Third parties’ anger at transgressors, and their intervention and punishment on behalf of victims, varied in real-life conflicts as a function of how much third parties valued the welfare of the disputants

When and Why Do Third Parties Punish Outside of the Lab? A Cross-Cultural Recall Study. Eric J. Pedersen et al. Social Psychological and Personality Science, December 16, 2019. https://doi.org/10.1177/1948550619884565

Abstract: Punishment can reform uncooperative behavior and hence could have contributed to humans’ ability to live in large-scale societies. Punishment by unaffected third parties has received extensive scientific scrutiny because third parties punish transgressors in laboratory experiments on behalf of strangers that they will never interact with again. Often overlooked in this research are interactions involving people who are not strangers, which constitute many interactions beyond the laboratory. Across three samples in two countries (United States and Japan; N = 1,294), we found that third parties’ anger at transgressors, and their intervention and punishment on behalf of victims, varied in real-life conflicts as a function of how much third parties valued the welfare of the disputants. Punishment was rare (1–2%) when third parties did not value the welfare of the victim, suggesting that previous economic game results have overestimated third parties’ willingness to punish transgressors on behalf of strangers.

Keywords: third-party punishment, anger, cooperation, bystander intervention

WTR = welfare trade-off ratio

Discussion
Here, we proposed that a major function of third-party punishment
is to deter aggressors from harming individuals with
whom the punisher shares a fitness interest and that the psychological
mechanisms that regulate punishment take into account
the punisher’s perceived welfare interdependence with the disputants
in a conflict (Pedersen et al., 2018). To test these
hypotheses, we asked U.S. students, U.S. Mechanical Turk
workers, and Japanese students to recall how they responded
the last time they observed a conflict. The recall study method
ensures a wide sampling of situations and thus high generalizability
to real-life conflicts. We found that third parties’ WTRs
for the victim in a conflict indeed predicted anger, intervention,
and punishment on behalf of the victim. We also found that
third parties’ WTRs for the transgressor were negatively associated
with anger toward the transgressor but not with intervention
or punishment as we had predicted. Besides the possibility
that WTR for the transgressor truly does not predict intervention
and punishment, one possibility for the lack of these associations
is that third parties who intervene or punish may
temporarily hold a negative WTR for the transgressor—that
is, they are willing to incur costs to inflict costs. Because our
WTR Scale only went down to 0, any negative WTRs would
have manifested as zeros and thus the variability of the scale
could have been restricted (see Figure S2, which suggests this
may have been the case), which would limit our power to detect
an effect.
These findings were generally consistent across our three
samples and never differed in kind, only magnitude. For intervention
and punishment, the effect of third parties’ WTR for
the victim was constant across all samples, though Japanese
students intervened and punished less often than either U.S.
sample. For anger, there were minor differences among the
samples in the magnitude of the effects of third parties’ WTRs
for the victim and the transgressor, but they remained in the
same, predicted directions in all samples. Thus, we have initial
evidence that our findings are at least somewhat generalizable
beyond a U.S. student population, both to a more general U.S.
population and to Japanese students.
The low model-predicted probabilities of punishment
( .02) we found when WTR for the victim was 0 suggest that
the frequency of third-party punishment has likely been overstated
in the literature that has focused on results from
laboratory-based experimental economics games (for similarly
low rates of punishment in naturalistic settings, see Balafoutas
et al., 2014, 2016). Thus, in addition to providing support for
our hypotheses that third-party anger, intervention, and punishment
vary as a function of the prospective punisher’s WTRs
toward disputants in a conflict, the present study adds to a
growing body of evidence suggesting that direct third-party
punishment on behalf of strangers is not a common feature of
human cooperation (Guala, 2012; Krasnow et al., 2012,
2016; Kriss, Weber, & Xiao, 2016; Pedersen et al., 2013,
2018; Phillips & Cooney, 2005).
We recognize that some might view our design choices here
as restrictive because we limited our scope to conflicts where
there was a direct harm to a victim and only considered intervention
and punishment that occurred in the moment. These
were intentional choices to mimic the types of interactions that
are created in the third-party punishment game (Fehr & Fischbacher,
2004), which typically shows that a majority of people
anonymously engage in immediate, uncoordinated, costly punishment
on behalf of victims. These findings have been generalized
to draw conclusions about humans’ willingness to
directly punish transgressions and what this implies for the evolution
of cooperation in humans (Fehr & Fischbacher, 2003;
Henrich et al., 2010, 2006; for review, see Pedersen et al.,
2018). Our results suggest that people are much less likely to
engage in this type of punishment than a direct generalization
of previous laboratory experiments would imply, though perhaps
future studies will show higher rates of after-the-fact punishment
with low-WTR parties than we found here. Thus, it is
important to note that our data cannot speak directly to a
broader range of social norm violations, some of which could
be likely to evoke punishment. Additionally, we did not focus
on indirect types of retaliation, such as gossip, or other mechanisms
that are likely important to maintaining cooperation and
social norms, such as partner choice.
Additionally, the higher rate of intervention than punishment
we observed here comports well with evidence suggesting
that people prefer alternatives to punishment (e.g.,
helping the victim) when they are available (Balafoutas et al.,
2014, 2016; Chavez & Bicchieri, 2013). It also suggests that
shifting focus beyond punishment could be a fruitful approach
to more fully understanding how third parties respond to conflicts
in the real world. We do notice that the amount of
reported intervention could have been inflated due to our asking
subjects to report whether they had “helped” either person
involved in the conflict, though this was asked after subjects
had already chosen a particular conflict to recall and thus
probably did not bias the choice of event in the first place.
It is also possible that our prompt elicited different recollections
between the U.S. and Japanese samples, which could
explain the difference intervention and punishment rates
between the countries.
This study had some limitations. First, memory limitations
may have prevented people from accurately recalling the
details of past events. For example, subjects’ WTRs for the victims
and transgressors were retrospective; consequently, they
might have been disproportionately reflective of their current
WTRs for the victims and transgressors. Indeed, it is possible
that choosing to intervene or punish increased subjects’ commitment
toward victims and thus could have increased their
WTRs. Although we cannot rule this possibility out given the
nature of our data, we do note that recalled WTRs varied
expectedly as a function of the relationship between subjects
and the victims (see Supplement Material), which suggests that
reported WTRs did at least moderately correspond to the existing
relationships.
Second, subjects’ reports might have been distorted by
socially desirable responding. The low levels of punishment
speak against this concern, but it might have played a role in
intervention responses. The possibility of socially desirable
responding in combination with our exclusion of cases a priori
from situations in which the costs of intervening were very
steep (e.g., conflicts involving guns, multiple transgressors)
leads us to believe that the current study did not underestimate
intervention and punishment frequency. Finally, we did not
code for consolation—attempting to make the victim feel better
after the conflict had ended—and instead treated it as the
same as doing nothing because it had no material effect on the
conflict as it was occurring. Although consolation is certainly
a much less costly helping behavior, it nevertheless may help
the victim and is an important area for future research (De
Waal, 2008).
To conclude, the present investigation moved beyond the
question, “do people punish on behalf of strangers,” to ask,
“when and why do people intervene on behalf of others?” Our
method sampled intervention and punishment decisions
across a wide range of situations and multiple populations,
complementing studies that have examined punishment (and
the desire to punish) in specific real-life situations (Balafoutas
et al., 2014, 2016; Hofmann et al., 2018). Our results converged
with results from these other studies, suggesting that
intervention is much more common than punishment in everyday
life. Perceived welfare interdependence with the victim
emerged as the strongest predictor of intervention and punishment,
signaling its promise as an explanation of involvement
of others’ affairs.