Aggregate Comparison

Result 4.10. When information costs are zero, the frequency of second party punish-ment is equal to the frequency of third party punishpunish-ment.

In particular, 73 percent (16 out of 22) of the second parties were willing to impose costly sanctions, exactly the same relative frequency as for third parties. The 16 subjects are not identical in the two conditions, respectively. Two subjects who did not impose sanctions as a third party did so as a second party. Likewise, two subjects who did not impose sanctions as a second party did so as a third party.

7The results of the SPMG have been dealt with in detail in chapter3. Here we just compare the results to the TPMG results.

Figure 4.2: Distribution of behavioral types at different monitoring cost levels for third and second parties.

0.2.4.6.8

M0 M5 M10 M0 M5 M10

1 2

Targeted Punishers (T) Blind Punishers (B) Non−Punishers (N)

Relative Frequency

Graphs by Cond

Result 4.11. When information costs are zero, the share of targeted punishment is about equal in third and second parties.

When monitoring was costless, all second parties (16 out of 16) that imposed a punishment acquired the information whether the target player cooperated or de-fected. In other words, all sanctions were conditioned on the target player’s first stage behavior. In terms of our behavioral types, there are only T -types and N-types, but no B-types among the second parties. This distribution is approximately equal to the one in the third party monitoring condition.

Result 4.12. When information costs are zero, the dominant direction and intensity of both third and second party punishment is cooperators imposing stiff sanctions on cheaters, but third party punishment is less targeted on defectors and weaker than second party punishment.

19 punishment actions were carried out in the baseline treatment by second par-ties of whom 15 (79 percent) were taken by cooperators and 4 (21 percent) by defec-tors. Recall that among the third parties, 78 percent of punishment actions were taken by cooperators and 22 percent by defectors. Taking again intensity into account, co-operating second parties (third parties) accounted for 84 percent (70 percent) of total investment into punishment while defecting second parties (third parties) accounted for 16 percent (30 percent). Thus, the distributions do not differ much: Among the class of both second party and third party punishers, cooperators clearly supplied most of the punishment action.

Whom were the second party punishments imposed on? In 84 percent (72 per-cent) second parties’ (third parties’) punishment decisions the target of a punishment action was a defector. Cooperators were the target of 14 percent (28 percent) of second party (third party) punishment actions. Again, taking intensity into account, defectors received 93 percent (77 percent) of all the second party (third party) pun-ishment imposed while cooperators received 7 percent (23 percent). Summarizing, both second and third party punishment originates predominantly with cooperators and arrives at defectors, but third parties are somewhat less targeting, imposing more punishment on cooperators.

Against the above benchmark comparisons, we now report on differential behav-ioral implication of endogenizing second and third parties’ information status at the time of the punishment decision through a costly choice, respectively.

Result 4.13. The propensity to punish decreases (relative to the baseline) slightly more strongly with information costs in second than in third parties.

Figure 4.2 illustrates this first finding: The share of second parties (third parties) that neither monitor nor punish (type N) increases from 27 percent (27 percent) at κm= 0 to 67 percent (62 percent) when κ^m= 5 and 74 percent (57 percent) when κm= 10. Thus, the share of N-types is slightly less increasing in third than in second parties. The ratio of the frequency of second (third) party punishers to the frequency of N-types decreases from 2.67 (2.67) at κm= 0 to 0.5 (0.61) at κ^m= 5 and to 0.35 (0.77) at κm= 10.

Result 4.14. The propensity to punish indiscriminately increases (relative to the baseline) more strongly in third than in second parties.

Figure 4.2 again reports, in second and third parties, the share of blind punishers (type B) that do not acquire information and yet impose sanctions. In second (third) parties this share increases from zero (5 percent) in the baseline to 12 (21 percent) and 13 percent (26 percent) as the monitoring costs increase to 5 and 10 tokens, respectively. Analogously, the share of targeted second party (third party) punishers (type T ) decreases from 73 percent (68 percent) for costless monitoring to 21 percent (17 percent) when κm= 5 and 13 percent (17 percent) when κ^m= 10, respectively.

The ratio of the frequency of indiscriminate second party (third party) punishers to the frequency of all second party (third party) punishers increases successively from zero (0.06) at κm= 0 to 0.36 (0.56) at κ^m= 5 and to 0.5 (0.6) at κ^m= 10. The ratio of the frequency of indiscriminate second party (third party) punishers to the frequency of targeted second party (third party) increases successively from zero (0.07) at κm= 0 to 0.57 (1.27) at κm= 5 and to 1 (1.5) at κ^m= 10. Thus, third parties have clearly a stronger tendency to punish indiscriminately if monitoring is costly.

Result 4.15. The incentives to cooperate are weaker in the third party than in the sec-ond party punishment csec-ondition at zero information costs. The presence of positive information costs increases this difference.

31 percent (35 percent) and 29 percent (44 percent) of cooperating second par-ties (third parpar-ties) imposed sanctions on (unilateral) defectors in M5 and M10, re-spectively, with respective average expenditures of 11.6 tokens (2.9 tokens) and 8.2 tokens (3.2 tokens). 10 percent (19 percent) and 11 percent (31 percent) of the co-operating second parties (third parties) in M5 and M10, respectively, also imposed punishments on other cooperators, with respective average expenditures of 6.8 to-kens (between 0.7 and 1.2 toto-kens) and 3 toto-kens (between 0.6 and 1.6 toto-kens). 61 percent (57 percent) and 82 percent (57 percent) of the defecting second (third) par-ties did not punish in M5 and M10, respectively. Of the remaining defecting second (third) parties in M5, 57 percent (70 percent) punished indiscriminately (B-type) with an average expenditure of 5.8 tokens (between 4.3 and 4.6 tokens), and 43 percent (30 percent) targeted (T -type) with an average expenditure of 6 tokens (between 5 and 6.7 tokens) on defectors and 5 tokens (1.7 tokens) on cooperators, plus the five tokens monitoring fee. Likewise, of the remaining defecting second (third) parties in M10, all (67 percent) punished indiscriminately (B-type) with an average expenditure of 5.8 tokens (between 1.8 and 4.3 tokens), and none (33 percent) targeted (T -type).

Despite the rise in monitoring costs, cheaters rather than cooperators remain the prevailing target of both second and third party sanctions on average, although less pronounced. Defection is still punished clearly more often by second parties than cooperation in both M5 (32 vs. 17 percent), and M10 (26 vs. 13 percent). The same has been shown above for third parties. Also, defectors continue to receive stronger punishment on average than cooperators by second parties in both M5 (9.8 vs. 3.3 tokens), and M10 (5.6 vs. 1.1 tokens). Again, the same has been shown above for third parties. The difference between average punishment of cheaters and average punishment of cooperators, however, decreases less strongly in second parties (from 20.1 tokens in M0 over 6.5 tokens in M5, a 68 percent decrease, to 4.5 tokens in M10, a further 31 percent decrease, for a total decrease of 78 percent) than in third parties (from 14.3 tokens in M0 over 1.2 tokens in M5, a 92 percent decrease, to 1.3 tokens in M10, for a total decrease of 91 percent). This reinforces the conclusion that while both second and third party punishment tends to become less targeted on cheaters through the introduction of monitoring costs, the effect is stronger in third parties. In other words, while third party punishment is provides weaker incentives to cooperate already at costless monitoring, introducing monitoring costs makes the difference even larger.

Result 4.16. In both the second party and the third party monitoring games, cheating is only unprofitable if monitoring is costless.

To sum up, while second parties respond to the introduction of monitoring costs more often than third parties with refraining from sanctioning altogether, third parties opt more often than second parties to indiscriminate punishment. Thus, third parties appear to be more willing to go for the risk of hitting a cooperator than second parties.

In consequence, while both second and third party punishment gets less targeted, and third party punishment provides weaker incentives to cooperate than second party

punishment already when information costs are zero, introducing monitoring costs has a more deteriorating effect on the incentive structure emerging from third party punishment than from second party punishment.