VOL 87 NO.5 PALFREY AND PRISBREY:PUBLIC GOODS EXPERIMENTS 841 15日 10 pooled separate 0 10 20 50 60 70 FIOURE 4.INDIVIDUAL WARM-GLOW EFFECTS:POOLED VS.SEPARATE ments sim measurements,the range of predicted contri- ilar to those of Isaac et al.(1984).but with nve-person groups and ri/v larcdtorec experienced subjects lie in the range of 10 per- findings by Andreoni(1995)in an experiment cent or less. which was also motivated by the problem of n than was d subi by we what th however.successfully replicate the Isaac et al. other where subjects were paid by a fixed for- results in an additional experiment using our mula based on the rank of their payoff.The protocol and a fixedr across subjects nent is assume t0 remove muc中g space See Palfrey and Prisbrey (1993)fo etails.That Palfrey ar ng ret of the expe nent.In the oken valucse of 80 ible fo did n Th VOL 87 NO. 5 PALFREY AND PRISBREY: PUBUC GOODS EXPERIMENTS 84t 15 - -5 - 30 40 subject i FIGURE 4. INDIVIDUAL WARM-GLOW EFFECTS: P(K)LED VS. SEPARATE Note: The esiimated individual wann-glow effects from Ihe separate estimations by endowment condition (Table 4 columns 2 and 3) compared to the pooled estimates (Table 4 column 1), r, = 20, which yields a wide range of predic￾tions, depending which one is used. Using our measurements, the range of predicted contri￾bution lies between 4 percent and 20 percent for inexperienced subjects, well over half of which is attributable to error. Predictions for experienced subjects lie in the range of 10 per￾cent or less. This estimated contribution rate is some￾what less than than was observed by Isaac et al., at least for inexperienced subjects. We did, however, successfully replicate the Isaac et al, results in an additional experiment using our protocol and a fixed r, across subjects.^^ This suggests that some of the overcontribution in past experiments may be due to the use of a degenerate distribution of private values.^^ Andreoni (1988) conducted experiments sim￾ilar to those of Isaac et al. (1984), but with five-person groups and r, IV = 2. Comparisons of our data to his lead to similar conclusions. These results can also be compared to recent findings by Andreoni (1995) in an experiment which was also motivated by the problem of differentiating errors. His design consisted of two treatments: one where subjects were paid what they earned in the experiment, and an￾other where subjects were paid by a fixed for￾mula based on the rank of their payoff. The latter treatment is assumed to remove much of the altruistic incentive for contribution,^•' Otherwise, the experiment is run in the usual ^^ This is not included in the preseni paper for reasons of space. See Palfrey and Prisbrey (1993) for details. That paper also presented results from a treatment in which subjects were informed of their fellow group members' token values, which was conducted to test whether the incomplete information in the present experiment was re￾sponsible for the slightly lower contribution rates. The finding was that this additional information did not in￾crease contribution rates, rejecting the hypothesis that in￾complete information leads to more free-riding. ^^ One theoretical explanation for this difference would be that warm-glow effects are subject to dimin￾ishing returns and that these effects are cumulative over the course of the experiment. In the nondegenerate de￾sign, subjects nearly always contribute when they have low values of r,. so diminishing warm-glow effects would lead to less contribution than in the degenerate designs where r, is always greater than V. This suggests possible future experiments. ••' It is not entirely clear why rank payoffs should nec￾essarily remove the warm-glow incentive for contribution