Research ArticleSOCIAL SCIENCES

Onymity promotes cooperation in social dilemma experiments

See allHide authors and affiliations

Science Advances  29 Mar 2017:
Vol. 3, no. 3, e1601444
DOI: 10.1126/sciadv.1601444
  • Fig. 1 Onymity brings out the best in people.

    Cooperation takes over defection as the dominant action if the anonymous treatment (T1) is replaced with the onymous one (T2). Pairwise comparisons indicate that the increase in the frequency of cooperation and the decrease in the frequency of defection between the two treatments are significant at the 5% level. The frequency of punishment decreases, but the difference between the medians is insignificant. Box-and-whisker plots with notches reveal the empirical distribution of the frequency of each action. Box height indicates the interquartile range with the median in between. Notches indicate the 95% confidence intervals for the median, thus permitting a visual pairwise comparison. Whisker height is such that 99.3% of the normally distributed data would be covered. The points drawn as outliers fall outside of the whisker coverage.

  • Fig. 2 Onymity elicits prosocial behavior, but punishment does not.

    The action in the first round (column 1st) and the first-order conditional strategies (columns C, D, and P, indicating the responses to cooperation, defection, and punishment, respectively) in later rounds are all significantly different between the two experimental treatments. The increase in cooperation and the decrease in defection by switching from the anonymous (T1) to the onymous (T2) treatment is particularly noticeable. Thus, onymity elicits prosocial behavior. In contrast, an immediate response to punishment—the preferred choice being defection followed by counter-punishment—is overwhelmingly antisocial even under onymity.

  • Fig. 3 Playing nice under onymity pays off.

    (A and B) When the opponent remains unknown (T1), the payoff per round does not correlate with the use of cooperation but correlates positively with the use of defection [in contrast with the study of Dreber et al. (22)], thus indicating that the prosocial action (that is, cooperation) is less desirable than the antisocial one (that is, defection). (D and E) When the opponent is known (T2), the payoff per round correlates positively with the use of cooperation and negatively with the use of defection, showing that the prosocial action is now more desirable than the antisocial one. (C and F) The only similarity between the two treatments is that the payoff per round correlates negatively with the use of punishment [the “winners don’t punish” effect (22, 24)]. Shown are the regression lines with the 95% prediction intervals (dashed curves). (G) In accompanying statistical analysis, the smaller font size indicates the 95% statistical confidence intervals.

  • Fig. 4 Onymity alters the factors of success in a social dilemma.

    (A and B) Under anonymity (T1), the use of cooperation does not affect the ranking, whereas the use of defection is advantageous, albeit in a limited fashion. (D and E) Under onymity (T2), the use of cooperation is decisively advantageous for the ranking of the participants. In contrast, the use of defection is disadvantageous. (C and F) The “winners don’t punish” effect remains unaltered between the two experimental treatments. The standard competition ranking is used, such that a lower ranking number is better. The vertical axes report the average use of each action per 100 rounds. For clarity, the scale for the use of punishment is doubled. The number of bins in the histograms is kept the same for both treatments, although the number of participants was slightly higher in T1 than in T2 (80 versus 74, respectively), thus causing the bin sizes to slightly differ between the treatments.

  • Fig. 5 Frequency of cooperation, defection, and punishment over the course of an interaction.

    (A) In the anonymous treatment (T1), action frequencies do not show a statistically discernible trend as the interaction progresses, thus contrasting the results in the study of Dreber et al. (22) and in the study of Wu et al. (24). (B) In the onymous treatment (T2), the frequency of cooperation (defection) seems to be decreasing (increasing), but again, the results are statistically insignificant. Shown are the data points (circles, squares, and x marks for cooperation, defection, and punishment, respectively) and the regression lines. The latter were simultaneously fitted to ensure that the frequencies add up to unity. (C) In the accompanying statistical analysis, the smaller font size indicates the 95% statistical confidence intervals. We additionally tested whether the angle between the regression lines for cooperation and defection was significantly different from zero, but we obtained a negative result (P = 0.58 and P = 0.072 for T1 and T2, respectively).

  • Fig. 6 Same beginnings but vastly different endings.

    Inner workings of anonymity and onymity through a direct comparison of some typical interactions that finished differently although they started the same. (A) In the anonymous treatment, even mutual cooperation occasionally gives way to defection and punishment. Other initial decisions make it only harder to reach and maintain a cooperative outcome. (B) In the onymous treatment, in contrast, even mutual preemptive punishment is reparable. Other initial decisions make it mostly easier to reach and maintain a cooperative outcome.

Supplementary Materials

  • Supplementary material for this article is available at http://advances.sciencemag.org/cgi/content/full/3/3/e1601444/DC1

    Supplementary Materials and Methods

    Supplementary Results

    fig. S1. Snapshot of the questionnaire used to test the basic understanding of PD games.

    fig. S2. Interface for playing the PD game in anonymous treatment.

    fig. S3. Interface for playing the PD game in onymous treatment.

    fig. S4. Control trials.

    fig. S5. Regression diagnostics.

    fig. S6. Computer-simulated recreations of the anonymous treatment (T1).

    fig. S7. Computer-simulated recreations of the onymous treatment (T2).

    fig. S8. Comparison with other similar studies.

    table S1. Basic information on the experimental sessions.

    table S2. Gender as a confounding factor.

    table S3. Academic background as a confounding factor.

    Reference (36)

  • Supplementary Materials

    This PDF file includes:

    • Supplementary Materials and Methods
    • Supplementary Results
    • fig. S1. Snapshot of the questionnaire used to test the basic understanding of PD games.
    • fig. S2. Interface for playing the PD game in anonymous treatment.
    • fig. S3. Interface for playing the PD game in onymous treatment.
    • fig. S4. Control trials.
    • fig. S5. Regression diagnostics.
    • fig. S6. Computer-simulated recreations of the anonymous treatment (T1).
    • fig. S7. Computer-simulated recreations of the onymous treatment (T2).
    • fig. S8. Comparison with other similar studies.
    • table S1. Basic information on the experimental sessions.
    • table S2. Gender as a confounding factor.
    • table S3. Academic background as a confounding factor.
    • Reference (36)

    Download PDF

    Files in this Data Supplement:

Stay Connected to Science Advances

Navigate This Article