University of Southern California Dissertations and Theses

Three essays on behavioral economics approaches to understanding the implications of mental health stigma

(USC Thesis Other)

Three essays on behavioral economics approaches to understanding the implications of mental health stigma

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content THREEESSAYSONBEHAVIORALECONOMICS APPROACHESTOUNDERSTANDINGTHE IMPLICATIONSOFMENTALHEALTHSTIGMA by Daehyun Kim A Dissertation Presented to the FACULTY OF THE USC GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulllment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (Economics) August 2020 Copyright 2020 Daehyun Kim Table of Contents List of Tables vi List of Figures vii Abstract viii 1 Introduction 1 2 A Decision Theoretical Model of Optimal Mental Health Belief Formation 4 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 2.2 Background Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 2.3 Conceptual Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 2.3.1 Optimism Bias: Choice of Optimal Belief . . . . . . . . . . . . . . . . . . . . 10 2.3.2 Information Avoidance: Reduced Demand for Mental Health Test . . . . . . 15 2.3.3 Response to Disappointing Partial Information: Mechanism for Developing Optimistic Bias . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 2.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 2.4.1 Expectation with Regard to Behavior . . . . . . . . . . . . . . . . . . . . . . 23 2.4.2 Behavioral Barriers to Mental Health Care Service Use . . . . . . . . . . . . . 24 2.5 Concluding Remark . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 2.6 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 2.6.1 Proofs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 3 Stigma Leads to Mental Health Information Avoidance: Experimental Analysis 30 ii 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30 3.2 Background Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 3.3 Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 3.4 Experimental Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 3.4.1 Sample . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 3.4.2 Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 3.5 Econometric Analysis and Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 3.5.1 Experiment 1: Correlational Observation . . . . . . . . . . . . . . . . . . . . 47 3.5.2 Experiment 2: Causal Eects of Reading Stigma-relevant Messages . . . . . . 53 3.5.3 Experiment 3: Eect of Partial Revelation of Information on Continuation of Information Seeking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 3.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 3.6.1 Implication on Theoretical Economics . . . . . . . . . . . . . . . . . . . . . . 58 3.6.2 Implication on Public Health Research . . . . . . . . . . . . . . . . . . . . . . 59 3.6.3 Relationship between Avoidance of Diagnosis and Diminished Willingness for Help-seeking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60 3.6.4 Dierent Eect Sizes across Subjects and Psychological Reactance Theory . . 63 3.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 3.8 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 3.8.1 Review of Theories on Mental Health Stigma and Help-seeking . . . . . . . . 67 3.8.2 Figures and Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71 4 Eect of Denial of Having Depression on Spouse's Mental Health Among Korean Elderly 84 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84 4.2 Literature Review: Spillover Eects, and Coping Strategy . . . . . . . . . . . . . . . 87 4.3 Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89 4.4 Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91 4.5 Description of Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93 4.6 Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96 iii 4.6.1 Eect of Spousal Acceptance of Having Depressive Symptoms . . . . . . . . . 96 4.6.2 Acceptance among Highly Depressed Spouses . . . . . . . . . . . . . . . . . . 96 4.7 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98 4.7.1 Suggested Interpretation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98 4.7.2 Relationship between Coping style and Choosing `Acceptance' . . . . . . . . 99 4.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 4.8.1 Appendix: Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102 Bibliography 108 iv List of Tables 3.1 List of variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 3.2 List of mental health related attitudinal variables . . . . . . . . . . . . . . . . . . . . 52 3.3 Underestimation of symptom severity . . . . . . . . . . . . . . . . . . . . . . . . . . 75 3.4 Information avoidance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 3.5 Baseline characteristics of each group in experiment 2 . . . . . . . . . . . . . . . . . 77 3.6 Self-estimation of symptom severity and demand for diagnostic information after treatments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 3.7 Eect of message provision on self-estimation of symptom severity and demand for diagnostic information (whole samples) . . . . . . . . . . . . . . . . . . . . . . . . . . 79 3.8 Eect of message provision on self-estimation of symptom severity (samples: married & employed as full time worker) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80 3.9 Eect of message provision on demand for diagnostic information (samples: married & employed as regular worker) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 3.10 Eect of message provision on the degree of stigma . . . . . . . . . . . . . . . . . . . 82 3.11 Eect of receiving CES-D test result on the willingness to receive the CES-D per- centile score . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83 4.1 List of the conditioning variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 4.2 Descriptive statistics (spousal denial vs. spousal acceptance) . . . . . . . . . . . . . 104 4.3 Descriptive statistics (adjusting for spousal depression severity) (spouse's self assess- ment: non-depression vs. depression) . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 4.4 Eect of spouse's acceptance of having depression . . . . . . . . . . . . . . . . . . . . 106 v 4.5 Eect of spouse's acceptance of having depression (Dierential eect size across spousal depressive symptom severity) . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 vi List of Figures 2.1 Timeline of decision-making process . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 2.2 Eects of psychological factors on risk underestimation . . . . . . . . . . . . . . . . . 14 2.3 Subjective belief and the choice of test type . . . . . . . . . . . . . . . . . . . . . . . 18 2.4 One-to-one correspondence between components of the model and behavioral barriers to mental healthcare use . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 3.1 Summary of experiment 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43 3.2 Procedure of Experiment 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 3.3 Self-estimation of percentile depression score ranking . . . . . . . . . . . . . . . . . . 71 3.4 underestimation of symptom severity . . . . . . . . . . . . . . . . . . . . . . . . . . . 72 3.5 Demand for depression test result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72 3.6 [Demand for test result] vs. [stigmaunderestimation] . . . . . . . . . . . . . . . . 73 3.7 Eects of message provision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74 4.1 Mechanisms of how the choice of coping strategy aects spousal depression . . . . . 89 vii Abstract In the rst essay (chapter 2), Based on the `optimal expectation model' by Brunnermeier and Parker (2005), I build a decision theoretical model of optimal mental health belief formation. The model assumes individuals optimally choose their subjective belief about their own mental health state by taking into consideration both the psychological felicity of believing they are mentally t and the potential future health cost from holding such optimistic view. Also, I add a component such that the relative importance of psychological felicity during this process is increasing in the individual level stigmatic attitude toward having mental illness. The implication is that individuals with high degree of mental health stigma develop positively biased subjective belief about their mental health state and tend to avoid knowing the true state in order to maintain the biased belief. The second essay (chapter 3) experimentally examines how individual level mental health stigma aects self-evaluation of mental tness and demand for diagnostic information. In my randomized priming experiment, to generate exogenous variation in stigmatic aect, two groups of subjects read some facts about depression with either negative or positive connotations. Next, they were asked to report their self-assessment of their own depression severity and willingness to receive the diagnostic information. Among some demographic group (married and also employed as a full time worker), under-assessment of symptom severity and avoidance of diagnostic information were more likely to be observed among those who read negative messages about depression compared to those who read positive messages. The results tentatively support a theory in which individuals are motivated to hold positively biased beliefs about their mental tness by being selectively acquire information, and mental health stigma increases these behavioral tendencies. These ndings provide a potential explanation regarding simultaneous manifestation of denial about their depression and reluctance in help-seeking among depressed individuals. In the third essay (chapter 4), using a panel data which consists of nationally-representative viii samples of Korean elderly, I analyze the relationship between denial of having depression and its impact on the spousal mental health state. The result implies that there is signicant association between denial of depression and their spouses' development of depression. Also, this association is stronger if the spouse is categorized as having a clinical depression. The potential mechanisms include rst, the direct negative emotional impact of being in denial on one's care-givers. Second, if couples share similar attitudes such as mental health stigma, these shared attitudes may cause one partner's depression and another partner's denial. ix Chapter 1 Introduction Mental health issues are as much prevalent as physical health issues are. For example, 30% of the American population is experiencing at least one diagnosable mental disorder [67]. Despite such prevalence of mental disorders, this topic has rarely been regarded as a research topic in health economics. This thesis aims to ll this gap by making use of previous ndings from theoretical microeconomics and a national-level survey dataset. Specically, the main theme of the current thesis is denial of having mental illness, including underlying mechanism of denial and consequence of denial on the family-level health. The issue of denial of having mental illness is particularly problematic in controlling negative eects of the disease because it results in avoidance or signicant delay in receiving treatment [112] despite the availability of treatments for such diseases [41]. The denial of having mental illness is closely related with mental health stigma which refers to the negative stereotype attached to those with mental health issue or those receive psychiatric treatment. According to previous research, individuals with highly stigmatic attitude tend to avoid seeking help in fear of experiencing discrimination or shame [35]. This thesis suggests one of the possible explanations in which individual level mental health stigma results in denial of having mental illness. Also, the suggested mechanism will be tested using a survey experiment. Denial of having mental illness not only negatively aects the future mental health state of those aected by the diseases, it also generates additional psychological burden to their family members who are playing caregiving roles. This hypothesis is tested using a national-level survey data. 1 In Chapter 2, it is theoretically demonstrated that mental health stigma interferes with demand for mental health care service by causing failure in recognizing one's own depression symptoms. In specic, three sub-processes are described during which stigma hinders self-recognition of hav- ing depression. First, having stigmatic attitude toward depression, potential patients motivatedly develop optimistically biased belief on their risk of having depression. Second, having this biased mental health belief, they avoid receiving diagnosis in order to maintain the belief. Third, even when the individual has decided to receive set of diagnostic information, if the earlier set of signal implies depression, he chooses not to receive the remaining set of signals to avoid additional dis- appointment. Importantly, mental health stigma increases the tendency of the behavioral patterns described in the models. Also, the three theoretical sub-processes will be discussed and related with observed behaviors of stigmatic mental health patients with respect to mental health care service use, including denial of having depression, treatment avoidance and treatment discontinuation. Chapter 3 tests the three sub-models in Chapter 2 by using Internet-based survey experiment with a particular focus on the roles of mental health stigma. First, to generate exogeneous variation in stigma level, I randomly divide subjects into two groups and provide them dierent messages, stigma inducing messages for one group and stigma reducing messages for another group. According to the theory in Chapter 2, subjects who read stigma inducing messages are expected to show more degree of under-assessment of depressive symptom and less willingness to receive the test result compared to those read stigma reducing messages. Also, in a distinctive session of experiment, I investigate the eects of receiving partial information on one's depression state on willingness to receive remaining information. Specically, I divide diagnostic information into two pieces, the brief result (whether the CES-D score is above the cuto for depression) and the detail of the result (percentile rank of CES-D score). Then the subjects are provided with brief result rst and asked whether they want to receive the detail of the diagnosis too. The results are consistent with the theory. In the rst session of the experiment, those read stigma-inducing messages are observed to make more optimistic self-assessment on their mental health states and be more likely to avoid receiving the diagnosis result. Also, in the second session of the experiment, if subjects experience disappointment when receiving the brief result, they are more likely to avoid receiving the detailed result. Chapter 4 focuses on the consequence of denial of having depression within a family. Specically, 2 based on the well-established observation of the concurrence of depressive symptoms in married couples [73], I explore whether acceptance of the illness mitigates the spillover eects of depression from individuals to their spouses using a panel survey on Korean elderly. This data set contains information of married couples, including extensive set of health and demographic variables. The main dependent variable has information on how individuals self-identify their mental health state. Specically, I test whether being in denial instead of acceptance of having depression is associated with higher degree of spousal depression severity. The results show that the individuals whose spouses are in denial have more severe depression. Moreover, this eect is more prominent if the spouse is categorized as having a clinical depression. Chapter 2 and 3 may contribute to theoretical economics by exploring the potential of informa- tion preference theories to explain real world problems. Previous theoretical literature has suggested various mechanisms to explain systematic belief bias and information avoidance, and many of them have been tested with lab experiments. However, there has been little number of previous studies that explore the applicability of those models to deal with impending socioeconomic issues. This study may have an importance as an earlier stage of attempt to show that information preference theories can provide frameworks for explaining some of the troubling issues in public health area. Also, Chapter 4 contributes to understanding the epidemiology of mental illness in family net- works. Previous studies mainly focus on the relationship between attitudes on mental illness or receiving treatment and one's own health consequences. However, there is a dearth of studies that explore how such attitudes have impacts on the health of their family members. This study provides initial motivation for future research on this topic. 3 Chapter 2 A Decision Theoretical Model of Optimal Mental Health Belief Formation 2.1 Introduction Mental health stigma is reported as one of the crucial barriers to mental healthcare service use among those might benet from available treatment [35]. The term mental health stigma includes related spectrum of concepts which are involved with fear, discrimination or contempt against those with mental disorder or those who use mental health service. Broadly, mental health stigma might be divided into two concepts, self-stigma and public stigma [8]. Among them, self-stigma refers to the negative stereotype against those mentally ill that each individual holds. Self-stigma is developed as an attitude of an individual when he accepts and internalizes the similar stereotype that is prevalent in the society. This term is the main topic of this study. This study will theoretically show how self-stigma interferes with mental health service use by leading to failure in recognizing their depression symptoms. In section 2.3, there will be three sub-processes introduced during which self-stigma plays important role in a way it hinders self-recognition of having depression. First, by having stigmatic attitude toward depression, potential patients have motivation of having erroneous (optimistically biased) belief that they are mentally t even when they experience symptoms of depression. Second, after developing this 4 biased mental health belief, they avoid receiving diagnostic information because it may force their belief bias to be unable to maintain. Third, even when the individual has decided to receive information which consists of a set of diagnosis signals, if the rst signal implies depression and it generates disappointment, he chooses not to receive the remaining set of signals to avoid further disappointment. Moreover, during all of these processes, self-stigma plays as a parameter that increases the tendency of the behavioral patterns described in the models. The idea of the model in this study is derived from the strand of theoretical literature on information preference and identity utility. There have been theoretical arguments that individuals motivatedly bias their belief about one's own traits such as ability or nancial perspective because overcondence may enhance their performance or it generates psychological felicity [23, 34, 70]. Also, individuals choose to ignore available information in order to develop or maintain their biased belief [45,65,82]. One of the possible ways to incorporate this phenomenon into utility framework is introduction of identity-utility according to which individuals derive utility directly from their belief about which group they belongs to [2,14]. Self-view about one's own mental health state might be one of the important objects that above literature can be applied to account for. However, there have been dearth of studies that investigate the behavior of those with mental disorder in the context of information preference theory. One exception a study about the denial bias and test avoidance behaviors of those with the genetic risk of Huntington's Disease [85]. However, even though this study explores the denial of the risk and consequent test avoidance, it does not deal with the mental health stigma which should be one of the most important factors that may aect those behaviors. to ll this gap, in this study, the framework of information preference, ego-utility and mental health stigma will be combined to draw a comprehensive description of the behaviors of those with the risk of developing mental disorder. The proceeding of the chapter is as follows. In section 2.2, the relevant previous literature will be reviewed. In section 2.3, the theory of under-(self)estimation of mental health risk and information avoidance will be described. Next, section 2.4 will discuss the implications of the theory. Lastly, section 2.5 will conclude the study. 5 2.2 Background Literature In this section, relevant theories from the previous literature are reviewed. According to neoclassical econpmics framework, information only has instrumental value. In other words, information itself does not directly aect utility, but it can help make better decisions. With given preferences, individuals make utility-maximizing decisions based on their beliefs about uncertainty about the state of the world. Thus, less biased or accurate belief will lead to better de- cision. For example, by being well-informed of one's own health state, one can make an appropriate treatment decision or choose suitable lifestyle such as paying more attention to the diet. Therefore, neoclassical framework implies valuableness of information. That is, because of the benet of having accurate belief in making decision, information always has positive or at least non- negative instrumental value as long as the source of information is trustworthy 1 . However, in various contexts, individuals are often observed to actively avoid information, which should be considered as anomalous behavior from the perspective of neoclassical economics [58]. Moreover, not surprisingly, information avoidance is likely to be more prominent when unwelcome news is expected 2 . For example, individuals tend to avoid monitoring their nancial portfolios when markets are falling [65]. Also, many researches empirically demonstrate or theorize medical testing avoidance especially when one may expect that the test results imply serious illness [55,69,85]. To explain information avoidance or the contrasting observations such as demanding informa- tion of no instrumental value, various mechanisms have been proposed 3 . Under the non-strategic decision making context 4 , disappointment or regret aversion, anxiety with regard to the contents of information and optimism maintenance are among the hypothesized mechanisms (see [58, 105] for review). Even though the deeper investigation of the psychological mechanisms is beyond the aim 1 The instrumental value of information is equivalent to the extent it can help decision making. Instrumental has zero value when the information does not aect decision, for example, when decision is already made or only one available option is presented to the decision maker. 2 When avoiding information is not viable option, individuals are also often observed to choose pay diminished attention or deny the credibility of information, thus minimally update their belief in response to the unwelcome information. For example, individuals tend not to fully update their beliefs about their intelligence when obtaining IQ test results that fall short of their a priori expectations, compared to the opposite case [45,82]. 3 Information seeking is the direct counterpart of information avoidance. Individuals often seek information even when the information does not benet for decision making (see [48,49]). 4 Information avoidance could be strategically driven decision in some contexts. For example, individuals might strategically choose not to know the health risk of cigarette smoking because overestimating the risk of smoking is even helpful to control their temptation to start smoking [31]. 6 of this study, it is important to note that many of the mechanisms posed assume the signicant roles of emotion in information acquisition decision. Given that neoclassical framework is silent about the roles of emotion on decision making, some revisions to the neoclassical utility theory are required in a way to incorporate emotion in preference. In their seminal study, Caplin and Leahy (2001) [29] extend classical utility theory and come up with a framework that allows for derivation of utility from emotions as well as consumption outcome. In their `anticipatory utility' framework, individuals form emotion based on their beliefs regarding uncertainty involved with unrealized outcomes of interest, and the emotion at the moment directly aects instantaneous utility experienced 5 . This framework has methodological importance because it builds a decision theoretical foundation of the idea that individuals derive utility from their belief, through mediating roles of emotion. Direct implication of deriving utility from belief is that individuals may have preference for biased belief over accuracy. For example, individuals would enjoy having optimistically biased belief about one's own ability, future wealth or future health state because optimism might be more emotionally pleasurable, or even materially benecial than objective or pessimistic perspective by enhancing performance 6 . Consequently, a majority of people are observed to answer they are better than average when evaluating themselves on positive traits such as competence [22], likelihood of business success [27] or one's life expectancy [91]. This view is consistent with dissonance avoidance theory and identity-based preference frame- work. Accodring to economic applications of those areas, individuals tend to form their self-images to be aligned with their preferred selves in order to avoid cognitive dissonance in dimensions such as ability, integrity, health, political ideology and religion. Moreover, the identity-based utility is maximized when the action is identity-congruent [1,2,14,54,89]. Aligning self-image with preferred identity might be achieved by choosing the most identity-congruent action. I reviewed two possible frameworks that optimal belief can be dierent from the objective probability. First, if `belief' itself results in utility (anticipatory utility), it can be easily modeled 5 It is important to note that the anticipatory utility is conceptually distinguished from expected utility. That is, anticipatory utility is experienced at the very moment based on current emotional state, whereas expected utility is simply an expectation about utility that will be experienced in the future. (see [29] for detail. Also, one of simple examples can be found from the study by Oster et al. (2013) [85]) 6 There are a lot of evidence and theoretical arguments about the benets of optimism. For example, while pessimism may cause anxiety, optimism can enhance condence and willpower, thus result in better performance [13, 34]. Also, there is a research evidence that optimism bias is associated with better mental health [106]. 7 that biased belief is preferred to the objective belief. Second, if individuals derive utility from identity-congruent action (identity-based utility), even without assuming the belief based utility, because individuals choose actions based on their belief, in order for the chosen action to be optimal action based on their belief, they might have motivations to have biased belief. In accordance with these ideas, a growing literature within economic theory has explored the idea that optimistic belief is optimally chosen one [13, 15, 30, 70{72]. Importantly, having some control over their belief is the underlying assumption of the idea that the biased belief is in fact motivated. Following the well-established framework that belief is formed and updated by receiving information, selective acquisition of information is regarded as a mechanism by which manipula- tion of their own beliefs can be possible. Namely, individuals may select the sources of information that could conrm or shift their current beliefs toward the preferred beliefs [51, 63]. The tac- tics that people may use to be selectively exposed to information include avoidance of potentially threatening information source [55], dismissing the credibility of unwelcome information [98], biased interpretation of information [45, 82, 93], motivated inattention [65] or motivated forgetting [17]. Taken together, when a person has a preference toward a specic belief, he might end up having belief with self-serving, often optimistic, bias, and simultaneously, is likely to show selective infor- mation exposure behavior in order to maintain or achieve the preferred belief. The consequence (biased belief) and underlying motivations (maintaining preferred self-image) of each tactics may be overlapped and I will focus on the avoidance of information in this study. Even if optimism is preferred, individuals rarely become unrealistically optimistic. Individuals also take into consideration the cost of inaccurate belief when forming belief because belief bias may lead to mistaken action. For example, overly optimistic belief about one's health state may result in unhealthy behaviors or missing timely treatment. Economics literature has built theoretical arguments on the course of `optimal belief' formation in which individuals choose the optimal level of belief bias by weighing the immediate anticipatory utility benets of having optimistic belief against the costs of having inaccurate belief [13,15,23,70,99]. Among this strand of studies, `optimal expectation' model by Brunnermeier and Parker (2005) [23] is notable in that it explicitly distinguishes objective belief and subjective (optimal) belief, while allowing for the possibility that individuals restrain their belief bias by exerting cognitive resources when the stakes are large. It is consistent with cognitive dual-process theory in which human de- 8 cision making relies on impulsive, automatic process of limbic brain system when doing habitual tasks or stakes are low, however starts activating rational, analytic process of prefrontal cortex (PFC) in brain when stakes are large enough to require some conscious reasoning [20, 64]. This model provides a parsimonious framework for prior belief formation, in which individuals rst start as if they have objective belief then choose the degree of optimal bias, nally, the optimal prior is determined as the sum of objective prior and optimal bias. Experimental study that directly demonstrates the validity of this model is rare, probably because of the diculty in measuring bias in prior beliefs 7 . Among empirical literature, Oster et al. (2013) [85] demonstrate that optimal expectation model is consistent with the behaviors of individuals at risk of Huntingon's disease such as optimistic bias or low screening test rate. In the next section, I describe this theory with application of mental health belief. 7 Only experimental study that I nd to directly test the optimal expectation model by Coutts (2019) [39]. However, this study does not nd support for optimal expectation model. According to the optimal expectation model, belief bias is expected to be reduced when the belief accuracy is associated with higher value of reward, but such anticipated eect is not observed in the study. 9 2.3 Conceptual Framework In this section, following the optimal expectation model by Brunnermeier et al. (2005) [23], I build a model of mental health belief formation according to which individuals tend to motivatedly underestimate the probability of having clinical depressive disorder, which is comparable to under- evaluation of depressive symptoms in self-assessment stage. Also, I add some modication of the optimal expectation model in a way that the agent's mental health stigma aects the degree of such bias, with more underestimation of the risk among those have higher level of stigma. Moreover, in order to maintain the biased belief, individuals intentionally choose to forgo relevant information on their mental health states such as depression test result which may force their belief to be updated. Before proceeding, dening the concrete meanings of some terminologies might be useful. First, the term `stigma' refers to the degree of negative attitude toward depressed people that an indi- vidual hold. Following the cognitive dissonance literature, this term may also represent the degree that the individual dislikes thinking oneself to be depressed. Second `objective belief' means the objective probability that an individual is diagnosed with depression based on his level of depressive symptom. Third, `subjective belief' refers to the self-estimated probability of being diagnosed with depression. Lastly, `belief bias' is dened as the dierence between the `objective belief' and the `subjective belief'. Specically, if an individual underestimates the possibility of being diagnosed with depression, that is, subjective belief is lower than objective belief, this bias is called `optimism bias'. 2.3.1 Optimism Bias: Choice of Optimal Belief First, I present a model of optimism bias. This model depicts the mechanism of why individuals underestimate the probability of being diagnosed with depression. Also, the role of stigma during this process will be described. The course of events and actions consists of two stages, t2f0; 1g. Individuals start with objective probability of having depression. In other words, they behave as if they rst accurately know this probability. In stage t = 0, individuals choose `optimal' subjective belief and then choose appropriate health behavior in accordance with this optimal subjective belief. In t = 1, the individual's health utility is realized. For example, if an individual starts to believe himself to be depression patient, he may decide to visit a psychiatrist (t = 0). The eect of 10 Figure 2.1: Timeline of decision-making process medication is observed later (t = 1). As in Figure 2.1, in the rst stage (t=0), the individual forms a subjective belief q 0 on the probability of being diagnosed with depression. This subjective belief is called `optimal' belief because the choice ofq 0 is determined by optimization process during which the individual considers both the loss of future utility resulting from biased belief and the current felicity of having optimistic belief. This argument implies the belief bias is motivated rather than being a random calibration error. The utility function used during the optimization process consists of identity-based utility as well as the standard expected utility. Let s2f0; 1g be an element of a binary state space where s = 1 represents the individual has depression and s = 0 represents being healthy. The objective prior 8 p is a probability of being diagnosed with depression if detailed assessment of mental health state is administered. Apart from the objective probability of having depression, individuals hold subjective belief q 0 about this probability, which might be formed through unconscious process of evaluation of various events experienced in daily life such as experience of low mood spell. In the course of forming subjective belief, the individual's stigma aects the degree of the belief bias. The level of stigma is denoted as stig. This setting can be summarized as below equations. For individual i, the objective prior (p i ) and the subjective probability (q i0 ) of having depression follow the relation : q i0 =p i B i B i =f(stig i ) (2.1) 8 I use the term `prior' instead of `probability' to capture the idea that individuals start with this objective probability p and make choices based on this. Those choices include the subjective probability q0 and action a. 11 where the variable B i (belief bias) represents the dierence between objective probability and subjective probability. The individual is dened as having optimism bias if B i > 0. Also, the function f summarizes the mechanism through which stigma generates optimism bias. Above setting comes from two assumptions. First, the discomfort from the belief that oneself has depression is higher among those have greater degree of stigma. In other words, becoming a depression patient might be deemed more threatening in maintaining positive self-image when the individual has negative view against depression 9 . Second, following the framework of optimal expectation model, individuals choose their subjective probability based on the objective prior. In detail, the optimal expectation model posits that individuals behave as if they, at rst, know the objective probability, then choose optimal level of optimism bias, and next, they forget the objective prior and live with the subjective belief. This framework is consistent with the idea of dual process theory in psychology, according to which human thought is, in the most cases, dominated by automatic, unconscious process of the limbic system which is related with biased belief, but when making important decision, by exerting some cognitive resources, controlled and conscious process of the prefrontal cortex become more prevalent so that individuals can recall the objective probability and make a better decision [20]. The below formulation is optimization problem an individual faces in t=0. The action chosen in this stage is denoted as a. Ex post Utility is a function of chosen action a2 [0; 1] and the realized state s, given the degree of stigmatizing attitude stig. The utility is denoted asu(a;sjstig). Utility loss is minimized when action is state-congruent but taking action suitable for controlling depression such as seeing a psychiatrist generates shame and is regarded as threatening for positive self-image. Formally, individuals choose q 0 2 [0; 1] to maximize U(^ a(q 0 )jp) = E p [Loss(^ a;s)] | {z } expected health loss stig ^ a | {z } identity-based utility where ^ a = ^ a(q 0 ) = argmax a E q 0 [Loss(a;s)] (2.2) Loss(a;sjD) is the health loss function and is the weight parameter for identity-based utility. 9 See [14,15] for more detailed argument on how identity belief is incorporated in utility framework. 12 Above formulation summarizes the decision making process including the choice of the subjective belief (q 0 ) and action (a). Assuming the individual initially knows the objective probability (p), he nds the optimal subjective belief by considering for both the expected health consequence and identity utility resulting from his action choice. The optimal action is the expected-loss minimizing action that is sought based on the chosen subjective belief. It is important to note that utility is calculated based on objective prior and action is based on the subjective belief. This is because action should be belief-consistent. The individual is assumed to forget the objective prior after choosing the subjective belief. The health loss function is dened as follows. Loss(a;s) = 8 > > < > > : (as) 2 if s = 0 W (as) 2 if s = 1 (2.3) where W is the self-assessed vulnerability to depression. I assume that W 1. Now, the optimal action ^ a( ^ q 0 ) and the optimal subjective belief ^ q 0 is obtained by backward induction manner. First, the optimal action as a function ofq 0 is obtained by nding loss-minimizing action from the second equation of equations (2). This yields ^ a(q 0 ) =q 0 W=fq 0 (W1)+1g which is increasing in q 0 and W, that is, the individuals put more eort to manage their mental health risk if they think they have a higher probability of developing depression or the cost of under-treatment is high. Next, I derive the optimal subjective belief and show that ^ q 0 always less than or equals to p. That is, individuals underestimate the risk of having depression. PROPOSITION 1 (underestimation of risk): Individuals always optimally choose ^ q 0 such that ^ q 0 p. PROPOSITION 2 (insucient healthy behavior): Due to stigma, the action chosen by an individ- ual is short for minimizing expected health cost of depression. The proposition 1 implies that individuals optimistically deviate their belief about the prob- 13 Figure 2.2: Eects of psychological factors on risk underestimation (a) Stigma ( = 1, W = 1:2) (b) Self-assessed Vulnerability to Depression ( = 1, Stigma = 0:3) ability of having depression from the objective probability in order to avoid applying stigma to themselves. This belief distortion might give them current felicity but result in sub-optimal action such as under-treatment. Also, it is important to note that the degree of optimistic distortion of belief is prominent when stig is high, or W is low, which can be shown by solving the equation (10) for ^ q 0 . Since the calculation is tedious and not an aim for this section, I skip the formula for comparative statics of ^ q 0 with respect to those two variables. Instead, Figure 2.2 displays these arguments graphically: when other variables are xed, (1) high degree of stigma (stig) leads to more underestimation bias (panel a), (2) low self-assessed vulnerability to depression (W ) leads to more underestimation bias (panel b). This result is intuitively straightforward: (1) high level of stigma is associated with reluctance to accept depression (2) if health cost of developing depression is high, individuals become less optimistic. 14 2.3.2 Information Avoidance: Reduced Demand for Mental Health Test One way to maintain biased belief is inhibition of belief update by not receiving relevant information. In this section, I describe a model according to which those with optimism bias and high level of stigma avoid diagnostic information. The decision problem that the individual faces is the choice of screening test for mental health state. There are three dierent test options available: 1) not taking the screening test, 2) simple test (receiving the screening test result only) and 3) detailed test (receiving the screening test result and additional information such as severity or curability of the recipient's mental health issue 10 ). The degree of detail of each test option is denoted as 0 , 1 and 2 , respectively. I start with dening a utility function of receiving information according to which the choice of diagnostic test is determined. In building the utility function, I hypothesize that individuals care both about the instrumental value of information, that is, the degree that individuals perceive the information is helpful for future decision making, and the emotional consequences of receiving information such as disappointment, relief or anxiety. The rst part, the perceived instrumental value of information is hypothesized as increasing in subjective risk of having depression (q 0 ) and the degree of detail of the information (). Therefore, the instrumental value of information with degree of detail is denoted as f(;q 0 ) where f q 0 > 0, f > 0 and f q 0 > 0. The rationale for increasing instrumental value of information among those more highly depressed (f q 0 > 0) is that the perceived cost of being uninformed of one's health state becomes riskier as the perceived severity of symptom increases. By analogy, failure in choosing appropriate disease-specic lifestyle including receiving treatment is more destructive for more serious disease such as cancer than lighter disease such as u. Therefore, as the individual perceives he is likely to develop depression (high value of q 0 ), he might feel more need of knowing his true mental health state. Next, it is plausible to assume that information of more detailed contents has more instrumental value (f > 0). Last inequality (f q 0 > 0) comes from the idea that the additional detail of diagnostic information might be more valuable among those highly depressed. Next, individuals also care for the emotional consequences of receiving information as well as the 10 By receiving only the test result, what the individual becomes to know is whether he is categorized as depressed person or not. Additional information is required to more rigorously assess the symptoms. 15 instrumental value of the information. To model the emotional value of information, I employ two well-documented hypotheses from information preference literature. First type is a version of gain- loss utility. Individuals experience disappointment or relief depending on the signal received [11,72, 77]. Second type is the attention-based utility. Receiving information leads individuals to pay more attention and it generates specic, in this case, unpleasant emotion [55, 59]. Additionally, I add a component that all of these emotional reactions are in uenced by the degree of stigma because stigma re ects the subjective undesirability of having depression. Formally, the utility of receiving information with the degree of detail is given as below: V (jq 0 ) = f(;q 0 ) | {z } 1) instrumental value of information 1 stig(sq 0 ) | {z } 2) emotional reaction: disappoint or relief 2 stig | {z } 3) emotional reaction: attention-based emotional cost (2.4) where s2f0; 1g is the test result such that s=0 and s=1 indicates healthy and depression, re- spectively. Also, q 0 is a prior belief which has been optimally chosen according to the model in proposition 1. In this equation, 1) f(;q 0 ) is instrumental value of information. 2) (y) is a psychological disappointment-relief function which activates when receiving diagnosis result and is weighted by constant and the degree of stigma. Also, this function is assumed to be (0) = 0, strictly increasing in y and has disappointment aversion property, that is, (y)<(y) for any y2 (0; 1]. The constant 1 is assumed to be non-negative and increasing in , which is relevant to the situation that more detailed diagnosis result entails shift of beliefs in more numbers of dimension not only whether or not it is depression, so the degree of emotional reaction becomes larger 11 . 3) Next, 2 is utility loss from paying attention to uncomfortable information which is also weighted by stigma. The value of 2 is also increasing in the value of . Overall, the emotional reaction to information becomes prevalent in the utility if the degree of stigma is high, which is consistent with the situation that stigma generates shame or anxiety if the individual comes to know or considering possibility that himself is a depression patient. The important premise from optimal expectation is that individuals can recognize himself to have biased belief when exerting cognitive resources. Consistent with this view, I assume that when facing the decision problem about choosing the degree of detail of information, individuals form 11 For example, detailed mental health assessment may include the diagnoses of mania or social anxiety disorder, not only depression. 16 expected utility based on the objective probability, not based on the subjective belief. In other words, individuals behave as if they know they are optimistically biased when making the choice of informativeness of the diagnostic test. They are willing to exert the cognitive resources because the contents of information might negatively aect their self-image and generate shame in the form of disappointment. With this assumption, the expected value of receiving information is: E p [V (jq 0 )] =f(;q 0 ) 1 stig [p(1q 0 ) + (1p)(0q 0 )] 2 stig (2.5) The implications of this formulation about the relation between subjective belief of having de- pression and demand for receiving the test is summarized in proposition 3. PROPOSITION 3 (subjective belief and demand for test): Avoiding test ( 0 is chosen) is more likely to happen among individuals with lower value of subjective belief about the probability of having depression (q 0 ), and higher level of optimistic belief bias (pq 0 ). This proposition predicts the eect of subjective belief on avoidance of mental health test. First, if an individual believes that he has a low possibility of having depression, he does not feel much need for diagnostic test because the information may have low instrumental value. Second, if an individual has optimism bias, receiving mental health test result is likely to provoke disappointment, thus, the test is avoided. The proposition 3 tells only about whether or not information is entirely avoided. If the individ- ual chooses to take a test, he should again make a choice between the basic test ( 1 ) and detailed test ( 2 ). The choice will be determined by the subjective belief of the individual, stigma level and the values of parameters in equation (5). I will illustrate this process with some simplications of the model. For simplicity, it is assumed that f(;q 0 ) = q 0 with being increasing in , and (y) = (1 + )y if y > 0 (disappointment) where > 0 is the disappointment aversion constant, and (y) =y if y< 0 (relief). Then, equation (5) becomes: 17 Figure 2.3: Subjective belief and the choice of test type (p = 0:5; = 1:2; stigma = 1:2) E p [V (jq 0 )] = q 0 1 stig [p (1 + ) (1q 0 ) + (1p) (q 0 )] 2 stig = q 0 1 stig [ p(1q 0 ) + (pq 0 )] 2 stig = q 0 1 stig [G( ;p;q 0 ) + (pq 0 )] 2 stig (2.6) where G( ;p;q 0 ) = p(1q 0 ) 12 . Also, the expected utility of receiving no information ( 0 ) is normalized to 0. Then, individuals choose the test that will give the highest expected utility. The decision problem that individuals face is given as below: = argmax 2f 0 ; 1 ; 2 g E p [V (jq 0 )] where E p [V ( 0 jq 0 )] = 0 E p [V ( 1 jq 0 )] = 1 q 0 1 1 stig [G( ;p;q 0 ) + (pq 0 )] 2 1 stig E p [V ( 2 jq 0 )] = 2 q 0 1 2 stig [G( ;p;q 0 ) + (pq 0 )] 2 2 stig (2.7) 12 The functionG( ;p;q0) is corresponding to the degree of fear about potential disappointment due to disappoint- ment aversion property. Apart from the eect of disappointment constant , the value of this function becomes larger when p is high and q0 is low, which is equivalent to the increased likelihood of disappointment. 18 Since the constants , 1 and 2 are increasing in the value of , the expectation from this formulation is that individuals are likely to seek to receive more detailed test as the value of q 0 is larger, and avoid detailed test when stigma is high or the the extent of optimism bias is large. Indeed, once the values of parameters satisfy aforementioned conditions ( 1 < 2 ; 1 1 < 1 2 , and 2 1 < 2 2 ), there exists a set of parameter values that results in increasing value ofq 0 generates increasing demands for more detailed test. Figure 2.3 provides a graphical illustrates of this model. In this illustration, if other variables including objective prior (p), disappointment aversion constant ( ) and stigma (sti) are set xed, the low value of q 0 results in avoidance of any test. Also, the medium and high level of q 0 lead to choose basic test and detailed test, respectively. This formulation is of empirical importance. If ^ denotes the type (detailedness) of test chosen based on equation (7), below inequalities hold: 1) (subjective belief): d ^ dq 0 > 0 2) (belief bias): d ^ d(pq 0 ) < 0 3) (stigma): d ^ d(sti) < 0 4) (disappointment aversion): d ^ d < 0 This comparative static might be used for reduced-form estimation of the model. If there is a data for choice of test and subjective belief, the model can be easily estimated by using ordered logit or multinomial logit method. 2.3.3 Response to Disappointing Partial Information: Mechanism for Develop- ing Optimistic Bias Above theoretical models oer expectation about the eects of stigma on having optimistic bias and information avoidance. Proposition 1 states that optimistic bias is optimal. Also, proposi- tion 3 describes individuals with optimistic belief tend to choose to avoid diagnostic information in order to maintain the optimistic bias. However, those models do not deal with mechanisms for how individuals, in the rst place, develop the bias in self-evaluation of one's mental health state. Assuming individuals start with objective prior and belief is updated in Bayesian manner, not receiving information implies stationary subjective belief, rather than evolving belief toward 19 optimistic state. In this section, I describe a simple model in which subjective belief can be evolved from objective prior to optimistically biased belief, also might be testable by experimental methods. In sum, when information is presented as a bundle, that is, information contain more than single content 13 , even when individuals rst have decided to receive the information bundle, they choose to opt out from receiving further information if the earlier revelation of part of the information bundle generates disappointment. This mechanism is consistent with the previous theoretical literature that individuals are selectively exposed to information in motivated way and thus developing belief bias, especially in the dimensions that aect their self-identity or self-condence [12,55]. First, suppose an individual already has made a decision to receive a bundle of diagnostic information, which contains both depression test result (signal A) and more detailed assessment of the symptom (signal B) 14 . Timeline is as follows. First, the individual has expectation about the realization of signal A, which is equivalent to prior belief. Next, he receives the test result (signal A) and derive psychological utility W A based on the result. Lastly, he forms psychological utility of receiving signal B which is denoted as W B , and then make a decision whether to opt out from receiving signal B. The value ofW B is a function of the individual's stigma level, the signal A, and W A . The signal A, that is, the test result which is denoted as A and can have two value: A = 0 represents not having clinical depression and A = 1 implies depression. The psychological utility function the individual derives at the time of receiving test result is given by W A ( A jq 0 ) =stig ( A q 0 ). The emotional reaction function () is the identical one with the emotional reaction function appeared in the equation (4). Especially, the positive and negative value of W A indicate disappointment and relief, respectively. Again, the emotional reaction is weighted by the degree of stigma. Next, the individuals recompute the expected utility of receiving signal B after receiving signal A. Formally, the utility of receiving signal B is given by: 13 depression diagnostic result can contain both test result regarding whether or not he has depression and also degree of severity 14 In my companion experimental study (Chapter 3), the signal B is the percentile ranking of the depression test score. The depression test result (signal A) only informs the participant of whether or not his score meets depression criteria without informing more detail regarding how severe his depression state is compared to other people. 20 E[W B j A ;p 0 ;stig] = A |{z} instrumental value + W A |{z} attention based psychological utility : disappointment-relief expereienced when receiving A + C |{z} utility of curiosity fulllment = A stig( A q 0 ) + C (2.8) On the other hands, the expected utility of not receiving signal B is normalized to be 0. Therefore, the individual proceed to receive the remaining set of information when E[W B j A ;p 0 ;stig]> 0. The rst term in the equation (10) is the instrumental value of receiving signal B. Inclusion of this term postulates that if the test result indicates depression ( A = 1), individuals might feel the need of better understanding of his symptom so that he would seek more detailed information. The constant represents the responsiveness of receiving the result of having depression on the perceived need of more detailed information. The second term is attention-based psychological utility which comes from the emotional state associated with paying attention to the signal B, which is simply assumed to be equivalent to W A . If the individual experienced disappointment when receiving signal A, he might want to divert his attention from the depression test and escape from the unpleasant situation. On the other hands, if he felt relief, he may want to continue to pay attention to the remaining information 15 . Finally, last termC > 0 represents the benet from fullling curiosity, or it can viewd as the disutility from forgoing the information that rst intended to receive. PROPOSITION 4 (Eect of stigma on the response to the partial revelation of information): With given value of q 0 , after conrming that signal A implies depression ( A = 1), the likelihood of opting out of the rst scheme of receiving signal B is increasing in stigma level. Equation (8) and proposition 4 oers expectation about the behavior of individuals when facing decision situation whether to proceed to receive the remaining part of information bundle after receiving the rst part of information. First, individuals feel the need of obtaining remaining set 15 The alternative explanation is based on the concept of disappointment aversion. Suppose that A and B are positively correlated given the true state s. It is a plausible assumption because both signals are released from the same data source (e.g., respondent's answers to the depression test). Therefore, if a subject experiences disappointment when receiving signal A, it is highly likely that signal B also generates disappointment so the subject might avoid receiving signal B. 21 of information bundle if the rst revelation of partial information implies depression because it may increase the perceived instrumental value of information. Second, however, the rst eect is likely to be oset if the rst set of information bundle induces disappointment, the individual might change the rst decision and opt out from receiving remaining set of information. Also, this decision change is more likely to occur for those who have high degree of stigmatizing attitude toward depression. It is important to note that this mechanism may contribute to the development of optimistic bias among those have high degree of stigma in the following way. First, individuals face decision problem whether to attend to or receive information that consists of a `set' of signals about the true state of the individual's mental health 16 . Even if those with stigma toward depression tend to avoid such information or choose less detailed information as discussed in the last section, it does not necessarily lead to systematic optimistic bias in self-evaluation of symptom. The belief bias, if Bayesian update is assumed, arises only when the individuals are selectively exposed to potentially relieving information. The theoretical argument discussed in this section suggests one of the potential sources of why such selective exposure to information takes place. Having been disappointed by signal A, by not receiving signal B, the further update of the belief is prevented. On the other hands, signal B is more likely to be sought if relief is experienced by receiving signal A, That is, individuals seek more signals from the same source because receiving them may allow them to update their belief in more desired way. Then the belief would be continually evolve into optimistically biased one by making similar choices in the course of life where individuals face the decision problems described in this section as various forms of activity participation decision as well as the decision of receiving treatment or diagnosis 17 . 16 Diagnostic information can contain more than one type of signal because there are many ways to design the signal structure of information. For example, depression can be assessed both as a form of a binary assessment (healthy/depressed) or degree of severity. 17 Similarly, K} oszegi (2006) [70] also assumes engaging in activities can generate signals about one's own ability so that individuals avoid an activity if it may force their optimistic belief or self-condence to be broken down. 22 2.4 Discussion 2.4.1 Expectation with Regard to Behavior The model in the last section describes the process of mental health belief formation among those have stigmatic attitude toward depression. This model consists of three distinctive sub-process. First part of the model explains that, due to the impact of perceived identity on utility, it is optimal subjective belief to underestimate the probability of having depression. Second, with this optimistically biased belief, individuals choose to forgo available diagnostic information in fear that the information forces the optimal belief to be unable to maintain. Last part of the model deals with the dynamic process of information choice which contributes to the movement of the subjective belief toward optimism. Taken together, the three parts of this model explain why (section 2.3.1) and how (section 2.3.3) the subjective belief moves toward optimism and how it is maintained (section 2.3.2). Also, all these processes are more prominent among individuals with higher level of mental health stigma. Not only this model oers the explanation about denial of mental disorder among potential pa- tients, it also provides framework for designing stratigies for empirical investigation 18 . The model consists of three sub-processes and each of them can be simplied and represented in reduced form equation with a parsimonious number of variables. The expected behavioral patterns that might be tested with experimental analysis are summarized as follows 19 : 1) Optimal subjective belief about the probability of having depression is lower than the ob- jective probability. Also, the degree of the belief bias is greater as having higher degree of stigma, also as having lower degree of perceived vulnerability to depression. q 0 <p q 0 =p +F [sti () ;W (+) ] 2) When facing decision problem of choosing mental health test type, the likelihood of choosing 18 The key variable `stigma' can be simply measured by using relevant psychological inventories such as Endorsed and Anticipated Stigma Inventory (EASI) [110]. 19 The signs at the lower subscripts in the functions represent the direction of impact of the variable. 23 more informative test is increasing in subjective belief (q 0 ), and decreasing in stigma and the degree of belief bias (pq 0 ). Demand(detail of the diagnosis) =G[q 0(+) ;sti () ;pq 0() ] 3) If a diagnostic information is composed of two signals, the likelihood that the subject con- tinues to choose to receive the second set of signal after the rst signal (A) is released is a function of stigma and the level of relief/disappointment. signal A = 1 (disappointment) : Demand(second signal) =H 1 [sti (1q 0 ) () ;sti () ] signal A = 0 (relief) : Demand(second signal) =H 2 [stiq 0(+) ;sti () ] 2.4.2 Behavioral Barriers to Mental Health Care Service Use The stage that the model in this paper depicts is prior to making a treatment decision when mental health belief with optimistic bias is formed. However, this model also can provide a potential explanation on the behavioral barriers for mental health patients to be engaged in treatment. Those barriers include denial of having depression, refusal to receive treatment and treatment non- adherence, and they could be matched with three sub-models described in the last section. This matching between the sub-models in this paper and those behavioral barriers is summarized in Figure 2.4. This section will illustrate how those behavioral barriers are related with the model in this paper. The rst barrier is denial of having depression. Depressed people are often observed to refuse to admit the possibility that they are in clinical condition and in need of help from mental health professionals, which operates as a barrier to make a treatment decision [57, 113]. For example, individuals who experience depressive episodes often dismiss the possibility that their condition may need to be handled by psychiatrists and tend to interpret the suering as the temporary low mood. Moreover, the likelihood of occurrence of the denial is largely in uenced by personal level mental health stigma. The model of optimism bias (section 2.3.1) provides straightforward explanation regarding this phenomenon of denial. The model describes that subjective belief about 24 Figure 2.4: One-to-one correspondence between components of the model and behavioral barriers to mental healthcare use the probability of having depression is subject to optimistic bias, which is comparable with the mental illness denial in clinical context. The second type of behavioral barriers for mental health service use is the avoidance of visiting formal mental health service such as psychotherapical or psychiatric clinic. Individuals who rec- ognize their emotional disturbances are due to depression still often avoid visiting mental health professionals in fear of feeling shame or being observed by others. Also, another aspect of visiting mental health clinic is receiving diagnosis, which may be threatening to maintaining positive self- view in case of receiving a result indicating depression. These types of fear is largely modulated by their personal level mental health stigma, thus stigma may increase the likeihood of treatment avoidance [8, 61]. Especially, avoidance of visiting a professional due to the fear of receiving dis- appointing diagnostic result is closely related with the model of information avoidance described in the last section (section 2.3.2). Considering that visiting mental health service involves receiv- ing diagnostic information, this model has a potential for explaining the observed behavioral of treatment avoidance among depressed individuals. The last type of behavioral barrier that will relate with the model in this paper is the discon- 25 tinuation of treatment or medication non-adherence even after the rst visit is made. Previous psychiatric literature reports that personal level stigma and under self-evaluation of the depres- sion severity are associated with antidepressant drug adherence [101]. However, the theoretical investigation on the relationship between medication adherence and stigma has been rare. The third model in this paper (section 2.3.3) might oer one possible explanation on this behavior. The model predicts that individuals who derive utility from their own self-view may choose to discontinue receiving diagnostic information if the earlier set of diagnostic information generates disappointment. One of the aspects of visiting a mental clinic on a regular basis might be learn- ing about one's mental health state in a repeated manner, which resembles the decision problem that a patient faces in the section 2.3.3. Based on the model, after receiving a diagnosis at the st visit to the clinic, conrming that he has depression which is disappointing, he might want to prevent further learning to occur by canceling their rst decision to engage in regular treatment, thus discontinuation of treatment happens. 26 2.5 Concluding Remark This paper provides a theoretical explanation on the behaviors of potential mental health patients which may interfere with their help-seeking from mental health professionals. Specically, the models of optimal subjective mental health belief and avoidance of diagnostic information are introduced. The basic setting of the model is that individuals care about their identity, that is, whether they belong to clinical depression patients. Moreover, the importance of this identity utility increases as they have greater degree of mental health stigma. In this setting, the model predicts that under-assessment of their severity of depressive symptoms and avoidance of depression test are observed among individuals who have stigmatic attitude toward mental disorder. Also, this study adds to the discussion on the potential of information preference theories to explain real world problems. Previous theoretical literature has suggested various mechanisms to explain systematic belief bias and information avoidance, and many of them have been tested with lab experiments. However, there has been little number of previous studies that explore the applicability of those models to deal with impending socioeconomic issues. This study may have an importance as an earlier stage of attempt to show that information preference theories can provide frameworks for explaining some of the troubling issues in public health area. 27 2.6 Appendix 2.6.1 Proofs PROPOSITION 1 (underestimation of risk): Individuals always optimally choose ^ q 0 such that ^ q 0 p. Proof. Because the action function ^ a(q 0 ) =q 0 W=fq 0 (W 1) + 1g is monotonically increasing inq 0 , there exists its inverse function q 0 (^ a). Then utility is: U(q 0 (^ a)jp) = [pW (1 ^ a) 2 + (1p)^ a 2 ] stig ^ a Within the possible range of action choice (a2 [0; 1]), The value of ^ a which maximizes above equation is given by: ^ a = maxf0; stig=2 +pWg p(W 1) + 1 (2.9) Because ^ a(q 0 ) =q 0 W=fq 0 (W 1) + 1g, maxf0; stig=2 +pWg p(W 1) + 1 = ^ q 0 W ^ q 0 (W 1) + 1 (2.10) Then, 1) if stigpW=2 : ^ q 0 = 0p. 2) if stig<pW=2 : Suppose ^ q 0 p, then ^ q 0 W ^ q 0 (W1)+1 pW p(W1)+1 > stig=2+pW p(W1)+1 , which is contradictory to the setting ^ q 0 W ^ q 0 (W1)+1 = stig+pW p(W1)+1 . Therefore, ^ q 0 <p. Especially, unless p = 0, q 0 <p always holds. PROPOSITION 2 (insucient healthy behavior): Due to stigma, the action chosen by an individ- ual is short for minimizing expected health cost of depression. 28 Proof. The action function is ^ a(q 0 ) =q 0 W=fq 0 (W 1) + 1g. Therefore, the loss-minimizing action is ^ a(p) =pW=fp(W 1) + 1g. By proposition 1, ^ a( ^ q 0 )< ^ a(p) as long as p6= 0. PROPOSITION 3 (subjective belief and demand for test): Avoiding test ( 0 is chosen) is more likely to happen among individuals with lower value of subjective belief about the probability of having depression (q 0 ), and higher level of optimistic belief bias (pq 0 ). Proof. The utility of avoiding test is normalized to be zero (E p [V ( 0 jq 0 )] = 0). Also, the utility of taking a basic test ( 1 ) isE p [V ( 1 jq 0 )] =f( 1 ;q 0 ) 1 1 stig[p(1q 0 )+(1p)(0q 0 )] 2 1 stig. The relative utility of choosing 0 instead of 1 isE p [V ( 0 jq 0 )]E p [V ( 1 jq 0 )] =E p [V ( 1 jq 0 )]. Performing comparative static analysis yields the results in the proposition. 1) d dq 0 fE p [V ( 1 jq 0 )]g =f q 0 1 1 stig [p 0 (1q 0 ) + (1p) 0 (0q 0 )]< 0 2) d d(pq 0 ) fE p [V ( 1 jq 0 )]g = dq 0 d(pq 0 ) d dq 0 fE p [V ( 1 jq 0 )]g> 0 Without loss of generality, the same analysis can be applied to the comparison between 0 and 2 (detailed test). PROPOSITION 4 (Eect of stigma on the response to the partial revelation of information): With given value of q 0 , after conrming that signal A implies depression ( A = 1), the likelihood of opting out of the rst scheme of receiving signal B is increasing in stigma level. Proof. The expected utility of receiving signal B is E[W B j A ;p 0 ;stig] =theta A stig( A q 0 ) +C. Since A = 1, ( A q 0 )> 0. Thus, d(E[W B ]) d(sti) < 0. 29 Chapter 3 Stigma Leads to Mental Health Information Avoidance: Experimental Analysis 3.1 Introduction It is reported that 30% of the American population is experiencing at least one diagnosable mental disorder [67]. Despite such prevalence of mental disorders, research shows that many people with mental disorder delay or fail to seek help from mental healthcare system and do not receive timely care although eective treatments for a wide range of mental disorders are available [68,111]. For example, according to a study based on WHO world mental health survey data, 56% of people in the USA as of 2010 who have depression did not seek help from the formal mental healthcare system [68]. Also, in the USA, even among those who eventually receive treatment for a major depressive disorder, it takes more than a decade for initial treatment contact to occur after the rst onset of the disorder [112]. Although quality of life and the ability to achieve personal goals or social roles can be signicantly impaired by symptoms that arise from mental disorders, those symptoms could potentially be eased by existing treatments [41]. Therefore, it is important to identify the discouraging factors for the service utilization and to design eective strategies to facilitate the use of the service for those in need. Research in public health and clinical psychology has shown that mental health stigma is one of the key factors that prevent people from seeking out timely treatment [6,35,47,97,109] 1 . Mental 1 Delay or failure of help-seeking is also associated with some socio-demographic factors: males, racial/ethnic minorities, married people, and the less educated tend to delay or avoid contacting mental health services [78]. In addition, there are situational barriers to accessing services including long waiting period, lack of time or nancial 30 health stigma is a negative stereotype attached to those with mental health issues or those receiving psychiatric treatment. According to research on mental health stigma, individuals rst internalize prevalent negative stereotypes about those with mental disorders, thereby become to agree with such view. As a result, they avoid seeking help in fear of negative consequences of visiting mental healthcare providers, such as discrimination or shame [35], to occur to themselves. Detailed review on the issue of mental health stigma is in the appendix. Despite its importance, mental health stigma has been largely neglected in economics. Instead, following the long-held tradition that emphasizes the importance of belief on decision making, models of help-seeking for health issues in economics largely focus on the roles of self-perceived seriousness, in other words, subjective belief on the probability of having `real' illness. For example, Oster et al. (2013) [85] build a model that optimistically biased belief about the probability of onset of disease in the future results in low likelihood of receiving screening test among individuals at risk for Huntington's disease. The current study aims to ll this gap experimentally testing a comprehensive framework where stigma plays a signicant role during the process of mental health belief formation 2 . In my frame- work, mental health stigma causes individuals to believe themselves to be mentally healthier than is the true state. It is important to note that not only mental health stigma refers to the negative stereotypes each individual attaches to those with mental disorders, but also it may be converted into the desire to keep social distance from them (See [9, 36]). Therefore, individuals with mental health stigma are motivated to believe that the probability that they currently have depression is low, which results in the optimistically biased belief about their mental states. An important behavioral implication is that, out of the desire of maintaining this biased belief, they choose to ignore or even avoid available diagnostic information because the information may force their bi- ased belief not to be feasible to maintain. This mechanism might partially contribute to treatment avoidance behavior. I test my mental health belief formation model using a series of survey experiments with sub- jects recruited from an Internet-based platform named Amazon Mechanical Turk (MTurk). The experiment consists of three sessions. In the rst session, to perform an exploratory correlation resources, and not knowing where to go for help [42]. 2 This framework is based on the theoretical arguments from the chapter 1. 31 analysis, I gathered individual level variables of depression level (CES-D score), degree of mental health stigma, self-assessment of depressive symptom (self-expected CES-D score) and willingness to receive the CES-D test result. In the analysis, I nd that, consistent with the theoretical expec- tation, subjects with high levels of stigma tend to report their self-assessed depression level to be lower than the true states are, while showing low levels of willingness to receive CES-D test results. While the rst session of the experiment is correlational investigation, the second session of the experiment aims to nd the existence of causal eects of stigma on under-assessment of depressive symptom, and on reduced willingness to receive the diagnosis result. To this end, I randomly divide subjects into two groups and provide them dierent messages, stigma-inducing messages for one group and stigma-reducing messages for another group 3 . If the eects of stigma are causal, it is expected that the subjects who read stigma-inducing messages show more under-assessment of depressive symptom and less willingness to receive the test result compared to those read stigma reducing messages. The result is partially consistent with this expectation. That is, the theoretically expected eect of reading stigma-inducing message compared to stigma-reducing message is observed among specic groups of subjects, but not among other subjects. With my data, it is not possible to identify the origin of varied eects of reading dierent messages across population group. I suggest the concept of `psychological reactance' to be a potential explanation. Psychological reactance refers to the unpleasant emotional reaction that emerges when people experience a threat to or loss of their freedom of choices or having specic attitudes [103], which might be able to explain the dierent levels of responsiveness to the messages across individuals with dierent characteristics. In addition to the above experiment, in the last session of the experiment, I investigate the eects of receiving partial information on one's depression state on willingness to receive information not yet disclosed. Specically, I divide diagnostic information into two pieces, the brief result (whether the CES-D score is above or below the cuto for depression) and the additional detail of the result (percentile rank of CES-D score). Then the subjects are provided with brief result rst and asked whether they want to receive the more detail. With the data, I analyze whether receiving the brief result rst aects the demand for detailed information. The result is consistent with the theory. If subjects experience disappointment when receiving the brief result, they are more likely to avoid 3 Also, there is a control group. 32 receiving the detailed result. In contrast, subjects who experience relief from the brief result tend to show greater demand for the detailed result. Also, this tendency is more prominent among highly stigmatic subjects. This result may suggest that individuals strategically choose to be selectively exposed to information in order to shift their beliefs and to attain the optimal level of belief bias. The current study has two contributions. First, I add to the theoretical literature on infor- mation preference by adding a component of personal attitude, namely mental health stigma, as a parameter in the theory 4 . Attitudinal traits have been largely neglected in economics perhaps because they have been assumed to be irrational part of human nature and also because it is hard to obtain good measures of attitude in available survey data. This study provides evidence that the addition of attitudinal variable might be relevant for research in economics. Second, in public health perspective, the current study may provide a systematic and quantitatively tractable way to explore the complicated network of stigma-related traits and its implications on behavior, such as, one's denial of having a mental disorder or having a low demand for mental health care services. The rest of this chapter is organized as follows. In the section 3.2, relevant previous literature will be reviewed. In section 3.3, the background theoretical arguments and hypotheses are described. Section 3.4 describes experimental design and section 3.5 reports the result of the experiment. Section 3.6 will discuss the result and section 3.7 presents some concluding remarks. 4 The detailed discussion on the theoretical contribution of this study is in the chapter 2. 33 3.2 Background Literature In this section, the relevant previous ndings about mental health stigma and its consequences discussed from previous literature are reviewed. First, mental health stigma is dened as the attitude that is involved with discrimination or contempt against those with mental disorders and/or those who use mental health services. It is useful for the present study to clarify the meanings of two sub-concepts which constitute stigma a broad term, self-stigma and public stigma [8]. Among them, self-stigma refers to the individual level negative attitudes against those with mental health issues, while public stigma means the societal level stereotype. The dynamic interaction between those two types of stigma follows a complicated psychosocial process but understanding this process is not an aim of this study. Between these two meanings, self-stigma is more relevant for this study. Previous psychiatry literature reports that denial of having mental health issue and reluctance in help-seeking are associated with self-stigma because self-stigma is converted into fear, expected discrimination or shame when one is aected by mental health problems and considering help-seeking as a means of coping. Therefore, it can be viewed that the individual demand for mental health care service is largely aected by self-stigma. Considering the importance of dealing with self-stigma to improve the outcome of community level mental health care policy, psychiatric literature has put eort into the understanding the mechanics of how self-stigma operates. However, due to the diculties in measuring and generating external variation on self-stigma level among subjects, studies that explore causal eects of self-stigma are rare 5 . In the literature within economics, there is a dearth of research on the eects of mental health stigma on economic outcome experienced by mental health patients. To my knowledge, only two published studies directly investigate the eects of stigma. Among them, one nds an evidence of mediating eect of mental health stigma on job market discrimination towards those with mental illness, wherein there is a greater likelihood for mental health patients to become unemployed, in countries with high levels of public stigma toward such people [50]. However, the meaning of stigma in this study is close to the concept of public stigma. Therefore, this analysis only could explore how stigma can make national/group level variation on the economic outcomes of mental health 5 More detailed review of psychiatry literature about mental health stigma is in the appendix (section 3.8.1) 34 patients, without being able to explain individual dierences in economic outcomes. Most relevant to the current paper is a study by Bharadwaj et al. (2017) [18] which empirically tests the existence of self-stigma among depression patients. Also, the authors test a hypothesis that, out of the fear of being stigmatized by survey administrators, those experiencing depression are likely to hide their mental health condition when answering survey question on their mental health state, and the probability of the under-report is negatively associated with the likelihood of receiving treatment in the future. This result is in line with the current study where individuals who have internalized public stigma and have developed self stigma tend to have positively biased beliefs about their mental health condition, and also, avoid receiving diagnosis results. However, there are considerable dierences between the hypotheses in the current study and those in the study of Bharadwaj et al. First, according to the hypothesis of Bharadwaj et al., individuals know their depression status but they want to hide their depression to others so they under-report it when answering survey questionnaires, whereas, in my hypothesis, individuals al- ready have biased beliefs about their own depression status regardless of the possibility for their depression being discovered by others. Second, according to the hypothesis of Bharadwaj et al., individuals choose not to seek treatment in order to avoid expected stigmatization. However, such broad term `stigmatization avoidance' can include dierent motivations such as fear of anticipated discrimination (public stigmatization) or anticipated internal shame (self stigmatization). In the current study, I propose a specic mechanism that can lead to treatment avoidance, that is, the motive for protecting one's preferred self-identity by not knowing one's own depression status (i.e., diagnosis avoidance), which will be detailed in the later section. Furthermore, by taking advantage of experimental research design, the variables used in the current study include self-stigma of each subject, allowing direct testing the hypothesis about the eects of self-stigma. However, because the study of Bharadwaj et al. lacks stigma measure, they rely on an indirect method such that the under-reporting of depression is used as a proxy variable for self-stigma, which might yield misleading results. 35 3.3 Hypotheses This section will introduce the hypotheses under study here and thier theoretical backgrounds. Those hypotheses come from the theoretical arguments in the companion paper (Chapter 2). The theory is based on literature on anomalous information preference and the origin of optimistic belief bias [23, 65, 70]. Even though the theoretical studies in this area do not particularly deal with behavior within the context of health or illness, they might oer some useful framework for understanding the behaviors of potential mental health patients. The detailed theoretical discussion is in the chapter 2. However, in words, the behavior of potential mental health patients is often characterized as denial of having mental illness and test avoidance, which is comparable with motivated belief bias [23] and information avoidance [58] in previous theoretical literature. According to the model, individuals derive utility directly from their expectation about future outcomes, therefore leading to optimistically biased beliefs, because optimism can yield current felicity. Moreover, since the acquisition of information may make it impossible to maintain the optimistically biased belief, individuals may want to avoid such information. Based on this general feature of the model, to apply this framework to understanding the belief formation of potential mental health patients, I assume that individuals derive `identity utility' from their belief about the probability of having depression. Also, I add some components in order for mental health stigma to be incorporated as a parameter in the model. That is, individuals derive greater degree of negative utility from this belief if they have greater level of mental health stigma, which captures the idea that self-stigma can be converted into shame if they realize that they are aected by depression. Based on this setting, there are three expectations with regard to the behavior of individuals who have mental health stigma and derive dis-utility from the belief that they have depression. All of these expectations are detailed in the chapter 2.3. Hypothesis (A): Mental health self-stigma causes optimistic (under-evaluation) bias regard- ing one's depression severity. This is because, assuming people have some degree of control over their beliefs, thinking positively about their mental health state generates higher identity- utility. Also, the importance of identity-utility when forming belief is greater when one has a 36 higher level of stigma. Hypothesis (B): Mental health self-stigma causes avoidance of knowing one's true mental health state. Also, as one has more optimistic bias, he is more likely to avoid access to in- formation. The rationale of this expectation is based on disappointment aversion tendency of human nature. Individuals feel disappointment when the diagnostic result implies that one has depression. The degree of disappointment is greater the greater the level of stigma is, and when one's prior belief is more optimistically biased. Hypothesis (C): (1) Assuming that one has received rst diagnostic test result, he tends to seek more detailed diagnosis if the test results indicated depression. This is because knowing this increases the perceived needs of knowing one's own mental health state. (2) Assuming that one has received rst diagnostic test result, he tends to avoid/seek more detailed diagnosis result if the rst diagnosis result has generated disappointment/relief. Dis- appointment occurs when rst diagnosis result unexpectedly implies depression. On the other hand, relief happens when the diagnosis result unexpectedly indicates. Moreover, mental health self-stigma increases both behavioral tendencies. Each hypothesis represents a specic stage in the process of mental health belief formation of individuals with mental health self-stigma. First, hypothesis (A) explains why positively (optimisti- cally) biased belief about one's own mental health state is optimal. Next, hypothesis (B) accounts for test (diagnostic information) avoidance among them as a strategy for maintaining the biased belief. Lastly, hypothesis (C), especially second part of the hypothesis, could be a possible mech- anism by which an initially unbiased belief about one's own mental health state gradually evolves into an optimistically biased one because individuals are more likely to be exposed to information that provides relief. It is important to note that hypothesis (A) deals with what is the optimal belief while hypothesis (C) represents how belief is dynamically evolving into the optimal belief. 37 3.4 Experimental Design Experiments in this study aim to test causalities in the above three hypotheses. Also, in a separate experiment (correlation experiment), I collect a cross-section of subjects in order to perform cor- relational analysis to explore whether, without considering causality, the behaviors of subjects are consistent with theoretical predictions, which are (1) the positive association between self-stigma and underestimation of one's own depressive symptom and (2) the negative association between self-stigma and demand for diagnostic information. In the second experiment (stigma experiment), I perform randomized controlled trial (RCT) analysis to explore the causal eects of stigma on underestimation of symptom severity and demand for diagnostic information. To generate exogenous variation in the stigma level among subjects, priming approach is employed. Specically, subjects are randomly divided into three groups and each of them is presented with dierent messages: stigma inducing messages (stigma group), stigma reducing messages (anti-stigma group) and stigma-irrelevant messages (control group), respectively. Next, subjects are asked to report their self-evaluation of depressive symptoms and demand for diagnostic information. The last experiment (disappointment experiment) is designed for testing hypothesis (C). In this experiment, diagnostic information consists of two separate signals (simple test result and detailed result). The subjects in the treatment group are presented with the simple result rst, and then asked if or not they still want to receive a more detailed test result. On the other hand, subjects in control group are asked, before they receive any information, if they want to receive only the simple test result or both of the results (simple and detailed). This design intends to test whether the disappointment has a deterring eect on them in seeking out further information. 3.4.1 Sample The protocol was reviewed by an IRB from University of Southern California. Subjects were recruited through Amazon Mechanical Turk (MTurk). The subjects volunteered to participate in the surveys by clicking the link I posted on MTurk web page. Because this study does not specically aim to explore the behavior of depression patients, I did not set any participation criteria about the level of depressive symptoms or depression treatment history. However, I limited the participation 38 only to the people living in the USA. Also, because of the possibility that some individuals might be concerned about reporting their depression-related information, I displayed the message at the beginning that they could freely close the survey page and cancel the participation at any time. The payment for participation was between $0.2 to $0.5. 3.4.2 Design 3.4.2.1 Experiment 1: Correlational observation This experimental procedure is designed to collect data on the observational patterns of behavior with particular focus on self-stigma and its relationship between underestimation of one's own depression symptom severity and willingness to receive diagnostic information. All the mental health related variables collected from this experiment are summarized in Table 3.1. Before collecting mental health-related information such as depression severity or stigma, since it is possible that their socioeconomic status might aect their behavior and those factors should be controlled for, I asked the participants to report their socio-demographic information including age, gender, marital status, race, level of education, level of income, employment status and the number of previous experiences with treatment for mental health issues. Next, I gathered depression-related information from the participants. This information consists of three variables. First, right after providing the socio-demographic information, participants reported their self-estimated probability of being diagnosed with depression if they want to take a depression screening test now (prior belief ). Next, I measured a more objective depression severity (objective depression score (CES-D)) of each individual by asking them to answer the Center for Epidemiologic Studies Depression Scale (CES-D) which is a commonly used questionnaire in screening for depression. When answering this questionnaire, the participants were not informed that the questions were taken from a screening test for depression because under-reporting might happen if they knew it, especially among participants with high degrees of mental health stigma. Last, after nishing the C-ESD test, each participant was asked to report their self-estimation of their percentile score of depression level among all the current survey participants (self-estimated depression (percentile) score). In order to prevent confusion, I displayed the instruction that a high value of percentile score implies more serious depression. Because self-estimated depression 39 score is one of the important variables, in order to improve the accuracy of the report, I incentivized the participants by oering bonus payment ($1) if their estimation was within the top 5% in terms of accuracy 6 . One of the aims of this study is to test whether mental health stigma induces under-estimation of one's depressive symptoms. The self-estimation bias is simply dened as the dierence between the self-estimated percentile score and true percentile score (true percentile score self-estimated percentile score). A positive value of self-estimation bias is equivalent to underestimation of de- pressive symptoms. Next, I measured the mental health self-stigma of the participants by asking them to answer below two questions: (1) \Some people think the cause of depression is lack of willpower. Do you agree with this statement?" and (2) \Some people think depressed people are not pleasant to be around. Do you agree with this statement?" The participants could indicate their degree of agreement to each of the question by choosing one of 4 choices which are `agree', `somewhat agree', `somewhat disagree' and `disagree'. The rst question is designed to measure the degree that each participant has internalized the public stereotype that depression is associated with weakness in personality (stigma-will), and the second question measures the degree of social undesirability each participant subjectively attaches to depression (stigma-social). The score of each type of stigma is computed by indexing the answers to number such as `agree'=3, `somewhat agree'=2, `somewhat disagree'=1 and `disagree'=0. Lastly, the composite score, stigma, is dened as the sum of both types of stigma measures (stigma = stigma-will + stigma-social) . In the next part of the survey, I measured the willingness of each participant to receive diagnostic information. Each participant could choose one of the three options: (1) \I want to know both my depression test result and my depression percentile score among all the participants.", (2) \I want to know my depression test result only" and (3) \I do not want to any information on my depression state". It is pointed out that the chosen information type will be displayed on the last page of the survey. Because there are no monetary cost associated with receiving information and time cost is also minimal (less than 15 seconds), those who do not choose option (1) might be regarded as showing information avoidance 7 . Choosing option (3) implies total avoidance 6 The accuracy of self-estimation is measured by the absolute dierence between the true percentile score based on the C-ESD test result and the self-estimated percentile score. 7 Information avoidance is dened as not wanting information that might have instrumental value even when no 40 Table 3.1: List of variables variable name description prior belief self-reported probability of having clinical depression objective depression score (CES-D) depression screening test (CES-D) score objective depression percentile score percentile score of CES-D score among all the participants self-estimated depression percentile score self-estimated percentile score of CES-D score among all the participants self-estimation bias objective depression percentile score minus self-estimated depression percentile score stigma degree of agreement to negative views against depressed people (range: 0,1,2,3,4,5,6) demand for information choice of informativeness of diagnostic signal (0: no, 1: simple result, 2: detailed result) self-ecacy self-reported ability to cope with depression by oneself (range: 0100) trust trust toward formal mental health system (range: 0100) of diagnostic information, while option (2) implies slight degree of avoidance. The demand for diagnostic information (demand for information) is indexed as integer values such as `want both results (option 1)'=2, `want only simple result (option 2)'=1 and `not want any result (option 3)'=0. As such, I collected the relevant variables including objective depression level, self-estimated depression score, self-estimation bias, stigma and demand for information. Apart from this set of information, I collected some additional information from the participants which might be inter- esting in exploring their relationships with the above variables. First, I asked the participants to material cost is associated with receiving information [58] 41 indicate their attitude toward mental health related topic. These questions include the expected ability to cope with depression, or, self-ecacy (How do you expect your ability of managing your emotion on your own without any help from psychiatrists if you develop depression? ), the degree of trust toward mental health care system (trust) (Do you believe that mental health care professionals can really help people with mental health issues? ). Answers for both questions can be ranged from 0 to 100. Lastly, on the last page of the survey, subjects were presented with the version of depression test result that they had chosen in the previous section (variable demand) 8 . 3.4.2.2 Experiment 2: Causal eects of message provision The next set of experiment is designed to test hypotheses (B) and (C). Those hypotheses state that stigma causally induces self-estimation bias in more optimistic ways and avoidance of diagnostic information as described in Figure 3.1. The subjects were recruited through Amazon Mechanical Turk again. As a design for causal investigation of the eect of stigma, I divided the samples into three groups and applied dierent treatment (stigma-inducing, stigma-reducing and stigma-neutral) in order to explore whether stigma-inducing treatment contributes optimistic bias in self-estimation of depression severity and avoidance of diagnostic information compared to other treatments. Every subject in three groups rst reported socio-demographic information and two of depres- sion related information, that is, prior belief and objective depression score which explained in the previous section 9 . Next, I provided dierent messages to subjects depending on the group they were in (stigma- induced, stigma-reduced, and control). The rst group is stigma-induced group, and was presented with three stigma-generating messages: (1) \People bullied as kids are less mentally healthy as adults. Kids don't easily outgrow the pain of bullying". (2) \Low self-esteem can cause depression. Because those with low self-esteem are prone to 8 The C-ESD test used in this survey experiment is one of the commonly utilized depression screening test. However, the accurate diagnosis only can be done through in depth interview procedure by certied psychologist or psychiatrist. I claried this to the participants when displaying the test result. 9 Provision of mental health stigma-related message might aect the subjects' C-ESD test score so I choose to display the messages after nishing the C-ESD test. 42 Figure 3.1: Summary of experiment 2 replay and focus on negative thoughts far more than those who have high self-esteem, putting themselves at higher risk for low moods". (3) \Being around depressed people may cause you to be depressed as well, because their nega- tive thinking style can in uence your own such that over time, you too become more vulnerable to depression". All of messages are taken from articles written by psychologists or psychiatrists. Also, the name of the authors and their aliations were also displayed in order to enhance the credibility of the messages. Also, after showing messages, in order to increase the attention to the contents of the messages, participants were asked to indicate their degree of agreement to each of the messages. Similarly, to the second group (stigma-reduced group), I provided messages which have positive or non-stigmatizing connotations about depression. Those are: (1) \Researches found that objective measures of physical attractiveness and depression are not signicantly associated with one another. It's just that depressed people tend to underestimate their own attractiveness". (2) \Depression can aect anyone regardless of physical or mental strength. Some of America's 43 most well-known citizens { including Abraham Lincoln, Terry Bradshaw, and Judy Collins - have experienced depression". (3) \"Creative people are more likely to have depression compared to non-creative people. There was a clear relationship between being creative and having a diagnosis of depression". The last group is control group. They were presented with three messages which are irrelevant of depression or mental health. Those messages are: (1) \Climate change is expected to have negative eects on human health although it is not clear what causes current climate change". (2) \Drinking warm water is healthier than drinking cold water". (3) \Fukushima nuclear disaster raised concerns as to whether eating contaminated seafood might impair human health|not just locally but across the Pacic". After reading these dierent messages, all the remaining procedures are the same as in ex- periment 1. First, the participants reported their self-estimation of their own depression severity in the unit of percentile score of CES-D result. Next, participants reported their stigma toward depression. These variables will be used to test whether or not the randomized message provision treatments in uence the participants' temporary stigma level in the intended way. After reporting stigma, the participants chose the informativeness of diagnostic result that they wanted to acquire: (1) both their depression test result and percentile score, (2) their depression test result only, and (3) no information. Finally, the survey ended with displaying the diagnostic information that each participant was willing to receive. 3.4.2.3 Experiment 3: The eect of the partial revelation of information on continu- ation of information seeking The third experiment is designed to test hypothesis (C). This hypothesis posits that, when in- formation is bundled with a set of partial signals, even when individuals rst choose to receive the information bundle, if the rst revelation of partial information induces disappointment, they opt out from receiving remaining signals. Also, the likelihood of opting out becomes larger if the 44 Figure 3.2: Procedure of Experiment 3 individual has a high degree of self-stigma toward depression. This mechanism may explain how those with mental health stigma develop optimistic bias. This experiment was performed with the subjects in the rst experiment. The subjects are limited to those had chosen to receive both the simple test results and detailed results. The procedure is as follows. First, I randomly divided the participants who had chosen to receive both information into two groups (treatment group vs. control group). The procedure applied to both group is summarized in gure 3.2. For the treatment group, at the end of the rst part of the survey (experiment 1), they were rst presented with a simple result, that is, their depression test score and the criteria for depression (C-ESD score 16). At the moment they received the depression test result, they might have felt either disappointed or relieved depending on the result. The emotion of disappointment is assumed to arise when the depression test score implies depression. On the other hand, the participants were assumed to feel relief if the their depression test scores were below the cut-o point. The emotional response is dened as [prior belief I CESD16 ] where the variable prior belief is the self-reported belief about the probability of being diagnosed with depression and I CESD16 is the indicator function whose value is 1 if the test result implies depression and 0 otherwise. The absolute value of the emotional response indicates disappointment or relief, respectively. Next, they were again asked if they still wanted to receive more detialed result (depression percentile score). If they chose 45 to opt out, the percentile score was not displayed. Whereas, for the control group, the treatment was designed in a way that their intention to receive the detailed information is not aected by disappointment or relief. Thus, they were asked if they wanted to receive detailed result before receiving the simple result. 46 3.5 Econometric Analysis and Result In this section, the results of three experiments are reported with interpretation of how those results establish the validity of three hypotheses in section 3.3. 3.5.1 Experiment 1: Correlational Observation 3.5.1.1 Self-estimation Bias The results of experiment 1 which is based on cross-sectional design t with the hypotheses. First, subjects' self-estimation of their percentile depression score is generally accurate. Figure 3.3 depicts the relationship between true percentile score and self-estimated percentile score, where those two variables approximately follow linear relationship. Such linear trend implies that participants are generally aware of their depression state relative to others. However, in the graph, there is also individual variation in the accuracy of self-estimation. Also, the self-estimation biases are observed to be skewed toward negative values. A majority of samples (63%) show underestimation bias in evaluation of their degree of depression compared to others. In other words, many of the subjects tend to believe they are more mentally t than they really are. In the theoretical discussion (Chapter 2), it is posed that such underestimation of the degree of severity might be motivated. The theory expects that the degree of underestimation bias would be greater for those that have a high level of mental health self-stigma or high level of self-ecacy. The rationale for this expectation is as follows. First, having higher stigma, the disutility of holding belief that one has depression becomes higher, so that they are motivated to have more optimistic view. Second, lower level of self-ecacy results in greater health cost if they are not receiving appropriate treatment, so a more objective belief is preferred. Equation (1) summarizes this argument, with the coecients 1 and 2 being expected to have negative values. q =p + 1 stig + 2 efficacy + 3 x (3.1) (q: self-estimated depression percentile score, p: objective depression percentile score, stig: stigma, efficacy: self-ecacy, x: vector of control variables) 47 Table 3.3 summarizes the OLS regression result. Consistent with the expectation, the partici- pants who have high value of stigma or high value of self-ecacy are observed to underestimate their percentile depression score even after controlling for possible confounding factors. This observation is also represented in Figure 3.4 which compares the marginal mean values of the self-estimation bias (with true score as the control factor) between high stigma group and low stigma group (panel a), also between high self-ecacy group and low self-ecacy group (panel b) 10 . The high stigma group estimates their percentile depression score about 2 rank lower (healthier) than low stigma group (-6.46 vs. -4.19). Also, the comparison between high and low self-ecacy group shows the similar pattern (-6.46 vs. -4.22). To begin the econometric analysis, using OLS, I regress the subjects' self-estimated percentile score on the true percentile score, stigma, self-ecacy and socio-demographic information as control variables. The results of analysis are presented in columns (1) to (3) in Table 3.3. Consistent with theoretical expectation and graphical represenatation in gure 3.5, the degree of underestimation bias is higher among those have higher value of the variable stigma and self-ecacy. First, in column (1), the self-estimated percentile score is highly associated with true percentile score which implies that individuals are generally accurately aware of their mental health state. This association is not attenuated either when psychological variables are controlled for as in column (2), or socio- demographic control variables are also controlled for as in column (3). Also, the variable stigma is negatively associated with the self-estimated percentile depression score. The standard deviation unit increase in stigma reduces the self-estimated depression score by around 1 percentile and this pattern is observed both when socio-demographic variables are controlled for(column 3) or not (column 2). Similarly the negative association is observed between self-estimated depression score and self-ecacy. Lastly, the column (3) shows that the socio-demographic variables are not associated with self-estimation of depression ranking except for non-Caucasians. Non-Caucasian subjects tend to estimate their depression percentiles score of being 3 rank lower than Caucasian subjects. This result is consistent with observed high level of self-stigma among ethnic minorities in the US [56,114]. 10 Marginal mean bias of each category (e.g., high stigma group) is the mean value of the bias averaged across all level of CES-D test score, which prevents the over-representation of samples within specic range of the CES-D score when computing the mean value. 48 3.5.1.2 Avoidance of Diagnostic Information This section reports the observed relationship between avoidance of diagnostic information and relevant variables including stigma and self-estimation bias. In summary, participants are less likely to seek information if they have higher degree of stigma and/or underestimation bias with respect to their depression percentile score. Especially, the information avoidance is more prominent when having higher value of stigma and more level of underestimation bias simultaneously. Even though the results do not establish the causal relationship due to the cross-sectional design of the study, they are consistent with the theoretical expectation which is posed in equation the hypothesis (B). Participants chose one of the option among (1. simple result + detailed result) \I want to know both my depression test result and my depression percentile score among all the participants.", (2. simple result) \I want to know my depression test result only" and (2. no information) \I do not want any information on my depression state". Among the whole sample (N = 1372), a majority (60%) chooses to receive both information (option 1). The portion of samples who choose option (2) and (3) is 32% and 8%, respectively. First, the theory (section 2.3.2) expects that individuals with more severe depressive symptoms derive more instrumental value from knowing one's true mental health state so they are likely to choose the information with more detail. The graph (a) in gure 3.5 shows that the observation from the data is consistent with this theoretical expectation. On this graph, `nonavoider' refers to the participants who choose to receive both the test result and percentile ranking (option 1), while `avoider' denotes the group of samples who choose option 2 or 3. It is apparent in the graph that the group of nonavoiders have the distribution of the variable objective depression score (CES-D) more skewed to right than avoiders, which implies there is a positive association between information seeking and symptom severity. The key component of the theory is that mental health self-stigma reduces willingness to know one's diagnosis result. The graph (b) of gure 3.5 shows the sample distribution of variable stigma among both the group of avoiders and nonavoiders. Consistent with the theory, the sample distri- bution of the variable stigma is more skewed to the right among avoiders. Also, the theory oers prediction that the deterring eect of stigma would be more prominent when the individual is displaying under-biased self-estimation of symptom severity, because the 49 disappointment experienced when biased belief is substituted by objective one is greater for them. Figure 3.6 depicts the observed relationship between demand for test result and the interaction vari- able of high or low stigma (lower half or upper half) and underestimation or overestimation. The y-axis represents the average of chosen information type among each group of participants 11 . As predicted by theory, information avoidance behavior is observed among (and only among) the par- ticipants who both having high degree of stigmatizing attitude and underestimated self-evaluation of symptom at the same time. In addition, I assume that self-ecacy reduces demand for information because the instrumen- tal value of information is lower if depression is thought to be well managed by oneself. Those theoretical arguments yield below equation which can be estimated using ordered logit or ordered probit model. The theory expects that the signs of the coecients 1 , 2 and 3 being expected to be negative. D =q + 1 stig (pq) + 2 stig + 3 efficacy + 3 x (3.2) D = 8 > > > > > > < > > > > > > : 0 if D 1 No information 1 if 1 <D 2 Simple information only 2 if 2 <D Simple information + detailed information (stig: stigma, p: objective depression percentile score q: self-estimated depression percentile score, efficacy: self-ecacy, x: vector of control variables) For estimation of equation (2), considering the ordinal nature of the variable choice of informa- tion type (D), ordered logit model is used. Mental health related attitude and socio-demographic variables are included as controls. The column (1) on Table 3.4 shows simple regression result where I use three key variables true percentile score, stigma and the degree of self-estimation bias (underestimation bias). Those three variables are strongly associated with demand for test result. 11 I assigned the value 2 to the variable demand for test result if a participant chooses to receive both simple test result and percentile score), value 1 if he/she chooses to receive the simple test result only and value 0 if refuses to receive any information. 50 More severe depressive symptom is positively associated with the demand for more detailed test result, whereas stigma and self-estimation bias are negatively associated with the demand for de- tail of the test result. In column (2), instead of assuming the coecient of self-estimation bias is universal among all the participants, I perform sub-group analysis regarding the eect of self- estimation bias among high stigma group and low stigma group. Consistent with the theory and the crude observation illustrated in Figure 6, the participants who have both high level of stigma and optimistic self-estimation bias are likely to avoid more detailed information. Columns (3)-(5) describe the regression results when other mental health-related variables are included as explanatory variables. First, column (3) shows that the variable self-ecacy is neg- atively associated with willingness to receive information. There are two potential explanations of this negative association. First, believing the cost of developing depression is low because high self-ecacy means self-condence in coping, implying that one may derive little instrumental value of knowing current mental health state. A second alternative explanation could be that high value of reported self-ecacy might re ect some aspect of mental health stigma. This second explanation is supported by the decreased coecient of stigma in column (3) compared to column (2) which implies the positive correlation between stigma and self-ecacy 12 . In column (4), I add other attitude-related variables including the intention to utilize mental healthcare, the intention to utilize physical healthcare, the perceived discrimination toward depressed people and the previous mental health service use. Brief descriptions of those variables are presented in Table 3.2. First, in column (4), the variable intention to utilize mental health care services is positively associated with demand for information. However, the column (5) shows that when the variables trust toward mental health services and trust toward depression screening test are included in the regression, the positive association between intention to utilize mental health care and demand for diagnostic information is reduced and no longer statistically signicant. This implies the positive association association between demand for diagnosis result in this survey and intention to utilize mental health care services occurs through subjects' attitude toward mental health care system. Indeed, those two variables measuring trust have strongly positive association with demand for information. Taken together, this observation implies that a positive attitude toward formal men- 12 The correlation between stigma and self-ecacy is 0.146 51 Table 3.2: List of mental health related attitudinal variables variable name question possible answers intention to utilize mental health care services How soon do you want to visit a mental health clinic if they oer free consultation? 0: never 1: within a year 2: within a few months 3: within a month 4: within a week 5: right now intention to utilize physical health care services How soon do you want to visit a local hospital if they oer a free cancer examination? 0: never 1: within a year 2: within a few months 3: within a month 4: within a week 5: right now perceived discrimination toward depressed people Out of 100 people, how do you estimate the number of people who think `depressed people are unpleasant'? 0100 previous experiences of mental health care service use How many times have you ever received treatment or consultation for mental health issues? 0: never 1: once 2: twice 3: more than twice tal health care system may act as an encouraging factor for individuals to receive mental health information in this survey experiment, as well as encouraging them to visit mental health care professionals if developing depression. Next, interestingly, participants who think depressed people experience discrimination from gen- eral public (perceived discrimination toward depressed people) want to seek more detailed informa- tion. This seemingly odd observation might be attributable to the Internet-based thus anonymous nature of the survey experiment. Namely, individuals with high levels of mental health stigma may consider internet-based informal depression tests as a substitute for actual visit to formal mental health services so as to avoid the possibility of being observed by others. This argument is sup- ported by the negative association between the perceived discrimination toward depressed people 52 and the intention to utilize mental health car services on the column (7). That is, individuals who consider visiting mental health professionals as potentially causing discriminating response from others avoid seek information from the anonymity guaranteed source such as the Internet instead of making a physical visit to a clinic. 3.5.2 Experiment 2: Causal Eects of Reading Stigma-relevant Messages In this section, I describe the ndings from the eect of randomized priming experiment. Even though the results described in the previous section oer supporting evidence of the theory, they are limited in establishing the causality because the experimental design of the rst experiment is cross-sectional. The priming experiment aims to investigate causal relationship about the patterns observed in cross-sectional data. Mainly, I test two hypotheses in this section: (1) mental health stigma causes individuals to display underestimation of symptom severity, (2) mental health stigma causes individuals to avoid diagnostic mental health information. To explore the causal relationship, after randomly assigning participants into three groups, dierent messages were presented to each group: stigma-reducing (N = 275), stigma-inducing (N = 265) and stigma-neutral message (N = 358). Also, it is shown that baseline characteristics of each group do not have signicant dierence (table 5. a). If the eects of stigma are causal, it is expected that participants who read stigma-inducing messages would display more underestima- tion of symptom severity bias, also less demand for receiving diagnostic information compared to participants who are provided with stigma-reducing messages or control group. In summary, the expected eects of reading dierent messages are weak among subjects as a whole. However, when the analysis is limited to certain group of participants, those eects are apparent. This group of subjects is identied as those married also employed as full time worker. The observed dierence of self-estimation and demand for diagnosis result across three groups (stigma-inducing, stigma-reducing and stigma-neutral) are displayed on the Figure 3.7. First, among the subjects as a whole (panel (a) and (b)), stigma-reduced group shows less underestimation of symptom severity and higher demand for diagnostic information than control (stigma-neutral) group. However, there is only a very slight dierence in behavioral patterns between stigma-induced group and stigma-reduced group. Moreover, even though it is theoretical expectation that stigma- induced group would show more underestimation bias and less demand for information, the result 53 is not in accordance with this expectation. On the other hand, among married/full time worker participants, stigma-induced group display apparently more underestimation bias than stigma-neutral group. Also, the stigma-reduced group shows higher demand for diagnostic information compared to stigma-neutral group. However, stigma-induced group does not show signicant dierent behavioral patterns with stigma-neutral group. Those patterns are numerically described on Table 3.6. This table summarizes the mean values of self-estimated percentile scores and demand for diagnostic information across treatment groups. Panel (a) uses the whole sample, where only little dierence could be found between groups with regards to self-estimated depression score and demand for information. However in panel (b) which uses only subjects who are the married/full time employees, even though stigma-reduced group has less degree of baseline depression severity compared to stigma-induced group, their self- estimated severity is higher. Also, the demand for diagnostic information is higher among the stigma-reduced group, which is consistent with the theoretical expectations. The regression results are consistent with these observations. First, the column (1) in Table 3.7 shows that both stigma-reduced and stigma-induced groups show less under-estimation bias and the dierences are statistically signicant. However, size of the eect is not great, and the there is no dierence between stigma-induced and reduced groups. This might be because, regardless of the contents of the messages, reading mental health related messages evokes one's own experi- ence of feeling depressed so both stigma-induced and stigma-reduced groups less under-estimate their depression severity. Also, as expected the stigma-reduced group has greater demand for the diagnostic information than other treatment groups as in the column (2). Table 3.8 shows that, when the analysis is limited to the married and full time employees, the participants who receive stigma-reducing messages are less likely to underestimate their depression severity compared to other groups. As for demand for diagnostic information, Table 3.9 shows that stigma-reduced group displays higher demand. The demand for such information among control (stigma-neutral) group is slightly higher than among stigma-induced group, even though the dierence is not statistically signicant. 54 3.5.3 Experiment 3: Eect of Partial Revelation of Information on Continua- tion of Information Seeking In this section, I present the experimental evidence of the theoretical expectation summarized in the hypothesis (C). The expectation about the choice of subjects in treatment group based on my theory is as follows: after receiving simple diagnostic information (CES-D score and cuto for depression) (1) they are likely to proceed to receive more detailed information (percentile score) if the CES-D test result indicates depression but (2) they are likely to choose not to receive the percentile score if the CES-D test result generates disappointment (the inverse prediction is that experiencing relief might increase the demand for further information.). The econometric specication is as follows. A subject's decision whether or not to continue to receive the percentile depression score (Stay = 1: Receive,Stay = 0: Not receive) is described with the below equation. Stay = 8 > > < > > : 1 if Stay 0 > 0 0 if Stay 0 0 where Stay 0 = 1 A + 2 ( A q 0 ) + 3 + (3.3) Here, A is the rst signal (simple CES-D test result) with the value of 0 and 1 representing `not depressed' and `depressed', respectively. Also, q 0 is the subject's self-reported prior belief about the probability of being diagnosed with depression. The disappointment happens when A = 1 and the degree of disappointment is 1q 0 , whereas relief is experienced when A = 0 and the degree of relief is q 0 0. According to the theory put forward in Chapter 1, the coecients 1 and 2 are expected to be positive and negative, respectively. Also, to investigate whether or not the response to disappointment or relief is dierent depending on mental health self-stigma, I perform the sub-group regression with respect to the coecient 2 by subjects with high levels of stigma (with values of the variable stig that are greater than or equal to 3) and low levels of stigma (with values of variable stig less than 3). In this case, the equation (3) is re-written as Stay 0 = 1 A + X gr 2;gr I gr ( A q 0 ) + 3 + (3.4) 55 where the indicator gr refers to the group the subject belongs to, which is one of among ld (low stigma/disappointment), lr (low stigma/relief), hd (high stigma/disappointment) and hr (high stigma/relief). The expectations about the value of the coecient 2;gr based on the theory are as follows. 1) 2;gr < 0 for all gr Experiencing disappointment/relief at the rst signal leads subjects to avoid/seek additional information. 2)j 2;hd j>j 2;ld j, andj 2;hr j>j 2;lr j Responsiveness to the emotion experienced from the rst signal with respect to the demand for second signal is greater among the subjects with higher degree of stigmatizing attitude. To estimate the model, I use the logit regression. The result is on Table 3.11. The value of dependent variable is 1 if the subject proceeds to receive the percentile score and 0 otherwise. The standard errors of estimated coecients are calculated by bootstrap resampling because the sample sizes in each of the four categories (disappointment/relief high/low stigma) are low and asymptotic normality assumption of estimated coecient may not be a good approximation. Also, the signs of the coecient of both disappointment and relief are expected to be negative according to the theory. This means disappointment leads to the avoidance of further information, whereas relief makes people want further information. Columns from (1) to (3) in Table 3.11 report the estimation results among those in the control group. The results support my theory. First, the rst row shows that receiving the rst set of information that implies depression leads the subjects in the treatment group to have stronger willingness to proceed to receive the remaining set of information, while subjects in control group in columns (4) to (6) do not show the same pattern ( 1 > 0 in equation (3.4)). Second, the column (2) shows that if the rst set of information generates disappoint or relief, the subjects tend to show decreased or increased willingness to receive the remaining information, respectively ( 2 < 0 in equation (3.4)). Also, the coecient of `rst signal' on the column (2) is 3 times as high as the coecient of same variable on the column (1), which implies the omission 56 of disappointment/relief variable in regression results in signicant underestimation of the positive impact of receiving a test result implying depression on the willingness to receive additional infor- mation. This happens because, according to the theory, although receiving rst signal implying depression can increase perceived instrumental value of additional information, this impact might be mitigated or even canceled out by the motivation of avoidance of further disappointment. Sim- ilarly, if test result is not depression, and if it is unexpected, individuals feel pleasant surprise. Then, they might be more likely to proceed to receive additional information. Third, the column (3) reports how the highly (stigm > 3) and mildly (stigm 3) stigmatic subjects respond dierently to the emotion experienced when receiving the rst signal (depression test result). As described on the table, the eect of disappointment is twice as high among individ- uals with high stigma compared to those with low stigma. Similarly, the eect of relief is greater among high stigma group. 57 3.6 Discussion 3.6.1 Implication on Theoretical Economics Findings from this study provide evidence for theoretical literature on optimism bias and its im- plication for information preference [23, 58]. Based on the observation that optimistic belief (or overcondence) is manifested in various situations such as nancial investment [65], health [69,85] or physical appearance [45]. Such biased perspectives are also commonly concurrent with avoidance of knowing reality. This phenomenon is called the `ostrich eect' which is a metaphor come from a common (but false) belief that ostriches bury their heads in the sand to avoid danger instead of escaping it [65]. The analysis in the current study also belongs to this research area: Individuals underestimate their risk of having depression and avoid knowing their objective mental health state. The common feature of the theories to explain this behavioral patterns is to assume that in- dividuals derive utility directly from their beliefs. Therefore, they consider the expected change of their beliefs when making a decision whether to collect information. If the negative emotional impact of bad news exceeds positive impact of good news, the information might be avoided [69]. Consequently, individuals who dislike disappointing news become biased in ther beliefs (e.g., opti- mistic or overcondent). Moreover, according to some studies, the size of belief bias is theorized to be optimally chosen by considering both the current felicity of having wishful beliefs and potential future utility loss by holding inaccurate beliefs [23]. The current study adds to this strand of models in two ways. First, this study provides evi- dence of the concurrence of belief bias and information avoidance. Specically, the size of mental health belief bias and the avoidance of diagnostic information are observed to be strongly posi- tively correlated. Second, this study nds evidence that there is optimal level of optimistic belief bias. The identity-based utility framework in this study predicts that under-estimation of one's own depression severity is optimal among individuals with mental health stigma. Experiments pro- vide tentative evidence that mental health stigma causally lead to an underestimation bias, which supports the existence of an optimization process for choosing mental health beliefs. Moreover, using mental health stigma as a parameter in the process in mental health belief formation complements existing studies. Individual dierences in the size of belief bias or tendency 58 of information avoidance might be originated from the individually varying importance of belief- based (or identity-based) utility. There has been a dearth of theoretical literature on individual attitudes aecting this dierence. This study suggests a theoretical framework in which attitudes of individuals, for example, mental health stigma, determine the importance of belief-based utility, along with providing experimental evidence of this hypothesis. Also, this study explores particular sub-stages of belief formation. First, individuals have opti- mal level of belief bias (hypothesis(a)). Second, individuals use particular mechanisms, such as the selective exposure to information, to control their belief and attain the optimal belief (hypothesis (c)). Lastly, after achieving the optimal belief, individuals choose to avoid information in order to allow their biased beliefs to be maintained (hypothesis (b)). 3.6.2 Implication on Public Health Research This study has implications for public health research and policy, especially on the topic of mental health stigma and how it interferes with treatment. First, stigmatic individuals often deny their having mental health issues and this results in inaccurate self-assessment of their own mental health states [84, 98, 101]. This has a harmful consequences for the patients because treatment decisions should only be made after their self-recognition of symptom is completed. Despite the apparent correlation in self-recognition failure and self-stigma, there is a lack of causal study. One reason might be diculty in measuring whether or not self-recognition is made. In this study, by making use of novel design for measuring self-assessment of depressive symptom, I nd an evidence that mental health self-stigma leads to self-recognition failure. Second, this study oers evidence for one of the possible mechanisms for how mental health self-stigma leads to the avoidance of help-seeking (hypothesis (B)). The previously suggested expla- nations include condentiality issues, anticipated discrimination or embarrassment [33]. Consider- ation that one of the most important reasons of visiting health care system is receiving a diagnosis, this study explores the causal eect of mental health self-stigma on the avoidance of diagnostic information. The nding from this study implies that stigmatic individuals avoid seeking help even when there is only a very little chance of condentiality issue to occur. Therefore, from the public health point of view, in order to improve the eciency in delivery of mental health care services, along with improving privacy issues involved with the service use, more needs to be done in terms 59 of de-stigmatizing among those in risk of mental health problems. Third, this study provides a possible explanation for the negative association between self- stigma and medication adherence among mental health patients. Previous studies have reported that depression patients with self-stigma, even after making a rst visit to a mental health clinic, often discontinue their treatment or medication [88,101]. Experiment 3 in this study may be seen as a experimental reproduction of early discontinuation of treatment among stigmatic depression patients. The subjects of experiment 3 consist of those who initially have chosen to receive both simple and detailed diagnostic results, but, after receiving the simple result rst, some subjects change the initial decision and choose not to receive detailed diagnostic information (percentile score). Also, as displayed on Table 3.11, such avoidance behavior is more prominent among subjects with higher degrees of stigma or subjects who receive simple test results that unexpectedly imply the depression. While it is unclear to what extent treatment discontinuation is related to avoidance of diagnostic information, the behavioral patterns observed in this study could supplement existing models of treatment discontinuation or medication non-adherence among depression patients. 3.6.3 Relationship between Avoidance of Diagnosis and Diminished Willingness for Help-seeking One of the aims of this study is to understand the insucient intention to seek help for mental health issues and its relationship to mental health self-stigma through considering the eect of stigma on the demand for diagnosis result. Use of the demand for diagnostic information as an alternative dependent variable in research may have a number of merits. First, this variable does not depend on self-reporting. Rather, it can be easily measured in experimental setting by observing the actual choices made by subjects. Second, understanding mental health care service use behavior is a tricky task because the decision to visit a clinic is aected by various factors. This study, by focusing on the impact of receiving diagnostic information on identity-based utility and its implication on the demand for diagnosis, may contribute to the understanding an aspect of help-seeking behavior. Importantly, even though the analyses in this study do not completely verify the relationship between the demand for diagnostic information and real service use, they provide supportive ev- idence for the association of these two variables. Column (4) in Table 3.4 shows that there is a statistically signicant positive correlation between demand for diagnostic information and the 60 intention to utilize mental health care services. Two explanations for this association might be posed. First, diagnosis avoidance may be the direct reason for diminished help-seeking intention. Second, there may be common factors aecting both help-seeking intention and demand for diagnostic information. Even though the rst explana- tion may be valid to some extent, additional regression analysis supports second explanation. As in the column (5) and (6) in Table 3.4, when the variables of attitudes toward mental health care system (trust toward mental health services and trust toward depression screening test) are included in the regression, the coecient of intention to utilize mental health care is no longer signicant 13 . This implies that it is through negative view about mental health care system that the association between diminished help-seeking intention and diagnosis avoidance is made. Interestingly, despite this close positive association between help-seeking intention and the de- mand for diagnostic information, mental health stigma is observed to have dierent association patterns with both variables. First, as in column (1) to (6) in Table 3.4, the variable stigma is ob- served to have negative association with demand for diagnostic information. However, as in column (7), this variable is positively associated with the intention of help-seeking from mental health care system, which is unexpected according to the model. One plausible explanation about this unex- pected result may be related with the design of questions measuring stigma levels of subjects. The variable stigma is a composite variable which combines answers to two dierent questions. The rst stigma question asks how much each subject agrees with the common belief that depression occurs for those lacking willpower 14 . Believing depression happens when one lacks willpower may translate into the thought that receiving help from experts will be more eective rather than dealing with depression by oneself, thus, the variable stigma is positively associated with willingness to seek help. Indeed, when running the regression as described in column (7) in Table 3.4 again, but substituting the composite variable stigma with the variable stigma will (answers to the rst stigma question, \do you agree that depression occurs for those lack willpower?"), the coecient has positive and 13 It may be clear that negative attitude toward mental health care system will cause decreased help-seeking inten- tion. However, there is not apparent explanation on association between attitude toward mental health care system and demand for diagnostic information. Possibly, rst, such negative attitudes might re ect internalized mental health stigma which reduces both demand for diagnostic information and help-seeking intention. Second, lack of trust toward mental health professional may provoke a fatalistic view about mental health issues, thus, result in reduced perceived usefulness of the diagnostic information. 14 Whereas, second question asks how much each of the subjects agree with the public's unwelcoming attitudes toward depressed people. 61 signicant ( = 0:257, p-value= 0:0001). Whereas, when substituting with stigma social (\do you agree that depressed people are unpleasant to be with?") the coecient is no longer signicant ( = 0:018, p-value= 0:757). The variable stigma in this analysis is a measure of self-stigma or, what might be termed stigma-endorsement 15 . Another aspects of stigma, that is, perceived stigma, is also observed to have dierent patters of association with demand for diagnostic information and the intention to use mental health care. Roughly, perceived stigma refers to one's perceived extent of public discrimina- tion against those with mental disorder regardless of whether he agrees with this public attitudes 16 . Also, according to previous studies, perceived stigma is known as one of the stigma-related attitudes of mental health patients which interferes with their treatment continuation because this percep- tion could be easily converted into anticipation of discrimination for receiving treatment [102]. The explanatory variable perceived discrimination toward depressed people in Table 3.4 is closely related with the concept of perceived stigma. As expected from the previous literature, column (7) shows that this variable is negatively associated with intention to use mental health care. However, there is a dierent pattern in the case of the relationship between perceived stigma and demand for diagnostic information. The results in column (4) to (6) of Table 3.4 report that perceived stigma (the variable perceived discrimination toward depressed people) is positively associated with the demand for diagnostic information. This is at odds with the its negative association with the intention to use mental health care service. One possible explanation is that some individuals seek Internet-based mental health care services instead of making a physical visit to a care provider's oce which is more likely to result in potential stigmatic experience. In sum, individuals who demand diagnostic information in this study also generally have intention to use mental health care, but some of them consider Internet-based diagnosis as a substitute for physical visits to a clinic. Therefore, in policy point of view, this result implies the need for expanding availability of Internet-based treatment for those have fear of anticipated discrimination and are reluctant to visit a care provider. 15 The term stigma-endorsement appears in the literature on mental health stigma, which is coined to describe individuals' endorsing attitudes toward public's negative stereotype toward mental health patients. 16 The detailed review on the concepts of mental health stigma is on the section 1 of appendix. 62 3.6.4 Dierent Eect Sizes across Subjects and Psychological Reactance Theory In the Experiment 2 of this study, randomized controlled trial design is used to explore the causal eects of self-stigma on self-assessment of depressive symptom and the demand for diagnosis results. The results of this experiment also may provide some implications for designing strategies for reducing negative impacts of mental health self-stigma. In particular, displaying stigma-reducing messages to some of subjects is like a de-stigmatizing campaign to increase mental health awareness and service use intention. The analysis of the results of experiment 2 is partially suggestive of the eectiveness of such an intervention. In this section, the variant eect size across subjects with dierent backgrounds will be discussed in relation with `psychological reactance theory'. The eects of reading stigma-reducing messages on the demand for diagnosis result are found only among the married and full-time employees, which implies the possibility that de-stigmatizing campaign might not be universally eective among people. Indeed, as in Table 3.10, when measuring the variable stigma after displaying each type of messages, married/full-time employee subjects show signicant reduction in stigma compared to stigma-neutral treatment group (1.98 vs. 2.67, p-value=0.01), whereas other subjects do not show such pattern (2.08 vs. 2.22, p-value=0.30). This observation implies that the persuasion occurred only among the married/full-time employees. This study does not explore which traits of the married/full time employees are associated with the likelihood of being persuaded. However, `psychological reactance theory' might be a candidate for explaining these dierent eect sizes across population groups [19,115]. This theory is a commonly used theoretical framework in persuasion research, such as marketing or public health campaign research, to explain why such campaign messages do not result in targeted behavioral changes 17 . According to the literature on the psychological reactance theory, individuals value their freedom of choice or behavior, and autonomy. Accordingly, if the persuasive message is perceived to be intrusive and hinder their freedom of choice, they may show reactance behaviors such as refusing to be persuaded [95]. Thus, if those married/employed as full-time worker tend to have lower level of reactance trait, the eectiveness of displaying stigma-reducing messages among them might be plausibly explained. Although there is lack of previous research that this assumption is valid, there is indirect evidence 17 Sometimes, individuals even display the opposite of the recommended action, which is called \boomerang ef- fects" [26,60]. 63 that reactance trait might be negatively associated with being married or employed as full-time worker. As for the job status, there is a research that reactance trait is a strong predictor of complaints about supervisors [96]. Thus, it could be possible that those with reactance trait choose to be part-time worker or or self-employed. Also, there are previous researches that heightened reactance may negatively aect dating or marital relationship [21,62]. Thus, those with high level of reactance trait might be more likely to choose to delay marriage or remain single compared to others. However, this explanation is not tested empirically, thus, for the future research, it would be required to investigate the traits of individuals which is associated with the eectiveness of the mental health de-stigmatizing intervention, along with putting extra eorts for designing intervention strategies for those have reactance trait. 64 3.7 Conclusion I have experimentally tested the hypothesis that those with high degree of self-stigma about having depression may simultaneously manifest denial of symptoms and avoidance of diagnostic test. This prediction is a direct implication of a model whose baseline assumption is that stigmatic people care about their self-appraisal of their own mental health state. This assumption leads to spontaneous manifestation of denial and diagnosis avoidance. Specically, they are motivated to have positively biased beliefs about their mental healthiness, and, in order to maintain the biased beliefs, they choose to forgo the chance of receiving a diagnostic test. I designed a novel experimental procedure through which I investigated whether there is causal- ity from self-stigma to denial and test avoidance. I employed subjects from Amazon Mechanical Turk. After asking them to answer to the CES-D screening test, I collected data on their guesses about their own CES-D scores and the degrees of self-stigma. Lastly, I asked about their intention of whether or not to receive the CES-D test result after the experiment. As expected, the subjects who had higher levels of self-stigma underestimated their depression severity, and also tended to choose not to receive the CES-D result. Moreover, when I had them read either stigma-inducing messages and stigma-reducing messages to test the causality, subjects who read the stigma-inducing messages were more likely to display the denial bias and diagnosis-avoidance behavior. Taken together, even though worry about condentiality might be one of the strong deterring factors for mental health care use, this study is supportive of internal shame or disappointment of admitting reality, that is, having depression, to be a reason for treatment avoidance. The nature of Internet based experiments makes the condentiality issue less important because anonymity is guaranteed. Moreover, even if the subjects dislike that experiment administrators know their depression state, whether or not they choose to receive test result does not aect this situation. Therefore, it could be concluded that avoidance of receiving diagnostic information among stigmatic subjects in this study is motivated from their internal process rather than because they are conscious of others. Also, the degree of such embarrassment or shame would be increasing in the degree of mental health self-stigma, which is related to the experimental result that stigma-induced subjects are more likely to avoid receiving diagnostic information. Lastly, I suggest two possible future study topics which are motivated by the ndings of the 65 current study. First, it would be important to understand the origin of heterogeneous eect sizes of public health intervention such as anti-stigma or mental health literacy enhancement campaign. In the previous section, I proposed that psychological reactance theory might be a candidate to explain this phenomenon. However, to establish the link between such psychological traits and responsive- ness to the intervention messages, more rigorous empirical investigation is required. Second, mental health self-stigma and avoidance of mental health related information may result in deterioration of mental health state. However, due to lack of survey data that includes such information, studies that explore this topic is rare. Therefore, it would be necessary to design experimental studies that may answer this question. 66 3.8 Appendix 3.8.1 Review of Theories on Mental Health Stigma and Help-seeking 3.8.1.1 Stages of Problem Recognition and Information According to theories of help-seeking pathways for mental disorders [4,32,46], help-seeking can be divided into three identiable stages which are problem recognition, the decision to seek help, and the selection of a help provider. Among them understanding the problem recognition stage is one of the objects of this study. In order for the decision to receive a treatment to occur, the individual rst needs to recognize one's emotional disturbance to be attributable to some kind of disease rather than temporary mood swings. This phase is problem recognition stage in help-seeking theories, and it includes the sub-stages of self-recognition of symptom, self-identication of the type of disease and perception of need for treatment [32]. The proposed factors that may in uence problem recognition and perception of need for treat- ment include the characteristics of the disorder such as the severity of symptoms, socio-demographic factors such as age, gender, race/ethnicity and attitudinal factors such as trust toward health care system, beliefs about the causes of mental illness [107]. Among the studies on the factors in uencing problem recognition and help-seeking, the work by Cause et al. [32] is particularly of interest in that they propose the importance of social network that the patient is in. Specically, they suggest that problem recognition for a specic type of mental disorder may not easily occur in a society that accepting of that psychiatric symptoms because the psychiatric state might be interpreted as normal in that social network [86]. However, there is a complication when it comes to consideration of the relationship between social norm and problem recognition. This is because if social norm denes having mental disorder as inferior personality trait such as lack of willpower then individuals might want to hide their symptom in public and it may interfere with help-seeking intention. Therefore, to explore the eect of social network on problem recognition and help-seeking on a certain mental disorder, it is required to consider not only the prevalence of such symptom in the society but also the public attitudes on the disease. Those public attitudes are closely related with concept of stigma, which will be reviewed in the next section. 67 3.8.1.2 Stigma: Concept and Its Implication on Help-seeking This study aims to explore the relationship between stigmatizing attitude toward those with de- pression and two types of mental health behaviors, that is, under-assessment of symptoms and avoidance of self-relevant mental health information. In this section, the relevant concepts, stigma, and its implications from literature are reviewed. A growing body of literature has studied the concept and health consequence of stigma which is mostly from the area of clinical and social psychology [33, 35, 38]. Mental health stigma is a negative stereotype attached to those with mental health issue, whether it is true or not, such as the statements `mental health conditions are caused by lack of willpower' or `depressed people are socially unpleasant'. According to Corrigan (2004) [35], mental health stigma is manifested twofold. The rst type of stigma is public-stigma which refers to negative stereotypes or prejudices that naive public has towards those with mental health issue. The second type is self-stigma which means the self-blaming or self-discriminating attitudes about one's own mental health condition such as the statements `I am incompetent' or `people will not like me'. Self-stigma often leads to negative emotional consequences including shame, diminished self-esteem or loss of hopefulness. Also, those emotional consequences of self-stigma further impair the social or vocational abilities of oneself by resulting in self-isolation from the social network and loss of motivation to pursue a job or education [37]. Between two types of stigma, self-stigma is the focus of this study. The self-stigmatizing process follow a complicated socio-psychological mechanism. In short, in the course of self-stigmatizing process, an individual rst internalizes the public stigma before becoming a patient (internalized stigma) and applies such negative views towards oneself when developing psychiatric symptoms (self-stigma) [76]. This framework implies that those who have highly internalized public stigma into their own attitudes are more likely to experience self-stigma when they develop mental health problem [37]. Stigma is picked as one of the crucial barriers for mental health care utilization. Specically, desire to avoid being a target of public or self stigmatization is considered as a potential cause for reluctance or delay in help-seeking [33, 56, 97]. In a meta analysis by Clement et al. (2015) [33], several sub-themes of mental health stigma are emerged to have negative association with help- 68 seeking, which include (1) desire to keep one's mental health state secret, and (2) desire to avoid dissonance between one's preferred self-identity and reality. The rst theme is interpreted as fear of anticipated discrimination by others such as employer or peer group, and might be termed as public-stigma avoidance intention. Whereas, the second theme implies individuals avoid mental health care service not only from the desire of non-disclosure but also from the desire to avoid internal shame which is purely psychological motives, which can be conceptualized as self-stigma avoidance intention [109]. These meta analysis results are consistent with the theoretical argument of Corrigan (2004) [35], which hypothesizes the two pathways how stigma can deter help-seeking: (1) the desire to avoid being labeled as a mental patient by others (public-stigma avoidance), and (2) the desire to avoid experiencing shame and embarrassment (self-stigma avoidance). 3.8.1.3 Self-stigma: Eects on Problem Recognition and Attention to Mental Health Information Between public-stigma and self-stigma, the latter is of focus in current study. As mentioned earlier, the precondition for help-seeking to occur is recognition of symptoms, self-identication of specic disease, and this stage involves gathering relevant information including symptomatology of the disease and informal self-administered diagnosis result. In this section, I review two potential mechanisms in which self-stigma might interfere with successful self-appraisal, which are the eects of self stigma on (1) denial or self-labeling avoidance, and (2) motivated inattention to mental health information. First, out of the fear of experiencing self-stigma which would be experienced as a form of shame or embarrassment, individuals might be reluctant to attribute the current emotional disturbances to mental illness [18, 84, 104]. For example, newly-admitted psychiatric patients are observed to display denial behavior, that is, they are likely to refuse to characterize themselves as having a negative mental health condition [84]. Moreover, according to a study with subjects of those potentially depressed but untreated, there is an association between internalized stigma (support for public's stigmatizing attitude) and reduced likelihood of self-identication as having depressive symptoms [104]. This association may imply the negative emotional consequence resulting from admitting depression to oneself would be greater for those have more discriminating view against those with depression. Thus, this process of self-identication failure might not happen just because 69 the subjects are irrational or lack sucient knowledge. Indeed, researchers have hypothesized that `rational' rejection of unpreferable-identity might explain denial of identifying oneself as a mental health patient [84]. The second potential mechanism of how self-stigma avoidance motive may prevent potential patients from recognizing their mental health symptoms is its negative eects on the willingness to attend to relevant mental health information such as treatment options and causes, symptoms of specic mental illness or even information on one's own experiences which may imply one's mental health state and be gathered from recollection of one's daily life. As mentioned earlier, the problem-recognition stage includes self-recognition of symptom, self-identication of the type of disease and perception of need for treatment, and each of them involves gathering relevant information. In order for the treatment decision to be made, individuals rst need to know whether their current emotional disturbance ts into the formal denition of mental disorder and whether it requires help from professionals [75]. For example, a majority of those eventually make a treatment decision are those who actively have sought out information [116]. For example, university students who highly internalized stigmatizing attitudes toward those with depression are less likely to seek counseling information even though the act of seeking information does not require them to see a mental health professional [75]. Also, it is observed that participants of a study are less likely to attend to health-promotion information regarding stigmatized health conditions (e.g., genital herpes or HIV) than stigma-neutral information (e.g., u) [43]. To explain the association between stigma and information avoidance, those studies suggest mechanisms according to which individuals with high level of internalized stigma are likely to perceive the information to be a threat in maintaining positive self-image (individual level eect), or information on stigmatized diseases could be deemed more fearful to attend do (disease-specic eect). However, those explanations require more theoretical basis regarding why personal level stigma or stigma attached to specic disease are positively associated with fear of being informed even though the information itself does not force individuals to take further action. 70 3.8.2 Figures and Tables Figure 3.3: Self-estimation of percentile depression score ranking Note. The higher value of percentile score implies more severe depressive symptom 71 Figure 3.4: underestimation of symptom severity (a) Stigma vs. Bias (b) Self-ecacy vs. Bias Figure 3.5: Demand for depression test result (a) objective depression score (CES-D) (b) Stigma 72 Figure 3.6: [Demand for test result] vs. [stigmaunderestimation] Note. The higher value of percentile score implies more severe depressive symptom 73 Figure 3.7: Eects of message provision (whole sample) (a) Self-evaluation Bias (b) Willingness to receive Information (married & full time employees) (c) Self-evaluation Bias (d) Willingness to receive Information 74 Table 3.3: Underestimation of symptom severity self-estimated severity percentile score (1) (2) (3) percentile score: true score 0.792*** (0.018) 0.794*** (0.019) 0.801*** (0.020) stigma -1.081** -0.986** (0.487) (0.496) self-ecacy -0.861* -0.834* (0.492) (0.495) age 0.065 (0.042) male 0.962 (1.048) married -0.141 (1.063) non-caucasian -3.094*** (1.138) years of education -0.187 (0.242) income ($1,000) 0.006 (0.013) full time job -0.021 (1.042) Observations 1372 1372 1372 Notes. In parentheses are robust standard errors of OLS regression. The dependent variable is self-estimated percentile ranking of one's own CES-D test score. The higher value represents more severe degree of depres- sive symptom. Two independent variables stigma and self-ecacy are standardized. ( p< 0:10; p< 0:05; p< 0:01.) 75 Table 3.4: Information avoidance demand for diagnostic information intention to utilize mental healthcare (1) (2) (3) (4) (5) (6) (7) percentile score: true score 0.015*** (0.002) 0.015*** (0.002) 0.013*** (0.002) 0.013*** (0.002) 0.014*** (0.003) 0.014*** (0.003) -0.001 (0.002) stigma -0.144*** -0.134** -0.112** -0.187*** -0.124* -0.113* 0.184*** (0.055) (0.057) (0.057) (0.062) (0.063) (0.065) (0.057) self-estimation bias -0.011*** (0.003) -0.003 (0.003) self-estimation bias high stigma -0.014*** (0.005) -0.014*** (0.005) -0.014*** (0.005) -0.012*** (0.005) -0.012*** (0.005) self-estimation bias low stigma -0.009** (0.004) -0.009** (0.004) -0.009** (0.004) -0.008** (0.004) -0.009** (0.004) self-ecacy -0.159*** -0.137** -0.139** -0.145** -0.213*** (0.058) (0.060) (0.061) (0.061) (0.054) trust toward mental health services 0.282*** (0.060) 0.293*** (0.060) 0.268*** (0.059) trust toward depression screening test 0.174*** (0.059) 0.182*** (0.059) 0.199*** (0.056) intention to utilize mental healthcare 0.169** (0.070) 0.095 (0.072) 0.061 (0.072) intention to utilize physical healthcare 0.022 (0.069) 0.022 (0.069) 0.025 (0.070) 1.410*** (0.067) perceived discrimination toward depressed people 0.222*** (0.060) 0.198*** (0.060) 0.193*** (0.061) -0.268*** (0.054) previous mental health services use 0.004 (0.048) -0.006 (0.049) 0.004 (0.050) 0.135*** (0.046) age -0.000 -0.010** (0.005) (0.005) male 0.018 -0.024 (0.123) (0.113) married -0.305** 0.003 (0.124) (0.114) non-caucasian -0.052 0.408*** (0.135) (0.125) years of education 0.014 -0.007 (0.028) (0.027) income ($1,000) 0.002 -0.004** (0.002) (0.001) full time job -0.343*** 0.037 (0.123) (0.113) Observations 1372 1372 1372 1372 1372 1372 1372 Notes. In parentheses are robust standard errors of ordered logit regression. The dependent variable is willingness to receive the diagnostic information. Mental health-related variables are standardized (stigma, self-ecacy, trust toward mental health service, trust toward depression screening test, intention to utilize mental healthcare, intention to utilize physical healthcare and perceived discrimination toward depressed people). ( p< 0:10; p< 0:05; p< 0:01.) 76 Table 3.5: Baseline characteristics of each group in experiment 2 (a) whole subjects characteristics stigma-inducing (N = 275) stigma-reducing (N = 265) control (N = 358) percentile score: true score 42.0 (21.9) 39.3 (21.5) 43.0 (22.4) age 35.9 (12.4) 35.2 (11.8) 35.6 (12.1) proportion of male 0.30 0.29 0.29 years of education 14.9 (2.0) 14.8 (2.0) 15.0 (2.1) household income ($1000) 61.2 (44.4) 59.3 (41.7) 60.2 (43.6) proportion of ethnic minorities 0.27 0.26 0.27 previous experiences of mental health treatment 1.3 (1.2) 1.3 (1.2) 1.3 (1.2) (b) married/full time employee characteristics stigma-inducing (N = 71) stigma-reducing (N = 69) control (N = 76) percentile score: true score 42.3 (21.9) 35.6 (20.5) 38.5 (24.3) age 37.4 (10.9) 38.7 (11.0) 36.6 (9.7) proportion of male 0.39 0.33 0.41 years of education 15.4 (2.1) 15.4 (1.8) 15.3 (1.9) household income ($1000) 75.3 (37.7) 75.3 (35.2) 85.5 (39.0) proportion of ethnic minorities 0.15 0.16 0.21 previous experiences of mental health treatment 1.4 (1.3) 1.3 (1.3) 1.3 (1.2) Notes. In parentheses are standard deviation. 77 Table 3.6: Self-estimation of symptom severity and demand for diagnostic information after treat- ments (a) whole subjects characteristics stigma-inducing (N = 275) stigma-reducing (N = 265) control (N = 358) percentile score: true score 42.0 (21.9) 39.3 (21.5) 43.0 (22.4) percentile score: self-estimated score 42.8 (23.8) 41.2 (24.0) 40.9 (24.6) demand for diagnostic information (max = 3) 1.61 (0.62) 1.63 (0.63) 1.58 (0.61) (b) married/full time employee characteristics stigma-inducing (N = 71) stigma-reducing (N = 69) control (N = 76) percentile score: true score 42.3 (21.9) 35.6 (20.5) 38.5 (24.3) percentile score: self-estimated score 39.0 (23.9) 42.4 (25.2) 38.8 (24.6) demand for diagnostic information (max = 3) 1.51 (0.65) 1.72 (0.51) 1.54 (0.60) Notes. In parentheses are standard deviation. 78 Table 3.7: Eect of message provision on self-estimation of symptom severity and demand for diagnostic information (whole samples) (1) (2) self-estimated severity (OLS) demand for diagnostic information (Ordered logit) stigma-inducing: (compared to control) 3.156** (1.470) 0.200 (0.169) stigma-reducing: (compared to control) 3.268** (1.490) 0.341* (0.175) percentile score: true score 0.718*** (0.027) 0.013*** (0.003) Observations 898 898 Notes. In parentheses are robust standard errors of each regression model. ( p< 0:10; p< 0:05; p< 0:01.) 79 Table 3.8: Eect of message provision on self-estimation of symptom severity (samples: married & employed as full time worker) model: OLS self-estimated severity (1) (2) stigma-inducing: (compared to control) -0.651 (3.012) -1.107 (2.998) stigma-reducing: (compared to control) 7.711** (3.031) 7.004** (3.036) percentile score: true score 0.754*** (0.059) 0.767*** (0.060) age 0.216* (0.122) male -0.691 (2.654) non-caucasian -3.225 (3.524) years of education -0.622 (0.687) income ($1,000) 0.042 (0.037) Observations 216 216 Notes. In parentheses are robust standard errors of OLS regression. The dependent variable self-evaluation bias is self-estimated percentile score minus true depression percentile score. The higher value of the bias represents more optimistic evaluation. ( p< 0:10; p< 0:05; p< 0:01.) 80 Table 3.9: Eect of message provision on demand for diagnostic information (samples: married & employed as regular worker) model: Ordered Logit demand for diagnostic information (1) (2) stigma-inducing: (compared to control) -0.064 (0.331) -0.032 (0.341) stigma-reducing: (compared to control) 0.743** (0.361) 0.825** (0.372) percentile score: true score 0.001 (0.007) 0.004 (0.007) age -0.018 (0.014) male -0.459 (0.305) non-caucasian -0.599 (0.396) years of education 0.122 (0.081) income ($1,000) 0.009** (0.005) Observations 216 216 Notes. In parentheses are robust standard errors of ordered logit regression. ( p< 0:10; p< 0:05; p< 0:01.) 81 Table 3.10: Eect of message provision on the degree of stigma (1) (2) married/ full-time employee others stigma-inducing N =71 2.63 (1.56) N =204 2.25 (1.45) stigma-neutral N =76 2.67 (1.61) N =282 2.22 (1.39) stigma-reducing N =69 1.98 (1.48) N =196 2.08 (1.38) Notes. (1) In parentheses are robust standard deviation. (2) Maximum value of stigma is 6. Minimum value of stigma is 0. 82 Table 3.11: Eect of Receiving CES-D test result on the willingness to receive the CES-D percentile score (model: Logit) receive the percentile score treatment group (N = 334) control group (N = 1566) (1) (2) (3) (4) (5) (6) rst signal: depression 0.750* (0.393) 2.310*** (0.790) 2.311*** (0.691) -0.185 (0.613) -0.211 (0.606) -0.198 (0.774) stigma -0.414** (0.197) -0.164 (0.166 ) disappointment (N treat = 193) -2.314*** (0.988) -0.037 (0.988) relief (N treat = 141) -5.086* (2.996) -0.239 (2.678) disappointment low stigma (N treat = 96) -1.709 (1.396) 0.038 (1.335) disappointment high stigma (N treat = 97) -3.730** (1.639) -0.315 (1.251) relief low stigma (N treat = 71) 4.873 (4.448) -0.366 (4.277) relief high stigma (N treat = 70) 5.907** (2.829) 2.227 (1.727) Notes. In parentheses standard errors computed by bootstrap. ( p< 0:10; p< 0:05; p< 0:01.) 83 Chapter 4 Eect of Denial of Having Depression on Spouse's Men- tal Health Among Korean Elderly 4.1 Introduction Depressive symptoms have destructive eects on achieving one's personal and social goals, often contributing to the deterioration of the nancial situation [81] and the physical health [7] of those aected. Moreover, studies have suggested that depressive disorder should be understood in social context, as there is abounding evidence that the negative impacts of depression are not limited to oneself but also aect the members of community surrounding the patients, especially toward their family members [100]. Specically, research shows that the likelihood for individuals to be depressed is positively associated with the depression of their spouses [24, 40]. It is important to understand the epidemiology of depression in the family context for developing mental health care systems capable of dealing with the contagious nature of depression and raising the stability of family lives. The depressive symptoms of each spouse are observed to be highly concurrent [73]. There are some underlying mechanisms proposed for explaining this concurrent occurrence of depression in couples. Among them is the `spillover eect', which refers to the situation in which the depression of one partner causes the depression of other partner in a couple 1 . A number of econometric studies 1 Other proposed mechanisms explain the positive spousal association of depressive symptoms with assortative mating [10] or shared social environment [90]. Assortative mating refers to the matching of personality traits in marriage market, including the predisposition to develop depression. Also, because married couples are likely to be exposed to common social environment, development of depression might be concurrent. 84 present supportive evidences about the existence of spillover eects from the mental health of one spouse to that of the other spouse [80,87,90]. A remaining question is whether the choice of a more adaptive coping strategy, such as openness to acknowledgement of having depression and seeking help, would help mitigate the undesirable spillover of depression between spouses. This question is important in public health perspective for three reasons. First, it may shed light on the social consequences of individual coping style instead of solely focusing on its individual consequences. Second, it may help in understanding the heterogeneity in spillover eects of depression across couples. Lastly, it could help in the design of eective intervention strategies for mental health improvement for families by promoting adapting coping strategies that may minimize negative family level consequences of depression. Despite its importance, studies that explore the eects of the choice of coping strategies on spouses' mental health are rare. In this study, I aim to ll this gap by providing empirical evidence showing that dierent coping strategies, that is, denial and acceptance coping, may have dierent impacts on the extent of spillover of depression among the spouses. Specically, This study inves- tigates the impact of denial and acceptance of depression on spousal mental health state, using a panel survey data of Korean elderly people. The hypothesis is that one spouse being in denial of having depression may be associated with development of depression in the other spouse. The rst potential pathway for this association is a causal eect, or, direct spillover eect. However, if both partners in a couple share similar attitudes such as mental health stigma, it may result in depression for one partner but denial for the other, thus generating correlation. To investigate denial and acceptance on spillover of depression among the spouses I use a panel survey data of Korean elderly. This data set contains information of married couples, including an extensive set of health and demographic variables. The estimation relies on a main variable that how they self-identify their mental health state. Specically, I test whether being in denial instead of acceptance of having depression is associated with higher degree of spousal depression severity. The results show that the individuals whose spouses are in denial have more severe depression. This eect is more prominent if the spouse is categorized as having a clinical depression. The rest of this chapter is organized as follows. In section 4.2, relevant previous literature is reviewed. In section 4.3, potential mechanism and hypothesis is introduced. Next, section 4.4 85 describes estimation strategy, and section 4.5 explains the data set used for the estimation. Section 4.6 presents the estimation result and the section 4.7 discusses the results. Lastly, section 4.8 concludes the chapter. 86 4.2 Literature Review: Spillover Eects, and Coping Strategy The empirical evidence of the spillover eect of depression to the spouses of those aected has been documented and analyzed in economics literature. For example, having a depressed spouse is observed to be positively associated with current or follow-up depressive symptoms [73, 100]. Moreover, studies using modern econometric techniques such as generalized method of moments (GMM) or instrumental viable (IV) suggest that the association between depressive symptoms in couples is partly due to the causal eects, or, spillover eects, of one partner's depression on the other partner's mental health [53,90]. The current study aims to explore the eect of the coping styles of those experience depression on the size of spillover. According to psychiatry literature, the coping strategy that depressed individuals employ may aect the likelihood of curability or the further development of the disease. The coping style could be categorized into two methods, denial and acceptance coping. First, the denial coping is the strategies to get rid of depressive symptoms by means of rumination or suppres- sion [3,44]. Denial coping is associated with the refusal to admit that one has depressive symptoms. This could be an eective method for preserving one's positive self-image or avoiding burdensome therapies. However, it may lead to maladaptive behavior such as delaying help-seeking [83] and can cause more serious psychological disturbances [52]. On the other hand, acceptance coping is dened as the openness to the internal experience of depressive symptoms [28], which is known to be more adaptive strategy in helping reduce the symptoms [16]. Moreover, healthy individuals as well may be more likely to develop depressive symptom when using denial coping compared to using acceptance coping [74]. Despite the observation that acceptance coping may help in dealing with depression, there are lack of empirical studies that investigate how the coping style of one partner aects the mental health of the other partner among married couples. However, there is ample narrative evidence that having a depressed partner who is in denial of the illness may provoke stress. Not only is it burdensome to persuade the depressed partner to start receiving a treatment, but those in denial often treat their caregivers with hostility instead of gratitude [66], which may increase mental health risk in their families. Moreover, it has been widely reported that denial of having mental health issues is positively 87 associated with mental health stigma. Especially, self-stigma against having depression or receiving treatment is known to be associated with choosing denial coping instead of acceptance [25, 108]. This association may have important implications for understanding marital stress from having a depressed spouse because endorsing mental health stigma is reported to be correlated with other personal traits that may undermine the marital relationship. For example, lesser endorsement of mental health stigma is associated with hospitality [5], or agreeableness [117]. Considering that the likelihood of choosing denial coping is correlated with mental health self-stigma, spouses of depressed individuals who choose denial coping style may also develop depressive symptoms due to marital stress. This study makes a new contribution to the prior economics and public health literature on family dynamics of mental health issues. To my knowledge, there are no previous studies that investigate the eects of choice of coping style on the extent of spillover of depression among married couples using large survey data sets. Specically, based on the previous ndings reviewed in this section, I hypothesize that denial and acceptance coping for depression will have dierent consequences to the partner's mental health among married couples, and then go on to test this hypothesis using national level survey data. 88 Figure 4.1: Mechanisms of how the choice of coping strategy aects spousal depression 4.3 Mechanisms In this section, brief possible explanations for how the spillover eect of depressive symptoms is less likely to occur if a depressed spouse uses acceptance coping compared to when using denial coping. A rst possible mechanism is the direct eect of choice of coping style on the spillover of depression between each partner. I assume that individuals may choose one of two types of coping styles when they feel depressive symptoms, denial and acceptance coping. Denial is associated with the refusal to choosr an appropriate lifestyle, such as to receive treatment, which is more adaptive to dealing with the emotional suerings. Whereas acceptance coping is likely to lead to readiness to actively seek help. Moreover, it is often observed that depressed individuals express hostility and blame toward spouses who try to persuade them to start treatment, thereby deepening the sense of helplessness for those spouses and weakening family cohesion [66]. Overall, having a partner in denial may generate extra stress because of the frustration of being unable to help them to deal with their symptoms. On the other hand, acceptance coping may result in less spillover of depressive symptoms to the spouse. This is because acceptance is associated with readiness to receive help from others. A second explanation is the correlation of the tendency to choose denial and other personal traits that may undermine the quality of marital relationship. According to previous studies, 89 denial coping is more likely to be manifested among individuals who have high levels of mental health self-stigma [25,108]. Again, self-stigma is observed to be a personal trait that may impede marital communication, and lead to lower levels of hospitality [5], or agreeableness [117]. Therefore, individuals who are in acceptance might have traits that enhance the marital satisfaction of their spouses, hence, mitigating the spillover of depressive symptoms. The two mechanisms discussed in this section are represented in Figure 4.1. Overall, if the level of depressive symptoms remains constant, choosing denial coping instead of acceptance coping may contribute to the occurrence of spillover eect of depression through a direct causal eect and also indirect correlation eect. In this study, I aim to test the existence of those two eects using national level panel survey data. 90 4.4 Estimation The main hypothesis is that the individuals whose spouses are in acceptance of their having depres- sive symptoms are less likely to be aected by spousal depression compared to individuals whose spouses are in denial. In other words, if the severity of the depressive symptom of the spouse holds constant, having spouse in denial results in the further development of one's own depressive symptoms. This hypothesis is represented as a positive valued coecient in the following panel equation. Y it = +Y pt + A pt + 0 1 X it + 0 2 X pt + 0 3 X ht +u ip + it (4.1) with i, p and h represent the individual, the spouse and household, respectively. The dependent variableY it is the severity of depressive symptom of the individual i in time t, while the explanatory variable Y pt is the severity of the spouse's depressive symptom. The main explanatory variable of interest is A pt : the choice of coping style for depression of the spouse of that individual. This is a binary variable whose value is 1 when the spouse chooses to have acceptance-coping and 0 when having denial-coping. Also, u ip is the combination of unobserved personality characteristics of the individual and spouse that may aect the depression severity of the individual. It is expected that u ip is corre- lated with the two explanatory variables, Y pt and A pt . First, based on the theory of assortative mating [10,79], individuals who have a predisposition for developing depression may have a spouse with a similar predisposition, which implies the positive correlation between u ip and Y pt . Next, as discussed in the previous section, individuals who choose to have acceptance-coping may have other personal traits that enhance marital satisfaction of their spouses, resulting in a negative correlation between u ip andA pt . These correlations will generate over-estimation and under-estimation of the coecients and , respectively, when using OLS or random eect model. Therefore, to adjust for this bias, I apply xed eect estimation model in estimation. Next, it is a time-varying error term. the common environment may aect both partner's depressive symptoms and the choice of coping styles, which generates correlation between this term ( it ) and the explanatory variables. For this reason, I include a wide set of individual and household characteristics, and physical health related variables of both partners as control variables in order 91 to minimize the bias due to the endogeneity in time-varying components in explanatory variables. According to the proposed mechanism in the previous section, the coecient of choosing accep- tance coping style in comparison with denial coping style ( ) is expected to have a positive sign. This estimated coecient of xed eect model might be interpreted as a direct eect of spousal coping style on the mental health of individuals for two reasons. First, using xed eect estimation, the personal traits of each spouse that may aect both depressive symptom of one spouse and the choice of coping style of the other, that are assumed to be time-invariant, are controlled for. Second, the time-variant error term is plausibly assumed to be minimally correlated with the variable A pt because, once the severity of depressive symptoms is hold constant, choice of coping style is mostly driven by the (time-invariant) personality traits of individuals. It is worth mentioning that, even though consistent estimates can only be obtained by using xed eects model, random eect coecient of also retains meaningful information. If the esti- mated coecient from the random eect estimation is greater than coecient from the xed eect estimation, it can be inferred that there exist personal traits of individuals which are correlated with their choice of coping strategies and also correlated with the depressive symptoms of their spouse. Overall, the xed eect model is used to estimate the direct eect of the choice of coping strategy for depression on the partner's mental health. Also, apart from the direct eect, there might be an indirect association between choice of coping strategy and the partner's mental health coming through some personality traits of individual which aect both variables. In this case, the estimate of is expected to be larger when using random eect model than when using xed eect model. 92 4.5 Description of Data The estimation of the model is performed by using data from six waves of the KLoSA (Korean Longitudinal Study of Aging). This data is nationally representative biennial panel survey of the Korean elderly (over 45 years old as of 2006), which started in 2006 and the most recent survey year is 2016. KLoSA includes an extensive set of economic, socio-demographic and health-related information of individuals. Moreover, this survey is designed to include both partners if the survey participant is married. this allows me to design a study that can investigate the eects of one partner's traits on the other partner's health. I limited the analysis to the married individuals whose spouses are also in the data, also both of them are not missing in any of the survey years. After eliminating samples in which one of or both of the partners have missing data or deceased in the middle, I identied 2048 men and 2048 women. Dependent Variable: CES-D 10 (depression severity) KLoSA contains various health information including physical and mental health, and health- related behavior. Information on mental health state in the data focuses on depression. Among them, this survey includes the scores of short-form (10-item) Center for Epidemiologic Studies Depression Scale (CES-D10) of samples at each wave. The CES-D10 is a commonly used screening instrument for assessing the depressive symptoms experienced during the most recent week before the test. Each of the 10 items assesses a specic symptom of depression. Among them, 8 items are negatively phrased (bothered, trouble keeping mind, depressed, everything is an eort, fearful, sleep restless, lonely, could not get going), and 2 items are positively phrased (hopeful about future, happy). Survey participants response to each item ranging from 0 to 3 depending on the frequency of experience described in the question (0: less than 1 day per week 3: more than 5 days per week). The total score is the sum of the answers to each items. Thus, the total score will be ranged from 0 to 30. 93 Acceptance vs. Denial The KLoSA also has information on the subjective appraisal of one's own depressive symptom, apart from the information of CES-D 10 scores of each participants. In the health section of the survey, participants answer a question on whether or not they have experienced depressive symp- tom spanning more than two weeks. The response is either `yes' or `no'. I interpreted the answer `yes' as the participant has acceptance coping style and `no' as denial coping style. Conditioning variables Lastly, in order to minimize the possibility that the time-varying error term is correlated with explanatory variables, I include various socio-demographic and physical health-related variables as controls. The information on both the individual of interest and the spouse is included in the estimation. The list of the controlling variables is presented Table 4.1. Descriptive statistics: comparison between when the spouse is in denial and in accep- tance Table 4.2 provides the descriptive statistics for the samples used in the analysis. The average values of the variables are computed separately for two groups: (1) individuals with spouses who self-assess that they do not have depression (denial) (2) individuals with spouses who self-assess that they have depression (acceptance). The notable dierence between two groups is in the CES-D scores (depression severity) of the spouses. The average value of the CES-D scores of spouses who are in acceptance have higher value (13.25) than those in denial (5.87). This dierence implies that the self-appraisal of one's mental health state re ects the true state. Also, the CES-D scores of individuals whose spouses are in acceptance (8.79) is higher compared to when their spouses are in denial (6.07). This observation captures the concurrence of depressive symptoms in couples, due to assortative mating, common environment and spillover eects. However, this comparison cannot oer an insight on how the choice of coping style may aect 94 spousal mental health, because the choice also re ects true severity of depressive symptoms. The question is whether or not the choice of acceptance coping style reduces the extent of spousal mental health deterioration. Thus, next, I make a comparison between the mental health consequences of having spouse in denial and in acceptance, within groups that have similar spousal depressive symptoms severity in each group. This analysis is on Table 4.3. When the spouses do not have clinical depression (CES-D < 10) there is little dierence in depression severity of individuals regardless of whether the spouses are in denial or in acceptance. However, when spouses are categorized as having clinical depression, having a spouse in acceptance is associated with lower level of the depression severity of the individual. When the CES-D score of spouses is between 11 and 15, less degree of depression severity is observed among individuals (10.12 vs. 8.95). Moreover, more noticeable dierence is observed when the spouses have severe depression (when CES-D score of spouses is between 16 and 20) (13.21 vs. 9.67). However, the similar pattern is not observed when spouses have extremely severe depression (when CES-D score of spouses equals to is above 21). This comparison oers an expectation that the positive eects of choosing acceptance instead of denial on the spouses' mental health might be prominent if the spouses have clinical depression. In the next section, the result of the regression analysis is presented. 95 4.6 Result 4.6.1 Eect of Spousal Acceptance of Having Depressive Symptoms Table 4.4 shows the coecients of the equation (4.1) estimated by using xed eect and random eect model, with and without including control variables. In all columns, individuals' depression severity is positively correlated with spouses' depression severity, which captures both concurrence of depression in couples (xed eect) and assortative mating (random eect). Next, eect of having a spouse being in acceptance (variable Acceptance (spouse)) on individuals' depression severity is negative once depression severity of the spouse is controlled for (p-value < 0.01). This negative association between depressive symptom and the spouse's acceptance is observed in both xed eect and random eect model. Whether including controlling variables or not does not make signicant change in this estimated coecient in both estimation models. Also, this coecient is estimated greater (in absolute degree) when applying random eect model. Random eect model does not adjust for the time-invariant individual characteristics. Thus, it can be inferred that the likelihood of choosing acceptance instead of denial is associated with time-invariant personal traits that may have positive impact on the partner's mental health. 4.6.2 Acceptance among Highly Depressed Spouses Based on the observation from Table 4.3, it can be expected that the positive impact of acceptance of having depression on spousal mental health state is more prominent among highly depressed individuals. Additional regression analysis yields the consistent pattern with this observation. For this regression analysis, I allowed that the coecient of the variable Acceptance (spouse) to be varied depending on the depression severity of the spouse. Specically, the coecient is separately estimated when spouses are screened to have clinical depression (CES-D (spouse) 10) or to have only slight degree of depressive symptoms (CES-D (spouse)< 10). Table 4.5 summarizes the results of this analysis. As expected, the positive impact of spousal acceptance of having depressive symptoms on individuals' mental health is greater when the spouse has severe degree of depressive symptoms. As in the column (1) The eect size estimated using xed 96 eect model is 1.71 (p-value < 0.01) when the spouse is categorized as having clinical depression but the eect size is lower ( = 0.45, p-value = 0.07) otherwise. As in the column (2), after including controlling variables, the eect size of spousal acceptance among those with severe spousal depres- sion remains statistically signicant, but it is no longer signicant if spousal depressive symptoms are slight. The same pattern is observed for the estimates using random eect model. This observation implies that the positive eect of choosing acceptance of having depression on spousal mental health is particularly relevant when the depressive symptoms are severe. 97 4.7 Discussion The results of the regression analysis show that acceptance of having depression may mitigate the spillover of the depressive symptoms to the marital partner, especially when having severe (clinical) symptoms is . Also, the eect persists under xed eect specication, which implies that the change of the attitude (denial or acceptance) in a person over time also aects the partner's mental health. In this section, the potential interpretation of the result is presented. 4.7.1 Suggested Interpretation This result should be interpreted with caution because it is uncertain if this eect is causal. First, there may exist a possibility of inverse causality: the lower level of partner's depressive symptom causes individuals to prefer acceptance over denial. However, this might be an unlikely explanation. To my knowledge, there is no previous study to theorize this hypothesis. Also, it might be dicult to nd a plausible mechanism. Second, their might be other unobserved factors that are correlated with both the tendency to choose acceptance or denial by an individual, and the spouse's mental health. Those factors may include both partners' attitudinal factors such as mental health stigma, or personality traits of the depressed partner such as hospitality or agreeableness. Mental health stigma as a personal attitude [92,94] is known to have a positive association of depression severity and also to result in denial [25]. Because the stigmatic attitudes might be shared by both partners out of assortative mating or in uences from one partner to another, it might be likely to occur that one partner develops less degree of depressive symptoms and another partner chooses to have acceptance coping if the low level of mental health stigma is shared in a couple. Similarly, if the depressed partner has a personal traits that facilitate acceptance, which also may enhance the marital relationship such as hospitality and agreeableness, their could be correlation observed between one partner's low level of depressive symptom and another partner's acceptance. Third, the result of the analysis in this study might re ect the direct positive eect of being in acceptance on the spousal mental health. Being in denial may cause frustration to the spouse who is likely to play a care-giving role. Also, denial is often accompanied with anger or hostility toward people who try to persuade them to seek treatment, which might exacerbate the spouse's mental 98 health. 4.7.2 Relationship between Coping style and Choosing `Acceptance' In this study, the variable `acceptance' is dened as whether an individual self-identify oneself to have experienced depressive symptom in the past year. There would be two dierent underlying factors behind this self-identication. First factor is the true severity of depressive symptoms. Specically, the likelihood of choosing `acceptance' is increasing in the true depression severity. Second, there would be a set of personal traits that is correlated with predisposition to the choice of acceptance or denial. Those two factors can be represented as follow equation. A it = 8 > > < > > : 1 (Acceptance) if A 0 it > 0 0 (Denial) if A 0 it 0 (4.2) A 0 it = 0 + 1 Y it +p it where Y i is the severity of depressive symptoms and p i represents the unobserved set of personal traits that in uence the choice. In the regression analysis (Table 4.4), spousal `Acceptance' is observed to positively aect in- dividuals' mental health only when spousal symptom severity is high (high CES-D score). Or, equivalently, spousal `Denial' negatively aects individuals' mental health. Because the symptom severity of the spouse is controlled for in the regression, the estimated eect mostly re ects the term p it in the the equation 4.2. The term p it may represent the coping style, or, individuals' predisposition to the choice of acceptance when depressed. The result that the eect of spousal `Denial' is manifested only among high spousal depression group implies that low value of the spouse'sp it is associated the individual's worse mental health state. This is because it can be interpreted that low depressive symptom severity and choosing denial might be merely driven by accurate self-appraisal of one's own mental health state but it is due to low value of p it among high depressive symptom group. On the other hands, high value of p it of spouses may not be associated with individual's mental health because the eect of spousal acceptance is not observed among individuals with less depressed spouses. 99 In sum, the result and its interpretation describes that the tendency to choose denial instead of acceptance especially among depressed spouses may negatively aect individuals' mental health. 100 4.8 Conclusion This study investigates the eect of spousal acceptance or denial of having depression on individuals' mental health. The xed eect and random eect estimation yield results that spousal denial of having depression negatively aects individuals' mental health. Potential mechanism of this observation is proposed. First, there would be direct deteriorating eect of being in denial on the partner's mental health. Second, there might be factors such as personality traits or mental health stigma that link spousal denial of having depression and individuals' mental health. The consequence of having mental health issues in a family context is one of the understudied areas in economics and public health, even though it has importance in designing health care system that may improve health in families. This study may provide a initial motivation for future research on this topic. 101 4.8.1 Appendix: Tables 102 Table 4.1: List of the conditioning variables characteristics description Demographic age age of the participant as of the year 2006 gender 0 = male, 1 = female years of education 1 = less than 6 years, 2 = 9 years, 3 = 12 years, 4 = more than 12 years area of residence 1 = large city, 2 = small and medium-sized city, 3 = countryside number of children total number of children social activity range:f1 to 10g from no social activity (1) to everyday (10) religion 0 = not have religion, 1 = have religion Income & Labor household income total income ($1000) employment status 1 = employed, 2 = unemployed, 3 = not in economic activity Health overall health range:f1, 2, 3, 4, 5g from very bad (1) to very well (5) disability 0 = not having disability, 1 = at least one disability ADL (activities of daily living) the level of functional assistance or support required range:f0 to 7g from normal (0) to totally dependent (7) MMSE (Mini-Mental State Exam) the level of functional assistance or support required range:f0 to 30g from severe dementia (0) to normal (30) Health behaviors exercise 0: not exercising, 1: exercising at least once a week smoking 0: non-smoker, 1: smoker drinking alcohol 0: non-drinker, 1: drinker 103 Table 4.2: Descriptive statistics (spousal denial vs. spousal acceptance) characteristics denial (N = 15632) acceptance (N = 692) CES-D score 6.07 (4.56) 8.79 (6.05) CES-D score (spouse) 5.87 (4.31) 13.25 (6.43) age 58.1 (8.45) 60.7 (8.42) fraction of male 0.49 0.64 years of education 9.50 (2.98) 8.34 (2.84) number of children 2.83 (1.28) 3.04 (1.42) social activity 3.50 (2.87) 3.06 (2.26) religion 0.51 0.50 household income ($1000) 28.21 (25.98) 17.86 (20.44) overall health 2.20 (0.86) 1.84 (1.01) fraction of disabled 0.18 0.06 ADL (activities of daily living) 0.52 (0.52) 0.19 (1.01) MMSE (Mini-Mental State Exam) 26.13 (3.69) 25.09 (4.33) fraction of ever exercising 0.40 0.32 smoking 0.18 0.26 drinking alcohol 0.41 0.48 Notes. CES-D score and ADL are coded in a way that lower value represents healthier state. MMSE is coded in a way that lower value represents more symptoms of dementia. 104 Table 4.3: Descriptive statistics (adjusting for spousal depression severity) (spouse's self assessment: non-depression vs. depression) characteristics denial acceptance CES-D score of spouses: 0 5 (N = 6469) (N = 69) CES-D score 3.23 (3.67) 3.56 (3.04) CES-D score (spouse) 1.93 (1.79) 3.06 (1.74) CES-D score of spouses: 6 10 (N = 7052) (N = 185) CES-D score 7.27 (3.30) 7.14 (4.21) CES-D score (spouse) 7.20 (1.42) 8.18 (1.46) CES-D score of spouses: 11 15 (N = 1720) (N = 194) CES-D score 10.12 (4.50) 8.95 (4.77) CES-D score (spouse) 12.35 (1.29) 12.96 (1.38) CES-D score of spouses: 16 20 (N = 313) (N = 150) CES-D score 13.21 (5.77) 9.67 (6.22) CES-D score (spouse) 17.43 (1.36) 17.87 (1.37) CES-D score of spouses: 21 30 (N = 81) (N = 91) CES-D score 13.98 (7.00) 14.33 (8.07) CES-D score (spouse) 23.16 (2.47) 24.35 (3.00) Note. CES-D score of 10 or higher is screened to be potentially having a clinical depression. 105 Table 4.4: Eect of spouse's acceptance of having depression (N = 16324) Depression Severity (CES-D score) Fixed Eect Random Eect (1) (2) (3) (4) CES-Dt (spouse) 0.508*** (0.008) 0.489*** (0.008) 0.583*** (0.007) 0.546*** (0.007) Acceptancet (spouse) -1.112*** (0.156) -1.090*** (0.153) -1.502*** (0.143) -1.649*** (0.137) Year 2008 0.286*** (0.067) 0.304*** (0.067) 0.222*** (0.067) 0.251*** (0.067) Year 2010 0.462*** (0.067) 0.258*** (0.070) 0.370*** (0.068) 0.132* (0.069) Year 2012 0.394*** (0.068) 0.175** (0.069) 0.310*** (0.068) 0.072 (0.068) age -0.023*** (0.006) female 0.429*** (0.092) education 0.006 (0.040) regiont (small city) 0.118 (0.368) 0.006 (0.080) regiont (countryside) 0.567 (0.402) 0.179** (0.091) number of childrent -0.209 (0.320) 0.005 (0.033) social activityt -0.059*** (0.015) -0.105*** (0.012) religiont 0.008 (0.081) -0.126** (0.060) incomet 0.000 (0.000) 0.000 (0.000) employment statust unemployed 0.612** (0.253) 0.926*** (0.230) employment statust not in economic activity 0.311*** (0.099) 0.365*** (0.070) overall healtht -0.843*** (0.043) -1.002*** (0.036) disabilityt 0.199 (0.195) 0.285 (0.185) ADLt 0.435*** (0.066) 0.569*** (0.053) MMSEt -0.097*** (0.011) -0.124*** (0.009) exerciset -0.181** (0.074) -0.207*** (0.060) smokingt -0.268* (0.154) 0.037 (0.091) drinkingt -0.385*** (0.133) -0.147** (0.074) 106 Table 4.5: Eect of spouse's acceptance of having depression (N = 16324) (Dierential eect size across spousal depressive symptom severity) Depression Severity (CES-D score) Fixed Eect Random Eect (1) (2) (3) (4) CES-Dt (spouse) 0.524*** (0.008) 0.495*** (0.008) 0.597*** (0.007) 0.556*** (0.007) Acceptancet (spouse) (CES-D (spouse)< 10) -0.446* (0.250) -0.264 (0.245) -0.427* (0.235) -0.356 (0.227) Acceptancet (spouse) (CES-D (spouse) 10) -1.714*** (0.190) -1.550*** (0.186) -2.226*** (0.173) -2.328*** (0.166) Year 2008 0.309*** (0.067) 0.258*** (0.067) Year 2010 0.263*** (0.070) 0.140** (0.069) Year 2012 0.181*** (0.069) 0.081 (0.068) age -0.023*** (0.006) female 0.434*** (0.091) education 0.004 (0.040) regiont (small city) 0.133 (0.368) 0.002 (0.080) regiont (countryside) 0.563 (0.402) 0.177** (0.090) number of childrent -0.223 (0.319) 0.004 (0.033) social activityt -0.059*** (0.015) -0.106*** (0.012) religiont 0.009 (0.081) -0.123** (0.059) incomet 0.000 (0.000) 0.000 (0.000) employment statust unemployed 0.626** (0.253) 0.941*** (0.230) employment statust not in economic activity 0.322*** (0.099) 0.368*** (0.070) overall healtht -0.843*** (0.043) -1.004*** (0.036) disabilityt 0.204 (0.195) 0.297 (0.185) ADLt 0.437*** (0.066) 0.575*** (0.053) MMSEt -0.096*** (0.011) -0.123*** (0.009) exerciset -0.178** (0.074) -0.2067*** (0.060) smokingt -0.260* (0.154) 0.047 (0.091) drinkingt -0.385*** (0.133) -0.146** (0.074) 107 Bibliography [1] Akerlof, G. A., and Dickens, W. T. The economic consequences of cognitive dissonance. The American economic review 72, 3 (1982), 307{319. [2] Akerlof, G. A., and Kranton, R. E. Economics and identity. The Quarterly Journal of Economics 115, 3 (2000), 715{753. [3] Aldao, A., and Nolen-Hoeksema, S. Specicity of cognitive emotion regulation strate- gies: A transdiagnostic examination. Behaviour research and therapy 48, 10 (2010), 974{983. [4] Andersen, R. M. Revisiting the behavioral model and access to medical care: does it matter? Journal of health and social behavior (1995), 1{10. [5] Ando, S., Nishida, A., Usami, S., Koike, S., Yamasaki, S., Kanata, S., Fujikawa, S., Furukawa, T. A., Fukuda, M., Sawyer, S. M., et al. Help-seeking intention for depression in early adolescents: Associated factors and sex dierences. Journal of aective disorders 238 (2018), 359{365. [6] Augsberger, A., Yeung, A., Dougher, M., and Hahm, H. C. Factors in uencing the underutilization of mental health services among asian american women with a history of depression and suicide. BMC health services research 15, 1 (2015), 542. [7] Bardone, A. M., Moffitt, T. E., Caspi, A., Dickson, N., Stanton, W. R., and Silva, P. A. Adult physical health outcomes of adolescent girls with conduct disorder, depression, and anxiety. Journal of the American Academy of Child & Adolescent Psychiatry 37, 6 (1998), 594{601. 108 [8] Bathje, G., and Pryor, J. The relationships of public and self-stigma to seeking mental health services. Journal of Mental Health Counseling 33, 2 (2011), 161{176. [9] Baumann, A. E. Stigmatization, social distance and exclusion because of mental illness: the individual with mental illness as a `stranger'. International review of psychiatry 19, 2 (2007), 131{135. [10] Becker, G. S. A theory of marriage: Part ii. Journal of political Economy 82, 2, Part 2 (1974), S11{S26. [11] Bell, D. E. Disappointment in decision making under uncertainty. Operations research 33, 1 (1985), 1{27. [12] B enabou, R. Groupthink: Collective delusions in organizations and markets. Review of Economic Studies 80, 2 (2012), 429{462. [13] B enabou, R., and Tirole, J. Self-condence and personal motivation. The Quarterly Journal of Economics 117, 3 (2002), 871{915. [14] B enabou, R., and Tirole, J. Identity, morals, and taboos: Beliefs as assets. The Quarterly Journal of Economics 126, 2 (2011), 805{855. [15] B enabou, R., and Tirole, J. Mindful economics: The production, consumption, and value of beliefs. Journal of Economic Perspectives 30, 3 (2016), 141{64. [16] Berking, M., Ebert, D., Cuijpers, P., and Hofmann, S. G. Emotion regulation skills training enhances the ecacy of inpatient cognitive behavioral therapy for major depressive disorder: a randomized controlled trial. Psychotherapy and psychosomatics 82, 4 (2013), 234{245. [17] Bernheim, B. D., and Thomadsen, R. Memory and anticipation. The Economic Journal 115, 503 (2005), 271{304. [18] Bharadwaj, P., Pai, M. M., and Suziedelyte, A. Mental health stigma. Economics Letters 159 (2017), 57{60. [19] Brehm, J. W. A theory of psychological reactance. 109 [20] Brocas, I., and Carrillo, J. D. Dual-process theories of decision-making: A selective survey. Journal of economic psychology 41 (2014), 45{54. [21] Brock, R. L., and Lawrence, E. Too much of a good thing: Underprovision versus overprovision of partner support. Journal of Family Psychology 23, 2 (2009), 181. [22] Brown, J. D. Understanding the better than average eect: Motives (still) matter. Personality and Social Psychology Bulletin 38, 2 (2012), 209{219. [23] Brunnermeier, M. K., and Parker, J. A. Optimal expectations. American Economic Review 95, 4 (2005), 1092{1118. [24] Burke, L. The impact of maternal depression on familial relationships. International review of psychiatry 15, 3 (2003), 243{255. [25] Byrne, P. Stigma of mental illness and ways of diminishing it. Advances in Psychiatric treatment 6, 1 (2000), 65{72. [26] Byrne, S., and Hart, P. S. The boomerang eect a synthesis of ndings and a preliminary theoretical framework. Annals of the International Communication Association 33, 1 (2009), 3{37. [27] Camerer, C., and Lovallo, D. Overcondence and excess entry: An experimental ap- proach. American economic review 89, 1 (1999), 306{318. [28] Campbell-Sills, L., Barlow, D. H., Brown, T. A., and Hofmann, S. G. Eects of suppression and acceptance on emotional responses of individuals with anxiety and mood disorders. Behaviour research and therapy 44, 9 (2006), 1251{1263. [29] Caplin, A., and Leahy, J. Psychological expected utility theory and anticipatory feelings. The Quarterly Journal of Economics 116, 1 (2001), 55{79. [30] Caplin, A., and Leahy, J. V. Wishful thinking. NBER Working paper (2019). [31] Carrillo, J. D., and Mariotti, T. Strategic ignorance as a self-disciplining device. The Review of Economic Studies 67, 3 (2000), 529{544. 110 [32] Cauce, A. M., Domenech-Rodr guez, M., Paradise, M., Cochran, B. N., Shea, J. M., Srebnik, D., and Baydar, N. Cultural and contextual in uences in mental health help seeking: a focus on ethnic minority youth. Journal of consulting and clinical psychology 70, 1 (2002), 44. [33] Clement, S., Schauman, O., Graham, T., Maggioni, F., Evans-Lacko, S., Bezborodovs, N., Morgan, C., R usch, N., Brown, J., and Thornicroft, G. What is the impact of mental health-related stigma on help-seeking? a systematic review of quan- titative and qualitative studies. Psychological medicine 45, 1 (2015), 11{27. [34] Compte, O., and Postlewaite, A. Condence-enhanced performance. American Economic Review 94, 5 (2004), 1536{1557. [35] Corrigan, P. How stigma interferes with mental health care. American psychologist 59, 7 (2004), 614. [36] Corrigan, P. W., Edwards, A. B., Green, A., Diwan, S. L., and Penn, D. L. Prejudice, social distance, and familiarity with mental illness. Schizophrenia bulletin 27, 2 (2001), 219{225. [37] Corrigan, P. W., and Rao, D. On the self-stigma of mental illness: Stages, disclosure, and strategies for change. The Canadian Journal of Psychiatry 57, 8 (2012), 464{469. [38] Corrigan, P. W., and Watson, A. C. Understanding the impact of stigma on people with mental illness. World psychiatry 1, 1 (2002), 16. [39] Coutts, A. Testing models of belief bias: An experiment. Games and Economic Behavior 113 (2019), 549{565. [40] Coyne, J. C., Kessler, R. C., Tal, M., Turnbull, J., Wortman, C. B., and Greden, J. F. Living with a depressed person. Journal of Consulting and Clinical psychology 55, 3 (1987), 347. [41] Crabb, R., and Hunsley, J. Utilization of mental health care services among older adults with depression. Journal of clinical psychology 62, 3 (2006), 299{312. 111 [42] Czyz, E. K., Horwitz, A. G., Eisenberg, D., Kramer, A., and King, C. A. Self- reported barriers to professional help seeking among college students at elevated risk for suicide. Journal of american college health 61, 7 (2013), 398{406. [43] Earl, A., Nisson, C. A., and Albarrac n, D. Stigma cues increase self-conscious emo- tions and decrease likelihood of attention to information about preventing stigmatized health issues. Acta de investigacion psicologica 5, 1 (2015), 1860{1871. [44] Ehring, T., Tuschen-Caffier, B., Schn ulle, J., Fischer, S., and Gross, J. J. Emotion regulation and vulnerability to depression: spontaneous versus instructed use of emotion suppression and reappraisal. Emotion 10, 4 (2010), 563. [45] Eil, D., and Rao, J. M. The good news-bad news eect: asymmetric processing of objective information about yourself. American Economic Journal: Microeconomics 3, 2 (2011), 114{ 38. [46] Eiraldi, R. B., Mazzuca, L. B., Clarke, A. T., and Power, T. J. Service utilization among ethnic minority children with adhd: A model of help-seeking behavior. Administration and Policy in Mental Health and Mental Health Services Research 33, 5 (2006), 607{622. [47] Eisenberg, D., Downs, M. F., Golberstein, E., and Zivin, K. Stigma and help seeking for mental health among college students. Medical Care Research and Review 66, 5 (2009), 522{541. [48] Eliaz, K., and Schotter, A. Experimental testing of intrinsic preferences for noninstru- mental information. American Economic Review 97, 2 (2007), 166{169. [49] Eliaz, K., and Schotter, A. Paying for condence: An experimental study of the demand for non-instrumental information. Games and Economic Behavior 70, 2 (2010), 304{324. [50] Evans-Lacko, S., Knapp, M., McCrone, P., Thornicroft, G., and Mojtabai, R. The mental health consequences of the recession: economic hardship and employment of people with mental health problems in 27 european countries. PloS one 8, 7 (2013), e69792. [51] Feiler, L. Testing models of information avoidance with binary choice dictator games. Journal of Economic Psychology 45 (2014), 253{267. 112 [52] Ferrario, S. R., Giorgi, I., Baiardi, P., Giuntoli, L., Balestroni, G., Cerutti, P., Manera, M., Gabanelli, P., Solara, V., Fornara, R., et al. Illness denial questionnaire for patients and caregivers. Neuropsychiatric disease and treatment 13 (2017), 909. [53] Fletcher, J. All in the family: mental health spillover eects between working spouses. The BE journal of economic analysis & policy 9, 1 (2009). [54] Frey, D. Recent research on selective exposure to information. In Advances in experimental social psychology, vol. 19. Elsevier, 1986, pp. 41{80. [55] Ganguly, A., and Tasoff, J. Fantasy and dread: The demand for information and the consumption utility of the future. Management Science 63, 12 (2016), 4037{4060. [56] Gary, F. A. Stigma: Barrier to mental health care among ethnic minorities. Issues in mental health nursing 26, 10 (2005), 979{999. [57] Goldman, L. S., Nielsen, N. H., Champion, H. C., and Council on Scientific Af- fairs, A. M. A. Awareness, diagnosis, and treatment of depression. Journal of general internal medicine 14, 9 (1999), 569{580. [58] Golman, R., Hagmann, D., and Loewenstein, G. Information avoidance. Journal of Economic Literature 55, 1 (2017), 96{135. [59] Golman, R., and Loewenstein, G. Curiosity, information gaps, and the utility of knowl- edge. Information Gaps, and the Utility of Knowledge (April 16, 2015) (2015). [60] Hart, P. S., and Nisbet, E. C. Boomerang eects in science communication: How motivated reasoning and identity cues amplify opinion polarization about climate mitigation policies. Communication research 39, 6 (2012), 701{723. [61] Henderson, C., Evans-Lacko, S., and Thornicroft, G. Mental illness stigma, help seeking, and public health programs. American journal of public health 103, 5 (2013), 777{ 780. 113 [62] Hockenberry, S. L., and Billingham, R. E. Psychological reactance and violence within dating relationships. Psychological reports 73, 3 suppl (1993), 1203{1208. [63] Holton, B., and Pyszczynski, T. Biased information search in the interpersonal domain. Personality and Social Psychology Bulletin 15, 1 (1989), 42{51. [64] Kahneman, D. A perspective on judgment and choice: mapping bounded rationality. American psychologist 58, 9 (2003), 697. [65] Karlsson, N., Loewenstein, G., and Seppi, D. The ostrich eect: Selective attention to information. Journal of Risk and uncertainty 38, 2 (2009), 95{115. [66] Karp, D. A., and Tanarugsachock, V. Mental illness, caregiving, and emotion manage- ment. Qualitative health research 10, 1 (2000), 6{25. [67] Kessler, R. C., Demler, O., Frank, R. G., Olfson, M., Pincus, H. A., Walters, E. E., Wang, P., Wells, K. B., and Zaslavsky, A. M. Prevalence and treatment of mental disorders, 1990 to 2003. New England Journal of Medicine 352, 24 (2005), 2515{2523. [68] Kohn, R., Saxena, S., Levav, I., and Saraceno, B. The treatment gap in mental health care. Bulletin of the World health Organization 82 (2004), 858{866. [69] K} oszegi, B. Health anxiety and patient behavior. Journal of health economics 22, 6 (2003), 1073{1084. [70] K} oszegi, B. Ego utility, overcondence, and task choice. Journal of the European Economic Association 4, 4 (2006), 673{707. [71] K} oszegi, B. Utility from anticipation and personal equilibrium. Economic Theory 44, 3 (2010), 415{444. [72] K} oszegi, B., and Rabin, M. Reference-dependent consumption plans. American Economic Review 99, 3 (2009), 909{36. [73] Kouros, C. D., and Cummings, E. M. Longitudinal associations between husbands' and wives' depressive symptoms. Journal of Marriage and Family 72, 1 (2010), 135{147. 114 [74] Kraaij, V., Pruymboom, E., and Garnefski, N. Cognitive coping and depressive symp- toms in the elderly: a longitudinal study. Aging & mental health 6, 3 (2002), 275{281. [75] Lannin, D. G., Vogel, D. L., Brenner, R. E., Abraham, W. T., and Heath, P. J. Does self-stigma reduce the probability of seeking mental health information? Journal of Counseling Psychology 63, 3 (2016), 351. [76] Link, B. G. Understanding labeling eects in the area of mental disorders: An assessment of the eects of expectations of rejection. American sociological review (1987), 96{112. [77] Loomes, G., and Sugden, R. Disappointment and dynamic consistency in choice under uncertainty. The Review of Economic Studies 53, 2 (1986), 271{282. [78] Magaard, J. L., Seeralan, T., Schulz, H., and Br utt, A. L. Factors associated with help-seeking behaviour among individuals with major depression: A systematic review. PLoS One 12, 5 (2017), e0176730. [79] Merikangas, K. R. Assortative mating for psychiatric disorders and psychological traits. Archives of General Psychiatry 39, 10 (1982), 1173{1180. [80] Michaud, P.-C., and Van Soest, A. Health and wealth of elderly couples: Causality tests using dynamic panel data models. Journal of health economics 27, 5 (2008), 1312{1325. [81] Mirowsky, J., and Ross, C. E. Age and the eect of economic hardship on depression. Journal of health and social behavior (2001), 132{150. [82] M obius, M. M., Niederle, M., Niehaus, P., and Rosenblat, T. S. Managing self- condence. NBER Working paper (2014). [83] M oller-Leimk uhler, A. M. Barriers to help-seeking by men: a review of sociocultural and clinical literature with particular reference to depression. Journal of aective disorders 71, 1-3 (2002), 1{9. [84] O'Mahony, P. Psychiatric patient denial of mental illness as a normal process. British Journal of Medical Psychology 55, 2 (1982), 109{118. 115 [85] Oster, E., Shoulson, I., and Dorsey, E. Optimal expectations and limited medical testing: evidence from huntington disease. American Economic Review 103, 2 (2013), 804{ 30. [86] Paradise, M., Cauce, A. M., Ginzler, J., Wert, S., Wruck, K., and Brooker, M. The role of relationships in developmental trajectories of homeless and runaway youth. [87] Pascual-S aez, M., Cantarero-Prieto, D., and Bl azquez-Fern andez, C. Partner's depression and quality of life among older europeans. The European Journal of Health Economics 20, 7 (2019), 1093{1101. [88] Pinto-Foltz, M. D., and Logsdon, M. C. Stigma towards mental illness: A concept analysis using postpartum depression as an exemplar. Issues in Mental Health Nursing 29, 1 (2008), 21{36. [89] Ploner, M., and Regner, T. Self-image and moral balancing: An experimental analysis. Journal of Economic Behavior & Organization 93 (2013), 374{383. [90] Powdthavee, N. I can't smile without you: Spousal correlation in life satisfaction. Journal of Economic Psychology 30, 4 (2009), 675{689. [91] Puri, M., and Robinson, D. T. Optimism and economic choice. Journal of Financial Economics 86, 1 (2007), 71{99. [92] Pyne, J. M., Kuc, E. J., Schroeder, P. J., Fortney, J. C., Edlund, M., and Sullivan, G. Relationship between perceived stigma and depression severity. The Journal of nervous and mental disease 192, 4 (2004), 278{283. [93] Rabin, M., and Schrag, J. L. First impressions matter: A model of conrmatory bias. The quarterly journal of economics 114, 1 (1999), 37{82. [94] Raguram, R., Weiss, M. G., Channabasavanna, S., and Devins, G. M. Stigma, depression, and somatization in south india. American Journal of Psychiatry 153, 8 (1996), 1043{1049. 116 [95] Rains, S. A., and Turner, M. M. Psychological reactance and persuasive health commu- nication: A test and extension of the intertwined model. Human Communication Research 33, 2 (2007), 241{269. [96] Sachau, D. A., Houlihan, D., and Gilbertson, T. Predictors of employee resistance to supervisors' requests. The Journal of Social Psychology 139, 5 (1999), 611{621. [97] Schomerus, G., and Angermeyer, M. C. Stigma and its impact on help-seeking for mental disorders: what do we know? Epidemiology and Psychiatric Sciences 17, 1 (2008), 31{37. [98] Schwardmann, P. Motivated health risk denial and preventative health care investments. Journal of health economics 65 (2019), 78{92. [99] Sharot, T. The optimism bias. Current biology 21, 23 (2011), R941{R945. [100] Siegel, M. J., Bradley, E. H., Gallo, W. T., and Kasl, S. V. The eect of spousal mental and physical health on husbands' and wives' depressive symptoms, among older adults: longitudinal evidence from the health and retirement survey. Journal of Aging and Health 16, 3 (2004), 398{425. [101] Sirey, J. A., Bruce, M. L., Alexopoulos, G. S., Perlick, D. A., Friedman, S. J., and Meyers, B. S. Stigma as a barrier to recovery: Perceived stigma and patient-rated severity of illness as predictors of antidepressant drug adherence. Psychiatric services 52, 12 (2001), 1615{1620. [102] Sirey, J. A., Bruce, M. L., Alexopoulos, G. S., Perlick, D. A., Raue, P., Fried- man, S. J., and Meyers, B. S. Perceived stigma as a predictor of treatment discontinuation in young and older outpatients with depression. American Journal of Psychiatry 158, 3 (2001), 479{481. [103] Steindl, C., Jonas, E., Sittenthaler, S., Traut-Mattausch, E., and Greenberg, J. Understanding psychological reactance. Zeitschrift f ur Psychologie (2015). 117 [104] Stolzenburg, S., Freitag, S., Evans-Lacko, S., Muehlan, H., Schmidt, S., and Schomerus, G. The stigma of mental illness as a barrier to self labeling as having a mental illness. The Journal of nervous and mental disease 205, 12 (2017), 903{909. [105] Sweeny, K., Melnyk, D., Miller, W., and Shepperd, J. A. Information avoidance: Who, what, when, and why. Review of general psychology 14, 4 (2010), 340{353. [106] Taylor, S. E., and Brown, J. D. Illusion and well-being: a social psychological perspective on mental health. Psychological bulletin 103, 2 (1988), 193. [107] Thurston, I. B., Phares, V., Coates, E. E., and Bogart, L. M. Child problem recognition and help-seeking intentions among black and white parents. Journal of Clinical Child & Adolescent Psychology 44, 4 (2015), 604{615. [108] Vogel, D. L., Wade, N. G., and Haake, S. Measuring the self-stigma associated with seeking psychological help. Journal of counseling psychology 53, 3 (2006), 325. [109] Vogel, D. L., Wade, N. G., and Hackler, A. H. Perceived public stigma and the willing- ness to seek counseling: The mediating roles of self-stigma and attitudes toward counseling. Journal of Counseling Psychology 54, 1 (2007), 40. [110] Vogt, D., Di Leone, B. A., Wang, J. M., Sayer, N. A., Pineles, S. L., and Litz, B. T. Endorsed and anticipated stigma inventory (easi): A tool for assessing beliefs about mental illness and mental health treatment among military personnel and veterans. Psychological Services 11, 1 (2014), 105. [111] Wang, P. S., Aguilar-Gaxiola, S., Alonso, J., Angermeyer, M. C., Borges, G., Bruffaerts, R., Chatterji, S., Chiu, W. T., De Girolamo, G., Fayyad, J., et al. Delay and failure in treatment seeking after rst onset of mental disorders in the world mental health survey initiative. In The WHO World Mental Health Surveys: Global perspectives on the epidemiology of mental disorders. Cambridge University Press, 2008, pp. 522{533. [112] Wang, P. S., Berglund, P. A., Olfson, M., and Kessler, R. C. Delays in initial treatment contact after rst onset of a mental disorder. Health services research 39, 2 (2004), 393{416. 118 [113] Warren, L. W. Male intolerance of depression: A review with implications for psychother- apy. Clinical Psychology Review 3, 2 (1983), 147{156. [114] Wong, E. C., Collins, R. L., Cerully, J., Seelam, R., and Roth, B. Racial and ethnic dierences in mental illness stigma and discrimination among californians experiencing mental health challenges. Rand health quarterly 6, 2 (2017). [115] Worchel, S., and Brehm, J. W. Eect of threats to attitudinal freedom as a function of agreement with the communicator. Journal of Personality and Social Psychology 14, 1 (1970), 18. [116] Ybarra, M. L., and Eaton, W. W. Internet-based mental health interventions. Mental health services research 7, 2 (2005), 75{87. [117] Yuan, Q., Seow, E., Abdin, E., Chua, B. Y., Ong, H. L., Samari, E., Chong, S. A., and Subramaniam, M. Direct and moderating eects of personality on stigma towards mental illness. BMC psychiatry 18, 1 (2018), 358. 119

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

The power of social media narratives in raising mental health awareness for anti-stigma campaigns

PDF

Cultural influences on mental health stigma in Asian and European American college students

PDF

Essays on health economics

PDF

Essays in empirical health economics

PDF

Mindfulness and resilience: an investigation of the role of mindfulness in post-9/11 military veterans' mental health-related outcomes

PDF

From “soul calling” to calling a therapist: meeting the mental health needs of Hmong youth through the integration of spiritual healing, culturally responsive practice and technology

PDF

Essays in health economics and provider behavior

PDF

Exploring mental health care utilization and academic persistence among college students: evaluating racial and stigma-related disparities

PDF

Three essays on health economics

PDF

Three essays on health & aging

PDF

Essays on development and health economics

PDF

Three essays on economics of early life health in developing countries

PDF

Behavioral approaches to industrial organization

PDF

Essays in labor economics: demographic determinants of labor supply

PDF

Essays on health insurance programs and policies

PDF

Addressing the mental health care gap in American youth: an evaluation study of character education

PDF

Essays on macroeconomics of health and labor

PDF

Essays on development economics and adolescent behavior

PDF

Essays on development and health economics: social media and education policy

PDF

Essays on the empirics of risk and time preferences in Indonesia

Asset Metadata

Creator Kim, Daehyun (author)

Core Title Three essays on behavioral economics approaches to understanding the implications of mental health stigma

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Economics

Publication Date 08/07/2020

Defense Date 05/26/2020

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

Tag disappointment aversion,Family,information preference,Mental Health,mental health stigma,OAI-PMH Harvest

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Coricelli, Giorgio (committee chair), Doctor, Jason (committee member), Nugent, Jeff (committee member)

Creator Email kimdaehy@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-360908

Unique identifier UC11666302

Identifier etd-KimDaehyun-8896.pdf (filename),usctheses-c89-360908 (legacy record id)

Legacy Identifier etd-KimDaehyun-8896.pdf

Dmrecord 360908

Document Type Dissertation

Rights Kim, Daehyun

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA