Evaluations of Action and Inaction Decision-makers in Risky Decisions Resulting in Negative Outcomes: Inaction Agents Are Preferred to and Perceived as More Competent and Normative Than Action Agents

Feldman and Albarracín (2017) conducted four experiments using variations of the classic action effect scenarios (Kahneman & Tversky, 1982) and showed that corporate, society, and family action-inaction norms weakened the action effect: As norms shift from inaction to action, action effect is weakened or even reversed into an inaction effect.

Similarly, expectations of agents to act or not act influences agents’ experienced regret when things end badly. For example, negative prior outcomes set expectations for agents to remedy the situation by doing something and taking action, resulting in an inaction-effect, the opposite of action-effect, where inactions are regretted more than actions (Zeelenberg et al., 2002).

Feldman (2020) combined the two experiments into a single design contrasting expectations and social norms, and found that both of these factors affected action-effect. Therefore, action-effect is sensitive to what is perceived as normal, regret is associated with stronger deviations from the normality reference point, and social norms and expectations both play an important role in what people interpret as normal.

In the classic action effect experiment, regret was the key outcome of interest. Follow-up studies attempted to generalize these initial findings to other domains. For example, the omission-bias literature examined action-inaction asymmetries regarding moral responsibility and decisions(Connolly et al., 1997; Yeung et al., 2021), exploring factors such as harm, blame, and intentionality (Feldman & Albarracín, 2017; Hayashi, 2015; Jamison et al., 2020).

Another extension of this work showing the generalizability of action-effect was the regret-action effect, that people tend to associate action rather than inaction when evaluating agents expressing regret (Feldman & Chen, 2019).

Action-effect studies mostly focused on attributions of and experiences of emotions as a result of taking or not taking action, for both self and others. Yet, how are agents who act or not act perceived by observers? Would acting or not acting affect the way an agent is perceived?

Trolley dilemma (Thomson, 1976) studies have traditionally examined moral decisions involving action and inaction that also include trade-offs between saving or hurting different number of lives. For example, the classic paradigm is that inaction would result in the death of five people, whereas taking action would mean the death of one other person. This is one of the most well studied paradigms in moral psychology, examining the manipulation of many cognitive, moral, and/or philosophical factors that may influence a person’s choice. Recent research in that domain has extended to investigate how observers perceive agents who make different choices in the trolley dilemma. Everett et al. (2016) found than an agent who made the deontological choice was perceived as more trustworthy and more moral, and was preferred to an agent that made a consequentialist utilitarian choice. Similarly, Bostyn and Roets (2017) generalized the effect a public good games, and showed boundary conditions regarding trust games.

Our research takes a similar step as these recent studies on the trolley dilemma to try and generalize the action-effect. In two studies, we had the following two goals: 1) replicate the classic action effect, and 2) examine preference and perceived competence regarding action and inaction agents.

In Study 1, we replicated the classic action effect paradigm and added extensions testing preference, competence, and alignment with broad norms for the action and inaction agents. In Study 2, we replicated the inaction effect paradigm (Zeelenberg et al., 2002) and manipulated outcomes prior to the decision (neutral, positive, and negative) with similar added extensions as in Study 1 regarding preference and competence. We also added extensions examining norms to investigate differences between injunctive and descriptive professional norms. Finally, we examined action effect for both regret and joy (Landman, 1987).

Our primary goal was to test whether the classic action-effect extends to evaluations of acting versus non-acting agents. The literature on omission bias showed that individuals tend to blame harm-inflicting agents more when harm was through commission than when through omission (Yeung & Feldman, 2023), suggesting a preference for inaction-omission over action-commission.

N’Gbala and Branscombe (1997) showed that when evaluating individuals whose behavior led to negative outcomes action agents were perceived as worse decision-makers and less wise than inaction individuals. Therefore:

Social norms can be broadly categorized into two types, descriptive norms are one’s own perceptions of what is considered normal whereas injunctive norms are norms that are socially upheld with punitive social actions against deviating individuals (Cialdini, 2003). Feldman and Albarracín (2017) showed that action effect is moderated by both descriptive and injunctive social norms, though these were tested separately in different studies. Feldman (2020) found that social norms moderated the action-effect across three studies with different manipulations of norms. In both articles, the perceived social norms for risky decision-making situations when norms were not manipulated were for not taking action. Therefore:

The classic action effect showing action-inaction asymmetries in regret has been widely replicated (Yeung & Feldman, 2023). Therefore, we expected to replicate the effect in our samples:

The vast majority of studies on action effect focused on the negative emotion of regret, with only a few examining positive emotions such as elation and joy. Landman (1987) modified Kahneman and Tversky’s (1982) original scenario by adding positive consequences and testing for joy, and showed a replication of the action effect regarding regret and an extension of the effect to joy, though. Meaning, that when things turn out favorably, those who acted experience stronger elation than those who did not act. However, the effect was much smaller for joy than for regret. Feldman (2020) tested action effect in both joy and regret and found that while norms impacted regret, they seemed to have little to no impact on joy. We therefore also set out to test both regret and joy and contrast their effects:

The action effect can be weakened and even reversed into an “inaction effect” if expectations are to take action, as demonstrated by Zeelenebrg et al. (2002) who manipulated prior outcomes showing that negative prior outcomes resulted in raising expectations to take action. We therefore expected that this too would generalize to evaluations of action and inaction agents regarding preference, competence, and alignment with norms:

We pre-registered the experiments on AsPredicted (Study 1: #14653, https://aspredicted.org/p8ir6.pdf, March 2018; Study 2: #20841, https://aspredicted.org/46x8w.pdf, March 2019). Materials, data, and code are available on: https://osf.io/a8e4d/

All measures, manipulations, exclusions conducted for this investigation are reported, all studies were pre-registered with power analyses and data collection was completed before analyses.

In Study 1a, we expected to detect a small-sized effect (d = .20) for a one-sample t-test with 95% power, and an α = .05. We planned to recruit 280 participants based on a power analysis, and 339 participants completed the study on Prolific (Mage = 37.54, SDage = 12.01; 93 males, 230 females, 7 others, 9 unknowns).

In Study 1b, we reran the study aiming for a sample of 330 participants and recruited a total of 339 participants from Prolific (Mage = 37.62, SDage = 13.30; 124 males, 208 females, 4 others, 3 unknowns).

Participants were presented with the scenario about an action agent and an inaction agent and were asked to make comparison evaluations. The only difference between Study 1a and Study 1b was that questions were made mandatory in Study 1b. Question order was randomized, and the scenario was presented before every question. Participants then answered a funneling section to allow participants the possibility of commenting on the study (“Please let us know if you have any comments or questions about this study”), and provided demographic information.

The scenario described two protagonists, Paul and George, who made financial decisions regarding an initial investment. Paul represented the inaction actor, and George represented the action actor. The scenario was as follows:

Participants answered questions regarding preference “Which advisor – Paul or George – would you prefer to hire in the future?”, competence “Which advisor – Paul or George – is more competent?”, perceived norms “Which advisor’s behavior – Paul’s or George’s – is more in line with norms?”, and regret “Which advisor – Paul or George – regrets their decision more?”, on the same scale (-5 = Definitely Paul for not switching, 5 = Definitely George for switching).

We summarized descriptive statistics of all measures in Table 1, with summary plots provided in Figures 1 and 2.

Our main hypothesis (H1a) was supported. We found that participants preferred Paul who did not act over George who acted. Our additional hypotheses were also supported, and inaction Paul was perceived as more competent than action George (H1b), and as behaving more in line with the perceived norms (H1c).

Finally, we replicated the classic action-effect findings from Kahneman and Tversky (1982). Inaction Paul was attributed as experiencing less regret than action George (H2).

We summarized all correlations in Table 2. We found positive associations between preference, competence, and norms (r = .27 to r = .48), and negative associations between regret and all other variables (from r = -.01 to r = -.39).

In Studies 1a and 1b, we found a preference for inaction over action agents and that inaction agents were perceived to be more aligned with general norms, more competent, and less likely to regret than action agents facing the same negative outcomes. The question concerning norms was ambiguous, as it only states “which protagonist is more in line with norms”, not explaining what these norms are. In Study 2, we examined two specific types of norms, descriptive norms – how common the protagonist’s behaviors are, and injunctive norms – how likely are the protagonist’s behaviors to be criticized by their peers.

We had multiple goals: 1) manipulating norms, examining the impact of prior outcomes following findings from “inaction effect” - prior negative outcomes shift social norms towards taking action, 2) again confirm the preference for inaction over action in the control condition (no prior outcomes indicated), and 3) explore different types of social norms, both descriptive norms, and injunctive norms.

We expected to detect a small-to-medium-sized effect (f = .15) with 80% power and alpha = .05. Our pre-registered minimum sample size was 432 and planned sample size of 450.

A total of 460 Prolific participants completed the study (Mage = 35.77, SDage = 11.71; 138 males, 308 females, 7 others, 7 unknowns). We assigned participants to one of the three prior conditions: no prior control condition, positive prior outcomes (eliciting inaction norms), and negative prior outcomes (eliciting action norms). The no prior condition was similar to the scenario in Study 1 and constitutes a replication.

Participants were randomly assigned to one of the three prior outcome conditions, presenting some backgrounds from an investment of the two investors.

Then, all participants were presented with a manipulation check of expectations for taking action and changing - “To what extent do you expect Paul and George to change their investment behavior in the future?” (-5 = definitely not changed their behavior, 5 = definitely changed their behavior).

Afterward, an adjusted version of the scenario in Studies 1a/b was presented as follow:

The scenario was followed by questions about preference “Which advisor – Paul or George – would you prefer to hire in the future?”, competence “Which advisor – Paul or George – is more competent?”, descriptive norms “Whose behavior – Paul’s or George’s – is more common among financial advisors?”, injunctive norms “Whose behavior – Paul’s or George’s – will be more criticized among financial advisors?”, regret “Which advisor – Paul or George – regrets their decision more?”, and joy “Which advisor – Paul or George - would have been likely to experience more joy if things had gone well?” on the same scale (-5 = Definitely Paul for not switching, 5 = Definitely George for switching). After completing the study, participants answered an attention check, a funneling section, and provided demographic information.

First, we repeated the same analyses as in Studies 1a and 1b using only the control condition. We presented the results in Figure 3 and Table 3.

We again found support for a higher attribution of preference (H1a) and competence (H1b) to the inaction than to the action investor. We separated norms (H1c) into descriptive and injunctive, yet only found support for an effect on injunctive norms but not on descriptive norms. Participants indicated action-George would be more criticized than inaction-Paul for his choices, but there was no indication of differences regarding whose behavior was more common among financial advisors.

We also replicated the action effect on regret when things ended badly (H2) yet found no support for attribution of joy in the case where things turned out to go well (H3). The effects of preference and competency attributions (ds between 0.05 and 0.56) were comparable to the effects of injunctive norms attributions and lower than effects for regret attributions (ds between 0.58 and 0.96).

We manipulated prior outcomes to see if the reversal of the action effect into an inaction effect in case of negative prior outcomes (Feldman, 2020; Zeelenberg et al., 2002) also extends to preference and attributions of competence, descriptive, and injunctive norms. We conducted an ANOVA and found support for differences between negative priors and the two other conditions for preference, competence, and injunctive norms (Figure 4). For the three variables, the means were below 0 for positive prior and control (no prior) conditions, and above zero for the negative prior condition. We did not find any support for an effect of priors on descriptive norms, regret, and joy. All effect sizes below are shown followed by 95% CIs.

In the no-prior outcomes or positive prior outcomes, the advisor who did not act was preferred over the one who did act (H4), t(303) = -6.69, p < .001; Cohen’s d = -0.38 [-0.50, -0.27]. He was seen as more competent, t(301) = -7.18, p < .001; d = -0.41 [-0.53, -0.30], more in line with injunctive norms, t(303) = 8.68, p < .001; d = 0.50 [0.38, 0.62], but not with descriptive norms, t(303) = 1.21, p = 0.223; d = 0.07 [-0.04, 0.18]. These findings are in line with results from Study 1, and with comparable effect sizes.

In the negative prior outcomes (H5), the advisor who did act was preferred over the one who did not act, t(148) = 2.92, p < .01; d = 0.24 [0.08, 0.40], was seen as more competent, t(148) = 1.98, p < .05; d = 0.16 [0.01, 0.32], but not as more in line with norms, injunctive t(148) = -0.86, p = 0.39; d = -0.07 [-0.23, 0.09]) or descriptive, t(148) = 0.49, p = 0.63; d = 0.04 [-0.12, 0.20]. This result indicates that, when the prior outcomes are negative, there is a preference for action over inaction, contrary to the no-prior or positive prior outcomes.

In addition to the pre-registered analysis above, we conducted additional post-hoc tests, and found another support for a higher preference in the positive prior condition than in the no prior condition, which we reported in the supplementary. Overall, our hypotheses concerning prior outcomes were supported, as we found support for a stronger attribution of preference, competence and injunctive norms to inaction when no prior exists, mirroring the results from Study 1, and for positive prior outcomes. On the contrary, for negative prior outcomes, action was preferred, and attributed more competence. The only exception were the results concerning descriptive norms, with no support for differences between the three conditions.

We presented all correlations in Table 4. Across the three conditions, preference and competence were positively correlated and negatively with regret and injunctive norms. In the negative prior condition, descriptive norms were positively correlated with competence and not with preference. We found no support for an association between joy with any of the variables.

We summarized our findings and support for the hypotheses in Table 5. In all studies, the inaction protagonist was preferred and found more competent than the action protagonist. Inaction was seen as more in line with norms in Studies 1a/1b, and in Study 2 we found a similar pattern for injunctive norms (less subject to criticism) but not for descriptive norms (more common). We found support for the classic action effect in that action decisions leading to negative outcomes were attributed more regret than inaction in both studies.

In Study 2 we manipulated norms as a proxy of prior outcomes, whereby negative prior outcomes were used to direct norms toward taking action. Negative prior outcomes led to a reversal of the action effect into inaction effect for preference and competence compared to the control and positive prior conditions, but no effect was found for regret. Overall, our results indicated that participants trusted more Paul’s decisions to not act than Georges’ decisions to act.

We extended the classic action-effect to examined preference for and perceived competence of action and inaction agents. We found that when outcomes were negative observers indicated preference for inaction over action agents, and perceived inaction as a more competent and norm-aligned behavior than action. Effects ranged from d = 0.22 to d = 0.48.

We also replicated the classic action effect on regret (Kahneman & Tversky, 1982) in both Studies 1a/1b and in Study 2 control condition, with a strong effect (d ~= 0.77) comparable to what was previously found in the literature (Feldman, 2020).

In Studies 1a and 1b, we asked a general question about perceived norms - “which advisor is more in line with norms?”. In these studies, and that framing, norms were related to competence and preference and the action effect applies. In Study 2, we divided this question into two types of norms, injunctive norms, asking which of the investors would be more criticized, and descriptive norms, asking which of the investors’ actions are more common (Cialdini, 2003; Feldman & Albarracín, 2017). We found support for our hypothesis for injunctive but not for descriptive norms, suggesting that action resulting in negative outcomes is perceived to elicit more criticism than inaction. Also, descriptive norms and injunctive norms were only weakly related, with descriptive norms going in the opposite direction to what we expected.

These findings suggest the need to clearly define what norms are being measured, and then to differentiate between the two types of norms when examining their associations with and impact on attributions and affect. Feldman and Albarracín (2017) showed that the impact of social norms on action is complex and depends on the reference point to which norms refer to and their importance for the individual (e.g., corporate norms, coworkers’ behavior, family norms, societal norms). Therefore, it is likely not only the type of norms, descriptive or injunctive, but also the reference group to which these norms are compared against. Future studies may aim to first measure the different types of norms and the different possible reference groups, and then to try and manipulate those in order to examine their impact on the action-effect.

In addition, it is possible that norms depend on the context. For example, our context was one of a financial investment, whereas the context in Zeelenberg et al. (2002) inaction-effect demonstration was in the context of sports. The two contexts differ, in their visibility, in their competitiveness, in the level of risk, and in other factors. Any of these factors may affect the way in which social norms impact the action effect, and therefore the variations of that impact in the different categories of norms. Overall, we see much potential for further theoretical developments regarding norms in the action effect by investigating types of norms, reference points, and context.

Action and inaction were widely studied in the judgment and decision-making literature as cognitive biases, or “systematic” asymmetries in judgements regarding protagonists that either acted or not acted when faced with a choice (Kahneman & Tversky, 1982). Studies examined evaluations or attributions of emotions (e.g., regret; Zeelenberg et al., 2002) and/or cognition (e.g., counterfactual thinking, N’Gbala & Branscombe, 1997) in hypothetical scenarios (for a review, see Feldman et al., 2020).

The attribution of regret, preference, or competence can also be seen as social evaluations, the assignment of positive or negative values to a behavior in social interactions (Abdai & Miklósi, 2016). Traditionally, action/inaction biases were focused on negative context – attributions of regret, responsibility, or intentionality in order to understand when someone is held accountable for a bad outcome. In our study, we changed the perspective to a positive social evaluation, the evaluations of trustworthiness and competence.

Our findings raise an interesting question as why inaction is attributed more trustworthiness, a positive evaluation, despite the same bad outcome? One possibility is that action may signal that someone may have taken too big of a risk, and therefore might be less trustworthy in similar risky decision. It is also possible that it is the social comparison between the acting and non-acting agents on a vignette in which the only provided factor is action to make the link between action and trust. Therefore, one possible line of future research would be to contrast between and within designs, such that agents would be rated independently to examine effects when there is no social comparison. Another potential line of research would be to assess perceived risk in that situation and rate each agent on perceived risk-taking tendencies, and examine how those are associated with trustworthiness.

In situations involving positive or neutral prior outcomes we found that the expected pattern that the inaction investor was perceived to be more competent than the action investor. In situations involving negative prior outcomes which elicit social norms for taking action, we found a reversal of the effect, in which the action investor was perceived to be more competent than the inaction investor. Curiously, our attempted replication of the inaction-effect to find this pattern regarding regret failed, even though our extension worked. It is possible that inaction norms in that context are so strong that prior outcomes do little to affect regret over a single bad decision. In recent research (Dorison et al., 2021), decision-makers who chose to invest more funds after a prior investment were perceived as warmer, more competent, and more confident than decision-makers who did not, and the results were the contrary in the absence of prior investment. Thus, researchers concluded that honoring sunk costs (or already having invested money in stocks) confer reputational benefits, which we confirmed in our study. Still, inaction-effect has been replicated multiple times with similar and other scenarios, and so more research is needed regarding the specifics of the current context that may have weakened the inaction-effect.

Investigating the roles of emotions in this preference, the effect size of joy was far weaker than the effect found for regret, supporting similar findings reported by Feldman (2020). Regret and joy also seemed to be unrelated to each other (r = .00, 95%CI [-.10, .11]).

A possible explanation for these differences may lie in effects such as “bad is stronger than good” (Baumeister et al., 2001) “bad is freer than good” (Feldman et al., 2016), and the Side-Effect Effect (Knobe, 2003). These reflect a similar idea that negative outcomes are more impactful, they elicit stronger attributions of responsibility, causality, and intent, and therefore likely to also elicit stronger emotions.

Another possible explanation may lie in our procedure. In the Elation effect for action and inaction (Landman, 1987), the regret/joy asymmetry was shown by reversing the scenarios. Half the participants were presented with a negative outcome and a regret question, and half with an opposite positive outcome and a joy question. In our study, this type of manipulation would have meant to randomly vary our scenario so that half the participants would instead read about financial advisors who won more money by acting and not acting, with a question about attributions of joy. We instead first provided all participants with the negative outcome scenario, and then asked participants to imagine joy felt had things turned out well. This procedure is different, and arguably more complex and cognitively effortful. It is possible that participants experienced difficulty in making that cognitive reversal, and so future studies can try and contrast the two procedures, and also provide a more rigorous test with an added option of some participants first being presented with a positive scenario joy and then asked to answer a regret question when things turned out badly.

Several constraints limit our possibility to generalize the effect. Our population consists of Prolific users and we used hypothetical scenarios and not in real environment decision-making. Prolific has been found to be a source of reliable data (Palan & Schitter, 2018; Pe’er et al., 2021) and we implemented various measures such as comprehension checks to ensure data quality, supported by evidence mostly consistent with previous literature. Yet future research can build on these findings to test robustness and generalizability to real-life decisions and wider populations.

Our scenarios were modified versions of the original study from Kahneman and Tversky (1982). This scenario has been replicated, extended, and modified many times, with both old and recent research testing various versions of the scenarios and extending to other domains to test the generalizability of the action effect (e.g., the Moral Sense Test, Cushman et al., 2006; sports coach, Feldman, 2020; Zeelenberg et al., 2002; factory rebate of a car, Tykocinski et al., 1995). We aimed to try and take the action effect in a different direction, to generalize the action effect to look at new dependent variable from different perspectives, examining preference, competence, and norms and rating perceptions of action and inaction agents rather than action or inaction behaviors. We see much potential in further extensions of this domain to examine other dependent variables and perspectives looking at agents and interactions rather than singular decisions.

We found support for an extension of the action effect to evaluations of agents looking at preferences, competence, and social norms. Evaluating agents facing negative outcomes over their decisions, people show preference for inaction, and find inaction more competent and more in line with norms, compared to action.

The author(s) declared no potential conflicts of interests with respect to the authorship and/or publication of this article.

None.

We thank Anthony M. Evans for initiating and guiding this project, and conducting the pre-registration, and data collections. We thank Philippe van de Calseyde for his valuable feedback regarding experimental designs and our pre-registrations in early stages of the project.

Please see the table. The pre-registration authors were primarily responsible for the empirical methods, results, and first drafts, and therefore were essential. All authors approved the final manuscript for submission.

All materials, data, and code are available on: https://osf.io/a8e4d/