Organ Donation Loopholes Undermine Warm Glow Giving: An Experiment Motivated By Priority Loopholes in Israel

Transcription

1 Organ Donation Loopholes Undermine Warm Glow Giving: An Experiment Motivated By Priority Loopholes in Israel By Judd B. Kessler and Alvin E. Roth This Draft: February 28, 2013 ABSTRACT Giving registered organ donors priority on organ waiting lists, as has been implemented in Israel and Singapore, provides an incentive for registration and has the potential to increase the pool of deceased donor organs. However, the implementation of a priority rule might allow for loopholes as is the case in Israel in which an individual can register to receive priority but avoid ever being in a position to donate his own organs. We experimentally investigate how such a loophole affects donation and find that the majority of subjects use the loophole when available. The existence of a loophole completely eliminates the increase in donation generated by the priority rule. When information about loophole use is made public, subjects respond to others use of the loophole by withholding donation, and the priority system with a loophole generates fewer donations than an allocation system without priority. Keywords: Organ Donation; Experiments; Public Goods; Warm Glow The authors thank the staff at the Wharton Behavioral Lab at the University of Pennsylvania. Both authors materially participated in the research and article preparation, and the authors have no conflict of interests related to this paper. Business Economics and Public Policy Department, The Wharton School, University of Pennsylvania, 3620 Locust Walk, Philadelphia, PA 19104; (215) ; judd.kessler@wharton.upenn.edu (corresponding author) Department of Economics, Stanford University, Stanford, CA 94305; alroth@stanford.edu 1

2 decade. 2 Under the current U.S. organ allocation system, which is similar to the systems in I. Introduction There are currently over 117,000 people on waiting lists for life-saving organ transplants in the United States. These individuals are waiting for an organ from a deceased donor, an individual whose organs are donated upon death. Deceased donors provide the majority of transplanted organs in United States (nearly 80% of organs in 2012) in part because deceased donors can provide multiple vital organs and other tissues whereas living donors overwhelmingly donate one kidney. 1 Even though one deceased donor can save the lives of up to eight people (and improve the lives of many more) and registering as an organ donor is relatively easy (it usually just requires checking a box on a form at the state department of motor vehicles) only 43% of Americans over the age of 18 are registered as organ donors (Donate Life America 2012). Due to federal legislation, there is no monetary incentive for organ donor registration (see Roth, 1997) and donors are motivated only by altruism (Becker 1974) and a warm glow from registering (Andreoni 1988, 1989, 1990). The only benefit a donor receives comes from the knowledge that one donor has the potential to save the lives of many people. This motivation alone has not generated enough deceased donation to halt the steady increase in the number of people on organ donor waiting lists. Table 1 lists the number people on the waiting list for a kidney, which has been growing over the past most other nations, priority on organ donor waiting lists is given to those who have been waiting the longest or those with the most immediate medical need. 3 One strategy to 1 Based on OPTN data accessed Feb. 25, 2013 ( A deceased donor can provide kidneys, liver, heart, pancreas, lungs, and intestine, as well as corneas, skin, heart valves, cartilage, bone, tendons, and ligaments 2 The kidney waiting list currently stands above 95,000, based on OPTN data accessed Feb. 25, 2013 ( The long waiting list for kidneys results in part from the ability for kidney dialysis to keep patients with kidney failure alive for many years. No dialysis exists for other organs. Waiting lists for other organs are shorter in part because many patients on those lists die while waiting. 3 The allocation rules vary by organ. In the United States, kidney allocation is primarily by waiting time while liver allocation is primarily by medical need. These policies are a function of feasible medical care: while kidney dialysis allows patients to survive for years without a kidney transplant, a patient whose liver fails will die very quickly without a new liver for transplant. 2

3 generate an incentive for organ donor registration without monetary payment is to allocate organs differently: to provide priority on organ donor waiting lists to those who previously registered as donors. Under such a priority system, individuals who register as organ donors but end up needing an organ are more likely to get an organ, or get one more quickly, than individuals who did not previously register as donors. This policy has been studied experimentally (Kessler and Roth 2012) and has been implemented in Singapore and, most recently, in Israel. In Israel, the policy appears to have increased the number of deceased donor organs and the organ donor registration rate, at least temporarily (Lavee et al. 2013), although the research into its effectiveness is ongoing. Table 1: U.S. Kidney Donors, Transplants, and Waiting List Deceased Donor Transplants New Wait List Additions Deceased Donors Living Donors All Wait List Patients ,638 8,539 6,241 50,301 23, ,753 8,668 6,473 53,530 24, ,325 9,359 6,647 57,168 27, ,700 9,913 6,573 61,562 29, ,178 10,660 6,436 66,352 32, ,240 10,591 6,043 71,862 32, ,188 10,553 5,968 76,089 32, ,248 10,442 6,387 79,397 33, ,241 10,622 6,277 83,919 34, ,433 11,043 5,770 86,547 33,568 The data for years are provided by OPTN as of February 22, New Wait-list Additions counts patients (rather than registrants) to eliminate the problems of counting multiple times people who register in multiple centers. All Wait-list Patients also counts patients rather than registrants. All Wait-list Patients data are from the 2008, 2009, and 2011 OPTN/SRTR Annual Reports. One concern with providing priority on the organ donor waiting list for registered donors is the possibility of gaming or loopholes that would allow an individual to receive priority if he needs an organ but avoid ever being in a position to donate his own organs. For example, one way an organ allocation system could be gamed is if it allowed individuals who registered as donors to immediately receive priority, letting them to wait until they needed an organ to register as a donor and effectively giving them priority without requiring anything in return. Careful implementation of allocation rules can 3

4 eliminate this scope for gaming. In Israel, for example, an individual who was not registered by April 1, 2012 must be on the registry for three years before receiving priority on a waiting list. While the Israeli legislation mitigated this particular type of gaming, it introduced a different loophole in the organ allocation system. One of the reported motivations for implementing the priority allocation legislation in Israel was widespread concern over free riding by ultraorthodox religious groups. These groups generally do not recognize brain death (i.e. when the brain ceases to function) as a valid form of death and consequently oppose providing deceased donor organs. 4 However, members of these religious groups do not oppose taking organs, even those recovered from brain dead donors. It has been argued that this group of explicit free riders those who will accept organs but not provide them is a major factor for the historically low rates of organ donation in Israel (Lavee et al. 2010, Lavee and Brock 2012). The priority allocation system was meant to minimize this free riding by rewarding registered donors and giving free riders lower priority on waiting lists. However, the implementation of the Israeli priority allocation legislation created a loophole that may allow this type of free riding to continue. The Israeli donor card gives a registrant the option to check a box requesting that a clergyman be consulted before organ donation occurs (see Figure 1). 5 An individual who wants priority but does not want to be a donor could check that box with the implicit or explicit understanding that his clergyman would refuse donation if the supposed donor were to die and be in a position to have his organs donated. 4 Most organ donation follows brain death, since the deceased patient can be left on a respirator, allowing the organs to be kept alive until they are recovered. Cardiac death (when there is an irreversible loss of circulation) requires fast action on the order of a few minutes for organ recovery to be possible. Data from the New England Organ Bank (NEOB) indicates that in New England, recovery rates are much higher among potential donors who died from brain death than cardiac death. Recovery rates were about 20 percentage points higher for registered donors and about 15 percentage points higher for non-registered donors in 2010, 2011, and (Personal communication, Sean Fitzpatrick, NEOB.) 5 During implementation, a number of people specifically advocated for this clergyman checkbox option amid suspicion that it was motivated by religious groups who wanted to receive priority without having to donate (see 4

5 Figure 1: Donor Card in Israel Translated into English the card (emphasis and color in original) reads: With the hope that I may be of help to another, I hereby order and donate after my death: () Any organ of my body that another my find of use to save his/her life. Or: () Kidney () Liver () Cornea () Heart () Skin () Lungs () Bones () Pancreas [] As long as a clergyman chosen by my family will approve the donation after my death. Even without an explicit checkbox there is still the potential for a loophole to be abused in the Israeli priority system. Signing the donor card in Israel is not binding, so next of kin are still asked about donation and can block the donation of a deceased who had signed a donor card (Lavee and Brock 2012). (This is also the case in the United States, where next of kin can refuse donation even if the deceased had previously joined a state registry; see Glazier 2006.) When next of kin are given a chance to make a final donation decision or block the donation of a registered donor, then individuals could register as donors to receive priority but instruct their next of kin to prevent their organs from being donated upon death, creating a loophole even if one is not explicitly available. What is the potential effect of such loopholes to the priority allocation rules? One potential downside is that the loophole might eliminate the incentive generated by a priority system, since anyone who wanted priority could simply ask for it and take advantage of the loophole rather than incur the cost of donation. Whether the loophole completely eliminates the potential increase in donors due to the priority allocation rule 5

6 would be a function of how the cost of being an organ donor compared to the cost of taking advantage of the loophole. Another potential downside, however, is that the existence of a loophole could poison the pool and make individuals who would have donated in the absence of the priority allocation rule decide not to donate, for example once they observe people undermining the allocation system by taking the loophole. We call those who donate in the absence of a priority incentive warm glow donors. 6 These warm glow donors already choose to donate knowing their organs might go to non-registered donors, but they may respond more negatively when they observe someone taking a loophole in a priority system rather than simply not donating. If a loophole can poison the pool, then introducing a priority system might lead to fewer donors as individuals who would have donated in the absence of the priority rule choose not to donate when there is a loophole. In this study, we use a laboratory game modeled on the decision to register as an organ donor to investigate how a priority allocation system impacts donation rates and how the existence of a loophole affects behavior. It will be years before we have data on actual donations and actual loophole use in Israel, but here we are able to study the loophole, understand what consequences it can have, and anticipate its effects. 7 We find that a priority allocation system generates significantly higher organ donation rates, increasing the number of organs recovered and overall efficiency. 8 However, providing a loophole that allows non-donors to take advantage of priority without paying the cost of donation completely eliminates the benefit created by the 6 The term warm glow donor is inspired by Andreoni (1990). Warm glow has been used to explain the motivation of organ donors in Kessler and Roth (2012). 7 Certainly some hypotheses about organ donation can only be investigated by asking for real organ donor registrations (see Kessler and Roth 2013). However, a number of important aspects about the organ donation decision and the organ allocation system cannot be easily manipulated in practice but can be manipulated and studied in the laboratory. We can use the laboratory to study the incentive issues involved in organ donation, abstracted away from the important but complex sentiments associated with actual organs. For example, in practice the costs of registering as an organ donor are difficult to identify in the field. Costs may include fears about differential medical care for registered organ donors, fear that organs will be removed at a time or in a manner that is inconsistent with religious beliefs, or simply discomfort from thinking about ones death. In the laboratory, we can impose monetary costs to model to some level of approximation the costs faced by donors and control them, e.g. giving some potential donors low costs and others high costs. 8 With our experimental parameters, the priority allocation rule substantially improves outcomes for low cost donors but harms high cost donors in the process. 6

7 introduction of a priority system. When a loophole is available, we find that almost all non-donors take advantage of the loophole; 96% of subjects in the loophole condition have priority. We also find evidence that providing a loophole can poison the pool, undermining warm glow giving by inducing individuals who would have given without priority to withhold donation when there is a loophole. This decrease in warm glow giving occurs primarily when individuals have information about how many people took advantage of the loophole and when they observe that other subjects are taking advantage of the loophole. The results of this study enter a rich literature on warm glow motivations for private provision of public goods and charitable behavior. A closely related literature focuses on the donation of blood and primarily investigates whether incentives for donation can cause a crowding out that might lead to less donation overall. This work has generally found that incentives increase donations without leading to a decrease in blood quality (see Mellstrom and Johannesson 2008; Lacetera and Macis 2010a,b; Lacetera, Macis and Slonim 2012). 9 In our experiment, we find that decisions are influenced by the choices of other subjects to donate or to take the loophole, particularly when those choices are observable. This finding relates to a vast literature on social information and conditional cooperation in public goods games in which subjects have been shown to conform to the public good contributions of others (for laboratory results, see Keser and van Winden 2000; Fischbacher, Gachter and Fehr 2001; and Potters, Sefton and Vesterlund 2005; for field results in the context of charitable giving, see Frey and Meier 2004 and Shang and Croson 2009). While this paper investigates priority allocation systems (and how they can be undermined by loopholes), there are number of strategies that might be employed to increase the number of individuals who register as organ donors or donate an organ while alive. One approach that has been heavily advocated is to change the way individuals are asked to register; in particular to switch from an opt-in protocol, in which individuals check a box to register and leave it blank not to register, to a mandated choice protocol, 9 For evidence of crowding out in other contexts, see Titmuss (1970); Deci, Koestner, and Ryan (1999); Gneezy and Rustichini (2000a,b); and Ariely, Bracha, and Meier (2009). For theory see Benabou and Tirole (2006). 7

8 in which individuals must choose between joining the registry or not joining the registry (see Thaler and Sunstein 2008, Thaler 2009). 10 Another approach is to facilitate kidney exchange, in which incompatible patient-donor pairs are matched. This process finds compatible patient-donor pairs where, for example, donor A gives a kidney to donor B s patient while donor B gives a kidney to donor A s patient (Roth, Sonmez and Unver 2004, 2005a,b, 2007; Roth et al. 2006; Saidman et al. 2006). 11 Results from our paper demonstrate that along with these other strategies, organ allocation policy may be a powerful tool to increase the number of deceased donor organs that are made available for transplantation. In particular, providing priority on organ waiting lists for registered donors has the potential to increase the number of registrations, but how such allocation rules are implemented can make a significant difference on their efficacy. Allowing loopholes in which individuals can receive priority without ever being in a position to donate can undermine the benefits of the priority system and could actually be worse than forgoing the priority allocation system all together. II. Experimental Design In the experiment, subjects played a game modeled on the decision to register as an organ donor. In the experiment, registering to be an organ donor always makes organ available when the health outcome allows it, and so we refer to the decision to register in the experiment as donating. In the instructions to subjects, the experiment was described in abstract terms rather than in terms of organs. Subjects started each round 10 This policy change has been implemented in Great Britain as well as a number of U.S. states, including Illinois and California (New York State just passed legislation to implement mandated choice). Recent research, however, suggests that changing the way individuals are asked to register can have a perverse effect on total donations, particularly from the next of kin of unregistered donors. In particular, individuals seem to treat the desire not to join the registry under mandated choice as more sacrosanct than failing to opt-in to the registry (Kessler and Roth 2013). 11 New institutions have been formed to organize these exchanges and to create chains of donation that start with a single undirected donor (see Roth et al and Ashlagi et al. 2011). As a consequence there have been over 2000 transplants due to kidney exchange since 2004 according to data reported to the Organ Procurement and Transplantation Network (see 8

9 with one A unit (representing a brain) 12 and two B units (representing two kidneys). 13 Each round of the game, the subject is endowed with $6, an A unit, and two B units. Each round, the subject must decide whether to pay a cost of donation that makes their B units available to others if their health outcome allows it (i.e. if they have A-unit failure, as described below). Subjects are randomly assigned a cost of donation (either $0.50 or $4.00, constant for a subject during the entire study) that is paid regardless of whether the subject will be in a position for his B units to be given to other players. 14 In each round, the subject then observes his health outcome. Each subject either has B-unit failure (i.e. both B units fail and the subject needs a B unit to earn more money in the round) or has A-unit failure, in which case he cannot earn any more money in the round and if he previously paid the cost of donation his B units are given to subjects with B-unit failure. 15 Subjects play in a fixed group of 8 players and are told that 2 of the 8 will be randomly selected to have A-unit failure in each round (and thus the probability of A-unit failure is 25%) and that the other 6 will have B-unit failure (and thus the probability of B-unit failure is 75%). In each round, 0, 2, or 4 B-units are made available depending on whether neither, one, or both subjects who ended up with A- unit failure paid the cost to register as a donor. Consequently, 0, 2, or 4 of the six players with B-unit failure receive a B-unit in a given round. Subjects with B-unit failure who receive a B unit from another player earn an additional $4 in the round. Subjects with A-unit failure, and subjects with B-unit failure who did not receive a B unit from another player, do not earn any additional money in 12 Under laws that require heart death for organs to be recovered, an A unit could represent a heart. 13 The design of the game bears similarities to the game in Kessler and Roth (2012) with some simplifications to the implementation and some different parameters. Consequently, our first result will be to replicate the results in Kessler and Roth (2012) to show this game generates the same pattern of behavior. 14 Note that we are modeling the cost of organ donation as a cost of registering to be a donor rather than of actually donating. Deceased donation occurs after death, when we generally assume that utility flows stop and an individual no longer incurs costs or benefits. We are implicitly assuming the costs of registering as an organ donor are psychological costs. 15 As noted above, subjects were always asked for their donation decision before they learned their health outcome. That is, they had to decide whether to pay the cost of donation before they knew whether they would have A-unit failure (in which case their B-units could be given to other subjects) or B-unit failure (in which case their B-units would be useless). 9

10 that round. 16 Since there are always six subjects with B-unit failure, a subject who pays the cost of donation and has A-unit failure always provides B units to two individuals who would not otherwise receive B units. Those individuals each earn an additional $4 from receiving the B unit. Consequently, each subject can calculate that paying the cost of donation generates $2.00 in expectation for other subjects (i.e. there is a 25% chance the donor will have A-unit failure and donate two B units, which each generate $4.00 of earnings for another subject in the group). The experimental design varied the organ allocation rules and the amount of information provided to subjects. There were three different allocation conditions and two different information conditions, generating six different treatments in a 3x2 design. We first describe the organ allocation conditions and then the information conditions. The organ allocation conditions differed in how B units were allocated to subjects with B-unit failure. In the control condition, any available B units were assigned randomly to the subjects with B-unit failure; all subjects with B-unit failure were equally likely to receive a B unit. The control condition models the first-come firstserved waiting list system in the United States with subjects arriving onto the waiting list in a random order and a limited number of organs available. Each round, the subject was reminded of his cost of donation and then asked to choose between two options: Yes, I want to donate my B units or No, I do not want to donate my B units. In the priority condition, subjects who had paid the cost of donation but ended up needing a B unit received priority for available B units. Subjects who paid the cost of donation and had B-unit failure were in a priority group, and any available B units were first assigned randomly among subjects in the priority group; all subjects in the priority group were equally likely to receive a B unit. Only if each subject in the priority group had received a B unit were any B units distributed to subjects who had not paid the cost of donation. In this case, any remaining B units were randomly assigned to subjects with B-unit failure who did not pay the cost of donation; each of these subjects were equally likely to receive one of the remaining B units. This is a very extreme form of priority in that no subject without priority ever receives a B unit unless all subjects with priority 16 To be precise, these subjects have round earnings equal to their initial $6.00 minus their cost of donation (if paid). Subjects who receive a B unit earn that amount plus $

11 have received one. Each round the subject is reminded of his cost of donation and then asked to choose between two options: Yes, I want to donate my B units and receive priority for a B unit if I need one or No, I do not want to donate my B units. In the loophole condition, organs were assigned as in the priority condition, but subjects could join the priority group either by paying the cost of donating or by asking to receive priority without paying the cost. Each round, subjects were reminded about their cost of donation and chose between three options, the two in the priority condition along with No, I do not want to donate my B units, but I do want to receive priority for a B unit if I need one. Throughout the paper, we refer to this latter option as receiving priority by taking advantage of a loophole (or just taking the loophole ). In addition to varying the organ allocation conditions, the experiment also varied the information provided to subjects about the costs of donation and decisions of other subjects in their group. In the low information conditions, subjects only knew their own cost of donation and (in each round) whether they had B-unit failure and, if so, whether they received a B unit. The low information conditions were meant to provide noisy feedback with regard to the number of registered donors and the number of people taking advantage of the loophole: participants could only infer this from their own experience of receiving B units when they had B-unit failure. It is easy to imagine policy makers being opaque about these statistics, making it difficult for individuals to learn the true numbers. In the high information conditions, subjects were also told the distribution of costs of donations of the subjects in their group and (in each round) the number of group members who paid the cost of donation and how many took advantage of the loophole when it was available. The high information conditions were meant to model a world in which policy makers provide more precise information about the number of people who register as donors and those who take advantage of the loophole when it is available. Figure 2 shows the six treatments in the 3x2 design. Subjects stayed in the same information condition (either low information or high information) for the entire study but played in two different organ allocation conditions. Subjects were not told how many rounds of the game they would play, but after their group had played 15 rounds in one of the condition, subjects were informed that the rules of the game had changed. The rule changes were explained and then the group played 15 11

12 rounds in another organ allocation condition. Each session had 16 subjects who played in one of two fixed groups of 8, and both groups in a session played in the same order of conditions so instructions could be read aloud. 3 x 2 Design Figure 2: The 3x2 Experimental Design Organ Allocation Condition Control Priority Loophole Information Condition Low Control, Low Info Priority, Low Info Loophole, Low Info High Control, High Info Priority, High Info Loophole, High Info After all rounds had been played, subjects were informed of the round that had been randomly selected for payment and all subjects were paid their earnings from that round in cash along with a $10 show-up fee. III. Experimental Results This paper reports results from 608 subjects who participated in one of 38 sessions run at the Wharton Behavioral Lab during the fall of Subjects were college students who participated for one hour and made decisions anonymously. Average earnings were $16.62 per subject, including a $10 show up fee. The experiment was conducted using z-tree (Fischbacher 2007). As explained in the previous section, subjects played in either the low information or high information condition. In that information condition, each group of 8 subjects first played 15 rounds in one organ allocation condition (either Control, Priority, or Loophole) followed by 15 rounds in a different organ allocation condition. Since subjects started in one of the three conditions and then switched to one of the other two, there are six possible organ allocation condition orders. Table 2 shows the number of sessions, groups, and subjects who participated in each order of the conditions under low information and under high information. 12

13 Table 2: Number of Sessions, Groups, and Subjects in Each Treatment Order Organ Allocation Condition Order Low Information High Information Control, Priority 4 sessions (8 groups, 64 Ss) 4 sessions (8 groups, 64 Ss) Control, Loophole 4 sessions (8 groups, 64 Ss) 3 sessions (6 groups, 48 Ss) Priority, Control 3 sessions (6 groups, 48 Ss) 4 sessions (8 groups, 64 Ss) Priority, Loophole 3 sessions (6 groups, 48 Ss) 2 sessions (4 groups, 32 Ss) Loophole, Control 3 sessions (6 groups, 48 Ss) 2 sessions (4 groups, 32 Ss) Loophole, Priority 3 sessions (6 groups, 48 Ss) 3 sessions (6 groups, 48 Ss) The main result of interest is the likelihood that subjects pay the cost of donation and make their B units available to others in the event of A-unit failure. Figure 3 displays, by round and treatment, the percentage of subjects that pay the cost of donation. The top panel, Panel A, displays the data from subjects playing in the low information conditions. The bottom panel, Panel B, displays the data from subjects playing in the high information conditions. Notice that in each panel, the data lines are broken after round 15. This gap is to indicate that for each organ allocation condition different groups comprise the data in Rounds 1-15 and the data in Rounds The following subsections will analyze the data presented in Figure 3. III.1. Priority The most striking result in Figure 3 is that subjects in the priority condition are much more likely to pay the cost of donation than in the control condition. 17 Combining data from the high and low information conditions, the donation rate across all subjects is 69.3% in the priority condition and 40.9% in the control condition. 18 This 28.4 percentage point difference represents a 70% increase in the donation rate. Result 1: The priority allocation rule substantially increases donation 17 Here we replicate the result from Kessler and Roth (2012). Even in this slightly different game, a priority allocation rule substantially increases the probability of donation. 18 There are no statistically significant differences in donation between the information conditions for any of the allocation conditions, so we pool high and low information conditions for some analysis. 13

14 Figure 3: Probability of Donation by Treatment and Round Panel A: Low Information Condition Percent of Subjects Who Donate Round Control (Low Info) Priority (Low Info) Loophole (Low Info) Panel B: High Information Condition Percent of Subjects Who Donate Round Control (High Info) Priority (High Info) Loophole (High Info) Lines are broken after round 15 to indicate that different groups comprise the data for Rounds 1-15 and for Rounds

15 There are two ways in which the priority allocation rule impacts the donation of subjects. First, there is an immediate response in the number of individuals who donate, even in the first round that priority is introduced (either Round 1 or Round 16; 432 and 416 observations, respectively; p<0.01 for both tests, data clustered at group level for Round 16 test). In addition, groups have a slower decline in donation over their 15 rounds in the priority condition than in their 15 rounds in either of the other two conditions. This result can be seen in Figure 3 and is confirmed in regression analysis in Table 3. Across all specifications, Table 3 displays a significant negative coefficient on Rounds played in condition, a variable that takes values from 0 to 14 denoting the number of previous rounds the subject has played in that condition. That Rounds played in condition is negative reflects the fact that our excluded group, the control group, displays the standard decrease in contribution over time that is routinely observed in public good games (Ledyard 1995). Additional results from Table 3 demonstrate the two effects of the priority allocation rule noted above. First, the coefficient on Priority is positive and significant, indicating that subjects are much more likely to donate in their first round in the priority condition than in the first round of the control condition. Second, the coefficient on Priority*Rounds played in condition is positive and significant, indicating that contribution in the priority condition declines much more slowly than in the control condition. While in control condition the probability of donation decreases by 1.2% to 1.4% per period (the magnitude of Rounds played in condition), in the priority condition the decrease is only 0.3% to 0.5% per period. Consequently, the difference between the donation rate in the priority condition and the donation rate in the control condition grows as subjects have more experience in the condition. Regressions (1) and (2) analyze data from the first 15 rounds, when subjects have only played in one condition. We see the same pattern of results in Regressions (3) and (4), which analyze data from all 30 rounds. Regressions (2) and (4) control for Info indicating that data came from the high information conditions and include interactions with Info. None of these are significant, demonstrating that these results are similar for both the low information and high information conditions. 15

16 Table 3: Organ Registration By Condition and Round Donation (0 or 1) Linear Probability Model (OLS) First 15 Rounds All 30 Rounds (1) (2) (3) (4) Rounds played in condition *** *** *** *** (0.002) (0.002) (0.002) (0.002) Priority 0.215*** 0.199*** 0.228*** 0.254*** (0.035) (0.046) (0.023) (0.029) Priority* 0.010*** 0.009*** 0.009*** 0.008*** Rounds played in condition (0.003) (0.003) (0.002) (0.002) Loophole ** (0.036) (0.043) (0.024) (0.032) Loophole* Rounds played in condition (0.003) (0.003) (0.002) (0.002) Info (0.045) (0.035) Info*Priority (0.066) (0.040) Info*Priority* Rounds played in condition (0.003) (0.002) Info*Loophole (0.066) (0.043) Info*Loophole* Rounds played in condition (0.005) (0.004) Last 15 rounds *** *** (0.015) (0.015) Cost *** (0.007) *** (0.007) *** (0.006) *** (0.006) Constant 0.713*** 0.704*** 0.702*** 0.692*** (0.028) (0.031) (0.023) (0.028) Observations Clusters R-squared Robust standard errors clustered by group are in parentheses: * significant at 10%; ** significant at 5%, *** significant at 1%. Rounds played in condition takes values 0 to 14 for the number of previous rounds the subject has played in that condition. Priority and Loophole indicate organ allocation condition and test for differences from the control condition. Info is a dummy variable equal to 1 if the data is from the high information conditions and interactions with Info test for differences between the low information conditions and high information conditions. Cost is a dummy variable equal to 1 if the subject has the high cost of donation (i.e. $4.00). Last 15 rounds is a dummy variable equal to 1 if the data is from the second set of 15 rounds of the game. 16

17 Table 4 combines the immediate difference in contribution and the changes in donation rate over time in a condition to test for the average effect of priority. Table 4 controls flexibly for round of the experiment by including round dummies. Table 4: Organ Registration By Condition Donation (0 or 1) Linear Probability Model (OLS) First 15 Rounds All 30 Rounds (1) (2) (3) (4) Priority 0.288*** 0.262*** 0.291*** 0.312*** (0.031) (0.041) (0.020) (0.024) Loophole * ** (0.035) (0.047) (0.022) (0.028) Info (0.045) (0.035) Info*Priority (0.061) (0.040) Info*Loophole (0.067) (0.043) Cost *** (0.007) *** (0.007) *** (0.006) *** (0.006) Round Dummies Yes Yes Yes Yes Observations Clusters R-squared Robust standard errors clustered by group are in parentheses: * significant at 10%; ** significant at 5%, *** significant at 1%. Priority and Loophole indicate organ allocation condition and test for differences from the control condition. Info is a dummy variable equal to 1 if the data is from the high information conditions and interactions with Info test for differences between the low information conditions and high information conditions. Cost is a dummy variable equal to 1 if the subject has the high cost of donation (i.e. $4.00). Round dummies include a dummy for each round of the game. Results from Table 4 show the magnitude of the priory allocation rule. The priority rule increases the probability of donation by approximately 29 percentage points over the control condition. Since two subjects from each group are randomly chosen to have A-unit failure, the increase in the donation rate from the priority rule has a direct effect on the number of B-units that are made available. Table 5 pools data from the high and low information conditions across all 30 rounds and shows, for each condition, the 17

18 donation rate, the average number of B units made available, the percentage of subjects with B-unit failure who receive a B unit, and average earnings. The priority condtion makes available and average of 2.80 B units per period, which is a 70% increase over the 1.65 B units available in the control condition (the differnece is statistically significant, p<0.01, data clustered at group level). The same pattern immerges for probability of receiving a B unit and for earnings (p<0.01 for both tests, data clustered at group level). Table 5: Outcomes by Condition Control Priority Loophole Donation Rate 40.9% 69.3%*** 35.9%** Number of B Units available *** 1.49 Percent who get B Unit when needed 27.5% 46.6%*** 24.9% Earnings $6.50 $6.87*** $6.46 This table pools data from the high and low information conditions over all 30 rounds. Stars in Priority and Loophole conditions indicate a statistically significant difference from the Control condition, robust standard errors clustered at group level: * significant at 10%; ** significant at 5%, *** significant at 1%. While the priority condition has a large positive effect on donation, the availability of B units, and earnings, the results in Table 5 mask two countervailing effects on subject outcomes. The introduction of priority affects different the low cost subjects (those who must pay $0.50 to donate) and the high cost subjects (those who must pay $4.00 to donate). Table 6 also pools high and low information conditions across all 30 rounds and shows, by treatment and cost of donation, the donation rate, the percentage of subjects with B-unit failure who receive a B unit, and average earnings. While the low cost subjects see a large increase in the probabilty of receiving a B unit and in earnings, the high cost subjects see a decrease in the probability of receiving a B unit and in earnings. Both the positive effects for low cost donors and the negative effects for high cost donors between the priority and control conditions are statically significant (p<0.01 for all tests, data clustered at group level). 18

19 Table 6: Outcomes by Cost and Treatment Cost = $0.50 Cost = $4.00 Control Priority Loophole Control Priority Loophole Donation Rate 50.1% 85.5%*** 43.8%** 13.1% 20.6%*** 12.0% Percent who get B Unit when needed 26.2% 55.5%*** 24.5% 31.3% 20.1%*** 25.9% Earnings $6.54 $7.24*** $6.51 $6.40 $5.78*** $6.32 This table pools data from the high and low information conditions. Stars in Priority and Loophole conditions indicate a statistically significant difference from the Control condition, robust standard errors clustered at group level: * significant at 10%; ** significant at 5%, *** significant at 1%. It is straightforward to see why the priority allocation rule would have a differential effect on the low cost and high cost donors. Priority rewards individuals who pay the cost of donation with a higher likelihood of receiving a B unit and the accompanying extra earnings. Since this incentive induces significantly more low cost subjects to donate (in the priority condition 85.5% of low cost subjects donate while only 20.6% of high cost subjects donate), high cost subjects are less likely to receive a B unit since any available B units will have been given away to low cost donors with priority. The priority rule, while encouraging donation and increasing efficiency, generates inequality between low cost subjects for whom priority makes outcomes substantially better and high cost subjects for whom priority makes outcomes substantially worse. While the net effect is positive, the inequality might still be a concern for policy makers. A policy maker who wants to mitigate this inequality might be tempted to provide a way for individuals with a high cost of donation to avoid the harsh outcome associated with the priority system, for example by providing a loophole for them a way for these individuals to receive priority without donating. In practice, however, the cost an individual faces to donate is not observable, and so such a loophole would have to be available to everyone. In the next subsection, we investigate the effect of introducing such a loophole into the priority allocation system. Of course, a priority allocation system might have loopholes that were not explicitly designed by policy makers for any particular end. 19

20 Loopholes may arise by accident due to the institutional details of the policy. Our loophole condition does not distinguish between the reasons that the loophole exists but rather investigates how such a loophole affects donation when it is available. III.2. Loophole What is the effect of adding a loophole to the priority allocation rule? We can see immediately from Figure 3 that having a loophole in the priority allocation system eliminates the increase in donation induced by priority. These results are statistically significant as shown in the results in Table 4. The Loophole coefficient has a negative sign for all specifications, indicating that not only are the donation rates in the loophole condition less than in the priority condition (tests of whether the coefficient on Priority is equal to the coefficient on Loophole are all rejected with p<0.01, data clustered at group level) they are also at least directionally less than the donation rates in the control condition. Consequently, the loophole has completely eliminated the beneficial effect of priority. Result 2: The loophole eliminates the increase in donation generated by the priority allocation rule Why does the loophole generate such a vast decrease in donation? Table 7 shows the choices subjects make in the loophole condition by information condition and cost of donation. Subjects overwhelming take advantage of the loophole when it is available. Among the low cost subjects, for whom donation only costs $0.50, the majority of actions are to take advantage of the loophole. Only 2.5% of actions of low cost subjects are to not donate and to not take the loophole. For high cost subjects, the vast majority of actions are to take advantage of the loophole (75% under high information and 83% under low information). Averaging across high and low cost subjects and high and low information conditions, only 4% of actions are subjects who choose neither to donate nor to take priority without donating. Put another way, across all rounds in the loophole condition, 96% of subjects have priority. 20

21 Table 7: Choices In the Loophole Condition Low Information Condition High Information Condition Cost = $0.50 Cost = $4.00 Cost = $0.50 Cost = $4.00 Donate 46.03% 12.05% 41.00% 11.83% Do Not Donate 2.52% 4.74% 2.50% 13.17% Take Loophole 51.45% 83.21% 56.50% 75.00% Looking back at Table 5, we see that introducing the loophole also eliminates the increase in B units and earnings associated with the priority rule. The percent of subjects with B-unit failure who get a B unit has fallen back to 24.9% and earnings are down to $6.46. Both of these are statistically smaller than their counterparts from the priority condition (p<0.01 for both tests, data clustered at group level) and statistically indistinguishable from their counterparts in the control condition. The introduction of the loophole also eliminates the inequality generated by the priority allocation rule. Looking back at Table 6, we see that in the loophole condition, both the likelihood of getting a B unit and earnings are back to control condition levels for both the low cost and high cost subjects. 19 Another striking fact from Tables 4, 5 and 6 is that donation rates are lower in the loophole condition than in the control condition. The coefficient on Loophole is negative and sometimes significant in Table 4 and the donation rate in the loophole condition is sometimes significantly less than the donation rate in the control condition in Tables 5 and 6. The difference is statistically significant when pooling all the data in the experiment, for example in Regression (3) of Table 4 and in the first row of Table 5. Subjects in the loophole condition are significantly less likely to donate than subjects in the control condition. Breaking down the data by information condition, as in Table 8, we see that this negative effect of the loophole is only directionally significant 19 The loophole condition results on both the percent who get a B unit and earnings are statistically different from the priority condition results for both the low and high costs donors (p<0.01 for all tests except the probability of getting a B unit among high cost donors where p<0.05, data clustered at group level). 21

22 for the low information condition but is statistically significant for the high information condition. Table 8: Organ Registration By Condition Donation (0 or 1) Linear Probability Model (OLS) First 15 Rounds Low High Low All 30 Rounds High Information Information Information Information (1) (2) (4) (5) Priority 0.262*** 0.312*** 0.310*** 0.271*** (0.041) (0.045) (0.025) (0.031) Loophole ** ** (0.048) (0.048) (0.029) (0.033) Cost *** (0.010) *** (0.010) *** (0.008) *** (0.009) Round Yes Yes Yes Yes Dummies Observations Clusters R-squared Robust standard errors clustered by group are in parentheses: * significant at 10%; ** significant at 5%, *** significant at 1%. Priority and Loophole indicate organ allocation condition and test for differences from the control condition. Info is a dummy variable equal to 1 if the data is from the high information conditions and interactions with Info test for differences between the low information conditions and high information conditions. Cost is a dummy variable equal to 1 if the subject has the high cost of donation (i.e. $4.00). Round dummies include a dummy for each round of the game. Result 3: In the high information condition, subjects are less likely to donate in the loophole condition than the control condition It is worth noting that do not get a significant interaction between the high and low information conditions when comparing the loophole condition to the control condition, as can be seen in Regressions (2) and (4) of Table 4, in which Info*Loophole is not statistically significantly less than However, we do get an interaction between the 20 One reason that we do not get a significant interaction between high and low information and the priority and control conditions is that in the presence of low information, the loophole generates directionally less donation than the control condition. 22

23 high cost and low cost conditions when comparing the loophole to the priority condition in the first 15 rounds before subjects have experienced any other treatments. Figure 5 shows donation rates for priority and loophole by information condition (the light lines are high information and the dark lines are low information) for the first 15 rounds. The interaction is partly driven by the higher donation rates in priority under high information than under low information (76.9% in the high information condition and 69.9% in the low information condition, p<0.1 clustered at group level). In addition, the donation rate in the loophole condition is directionally lower under high information than under low information. When combined, this leads to a significantly negative interaction. Adding the loophole to the priority rule is significantly more damaging under high information than under low information. Under high information, the donation rates drops from 76.9% under priority to 35.1% under loophole (41.8 percentage points); under low information the donation rate drops from 69.9% under priority to 41.3% under loophole (28.6 percentage points). The interaction term reflects a 13.2 difference in difference in donation rates, p<0.05 clustered at group level). 0.9 Figure 5: Priority and Loophole for First 15 Rounds Percent of Subjects Who Donate Priority (High Info) Round Loophole (High Info) Priority (Low Info) Loophole (Low Info) 23