-test Neither C nor D (NS) Weaker than trickling-down C(S) Neither C nor D (NS) Neither C nor D (NS) H—k Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. Table 6.1 (continued). 31-Year Period Initial Period Recent Period Unit of analysis Method Pattern Comparison Pattern Comparison Pattern Comparison Observation Weak C (S) Weaker than Decelerating C Accelerating C trickling-down & accelerating & decelerating C(S) D (S) D (S) C = convergence; D = divergence; S = support for the hypothesis; NS = no support for the hypothesis; N/A = not applicable 4^ to 143 to convergence, and the county divergence slowed down in the IIC Y period. This might be due to the fact that in the previous period the gasoline shock dragged the income distributions into a strong divergence, as discussed in the literature. The divergence caused by the energy crisis had been even stronger than the divergence associated with the initial emergence o f IT. Another glitch is that the county p convergence finding is less supportive for the research hypothesis than the state convergence. This may be demonstrating the role of the spatial scales in the convergence picture. More importantly, the p tests for states, for the counties within the states and all of the a tests show long-time-period convergence, providing support for the research hypothesis. In addition, the OLS regression tests of convergence on spatial IT indices and on productivity growth show further statistical evidence for the hypothesis that the development and application of IT is relevant to the income-convergence patterns. The results are summarized in Table 6.2. Convergence patterns are related to the development of the total IT industries in states. In addition, it appears that the software and IT services sub-sector has a weaker relationship with convergence patterns than the hardware and IT equipment sub-sector or the total IT industry. Via productivity, an indirect link between IT presence and convergence patterns is found, as well. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 144 Table 6.2 Summary o f Findings in the Statistical Relationship Tests Convergence Pattern Method ITIA/ITLIA areas (total IT index) Significantly related Linear regression test Hardware & IT equipment areas Significantly related Linear regression test Software & IT services areas Not significantly related Linear regression test Productivity Significantly related with both state and county sigma convergence Linear regression test Furthermore, empirical evidence discloses an additional feature of convergence. Compared to the historical trends of convergence in the literature, speeds of convergence appear to be slower with the development of an economy. For instance, the p convergence speeds from 1969 to 1999 were much slower than those between 1800 and 1969, as shown in Barro and Sala-i-Martin’s (1991) paper. This feature might be explained this way: with the development of an economy, income differences and the room for the changes of those differences become smaller and smaller. 6.2.2 Findings of Spatial Scales Income convergence shows different features on different spatial scales or in different geographic units. All of the p and a tests show that convergence speeds were faster for the longer periods of time and fluctuated more intensively during the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 145 short term at the state level than at the county level. A finding of income convergence on comparatively high spatial scales does not necessarily stand for income convergence on low scales. Convergence at the county level was relatively slower in the longer time period and more stable during the short term. Income inequalities at the county level were much more intensive, probably as a result of slower convergence. 6.2.3 Center-Suburb Testing Results and Observations The tests on the spreading-out convergence— central and surrounding counties— do not show statistically significant support for the research hypothesis. Statistically speaking, there was neither significant convergence nor significant divergence for all compared pairs and for all temporal units. Even with the exogenous shocks, not only the IT but also the gasoline crisis had no statistically significant impact on the income redistributions between these spatial pairs. However, some evidence that was not statistically tested supports the research hypothesis. During the initial years of the Information Age, the divergence cases increased but leveled off later. Simultaneously, the convergence cases dropped initially but increased later. The negative impact on convergence was likely short term. For the longer periods, more convergence cases were likely to be seen. For the time span of three decades, there were more convergence cases than short sub periods o f the IICY and the IBP. Furthermore, the trickling-down convergence was stronger than the spreading-out convergence. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 146 There are two additional important observations. First, the income distributions between central places and outer suburbs could be affected more dramatically in the initial phase of the exogenous IT shock in a form of divergence, but tended to converge more strongly for a longer time period than the income distributions between central places and inner suburbs could. Second, the exogenous IT shock to smaller places seemed to lag behind the shock to larger places. 6.3 Theoretical Conclusions 6.3.1 Income Convergence Patterns, IT Impacts, and a New Model The statistical evidence in this research provides support for the following points. There is almost no evidence showing that income convergence is not a relatively long time or probably long-range trend. However, during the short term, this trend can fluctuate. That is, we may see some short-term divergence fragments. All o f the divergence observations in this research and in the literature appear transitory. In fact, if combining the observations in this research and the findings in the literature on convergence, a conclusion can be drawn that there has been no evidence for long-run income divergence. Because for a longer period of time, it is more likely to observe income convergence, spatial incomes should become more and more evenly distributed. One caution is that, even though this research extends the inquiry on convergence from states into counties and center-suburbs, the conclusion regarding the convergence patterns may still be limited to the units Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 147 examined here and not be generalized for other geographic units, such as cities in the U. S. Also, combined findings in this dissertation and in the existing literature support a point that income convergence patterns vary as an economy develops into different stages. When convergence starts, its speed can be relatively fast. However, later, the convergence speed becomes slow. The reason can be that, with the maturation of an economy and the equalization of spatial incomes, the income inequality becomes smaller, and there is not much room for further equalization. The temporal, spatial, and regression tests conducted in this research have presented supportive evidence for an assumption that IT has an impact on convergence. For the two short timeframes, the data show that the IT impacts are significant, though long-range data are required to confirm this finding. Furthermore, for the special impact of IT, the evidence disclosed in this research can support the following points. The early period of divergence or slower convergence may indicate stronger centripetal forces and concentration behavior. Later, the convergence evidence can demonstrate that a centrifugal dynamic and decentralization location behavior may prevail. This is another demonstration of theoretical wisdom regarding technological impacts in Barro’s (2000) and some others’ papers: an early phase competitive advantage limited to small group of people in some poles, but later phase ubiquitous advantages due to knowledge spillovers and technological diffusions. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 148 Even though the empirical evidence supports the hypothesis that IT has an impact on convergence, the changes o f convergence patterns can also be affected by other factors. The testing in this research does not exclude other factors. That is, the findings in this research do not rule out alternative explanations. So far, convergence advocates have been seeking evidence for an all time convergence story, but have not recognized that the impact from later technological shocks like IT can be so significant that the convergence may not hold during some shorter periods. This implies that they assume there is only one significant technological shock, and that only one industry dominates a country’s or regional convergence pattern forever. Yet, there are other significant exogenous shocks, such as information technology and definitely more in the future. These shocks can interfere with the growth path dominated by one industry, such as manufacturing. Hence, the real growth pattern is much more complicated than a single Bell-shaped path can predict. Simplicity is a beauty of theorization. However, over simplification hamstrings the explanation power of a theory. The single Bell-shaped and the inverted U-shaped growth path assumptions are more descriptive for economies in the early industrial stages, but too simple for advanced economies in late economic stages, when there are more and more technological revolutions. This dissertation assumes that the path is an aggregation o f overlapping multiple Bell-shaped curves. The overlapping Bell-shaped paths can explain the fluctuation pattern of income convergence. This new model of overlapping multiple Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 149 Bell-shaped curves or multiple inverted-U-shaped curves, offers a better explanation for the current convergence or divergence patterns in the real world. Each Bell shaped or inverted U-shaped curve can result from each important technological renovation that significantly contributes to economic growth and productivity acceleration, such as industrial revolution, information revolution, and those in the future. Aggregately, the overlapping curves may demonstrate a quasi-cyclical pattern of growth curve. 6.3.2 Exogenous or Endogenous Assumption In terms of over 3 decades, the evidence of convergence supports the assumptions on diminishing returns to capitals and on exogenous technological growth. In summary, two key observations in this research support the “neoclassical growth model” rather than the “new growth model.” The first one, as already mentioned, is that there has never been any finding for the long run or even relatively long-time divergence, but for long run convergence. The second is the slower convergence speed in the “Information Age” than that of the Industrial Revolution. This second point needs some elaboration. According to the “new growth model,” because technological changes are endogenous, the spillover speed determines the convergence speed, providing there is any convergence. Based on this logic, convergence should be faster in the “Information Age,” since evidently the spillover speed during this timeframe is much faster than that during previous technological revolutions, such as the Industrialization era. However, the contrary evidence— a Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 150 slower convergence in the faster spillover “Information Age” has been observed. Although there are some glitches in explaining the income convergence or the uneven growth pattern with the “neoclassical theory,” the “new growth theory” offers a weaker explanation power, first, for the long-time convergence pattern, and second, for the slower convergence in a faster technological diffusion age. Notwithstanding the two key points in the long run, the short-term divergence finding could support the endogenous assumption on increasing returns to capital. The evidence of initial divergence can suggest that, when IT is in the early phase of development, there are increasing returns to capital. However, in the end, the investment return feature appears to switch from increasing returns to diminishing returns. The researcher titled this feature “development phase-based increasing or diminishing returns to capitals,” which stands for whether there is increasing or diminishing returns depend on the development phase of a technology. Take personal computers, for instance. When PC kits were first invented, during the initial several years, companies had to stay in some poles; because if those companies had invested in other lagging behind regions or developing countries, their investments might have brought much lower returns. In those lagging behind regions or developing countries, people lacked even basic computer knowledge and skills, so training and other monetary non-production costs must be very high, let alone time and psychological costs. Also, the small pool of human resources who possessed the computer technologies were not willing to move to any lagging-behind Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 151 regions or developing countries, since their opportunities in several poles were good enough. It was even hard to move human capital. Consequently, return on capital invested in several poles, could be higher than those that went to any lagging behind regions or developing countries. These initial increasing returns to capital can also provide an explanation for Keller’s (2002) finding, that initially, technology is to a substantial degree local, not global. In later phases, people even in lagging behind regions or developing countries had technological preparation for computers, from either educational resources or any other noncompany-offered channels. The nonexploitable differences between the leading poles and the lagging behind regions rise from initially below an investment threshold (due to too little technological preparation) to exploitable ones. Costs in lagging behind regions decline, but costs in those poles must rise owing to increasingly intensive competition. Then, the returns could switch to diminishing ones, if investment goes to the same pole where companies initially start. This might offer an explanation for Keller’s (2000) findings that over time, technological knowledge has become considerably more global. Also, the feature of “development phase-based increasing or diminishing returns on capital” reconciles with the argument on initial concentration and later decentralization, or the research hypothesis on initial divergence and later convergence pattern. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 152 6.3.3 Spatial Scales and Convergence There is evidence showing that spatial scales play a part in income convergence. Three theoretical points are summarized here. First, if we presume spatial scales to be existing fixes or entities theoretically, one observation is that convergence is faster on higher spatial scales. This finding indicates that the three assumptions based on the “neoclassical logic”—Marston-Smith’s (2000) analyses of uneven development and diffusion patterns of IT—may be reasonable. Second, convergence trends on higher spatial scales are more sensitive to exogenous shocks. Swings of convergence, i.e., divergence or reversing convergence, caused by an exogenous shock at higher geographic levels are more intensive. This observation is not alone in the research. It agrees with Cashin and Strappazzon’s (1998) finding that stronger divergence was found at a higher spatial level— state—than at a lower level, such as substate areas (SDs) in Australia. The findings support an argument that the existing literature could overstate the convergence intensity and might overestimate the recent divergence as well, because all of those studies were conducted on higher spatial scales. Third, an additional observation beyond the research hypothesis from the a testing indicates that incomes are more evenly distributed, if we review larger spatial units rather than smaller ones. In other words, on a lower spatial scale, such as counties in the United States, income distributions are more uneven than on a higher scale, such as U. S. states. This observation reconciles with the research hypothesis Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 153 that convergence is slower on lower spatial scales. That is, it can be the slower convergence at the county level that has led to more intensive income inequality among counties. If this holds, then in the future, the county income inequality issue will be more severe than the state’s. Moreover, the differences of convergence and reversed convergence among different levels suggest that spillover and diffusion of information technology is relevant to spatial scales. For instance, for the long-term, the stronger convergence at a higher spatial scale indicates that recent spillover and diffusion may be crossing greater distance now. This can support the argument for an across-large-distance distribution pattern o f firm locations and branching. 6.3.4 Spreading-out Type of Income Convergence The statistical evidence of neither convergence nor divergence for all of the center periphery pairs poses a question mark to the research hypothesis. One explanation can be that the finding is subject to the datum bias, because this research uses “county” to look at income distributions within an area with only a few counties. That is, the conclusion drawn from statistical tests may underestimate inequality owing to the “aggregation and averaging effects.” Even so, data show a stronger convergence trend in the long run than during the short term. Also, the divergence or nonconvergence trends were stronger in the initial years of the “Information Age” than in the later years. These nonstatistical observations still Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 154 support the research hypothesis for the longer time spreading-out convergence and the short-term reversals. Moreover, be cautioned here. The finding of a weak spreading-out convergence may not mean weak decentralization for the whole economy. As discussed previously, both spreading-out and trickling-down convergence stand for decentralization, but the spreading-out convergence corresponds with suburbanization, and the trickling-down convergence implies across large distance decentralization. One explanation can be that an across-big distance decentralization pattern may substitute for the across-short distance pattern. In other words, a great distance global/regional decentralization may be replacing the short distance suburban decentralization from the combined evidence of the relatively strong trickling down and the weak spreading-out convergences. Both stronger trickling- down convergence and stronger high geographical level convergence suggest across- large distance decentralization. Actually the findings in this regard agree with the two features of the new geo-technological movements, moving away from the poles to places in less developed states in the United States, and migrating offshore to developing or low-cost countries, as discussed in the “3.2 Analysis” section o f chapter 4. Be cautioned here that this assumption is only implied from the findings in this dissertation and needs further investigation. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 155 6.4 Further Research Issues The shortcomings and limitations of our understanding of convergence patterns analyzed in previous parts of this dissertation suggest that we should extend our research into three dimensions. The first is the temporal dimension. One concern in this research is that the findings here still cannot account for long-term observations. The long-run patterns in the presence of information technology should be examined. This can be conducted when long-range data become available. Second, more work should address why convergence is faster on higher spatial scales, such as states, than on lower scales, counties. We may seek explanations from the distribution patterns of firms and capital. That is, across-large distance decentralization of firms and across higher-spatial scale diffusions of technologies may be a reason. This needs further investigation into whether there are significant across bigger-distance relocations of firms— a phenomenon that is not studied in the literature, but only reported in mass media. The across-bigger distance decentralization could lead to faster convergence on higher spatial scales. Finally, the income convergence trends in comparatively micro spatial units or in lower spatial scales should be evaluated. The investigation at the city or zip code zone level should show a better picture on this issue. Currently, there are no adequate data for income convergence testing at those levels from the Bureau of Census. Because of this datum limitation, ad hoc datum collection is needed. Also, investigations into convergence for some units, such as Sunbelt vs. Snowbelt and Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 156 rural vs. urban areas, can help to understand the convergence patterns from other perspectives. 6.5 Implications for Social Income Inequality Although a discussion on social income inequality is beyond the scope of this dissertation, the findings in the spatial income inequality could be extended to some possible implications on social income inequality. These points are hints rather than any conclusions extended from the findings. The goal of going the extra mile here is to raise interests in the social income-convergence research. It is hoped that some people can do research on the hypotheses, as the researcher believes to be interesting and meaningful for policymaking. Social income convergence behavior may share some similar features with spatial income convergence. Therefore, it is likely that exogenous shocks may have an impact on social income convergence, too. Similarly, significant technological shocks may initially drive a short-term divergence or slower convergence; but in later phases, the shocks should contribute to long-run convergence o f social inequalities. In addition, there might be a possibility that social income convergence occurs faster in the early stages, but slower in the late stages of an economy. Furthermore, social convergence trends in relatively macro scales or larger aggregation units may appear stronger than those in micro scopes or smaller aggregation units. Also, social inequality issues might be more severe, if we review them in comparatively micro scopes or by smaller groups. There can be aggregation Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 157 and averaging effects on evaluating convergence of social income inequalities as well. That is, if we evaluate income inequality on lower spatial scales or in comparatively micro scopes, the inequality may be more severe than if evaluated on higher scales or macro scopes. 6.6 Policy Implications Some policy implications can be drawn from the findings of this research. First, since for a long run or a relatively long time period spatial incomes converge, divergence appears to be short-term. These divergences caused by exogenous shocks may be self-corrected automatically after a long period of time. However, if those short-term divergences could cause social instability, such as riots like the one in Los Angeles in the 1990s or economic inefficiencies, policymakers may consider ameliorating those impacts with income redistributive treatments. Even so, they should keep in mind that those treatments should be short-term rather than permanent. Second, the empirical evidence in this research indicates that the convergence features vary on different spatial scales. The income inequalities are more severe on lower spatial levels than at the higher levels. Moreover, the convergence speeds are slower at lower levels, such as U. S. counties, than at higher levels, e.g., the U. S. states. Therefore, if there are any financial treatments, they should be considered in comparative micro spatial units rather than in macro units. For Gyourko’s (1998) argument for place-based aid, there might be a revision. A small-place-based aid Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 158 policy may be more effective in ameliorating income inequality than place-based aid. In general, the redistribution at higher levels is less necessary than at lower levels. Also, convergence findings cannot be reasonable evidence for policymaking on different levels. Finally, the spatial income inequality is more severe at the beginning stages of an economy than in later stages, if we use Rostow et al.’s (1960) definition on stages o f growth for temporally dividing the economic development process. This implies that, in developing economies, there could be more and stronger income redistribution policies than in developed economies. One example is that governments in China, may implement stronger income redistribution policies and measurements than governments in the U. S. should. This also means that there should be less policy intervention in the future than at present and in the past. It sounds like, if a baby is too young, she needs more help. As soon as she grows up, we should let her make up her mind and do her job on her own. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. REFERENCES Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 160 REFERENCES Alesina, A., & Rodrik, D. (1994). Distributive politics and economic growth. Q u a r t e r ly J o u r n a l o f E c o n o m ic s , 1 0 9 , 465-490. Alonso, W. (1968). Urban and regional imbalances in economic development. Economic Development and Cultural Change, 77,1-14. Alonso, W. (1980). Five bell shapes in development. Papers o f the Regional Science Association, 45, 5-16. Amos, O. (1988). Unbalanced regional growth and regional income inequality in the latter stages of development. Regional Science and Urban Economics, 18, 549-566. Anonymous. (2003, July 19). The new geography of the IT industry (special report—information technology). The Economist, 47-49. Auchard, E. (2003). Reuters: One in 10 U.S. tech jobs may move overseas, report says. CNET. Retrieved August 11,2003, from http://www.cnet.com/ investor/news/newsitem/0-9900-1028-21357656-0.html?tag=ltnc, Outsourcing-Russia.com, retrieved August 11,2003, from http://www.outsourcing-russia.com/news/2003/08/05/01/, & Yahoo! News, retrieved August 11,2003, from asia.tech.yahoo.com/030729/3/1 ly6r.html. Audretsch, D. B., & Feldman, M. P. (1996). R&D spillovers and the geography of innovation and production. The American Economic Review, 86 (3), 630- 640. Austin, J. S., & Schmidt, J. R. (1998). Convergence amid divergence in a region. Growth and Change, 29, 67-89. Baily, N. N. (2002). Distinguished lecture on economics in government, the new economy: Post mortem or second wind. Journal o f Economic Perspectives, 16(2), 3-22. Baily, N. N., Bartelsman, E. J., & Haltiwanger, J. (2001). Labor productivity: Structural change and cyclical dynamics. The Review o f Economics and Statistics, 83 (3), 420-433. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 161 Baily, N. N., & Gordon, R. J. (1988). The productivity slowdown, measurement issues, and the explosion of computer power. Brookings Papers on Economic Activity, 19 (2), 347-420. Baily, N. N., & Lawrence, R. Z. (2001, May). Do we have a new e-economy? AEA Papers and Proceedings. Bakos, Y. (2001). The emerging landscape for retail e-commerce. Journal o f Economic Perspectives, 75(1), 69-80. Barkley, D., & Keith, J. (1991). The locational determinants of western non-metro high tech manufacturers: An economic analysis. Western Journal o f Agricultural Economics, 16 (2), 331-344. Barro, R. (2000). Inequality and growth in a panel of countries. Journal o f Economic Growth, 5, 5-32. Barro, R., & Sala-I-Martin, X. (1991). Convergence across states and regions. Brookings Papers on Economic Activity, 1 ,107-182. Barro, R., & Sala-I-Martin, X. (1992). Convergence. Journal o f Political Economy, 100 (2), 223-251. Bauer, P. W. (1992, January 1). Unbalanced growth and the U. S. productivity slowdown. Economic Commentary. Benjamin, M., & Perry, J. (2003). The new job reality. U. S. News & World Report. Retrieved August 11,2003, from http://www.usnews.com/ usnews/issue/030811/biztech/l 1jobs, htm Bimie, J. E., & Hitchens, N. (1998). Productivity and income per capita convergence in a peripheral European economy: The Irish experience. Regional Studies, 32(2), 223-234. Blanchard, O. J. (1991). Comments and discussion. Brookings Papers on Economic Activity, I, 159-174. Bliss, C. (1999). Gabon's fallacy and economic convergence. Oxford Economic Papers, 57(1), 4-15. Borenstein, S., & Saloner, G. (2001). Economics and electronic commerce. J o u r n a l o f E c o n o m ic P e rs p e c tiv e s , 75(1), 3-12. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 162 Borts, G. H. (1960). The equalization of returns and regional economic growth. American Economic Review, 51, 319-347. Brenner, N. (2001). The limits to scale: Methodological reflections on scalar structuration. Progress in Human Geography, 25 (4), 591-614. Browne, G. H. (1989, May/June). Shifting regional fortunes: The wheel turns. New England Economics Review, 27-40. Brynjolfsson, E., & Hitt, L. (1996). Paradox lost? Firm-level evidence on the returns to information systems spending. Management Science, 42, 541-555. Brynjolfsson, E., & Hitt, L. (2000). Beyond computation: Information technology, organizational transformation, and business practices. Journal o f Economic Perspectives, 14(4), 23-48. Cardenas, M., & Porton, A. (1995). Growth and convergence in Colombia: 1950- 1990. Journal o f Development Economics, 47,5-37. Carlino, G. A. (1992, March/April). Are regional per capita earnings diverging? Business Review (Federal Reserve Bank o f Philadelphia), 3-12. Carlino, G., & Mills, L. (1996). Convergence and the U.S. states: A time-series analysis. Journal o f Regional Science, 36 (4), 597-616. Carlson, J. B. (2002, April 1). Why the optimism? Economic Commentary. Cashin, P. A., & Strappazzon, L. (1998). Disparities in Australian regional incomes: Are they widening or narrowing? The Australian Economic Review, 3/(1), 3-26. Chen, J., & Fleisher, B. M. (1996). Regional income inequality and economic growth in China. Journal o f Comparative Economics, 2 2 ,141-164. Cheshire, P., & Magrini, S. (2000). Endogenous processes in European regional growth: Convergence and policy. Growth and Change, 3 1 ,455-479. Clarke, G. (1995). More evidence on income distribution and growth. Journal o f Development Economics, 47, 403-427. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 163 CNN.com. (2003). Report: 1 in 10 tech jobs may go overseas. CNN.com. Retrieved July 31,2003, from http://www.cnn.com/2003/ TECH/internet/07/30/jobs. oversees, reut/ Commission o f the European Communities. (1993). European economy: The economics o f community public finance. Luxembourg: Commission o f the European Communities. Comford, J., Gillespie, A., & Richardson, G. R. (2000). Regional development in the information society. In K. Ducatel, J. Webster, & W. Herrmann (Eds.), The information society in Europe: Work and life in an age o f globalization (pp. 21-44). Lanham, MD: Rowman and Littlefield. Coughlin, C., & Mandelbaum, T. (1988). Why have state per capita incomes diverge recently? Federal Reserve Bank o f St. Louis Review, 70(5), 24-36. Coulombe, S. (2000). New evidence of convergence across Canadian provinces: The role of urbanization. Regional Studies, 34(8), 713-725. Cuadrado-Roura, J. R. (2001). Regional convergence in the European Union: From hypothesis to the actual trends. Annuals o f Regional Science, 35, 333-356. David, P. A. (1990). The dynamo and the computer: An historical perspective on the modem productivity paradox. American Economic Review, 80, 355-361. Dluhosch, B. (1997). Convergence of income distributions: Another measurement problem. Constitutional Political Economy, 8, 337-352. Drennan, M., & Lobo, J. (1999). A simple test for convergence o f metropolitan income in the United States. Journal o f Urban Economics, 46, 350-359. Drennan, M., Tobier, E., & Lewis, J. (1996). The interruption o f income convergence and income growth in large cities in the 1980s. Urban Studies, 35(1), 63-82. Easterlin, R. (1994). Twentieth century American population growth. In S. Engerman & R. Gallman (Eds.), The Cambridge economic history o f the United States, III. The twentieth century. Cambridge, England: The Cambridge Press. Ferreira, A. (2000). Convergence in Brazil: Recent trends and long-run prospects. Applied Economics, 32, 479-489. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 164 Frantzen, D. (2002). Intersectoral and international R & D knowledge spillovers and total factor productivity. Scottish Journal o f Political Economy, 49(3), 280-303. Fraumeni, B. M. (2001). E-commerce: Measurement and measurement issues. AEA Papers and Proceedings, 91(2), 318-322. Gamick, D. H. (1990). Accounting for regional differences in per capita personal income growth: An update and extension. Survey o f Current Business, 70, 29-40. Gamick, D. H., & Friedenberg, H. L. (1882). Accounting for regional differences in per capita personal income growth, 1929-79. Survey o f Current Business, 62, 24-34. Gaspar, J., & Glaeser, E. (1998). Information technology and the future of cities. Journal o f Urban Economics, 4 3 ,136-156. Glaeser, E., Kallal, H. D., Scheinkman, J. A., & Shleifer, A. (1992). Growth in cities. Journal o f Political Economy, 1 0 0 ,1126-1152. Gong, F., & Gyourko, J. (1998). Evaluating the costs of increased lending in low and negative growth local housing markets. Real Estate Economics, 26(2), 207-234. Gordon, P., Richardson, H., & Yu, G. (1998). Metropolitan and non-metropolitan employment trends in the U. S.: Recent evidence and implications. Urban Studies, 35(7), 1037-1058. Graham, S. (1998). The end of geography or the explosion of place? Conceptualizing space, place, and information technology. Progress in Human Geography, 22(2), 165-185. Greenspan, A. (2000, October 19). Challenges fo r monetary policy-makers. Speech, Board of Governors of the Federal Reserve System. Griliches, Z. (1994). Productivity, R. & D. and the data constraint. American Economic Review, 8 4 ,1-23. Griliches, Z. (1999). R & D and productivity growth: Recent evidence and the uncertain future. In R. Barrel, G. Mason, & M. O’Mahoney (Eds.), Productivity and competitiveness. NIESR/Cambridge University Press. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 165 Grossman, G. M., & Helpman, E. (1994). Protection for sale. American Economic Review, 84 (4), 833-850. Gyourko, J. (1998). Place-based aid versus people-based aid and the role o f the urban audit in a new urban strategy. Cityscape: A Journal o f Policy Development and Research, 3(3), 193-203. Hagel, J., & Armstrong, A. G. (1997). Net gain: Expanding markets through virtual communities. Cambridge, MA: Harvard Business School Press. Haynes, M., & Thompson, S. (2000). The productivity impact o f IT deployment: An empirical evaluation of ATM introduction. Oxford Bulletin o f Economics and Statistics, 62(5), 607-619. Hedley, R. A. (1999). The information age. Social Science Computer Review, 17 (1), 78-87. Hibbs, D. (1973). Mass political violence: A cross-sectional analysis. New York: Wiley. Hicks, D. A., & Nivin, S. R. (2000). Beyond globalization: Localized returns to IT infrastructure investments. Regional Studies, 34(2), 115-127. Hirschman, A. (1958). The strategy o f economic development. New Haven: Yale University Press. Hoffer, H., & Worgotter, A. (1997). Regional per capita income convergence in Austria. Regional Studies, 37(1), 1-12. Hoover, E. (1936). The measurement o f industrial localization. Review o f Economics and Statistics, XVIII, 162-171. Jones, B. (2002). Economic integration and convergence of per capita income in west Africa. African Development Review, 74(1), 18-47. Jorgenson, D. W. (2001). Information technology and the U. S. economy. The American Economic Review, 91(1), 1-32. Kangasharju, A. (1998). p convergence in Finland: Regional differences in speed of convergence. Applied Economics, 30, 679-687. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 166 Kauffman, R. J., & Walden, E. A. (2001). Economics and electronic commerce: Survey and directions for research. International Journal o f Electronic Commerce, 5(4), 5-111. Keller, W. (2002). Geographic localization of international technology diffusion. The American Economic Review, 92 (1), 120-142. Kuznet, S. (1955). Economic growth and income inequality. The American Economic Review, 4 5 ,1-28. Lande, P., & Gordon, P. (1977). Regional growth in the United States: A reexamination of the Neoclassical model. Journal o f Regional Science, 17 (1), 61-69. Lau, T., Wong, Y. H., Chan, K. F., & Law, M. (2001). Information technology and the work environment—Does IT change the way people Interact at work. Human Systems Management, 20, 267-279. Leebaert, D. (Eds.). (1998). The future o f the electronic marketplace. Cambridge, MA: MIT Press. Lichtenberg, F. R. (1995). The output contributions of computer equipment and personnel: A firm-level analysis. Economics o f Innovation and New Technology, 3 ,201-217. Litan, R. E., & Rivlin, A. M. (2001). Projecting the economic impact of the Internet. AEA Papers and Proceedings, 91(2), 313-317. Livemier, W., Rickman, D. S., & Partridge, M. D. (1995). Variation in U. S. state income inequality: 1960-1990. International Regional Science Review, 18 (3), 355-378. Lucas, R. (1988). On the mechanics of economic development. Journal o f Monetary Economics, 22, 3-42. Lucking-Reiley, D., & Spulber, D. F. (2001). Business-to-business electronic commerce. Journal o f Economic Perspectives, 75(1), 55-68. Lum, S. K. S., Moyer, B. C., & Yuskavage, R. E. (2000). Improved estimates of gross product by industry for 1947-98. Survey o f Current Business, 80(6), 24-54. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 167 Mallick, R. 1993). Convergence of state per capita incomes: An examination of its sources. Growth and Change, 24, 321-340. Marston, S. A. (2000). The social construction of scale. Progress in Human Geography, 24(2), 219-242. Marston, S. A., & Smith, N. (2000). States, scales and households: Limits to scale thinking? A response to Brenner. Progress in Human Geography, 25(4), 615-619. Maudos, J., Pastor, J., & Seranno, L. (1999). Economic integration, efficiency and economic growth: The European Union experience. Applied Economics Letters, 6, 389-392. Maudos, J., Pastor, J., & Seranno, L. (2000). Efficiency and productive specialization: An application to the Spanish regions. Regional Studies, 34(9), 829-842. Maxwell, P., & Hite, J. (1992). The recent divergence of regional per capita incomes: Some evidence from Australia. Growth and Change, 23, 37-53. McCarthy, L. (2000). European integration, urban economic change, and public policy responses. Professional Geographer, 52(2), 193-205. McCoskey, S. K. (2002). Convergence in sub-Saharan Africa: A nonstationary panel data approach. Applied Economics, 34, 819-829. Morrill, R. (2000). Geographic variation in change in income inequality among U. S. states, 1970-1990. The Annals o f Regional Science, 3 4 ,109-130. Morrison, C. J. (1997). Assessing the productivity of information technology equipment in the U. S. manufacturing industries. The Review o f Economics and Statistics, 471 -481. Moulaert, F., & Djellal, F. (1994). Information technology consulting firms: Economies of agglomeration from a wide-area perspective. Urban Studies, 32(1), 105-122. Myrdal, G. (1957). Rich land and poor. New York: Harper. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 168 Nagaraj, R., Varoudakis, A., & Vaganzones, M. (2000). Long-run growth trends and convergence across Indian states. Journal o f International Development, 1 2 ,45-70. Nissan, E., & Carter, G. (1993). Income inequality across regions over time. Growth and Change, 24, 303-319. Nissan, E., & Carter, G. (1999). Spatial and temporal metropolitan and nonmetropolitan trends in income inequality. Growth and Change, 3 0 ,407- 429. Oliner, S., & Sichel, D. (2000). The resurgence of growth in the late 1990’s: Is information technology the story. Journal o f Economic Perspectives, 14 (4), 3-22. Oshima, H. (1992). Kuznet’s Curve and Asian distribution trends. Hitotsubashi J. Economy, 33(1), 95-111. Oxley, L., & Greasley, D. (1999). A Nordic convergence club? Applied Economics Letters, 6 , 157-160. Pastore, M. (2003). Gartner says tech jobs will continue to move overseas. Datamation. Retrieved August 11, 2003, from http://itmanagement. earthweb. com/career/article.php/2241641, & InternetNews.com, retrieved August 11,2003, from http://www. internetnews, com/bus-news/article.php/2241641. (Cited from the above two sites: “By the end o f 2004, one out o f every 10 jobs with U.S.- based IT vendors and IT service providers will move to emerging markets, according to Gartner Inc., as will one out of every 20 IT jobs within user enterprises.”) Perloff, H. S. (1963). How a region grows. New York: Committee for Economic Development. Persson, T., & Tabellini, G. (1994). Is inequality harmful for growth: Theory and evidence. American Economic Review, 84, 600-622. Pompili, T. (1994). Structure and performance of less developed regions in the EC. Regional Studies, 28(7), 679-693. Quah, D. T. (1993). Gabon’s fallacy and tests of the convergence hypothesis. Scandinavian Journal o f Economics, 95, 427-443. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 169 Quah, D. T. (1996). Empirics for economic growth and convergence. European Economic Review, 40, 1353-1375. Ray, P. M. (1996). Information economics and libraries in the digital age. The Bottom Line: Managing Library Finances, 9(2), 29-34. Rey, S. J., & Montouri, B. D. (1999). U. S. regional income convergence: A spatial econometric perspective. Regional Studies, 33(2), 143-156. Richardson, H. (1973). Regional growth theory. New York-Toronto: John Wiley & Sons. Romer, P. M. (1986). Increasing returns and long-run growth. Journal o f Political Economy, 9 4 ,1002-1037. Romer, P.M. (1990). Endogenous technological change. Journal o f Political Economy, 98 (5), S72-S101. Romer, P. M. (1994). The origins of endogenous growth. Journal o f Economic Perspectives, 5(1), 3-22. Rostow, W. W. (1960). The stage o f economic growth. Cambridge, MA: The Cambridge University Press. Sala-I-Martin, X. (1996). Regional cohesion: Evidence and theories o f regional growth and convergence. European Economic Review, 4 0 ,1325-1352. Schoen, J. W. (2003). Finding it harder to make ends meet. MSNBC. Retrieved August 11,2003, from http://www.msnbc.com/news/931172.asp?0cb=- 216166514. (Cited from the site: “A lot of companies have left San Diego and they’ve left California in fact because the taxes are so high here in the state.”) Smith, N. (1984). Uneven development: Nature, capital and the production o f space. Oxford and Cambridge, MA: Blackwell. Smith, N. (1992). Geography, difference and the politics of scale. In J. Doherty, E. Grahan & M. Mallek (Eds.), Postmodernism and the social sciences (pp. 57- 79). London: Macmillan. Smolny, W. (2000). Post-war growth, productivity convergence and reconstruction. Oxford Bulletin o f Economics and Statistics, 62(5), 589-606. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 170 Stiglitz, J. E. (2002). Information and the change in the paradigm in economics. The American Economic Review, 92(3), 460-501. Stiroh, K. J. (2002). Information technology and the U. S. productivity revival: What do the industry data say?” The American Economic Review, 92(5), 1559-1576. Sylwester, K. (2000). Income inequality, education expenditures, and growth. Journal o f Development Economics, 63, 379-398. Talor, P. (1982). A materialist framework for political geography. Transactions, Institute o f British Geographers, 7 ,15-34. Talor, P. (1984). Introduction: Geographical scale and political geography. In P. Talor & J. House (Eds.), Political geography: Recent advances and future directions (pp. 1-7). London: & Sydney: Croom Helm. Talor, P. (1987). The paradox of geographical scale in Marx’s politics. Antipode, 19,287-306. Taskforce of Economics and Statistics Administration, U. S. Department of Commerce (1998). The emerging digital economy. Washington, DC: U.S. Government Printing Office. Thrift, N. (1996). New eras and old technological fears: Reconfiguring the goodwill of electronic things. Urban Studies, 33(8), 1463-1493. Tsionas, E. (2000). Regional growth and convergence: Evidence from the United States. Regional Studies, 34 (3), 231-238. Tsionas, E. (2001). Regional convergence and common, stochastic long-run trends: A re-examination of the U. S. regional data. Regional Studies, 35(8), 689- 696. U. S. Government Working Group on Electronic Commerce. (1998). First annual report. Washington, DC: U. S. Government Printing Office. Varian, H. R., & Lyman, P. (2000). How much information? Berkeley, CA: School o f Information Management and Systems at the University o f California at Berkeley. Retrieved May 21,2003, from http://www.sims.berkeley.edu/how- much-info/ Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 171 Weaver, J. (2003). Jobless tech exec keeps applying. MSNBC. Retrieved August 16, 2003, from http://www.msnbc.com/news/931060.asp70cb~316166514. (Cited from the site: “Nearly 5 percent of IT jobs in the U. S. have evaporated in the last year, according to the Information Technology Association o f America. What’s worse, many of those jobs are gone forever as more corporations outsource computer functions to firms overseas.” “A recent report by Forrester Research predicted that outsourcing to countries like India, the Philippines and China would cost 3.3 million American jobs and result in $136 billion in lost wages by 2015.”) Webber, D. J. (2001). A slowing of national income convergence. Applied Economics Letters, 5,709-711. Whelan, K. (2002). Computers, obsolescence, and productivity. The Review o f Economics and Statistics, 84(3), 445-461. William, D. (1999, May 25-26). Remarks by Secretary o f Commerce William M. Daley— Digital economy conference. The Conference on “Understanding the Digital Economy: Data, Tools, and Research,” Washington, DC. Xie, Q. (2002). Convergence or divergence? A revisit to the debate on U.S. regional divergence in the 1980s. The 41st Annual Meeting of the Western Regional Science Association, Monterey, California. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 172 APPENDIX MEASUREMENT AND DATUM ISSUES ON INFORMATION TECHNOLOGY INDUSTRIES There is a great challenge in analyses using the U. S. statistics for IT industries. This is because the classification for the U. S. governmental statistics uses the Standard Industrial Classification (SIC) codes, which began when Franklin Roosevelt was President. Not only were there no computers then, but the adding machines had cranks on them (Daley, 1999). The IT-related activities fall into difficult to measure categories. Even up until 1998, computers were classified in a category called nonelectronic machinery. Fraumeni (2001) argued that, as a practical matter, it does not seem feasible to separate e-commerce activities, let alone the broader IT sector from other activities. However, the U. S. is in a transition from this old SIC classification to a new one—North American Industry Classification System (NAICS). NAICS thoroughly modernizes the industry groupings and provides for a new Information Sector. Under NAICS, businesses engaged primarily in Internet commerce are classified separately from traditional retailers (Haltiwanger & Jarmin, 1999). The replacement of SIC with NAICS will improve the measurement of IT and relevant activities. With NAICS, several issues exist. First, the focus of NAICS is primarily on industries that produce information and not on hardware items, such as computers or communications equipment (Taskforce of Economics and Statistics Administration, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 173 1998). While NAICS offers much detail in terms of industries in the information and service sectors, it is unclear how easy it will be to track key aspects of the digital economy without additional modification to American industry codes (Haltiwanger & Jarmin, 1999). For example, there are no current plans to classify businesses that primarily sell by e-commerce in a separate category. Second, the transition process is slow, so we cannot expect adequate data to be available soon. As recognized by Haltiwanger and Jarmin, making changes in the basic data collection plans by the U. S. statistical agencies is a very slow process. Then U. S. Secretary of Commerce Daley claimed that “in order for it to take hold across all agencies and businesses— it will take time, even years or far longer than it takes to develop Internet technology” (Daley, 1999). NAICS is being implemented by the statistical agencies over a 7-year horizon. Finally, even though it is a great advancement over the prior system, it does not adequately capture the changes emerging from the growth of e-commerce. Thus, several points should be addressed here. First, all of the current understanding on spatial income inequities and their changes drawn from the existing U. S. statistics have been constrained due to the SIC measurement issues, but these are the best that we can do at present. Second, using the statistics under NAICS in the future should improve our understanding, since NAICS provides a better picture on IT and relevant industries. However, some problems exist, just as mentioned above. Finally, researchers should caution about any comparison between new Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. findings using NAICS with those using SIC. Any carelessness can lead to a comparison of oranges with apples. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Travel patterns, land use, and the elderly

PDF

Suburbanization of the mind: The Hollywood urban imaginarium and the rise of American suburbia in the Colombian Caribbean

PDF

Orientalism and the other: The case of Hassan Fathy

PDF

Intra-metropolitan variation in residential mortgage prepayment risks: Unobserved factors and estimation bias

PDF

Technical efficiency and total factor productivity of United States states, 1977--1986: Multi-output distance function approach

PDF

Non-survey regional freight modeling system

PDF

Leasehold, ownership, and urban land use

PDF

An essay on Korean reunification: Spatial effects of regional integration and its implications for regional and transportation policies

PDF

Three essays on asymmetric information and public policy: Information, the environment, and the provision of defense goods

PDF

Los Angeles: The greatest experiment of modernity

PDF

Spatial dispersion of economic activity and trade in noncontiguous regions

PDF

Immigrants and housing attainment: Different forms of residential assimilation

PDF

Social capital and community economic development in Los Angeles Koreatown: Faith -based organizations in transitional ethnic community

PDF

Alberta's economic development of the Athabasca oil sands

PDF

Institutional analysis of evolution of housing cooperatives in India

PDF

Essays on the application of time series analysis to regional growth

PDF

Locating gender in the planning process: Municipal housekeeping in Santa Ana, California

PDF

New Urbanism and sustainable growth: The making of a design paradigm and public policy

PDF

The role of multifamily housing and its declining construction: How can it be judged a problem?

PDF

The political economy of Turkish Islamism and secularism: The role of preference falsification in the establishment of Kemalist secularism and the polarization of Turkish public opinion on the pu...

Asset Metadata

Creator Yu, Gang (author)

Core Title Spatial convergence patterns in the presence of information technology

Contributor Digitized by ProQuest (provenance)

Degree Doctor of Philosophy

Degree Program Planning

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag economics, general,information science,OAI-PMH Harvest,Urban and Regional Planning

Language English

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c16-500256

Unique identifier UC11340097

Identifier 3133358.pdf (filename),usctheses-c16-500256 (legacy record id)

Legacy Identifier 3133358.pdf

Dmrecord 500256

Document Type Dissertation

Rights Yu, Gang

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 au...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus, Los Angeles, California 90089, USA