Insurance and Solidarity

Transcription

1 Insurance and Solidarity First draft. Please do not circulate. Friederike Lenel, Susan Steiner February 15, 2016 Abstract Guided by the literature on fairness and accountability, we study the relationship between accountability and solidarity with a lab-in-the-field experiment in rural Cambodia. We investigate whether solidarity transfers are reduced when individuals can be held accountable for their misfortune. We designed a game which borrows from both the dictator game and the solidarity game. All subjects receive the same endowment. Half of the subjects can lose a large proportion of their endowment due to a random shock, the other half can decide to transfer part of their endowment to those who lost. We vary whether recipients have the option to purchase an insurance which covers the loss from the shock and whether recipients are aware of the transfer possibility. We investigate to what extent providers reduce their transfers when recipients can be held accountable for the outcome and how much of this reduction is driven by a response to the choice of the recipients per se and how much by a response to the recipients intentional reliance on the solidarity transfers. We find a significant reduction in transfers provided when the recipients can be held accountable for their loss: providers reduce their transfers by 30% when recipients forewent the option to insure. This reduction is mostly driven by a response to the recipients choice and less by a response to the recipients intention. This paper is an outcome of the research project Insurance and Private Transfers - Experimental Evidence funded by the German Research Foundation (DFG). We are indebted to Georg Weizsäcker for his decisive advice on the design of our game. We also thank Yves Breitmoser, Dirk Engelmann, Nathan Fiala and Krisztina Kis-Katos for their crucial feedback. Friederike Lenel would like to thank her doctoral colleagues Nico De Silva, Hannah Liepmann and Arne Thomas. Last but not least, we are thankful to the Social Health Protection Association, Tillmann Eymess, Prem Chap and the Cambodian team of research assistants for their valuable support in the field. DIW Berlin; Humboldt Universität Berlin Leibniz Universität Hannover 1

2 1 INTRODUCTION 2 1 Introduction Assume an individual in your neighborhood has just lost a considerable part of his belongings due to a large fire. You are asked for donations. How much will you donate? Will you donate less in case you learn that your neighbor had been offered an affordable and fair insurance some weeks before the fire; that this insurance would have covered the loss but the neighbor had declined to purchase it? What if you have reasons to assume that your neighbor had declined the insurance because she expected she would receive sufficient support from her neighbors in case of a loss; will that affect your donation decision? The relationship between accountability and solidarity forms the core of our paper. We investigate whether solidarity transfers are reduced when individuals can be held accountable for their misfortune. We analyze this question in the insurance context. This context is of scientific and political relevance. Over the last years, there have been considerable efforts, both from the private sector as well as from the donor community, to expand formal insurance into previously unattended markets in developing countries (for a documentation of the rapid expansion, see Churchill and McCord 2012). For lowincome people in these countries one major risk-management tool are informal support arrangements. Relatives, friends and neighbors assume insurance functions by providing monetary and in-kind transfers, shelter and labor assistance in times of need, as has been well-documented in a number of studies (Ligon, Thomas, and Worrall 2002; Fafchamps and Lund 2003; De Weerdt and Dercon 2006). The motives behind such support arrangements are separated into selfish, incentive related motives (such as reciprocity) and altruistic, social preference related motives (Cox and Fafchamps 2007; Ligon and Schechter 2012). The question emerges how these informal support arrangements change when formal insurance becomes available. Evidence is scarce and not clear-cut (Hintz 2010; Landmann, Vollan, and Frölich 2012; Mobarak and Rosenzweig 2013). It is relatively straightforward to expect that incentives to engage in these arrangements decline as the autarchy value increases, i.e., when insurance becomes available (for a theoretical and empirical analysis, see Lin, Liu, and Meng 2014); it is less clear how social preference related motives are affected. This is where this paper sets in. Guided by the literature on fairness and accountability (Roemer 1998; Konow 2003), we study the relationship between accountability and solidarity with a lab-in-thefield experiment conducted in Cambodia. We designed a game that borrows both from the dictator game and the solidarity game. Players are randomly assigned the role of provider or recipient. Each provider is anonymously matched with one recipient. Both receive the same endowment. The recipient can lose a large proportion of her endowment due to a random shock. The provider is asked how much of his endowment he would transfer in case the recipient lost. We vary whether the recipient has the option to purchase an insurance which covers the loss from the shock and whether the recipient is informed about the provider s support possibility. We investigate to what extent the provider reduces his transfers when the recipient can be held accountable for the outcome as she forewent the insurance option. Specifically, we analyze how much of this reduction can be attributed to a response to the choice per se of the recipient, in line with the accountability principle (Konow 1996), and how much to a response to the recipient s informed choice, i.e., his intentional reliance on the solidarity transfers, in line with intention-based reciprocity (e.g.,charness and Rabin 2002). We complement the results with detailed survey data on subjects characteristics and attitudes. We present a theoretical model on conditional solidarity based on Konow (2010) where a provider incurs an internal cost when deviating from what he perceives as the adequate transfer to a less well-off

3 1 INTRODUCTION 3 recipient. We allow this adequate transfer to depend not only on the recipient s endowment but also on the recipient s choice set and information (i.e., his decision calculus). We define different types of solidarity behavior based on the extent to which transfers are conditioned on the recipient s decision calculus. The anonymous, within-subject design of the experiment allows us to classify the subjects transfer behavior according to these types and to contrast the types with subjects characteristics and attitudes. The experiment and the survey were conducted in 21 villages in two neighboring provinces, Siem Reap and Banteay Meanchey, in Northwestern Cambodia from July to October The context of Cambodia is interesting as insurance for the low-income population (more specifically, health insurance) has recently been introduced, yet is still not available in all provinces. In one of our two provinces, Siem Reap, health insurance was introduced in the rural areas in In the other province, Banteay Meanchey, health insurance is not yet available. We first conducted a comprehensive household survey with close to 1,300 households in these villages. From the surveyed households, 672 respondents then took part in the experiment. Between the household survey and the experiment there was a time gap of two weeks in order to reduce confounding factors. Data collection was finalized at the end of October All findings presented in this version of the paper are preliminary. Data analysis is ongoing. On average, providers transfer 15% of their endowment to a recipient who had no option to insure. There is a significant reduction in transfers provided when the recipient can be held accountable for her loss: providers reduce their transfers on average by 30% when a recipient forewent the option to insure. This reduction is mostly driven by a response to the recipient s choice per se and less by a response to the perceived intention behind this choice. Interestingly there is a large proportion of providers (42%) that transfer the same amount, independent of the option the recipient had available. This proportion is comparable to the findings by Trhal and Radermacher (2009). Potentially, individual accountability receives more consideration with a higher valuation of autonomy. When we contrast the participants behavior in the experiment with their self-reported autonomy elicited in the survey, we find support for this proposition. Participants who see their lives to depend on themselves punish the foregoing of insurance significantly stronger in the experiment. Also, participants who reportedly rely less on the support of others punish the foregoing of insurance stronger. Surprisingly, the response to a recipient who decides informedly against the insurance (suggesting a willful reliance on the provider s support) is less clear-cut. The majority of the providers (80%) do not punish the recipient s willful reliance. In fact, 20% transfer even more (or, punish less), in case the recipient was informed about the support possibility when deciding against the insurance. We can only speculate about the reasoning behind this behavior. One possible explanation is that these providers wish to fulfill the expectations of a recipient who intentionally did not take up insurance specifically because she counted on the provider s transfer in case of loss, i.e., they wish to fulfill the expected transfer. Additional analyses will be done. The paper is structured as follows, in Section 2 we introduce the central concepts of this study, discuss the relevant literature and present a theoretical framework that guides our research question. Section 3 presents the design of the game that we have developed and how this game was implemented in the field. In Section 4 we present the empirical methodology and the data. The results are presented in Section 5. Section 6 concludes.

4 2 RELEVANT LITERATURE AND UNDERLYING THEORETICAL FRAMEWORK 4 2 Relevant Literature and Underlying Theoretical Framework 2.1 Solidarity and Accountability We use the concept of solidarity as defined by Selten and Ockenfels. Solidarity means a willingness to help people in need who are similar to oneself but victims of outside influences such as unforeseen illness, natural catastrophes, etc. (Selten and Ockenfels 1998, p. 518). As stressed by Selten and Ockenfels, solidarity does not involve direct reciprocity, in the sense that every transfer given is (expected to be) returned at some point in time. 1 Solidarity transfers reflect a concern for people in need and are thus driven by altruistic preferences. Altruism is here understood in a broad sense as a motivation to help others when it does not result in material benefits to oneself and may come at a material cost. Following Fong (2007), we distinguish between unconditional and conditional altruism. Unconditional altruism is a preference to help others independent of the characteristics, choices or intentions of the person in need; at most, the material outcome of the recipient is taken into account. 2 Under conditional altruism, on the other hand, help is conditioned on the characteristics, choices, or intentions of the person in need - for example, their ethnicity (Habyarimana et al. 2007), the relationship between them and the helping person (Leider et al. 2009), or their effort (Cappelen, Sørensen, and Tungodden 2010; Jakiela 2015) and choices (Cappelen et al. 2013a; Mollerstrom, Reme, and Sørensen 2015). 3 This takes us to the second main concept of our study, the concept of accountability. It is summarized in Konow s formulation of the accountability principle which requires that a person s fair allocation (e.g., of income) vary in proportion to the relevant variables that he can influence (e.g., work effort) but not according to those that he cannot reasonably influence (e.g., a physical handicap) (Konow 1996, p. 14). People who act according to the accountability principle thus condition their help to a needy person on the extent to which this person can be held accountable for her neediness. Closely related is Dworkin s (1981) distinction between brute luck and option luck. Brute luck stems from random processes and cannot be influenced; option luck stems from choice or deliberate gambles (Dworkin 1981, p. 293). Whereas a person that becomes needy due to bad option luck can be held accountable for her neediness; this is not the case for bad brute luck. 2.2 Relevant literature Empirical evidence, typically from dictator and solidarity games, shows that individuals tend to perceive inequalities that are the result of bad brute luck as more unfair than inequalities that are the result of effort or deliberate choice. They are therefore more in favour of redistribution in the first case than in the latter. For example, Fong (2007) plays dictator games in which dictators are university students and recipients are real-life welfare recipients. Dictators receive information about the welfare recipients, according to which they are similar in terms of socio-demographic characteristics but differ in the degree to which they are willing to participate in the labour market. This difference in 1 The authors emphasize that solidarity touches on a more subtle form of reciprocity, in the sense that the provider expects similar behavior by the recipient in case the provider was in need. 2 Note that in the case of warm glow, which is part of unconditional altruism, the material outcome is not relevant. Warm glow describes the utility gained from helping others through the mere feeling of having done good unto them. 3 Konow (2010) also distinguishes between unconditional and conditional altruism but, different from Fong (2007), he includes adherence to social norms in conditional altruism. For Fong (2007), this is not necessarily the case. For example, people who prefer an egalitarian distribution behave in her case in line with unconditional altruism, because an egalitarian distribution is entirely independent of the recipient s traits. For Konow (2010), the same preference would be included in conditional altruism because the provider here behaves in line with an egalitarian norm. Our definition of unconditional and conditional altruism follows that of Fong (2007).

5 2 RELEVANT LITERATURE AND UNDERLYING THEORETICAL FRAMEWORK 5 the willingness significantly affects the transfers provided by the dictators to the recipients. Subjects provided, on average, 34% less to a recipient who was reportedly less willing to work than to a more industrious recipient. The proportion of subjects who transferred nothing was nearly twice as high (43% vs. 23%) for the less willing than for the industrious. Trhal and Radermacher (2009) conduct solidarity games, in which the endowment of their subjects is either determined randomly or by deliberate choice between a safe amount and a risky lottery. Subjects are willing to transfer significantly more to others when they become needy due to bad brute luck than when they become needy because they chose the risky lottery and lost. Among those subjects, who do not behave purely selfish and never transfer anything, 39% do not take the choice of the needy into account, while 61% do. Experimental evidence from other studies confirm that a substantial proportion of people condition their support on the extent recipients are accountable for their outcome (Cappelen et al. 2013a; Bolle and Costard 2015). Related studies analyze accountability in the context of effort, typically with the help of dictator games with a production stage in which subjects need to earn their endowment in a so-called real effort task. These studies show that a large proportion of people (both, in the role of dictators and in the role of spectators) condition their redistribution decisions on the effort that recipients put into earning their endowment (Cherry 2001; Cherry, Frykblom, and Shogren 2002; List 2007; Jakiela 2015). A study that is close to ours is Mollerstrom et al. (2015), which relies on Dworkin s distinction between brute luck and option luck. The authors implement dictator games in which they provide their subjects with an endowment and inform them that one of three scenarios will be drawn with an equal probability. In scenario A, they could keep the endowment; in scenarios B and C, they will lose the endowment (brute luck). Before the scenario is determined, subjects are given the option to purchase an insurance (option luck) which covers the loss if scenario B were drawn, but does not cover the loss if scenario C were drawn. After making their decision, subjects are teamed up. For each team that ends up in two different scenarios (and, hence, with different individual incomes), another subject decides how to distribute the total income between its members. The authors find that spectators were significantly more likely to equalize income if the outcome was a consequence of brute luck compared with a consequence of option luck (64% vs. 45%). They estimate that 34% of the subjects in their sample followed a fairness view that conditioned on the available choices. 4 Our study makes two important contributions to the literature. First, we investigate the role of informed choice versus choice per se. In particular, we analyze whether people hold others differently accountable for their choices depending on whether or not those others knew about the possibility of redistribution at the time of making their choices. In the above mentioned studies, this is impossible to analyze because subjects are either informed about the later redistribution stage before they make their choices (Thral and Rademacher, 2009; Cappelen et al., 2013; Bolle and Costard, 2015) or not (Fong, 2007; Mollerstrom et al., 2015). 5 Hence, it remains an open question whether people s response to choice depends on the level of information available to recipients when making their choice. We assume that this may be the case. When recipients are not informed about potential redistribution, they can only be held accountable for their choice per se. When recipients are informed about potential redistribution, however, they can be held accountable for the choice and for their deliberate reliance on 4 Interestingly, subjects are held accountable for not buying insurance even when scenario C was drawn. In other words, subjects hold others accountable for a particular choice even if this choice has no impact on the outcome. 5 To be precise, Fong (2007) is a special case because the real-life welfare recipients did not know about the dictators when deciding about their labour market participation. However, they may have had different levels of information about welfare programs, which may have influenced their willingness to work. Dictators in the experiment do not know about this.

6 2 RELEVANT LITERATURE AND UNDERLYING THEORETICAL FRAMEWORK 6 the support by others. It does not matter whether recipients indeed deliberately relied on this support; it only matters that the person deciding about the redistribution perceives this to be the case. This is in line with models on intention-based reciprocity (see Charness and Rabin, 2002; Dufwenberg and Kirchsteiger, 2004; Cox et al; 2008), which, in a nutshell, describe that people tend to reward behavior they perceive as kind and to punish behavior they perceive as unkind. Second, we conduct our experiment with villagers in a developing country. We are not aware of any studies that investigate in the development context the prevalence of the accountability principle. The experiments in Fong (2007), Thral and Radermacher (2009), Cappelen et al. (2013), Bolle and Costard (2015), and Mollerstrom et al. (2015) are all conducted with students in university labs of developed countries. 6 Yet, it appears reasonable to expect that preferences differ systematically between the population of developing countries and students of universities in such countries as Germany, Norway, and the US, due to historical and cultural legacy (Guiso, Sapienza, and Zingales 2006; Henrich et al. 2010). In the above mentioned dictator games with a production stage, university students typically condition their redistribution decisions on the effort of the recipient. In contrast, Jakiela (2015) finds that villagers in Kenya make no difference between earned and unearned income. She argues that this is due to the persistence of egalitarian norms in poor communities, where luck, rather than individual effort, is regarded to be the reason for success. Similarly, Cappelen et al. (2013b) show that students in Tanzania and Uganda are less responsive to individual players contributions to total income than students in Germany and Norway. These findings suggest that there may be different fairness norms and, consequently, preferences for redistribution across cultures and levels of economic development. 2.3 Theoretical Framework To define the concepts that guide our analysis we use a simple theoretical framework that we base on Konow s model of conditional altruism (Konow, 2010). We analyze the transfer decisions y of the provider i with the endowment x e i to the recipient j with the endowment x j, with x e i > x j 0. The provider s utility can be described as follows U( ) = v(x e i y) f(y φ) (1) v( ) is the provider s utility from her final material payoff. φ( ) captures what a provider believes to be the adequate transfer to j; we call it his personal transfer norm. We allow providers to be heterogeneous in their personal transfer norm φ( ), i.e. providers can differ in the amount of transfer they perceive as adequate for a specific situation. We call a provider i s preferences to be in line with unconditional altruism, if i s personal transfer norm takes into account at most his own and the recipient s endowment and no other factors concerning j. 7 We call a provider i s preferences to be in line with conditional altruism, if i s personal transfer norm takes into account other factors besides his own and the recipient s endowment (such as effort, choice or personal traits). 8 We are interested in the extent to which providers condition their solidarity transfers on a recipient s 6 The recipients in Fong (2007) are indeed welfare recipients in real life. However, the transfer decisions are made by students in a lab-like setting. 7 Note that this definition includes egalitarian fairness norms that only take into account differences in endowments as well as norms following the basic need principle; in the latter case, φ = max[( n x j ); 0], where n is a fixed amount that is perceived to cover the basic needs. 8 However, as noted above, Konow (2010) uses a slightly different definition of conditionality. For Konow adherence with any social norm already characterizes conditional altruism; i.e. for him f(y φ) represents the conditionality while for us f(y φ) only represents conditionality if φ includes other factors concerning j besides his endowment.

7 2 RELEVANT LITERATURE AND UNDERLYING THEORETICAL FRAMEWORK 7 choice to avoid his loss. We look at the following two situations: 9 1. Ex-ante, i and j have the same endowment (x e i = xe j ), but j faces a positive probability of a shock which would result in a considerable loss of his endowment. If j is hit by the shock, j has an endowment of x j < x e i and i will be asked for support. 2. Ex-ante, i and j have the same endowment (x e i = xe j ), but j faces a positive probability of a shock which would result in a considerable loss of his endowment. Before the shock occurs j has the option to purchase insurance which would cover the loss resulting from the shock. If j purchases insurance, he will not need the support of i. If j decides not to purchase insurance and is hit by a shock, j has an endowment of x j < x e i and i will be asked for support.10 Let O j be an indicator for whether the recipient had ex-ante the option to avoid the loss. If O j = 0 the recipient had no option to avoid the loss; or, to speak in Dworkin (1981) terms, the loss is a result of brute luck. If O j = 1 the recipient had the option to avoid the loss; the loss is a result of option luck. We then define i s preference in line with choice conditional altruism if c.p. φ i (x e i, x j O j = 1; F j ) φ i (x e i, x j O j = 0; F j ). Fj is a wildcard for other factors concerning j which remain fixed. We further specify the concept of choice conditional altruism by incorporating the level of information of the respondent j. Let I j be an indicator for whether the recipient had ex-ante the information about i s potential support in case of loss. I j = 1 if the recipient is informed about the support possibility. I j = 0 otherwise. In a situation where the recipient j had ex-ante no information about the support possibility by i (i.e., I j = 0), let δi NI be the change in i s personal transfer norm φ i for the case j had the option of insurance vs. the case where j had no option of insurance; that is δ NI i = φ i (x e i, x j I j = 0; O j = 1; F j ) φ i (x e i, x j I j = 0; O j = 0; F j ) In a situation where the recipient j had ex-ante the information about the support possibility by i (i.e., I j = 1), let δi I be the change in i s personal transfer norm φ i for the case j had the option of insurance vs. the case where j had no option of insurance; that is δ I i = φ i (x e i, x j I j = 1; O j = 1; F j ) φ i (x e i, x j I j = 1; O j = 0; F j ) Hence, δi NI describes the provider i s response to the choice per se, the pure choice, and δi I describes i s response to an informed choice. We define i s preference to reflect pure choice conditional altruism if δi NI preference reflects informed choice conditional altruism if δi I δni i We focus on three types of preferences and the resulting solidarity transfers: 0. Furthermore, i s Unconditional altruism: the transfer that is perceived as adequate is independent of the option and the information the recipient had available ex-ante; i.e., δ NI i = δ I i = 0. A provider i s 9 Note that the preferences are specified by the personal transfer norm φ while the actual solidarity behavior is characterized by the transfers that maximize each individual s utility specified in (1). 10 That is, i will only be asked for support in case j did not purchase the insurance.

8 3 EXPERIMENTAL DESIGN 8 transfer behavior can be characterized by unconditional solidarity if i transfers the same amount independent of the option and the level of information of the recipient j. Pure choice conditional altruism: the transfer that is perceived as adequate depends on whether the recipient had the choice to avoid the loss, i.e., δi NI 0. A provider i s transfer behavior can be characterized by pure choice conditional solidarity if i transfers a different amount to an uninformed recipient j, when j had the option to avoid the loss. Informed choice conditional altruism: the transfer that is perceived as adequate depends on the informedness of the recipient when he made the choice; δi I δi NI. A provider i s transfer behavior can be characterized by informed choice conditional solidarity if i transfers a different amount to the recipient j when j made an informed choice vs. when he made an uninformed choice. Notably, preferences in line with pure choice conditional altruism and preferences in line with informed choice conditional altruism are not mutually exclusive, a provider might condition the transfer he perceives as adequate both on the choice per se as well as on whether this choice was made informed. We analyze the extent to which choice conditional altruism can explain solidarity transfer behavior. More specifically, we analyze to what extent preferences in line with pure choice conditional altruism and preferences in line with informed choice conditional altruism can describe transfers in a solidarity context. Based on the literature (Cappelen et al. 2013a; Mollerstrom, Reme, and Sørensen 2015), we expect that a large proportion of providers transfer less when the recipient had the choice to avoid the loss (choice conditional solidarity). We expect that this is mainly driven by a response to the choice of the recipient (pure choice conditional) but that a considerable proportion of providers also responds to the informedness of the recipient when making this choice (informed choice conditional) Experimental Design 3.1 The basic design A T 1 T 2 No Option B1 B2 Option to insure Figure 1: Basic Experiment Design In order to analyze the existence of choice conditional solidarity, we developed a game that is a mixture of the dictator game and the solidarity game. The basic design of the game is depicted in Figure Notably, providers might respond to both conditions: a provider might lower the transfer he perceives as adequate in case the recipient had the choice to avoid the loss, and might lower it even more if this choice was made informed; i.e., δ I i < δni i < 0.

9 3 EXPERIMENTAL DESIGN 9 There are three different player types, player A (provider) and players B1 and B2 (recipients). Each player is equipped with the same endowment Y. Players B1 and B2 can lose part of their endowment (L) with probability π (a random shock). Player B2 has the option to insure himself against the loss. More specifically, player B2 is offered an insurance for price P which covers all loss. If player B2 purchases the insurance, he pays the price P and always keeps (Y P ) independent of whether a shock occurs or not. Player B1 does not have this option. Player A keeps his endowment. He is informed about the situation of B1 and B2. Player A is asked to make two strategic transfer decisions: 1. The amount he would transfer to player B1 in case this player lost [T 1 ]. 2. The amount he would transfer to player B2 in case this player lost [T 2 ]. He is then randomly matched either with player B1 or with player B2. In case the matched player loses, the respective transfer decision is implemented. The transfer decisions T 1 and T 2 of player A form the core of the analysis. They reflect the provider s personal transfer norm for each situation of B1 and B2. T 1 can also be interpreted as an indication for the intensity of compassion that a provider feels towards the recipient. 12 The difference in the transfers (T 2 T 1 ) then describes how this feeling changes when the recipient had the choice to avoid the loss. If a provider feels less compassion with a recipient who forewent the option to insure himself, we should expect T 2 T 1 < 0. In this case the provider behaves in line with choice conditional solidarity. On the other side, if a provider only takes into account the eventual outcome and not the option the recipient had available, we should expect T 2 T 1 = 0, the provider shows unconditional solidarity. Case (1) Case (2) A1 A2 T 1 T 2 T 1 T 2 B1 B2 C1 C2 Informed Informed Uninformed Uninformed No Option Option to insure No Option Option to insure Figure 2: Extension: Varying information on support As an extension we vary whether a recipient is ex-ante informed about the provider. This extension is depicted in Figure 2. There are now two different cases. In Case (1), the recipients B1 and B2 are informed ex-ante about the provider (player A1) and the support possibility (informed recipients). Likewise, player A1 is informed that the recipients are aware of the prospect of transfers. Player A1 makes the strategic transfer decisions T 1 and T 2, for recipient B1 and recipient B2, respectively. In Case (2), the recipients C1 and C2 are not informed about the provider (player A2) and the support possibility (uninformed recipients). Player A2, in turn, is informed that the recipients do not know 12 Similarly, it can also be interpreted as an intensity of the feeling of guilt (as the higher endowment of the provider was a result of pure chance).

10 3 EXPERIMENTAL DESIGN 10 about the support possibility. Player A2 makes the strategic transfer decisions T 1 and T 2, for recipient C1 and recipient C2, respectively. 13 This extension allows us to analyze whether providers react differently when the recipient made the choice to forego insurance uninformed (i.e., not aware of the support possibility in case of loss) compared to when the recipient made the choice informed. That is we can analyze the prevalence of pure choice conditional solidarity and informed choice conditional solidarity. 14 If we believe providers disapprove the choice of foregoing insurance per se, we should expect that player A2 transfers less to C2 than to C1, i.e., (T 2 T 1) < 0; or in other words A2 shows pure choice conditional solidarity. If we believe providers disapprove foregoing insurance more when the choice was made informed (e.g. because providers perceives this as a free-riding on their solidarity) we should expect player A1 to reduce his transfer to a larger extent than player A2, i.e., (T 2 T 1 ) < (T 2 T 1); this would be an indication for informed choice conditional solidarity. If we believe providers to be indifferent on whether the recipient made his choice informed or not, we should expect (T 2 T 1 ) = (T 2 T 1). 3.2 Experimental procedure Table 1: Experiment Overview Subject Group Round 1 Role A1 A2 B1 B2 C1 C2 Room No. of subjects per session Transfer Decisions B1 + B2 C1 + C Insurance Option - - no yes no yes Information - - yes yes no no Round 2 Role A2 A1 B2 B1 C2 C1 Room No. of subjects per session Transfer Decisions C1 + C2 B1 + B Insurance Option - - yes no yes no Information - - yes yes no no Total No. of Subjects In the following, we describe the implementation of the game in the field. In each session, the experiment was conducted with 32 subjects: 16 providers and 16 recipients. All subjects played two rounds.before the game, subjects were randomly allocated to one of six groups; the group determined the role each subject would play in Round 1 and Round 2 (see Table 1). There were two groups (à 8 subjects) of providers, Group 1 and Group 2; and four groups (à 4 subjects) of recipients, Group 3 - Group 6. Subjects in Group 1 played the role of player A1 in the first round and the role of player A2 in the second round. Subjects in Group 2 played the role of player A2 in the first 13 The outcomes for the players in Case (1) and Case (2) are depicted in Figure 5 in the Appendix A Note that the design does not allow to fully disentangle the response to the choice per se from the response to the informedness only. With the difference in information also the expected loss and thus the assumed risk differs between B2 and C2 players. If players expect positive transfers, the expected loss without insurance of B2 players is smaller than the expected loss without insurance of C2 players, i.e., the risk to remain with (Y L) that B2 players assume when foregoing insurance is smaller than the risk that C2 players assume. That means, strictly speaking, also the type of choice differs between recipient B2 and C2

11 3 EXPERIMENTAL DESIGN 11 round and the role of player A1 in the second round. Subjects in Group 3, Group 4, Group 5 and Group 6 played in the first round the role of player B1, B2, C1 and C2, respectively, and in the second round the role of player B2, B1, C2 and C1, respectively. Each group played in a separate room (i.e., there were six rooms). No communication nor interaction between the subjects within a room and between the rooms was allowed. Subjects were at no time told the purpose of the experiment. The experiment was implemented in a semi-doubleblind setting: subjects did not know the identity of their partners; 15 and the research assistants supervising the games did not observe the subjects decisions; 16 those research assistants, who recorded the decisions and could link them to the identity of each subject, did not interact with the subjects until the final payout. This setting was explained in the introduction. The procedure for type B1, B2, C1 and C2 was as follows: 1. Each subject received the endowment: 16,000 Riel in sixteen 1,000 Riel bills in play money Subjects were explained the dice game: all subjects would roll the dice. The outcome would determine how much they could keep of the endowment; if the dice showed 1, 2 or 3 they would lose 14,000 Riel, if the dice showed 4, 5 or 6, they would keep the 16,000 Riel. 3. Subjects of type B2 and C2 were explained the insurance option: they had the option to privately purchase an insurance for the price of 6,000 Riel; if they decided to purchase the insurance, they would keep 10,000 Riel independent of the outcome of the dice Subjects of type B1 and B2 were informed that each of them were matched with a player in a different room, that these players had a safe endowment of 16,000 Riel, but could decide to transfer part of it to their partner (i.e., to the B1 and B2 players) in case this person lost; subjects were asked to note down how much transfer they expected from this partner player in case of loss All subjects were asked questions to test their understanding of the game. 6. Subjects of type B2 and C2 were asked to go outside the room one by one to make their insurance purchase decision with a research assistant sitting outside. Subjects were not allowed to reveal their decision to the other subjects in the room. 7. Each subject rolled the dice. The outcome was noted down. In case subjects lost, they handed in 14,000 Riel of their play money to the research assistant. The remaining money was inserted in an envelope and collected; subjects were told that this money would be transferred to their personal game accounts. This money together with any potential transfers of the provider determined the payout of the recipients for this round in case this round was selected for payout. 15 Subjects saw each other during introduction before the game, but they did not know who played which role except for those who were in the same room. Thus, if providers wished to form expectations about the identity of the recipient, they had to take into account the pool of all 24 subjects who were not in their group. 16 Exceptions had to be made for those participants who could not write numbers. We account for this in the analysis. 17 4,000 Riel are worth approximately 1US$. As a benchmark: the average daily household income of the participants in our experiment is 6US$ (including remittances, state assistance etc.), the median daily household income is just about 2US$. The information is based on the data from our household survey. Nevertheless, as typical in this context, caution is advised when using income measures. 18 For the insurance option, we intentionally did not use the Khmer word for insurance but the more general word bankapie ( guarantee ) in order to not evoke any associations with existent insurance schemes. 19 The beliefs were noted down in private behind a cardboard and then collected. Subjects were ensured that their partner would not see these beliefs and that they had no impact on the transfer decisions.

12 3 EXPERIMENTAL DESIGN 12 Figure 3: Illustrations for A1 and A2 players, respectively (player roles not in the original) The procedure for type A1 and A2 was as follows: Each subject received the endowment: 16,000 Riel in sixteen 1,000 Riel bills in play money. 2. Subjects were explained the situation of the recipients. Specifically, subjects of type A1 were explained the situation of B1 and B2 players, and subjects of type A2 were explained the situation of C1 and C2 players. Subjects were shown one of the overview illustrations depicted in Figure 3 as well as a detailed illustration for each player type (see Figure 7 and 8 in the Appendix B.5). 3. Subjects simulated the situations of the recipients in the two rooms, first of type B1 [C1], then of type B2 [C2]. During this simulation, subjects were asked test questions to test their understanding of the game Subjects were explained the random partner matching and the following transfer procedure. 22 Again, all subjects were asked questions to test their level of understanding of the transfer procedure. 5. All subjects were asked to write down in private (using cardboards) on two separate sheets the following transfer decisions (see Figure 9 and 10 in the Appendix B.6): 23 In case your partner was of type B1 [C1] - how much of your 16,000 Riel would you transfer if your partner lost? In case your partner was of type B2 [C2] - how much of your 16,000 Riel would you transfer if your partner lost? 20 For the script of the instructions for players A1 and A2 see in the Appendix B.2 and B.3 21 For each room the same set of test questions was asked in each session. For the instructions of the test questions see in the Appendix B.4. Research assistants recorded the correctness of the response on a four-point scale. The results of the test questions can be used as an indicator for the level of understanding. 22 It was emphasized that transfers would only take place in case the partner had lost in the dice game and, for a partner of type B2 and C2, had not bought insurance. 23 Instead of the types, in the experiment two Khmer letters were used, letter kâ for type B1 and C1, and letter khâ for type B2 and C2.

13 4 EMPIRICS After decisions were noted down, subjects had time to check both decisions and to make final changes; then, pencils were collected. 7. Each subject was asked to draw an envelope from a box. On the envelope was a sign indicating the player type of the partner and a unique ID for the partner. 24 The irrelevant decision sheet (i.e., the decision for the other player type) was collected. 8. Subjects were asked to insert in the envelope the remaining, relevant decision sheet and the amount of bills, they had noted on the sheet Subjects were provided a second envelope where they inserted the remaining amount of bills. They were told that this money would be transferred to their personal game accounts and that in case their partner had not lost they would also receive back the amount they had transferred. 10. All decision sheets and envelopes were collected by the research assistants. 26 After Round 1, subjects switched rooms and player types for Round 2; the research assistants remained in the rooms. The procedure of Round 2 was the same as the procedure of Round 1, only the simulation of the partners situation and the related test questions were skipped. Over the course of Round 1 and Round 2 subjects who played the provider role made in total four transfer decisions, two transfer decisions as A1 player, i.e., for a partner who was aware of their potential support, and two transfer decisions as A2 player, i.e., for a partner who was unaware of their potential support. 4 Empirics 4.1 Empirical Methodology For our research question the transfer decisions of the providers (the A players) form the core of our analysis. We have four transfer decisions for each subject i who plays the role of the provider. We can split-up these four transfer decisions as depicted in Table 2. Table 2: Treatments and Transfers of Provider i Recipient informed of provider no yes Recipient had no θ i θ i + β i option to insure yes θ i + γ i θ i + β i +γ i + η i We model the transfer decision trnsf of subject i in treatment t as follows: trnsf i,t = θ + βinf t + γopt t + ηinfopt t + ɛ i,t (2) 24 Based on this ID, we identified the partner of each subject. However, subjects could not make any inference from the ID on the identity of their partner. 25 Subjects were told that the amount they inserted would be double-checked with the amount indicated on the decision sheet and that in case there was a difference, the amount indicated on the decision sheet would determine the transfer. 26 It was ensured that neither the decisions nor the amount of bills in the envelopes could be observed by the assistants. However, a number of participants were not able to write and needed assistance. In case assistance was needed, this information was noted down. Details are discussed in Section 4.2.

14 4 EMPIRICS 14 Inf, Opt and InfOpt describe the specific features of each treatment. Inf is equal to one for transfer decisions to a recipient who is informed about the potential transfers, and zero otherwise. Opt is equal to one for transfer decisions to a recipient who had the option to take up insurance, and zero otherwise. InfOpt is equal to one for transfer decisions to a recipient who is informed and had the option to take up insurance, and zero otherwise. θ describes the average so-called baseline solidarity, i.e. how much is provided to a recipient who had no option of insurance purchase and is not informed about the support possibility. β describes the average change in the baseline solidarity when the recipient is informed about the support possibility. γ describes the average change in transfers towards an uninformed recipient when he had the option to take up insurance, i.e. the extent providers condition their transfers on the recipient s choice per se. γ thus captures pure-choice conditional solidarity. γ + η describes the average change in transfers towards an informed recipient when he had the option to take up insurance, i.e. the extent providers condition their transfers on the choice the recipients had when this choice was made under the knowledge of potential subsequent transfers. η thus captures informed-choice conditional altruism. Of importance for our research interest are γ and η. We expect the following: γ < 0 - On average, there is a reduction in transfers to a recipient who decided against the option of insurance compared to a recipient who had no option of insurance η < 0 - On average, there is a reduction in transfers to an informed recipient who chose against the option of insurance compared to an uninformed recipient who chose against the option of insurance. We test our hypotheses, estimating Equation (2) using a GLS fixed effect model. Furthermore, as discussed in Section 2.3 we assume heterogeneity in transfer types which cannot be observed when analyzing average transfers. Therefore we analyze the distribution explicitly. In particular, we examine the distribution of pure choice conditional and informed choice conditional solidarity. In line with the literature, we expect the majority of providers to behave in line with pure choice conditional solidarity and a considerable proportion of providers to behave in line with informed choice conditional solidarity. 4.2 Data Implementation of the Experiment in the Field The experiment was conducted in 21 villages in Northwestern Cambodia between August and October The villages are located along the river Stong-Sreng which separates the two provinces Banteay Meanchey and Siem Reap (see Figure 4) The experiment was conducted in eleven villages in Siem Reap and in ten villages in Banteay Meanchey. The location is particularly interesting for the research project, because health insurance was introduced in Siem Reap province in 2012 while it was planned to be introduced in the province of Banteay Meanchey in late 2015 (it had not yet been introduced when data collection was completed). Villages were selected to be comparable within and across the two provinces. Selection criteria included the size of the village, the level of migration and remoteness.

15 4 EMPIRICS 15 Figure 4: Map of the provinces Banteay Meanchey and Siem Reap; Ministry of Health, 2008 Two weeks before the experiment took place in a village, a detailed household survey was conducted with 60 randomly selected households of the village as well as a community survey with the village head.in total, 1,272 households were interviewed. The survey focused on support networks within and outside the village, work migration, access to formal risk management tools, such as insurance, savings and credit, and attitudes on the value of autonomy. At the end of each survey, the respondent was asked whether he or she was able and willing to participate in an upcoming workshop on financial decision making. 28 If the respondent answered affirmatively he or she was included in the pool of potential experiment participants for this particular village. Our original target participant was literate and between 18 and 65 years old. However, as the literacy rate in this region is very low and migration of the young in some areas particularly high, illiterate and older respondents had to be included. We sorted the list of potential participants according to their literacy level and age and sampled from this list in the resulting order. 29 Two days before the experiment (around twelve days after the survey was conducted), research assistants handed out personal invitations for the experiment to 32 respondents from the pool. In case a respondent indicated not to be able to participate or could not be reached, he was replaced by a reserve respondent 28 In particular, respondents were told the workshop would be conducted in the local school in two weeks time, that it would last 4 hours and that they could earn some money. In each village, on average, 75% of the survey respondents had time and were willing to participate. 29 Eventually, 16 of the 672 experiment participants were above 65, and 230 could not read and write Khmer properly; however, of those the majority could write numbers which was necessary for the experiment; the participants who could not write numbers received help from the research assistants during the experiment.