Introduction

How do we make decisions in the face of conflicting information? Imagine a doctor recommending a treatment as highly likely to improve a patient’s medical condition. Seeking a second opinion, the patient turns to another physician, who recommends the same treatment but provides a much lower estimate of success likelihood. This type of uncertainty – conflicting information about outcome probability – frequently emerges in everyday decisions. How do decision-makers interpret such conflicting information, and what attitudes does it evoke?

A broad body of literature in the cognitive sciences has demonstrated that conflicting information, broadly defined, has adverse effects on cognition. Cognitive dissonance1 – a conflict between two perceptions – can lead to sub-optimal behaviors, such as changing the perceived desirability of products to justify, post-hoc, purchasing decisions2. The Stroop task presents a conflict between different dimensions of a stimulus (e.g., the word ‘green’ colored in yellow), causing reduced accuracy and increased reaction times in performing the task3. Similarly, in the case of the Flanker effect, incongruent information leads to slower reaction time in perceptual4 and higher-level cognitive tasks5. Importantly, most previous literature focused on the interference of informational conflict in the processing of an objective state (e.g., the true color of the yellow letters spelling the word ‘green’). Here, in contrast, we focus on subjective value-based decisions, and how conflicting information about outcome likelihood affects these decisions. Given their ubiquitous presence, elucidating how humans interpret informational conflicts is essential for understanding the cognitive processes involved in decision-making under uncertainty, with direct implications for our everyday lives.

Information conflict may be a special case of ambiguity6,7, a highly studied type of uncertainty, where outcome probabilities are not precisely known. Ambiguity exists, for example, when the success likelihood of a medical treatment is between 25% and 75% Ambiguity is often considered aversive7,8, although the degree of aversiveness depends on the source and context of ambiguity9,10 and varies substantially across individuals11. Like ambiguity, conflict encompasses multiple potential future outcomes. However, while ambiguity suggests a spectrum of possibilities, conflict expresses mutually exclusive situations.

How do decision-makers interpret and respond to conflict? One possibility is that they equate conflict with ambiguity. For example, if one source states that the probability of a particular outcome is 25%, and another source asserts it is 75%, a decision-maker might interpret this as indicating a range of potential probabilities between 25% and 75%. Another possibility is that conflict is an irreducible form of uncertainty. If uncertainty attitudes are driven by factors that are more pronounced with conflict, for example cognitive dissonance, conflict and ambiguity may be associated with distinct individual attitudes12,13.

Only a handful of studies have explored how conflicting information is processed and how it affects choice. Smithson14,15 presented participants with scenarios involving two pieces of evidence that either agreed or conflicted with each other, and found that participants were generally more averse to conflicting scenarios. For example, as members of a hypothetical jury, participants indicated that they would prefer to receive testimonies from two witnesses agreeing that a robber’s car was either green or red, over testimonies from one witness stating the car was red and another stating it was green. Similarly, risk professionals, insurers, required a higher premium when facing a conflict, compared to ambiguity16,17. Conversely, Baillon et al.18 estimated uncertainty attitudes by quantifying the certainty equivalents of various forms of uncertainty and found that ambiguity, but not conflict, was aversive. Chew et al.19 addressed similar topics regarding conflict and ambiguity attitudes (termed there two-point and interval ambiguity, respectively). Using a valuation task, the researchers estimated participants’ conflict and ambiguity attitudes, but did not directly contrast the two types of uncertainty.

Taken together, previous studies generally suggested that conflict is aversive, but the findings supporting this inference are inconclusive. Furthermore, many prior studies employed experimental designs with a limited number of questions, restricting the ability to quantitatively estimate individual aversion to conflict and ambiguity. In this study, we aim to compare the responses to conflict and ambiguity within individuals by quantitatively characterizing individual conflict attitudes, using a decision-making task. To achieve this, we developed a novel paradigm in which participants make choices under systematically manipulated levels of ambiguity and conflict. We assessed conflict attitudes using two complementary approaches:(1) comparing participants’ preferences for conflict and ambiguity when each was contrasted with a certain outcome, and (2) directly contrasting conflict with ambiguity (see below, Tasks 1 and 2).

Interestingly, our findings reveal a shift in preference between these two methods. While participants were generally indifferent or exhibited a slight preference for conflict over ambiguity when each was separately contrasted with a certain option, they strongly preferred ambiguity over conflict when the two were directly compared.

Results

We recruited online participants for a behavioral experiment which consisted of several parts (Fig. 1.a). 237 participants met our inclusion criteria (see Methods) and are included in the analysis. Using a decision-making task, we assessed each participant’s individual conflict and ambiguity attitudes through two distinct methods (see below).

Fig. 1
figure 1

The conflict task. a. Experimental design. In four consecutive blocks, participants first chose between lotteries contrasting certain outcomes with conflicting, ambiguous, and risky outcomes (Task 1, see Fig. 1.b). Next, in Task 2, participants chose between conflicting and ambiguous options, with winning chances centered around 15% (as shown in the figure), 50%, and 85% (see Table S2). Finally, participants rated their anticipated regret for different outcomes and completed a personality test. b. Choice under conflict, ambiguity, and risk. The lotteries presented in Task 1 contrasted certain fixed outcomes with different types of uncertainty. Different reward amounts (shown in the gray box) were presented in uncertain options. Winning chances in conflicting and ambiguous trials were centered around 50%, with different probability ranges (presented in increasing magnitude from left to right).

Full size image

Additionally, we examined the relationship between conflict attitudes and various individual factors, including personality traits (Supplementary Text 1), anticipated regret (Supplementary Text 2), and subjective probability perceptions (Supplementary Text 3). However, these factors were not substantially predictive of conflict attitudes (Supplementary).

Task 1: Indirect comparison of conflict and ambiguity attitudes

In the first part of the experiment, participants made a series of 2-alternative forced choices between a certain small reward (100% chance of receiving $5) and an uncertain option that offered a chance for a greater outcome (for example, $12; Fig. 1.a, Table S1), but also a chance for a zero outcome. Uncertainty was framed either as conflict (e.g., “the chance of winning $12 is either 25% or 75%”, Fig. 1.b top), or ambiguity (“the chance of winning $12 is between 25 and 75%”; Fig. 1b middle). Each conflict level was matched by an ambiguity level spanning the range between the two probabilities (Fig. 1.b). Because behavior under both ambiguity and conflict is also influenced by decision-makers’ attitude to uncertainty with known probabilities (“Risk”, see Methods), we included risk trials, where the uncertain option offered a precise 50% chance of winning a reward (Fig. 1.b bottom). The same 21 reward magnitudes were presented in the Conflict, Ambiguity, and Risk conditions (Table S1), allowing a clear comparison of behavior under conflict and ambiguity. To incentivize participants to choose according to their true preferences, participants were informed that at the end of the experiment, they would receive a bonus payment determined by their choices in the task (see Methods).

In the risk trials, we found that participants were more likely to choose risky options that offered greater rewards, suggesting that participants were sensitive to task conditions and that their choices were aligned with subjective value maximization (Fig. 2.a).

Fig. 2
figure 2

Proportion of choices in uncertain options. a. Proportion of choices in risky lotteries. As expected, preferences for risky lotteries increased with the reward offered by the lotteries, suggesting participants understood and were sensitive to the incentives offered by the experiment. b. Proportion of choices of conflicting and ambiguous lotteries. For each uncertainty level, the red and blue bars summarize the mean proportion of choices of conflicting and ambiguous lotteries, respectively. The red and blue lines connecting the bars show that across uncertainty levels, conflicting lotteries are slightly but consistently selected more frequently than ambiguous lotteries. In both lottery types, greater uncertainty levels are associated with lower choice proportion (observe the slope of the red and blue lines), suggesting a general aversion to conflict and ambiguity. Each circle represents the choice proportions of a single participant. In all panels, error bars are standard errors.

Full size image

Observing the overall proportion of choices of conflict and ambiguity trials, participants chose the conflicting alternative in 46.9% ± 1.7% (mean ± standard error) of the trials, significantly more frequently than they chose the ambiguous alternative, 42.7% ± 1.7% (2-way ANOVA, uncertainty level [10..90, 25..75, 40..60] X uncertainty type [conflict, ambiguity]; uncertainty type main effect: p = 0.003, DF = 1; Fig. 2.b). Notably, this preference was expressed across all uncertainty levels (compare the red and blue lines in Fig. 2.b).

Additionally, under both conflict and ambiguity, greater levels of probability spread were associated with a lower likelihood of choosing the uncertain option (bootstrap test, p < 0.001; observe the slope of the red and blue lines in Fig. 2.b), suggesting general aversion to conflict and ambiguity. 

The preference for conflict compared to ambiguity was observed not only at the population level but also across individual participants. A small majority of the participants, 52.3%, chose conflict more frequently than ambiguity, 16.9% chose the two in equal proportions, and the remaining 30.8% chose the ambiguous options more frequently than the conflicting ones.

Because the rewards and probability ranges of the conflicting trials were replicated in the ambiguity trials, such a comparison of raw proportions is valid. However, to estimate conflict and ambiguity attitudes, choice proportions cannot be used on their own. This is because decision-making in conflicting and ambiguous trials depends both on the subjective interpretation of the presented probabilities and the attitude to this interpretation. For example, a participant may perceive a 25%-75% probability range optimistically, as representing a 75% success chance, or pessimistically, as representing a 25% success chance. Independent of these perceptions, a lottery comprising a 25% or 75% success chance may be attractive or unattractive based on participants’ risk attitudes.

To disentangle conflict and ambiguity attitudes from risk attitudes, we corrected the proportions of conflicting and ambiguous choices by subtracting the proportion of choices in the risk condition (Methods). The corrected attitude is a score ranging from -1 to 1, corresponding to maximal aversion to maximal seeking, respectively, of conflict or ambiguity. A corrected score of 0 reflects conflict or ambiguity neutrality. Participants’ corrected-conflict attitudes were significantly correlated with their corrected-ambiguity attitudes (r = 0.62, p < 0.001; Fig. 3.a). Additionally, there was a weak but significant correlation between the proportion of choices in risky options and conflicting (r = -0.21, p = 0.001) and ambiguous (r = -0.20, p = 0.002) options, such that greater risk-seeking was associated with greater ambiguity and conflict aversion.

Fig. 3
figure 3

Conflict and ambiguity attitudes comparison. a. Corrected attitudes, indirect comparison (Task 1). Proportions of choices of ambiguous and conflicting alternatives are highly correlated. Each circle represents a single participant’s proportion of choices of conflicting and ambiguous options, corrected by the participant’s proportion of choices of risky options. Shaded area is 95% confidence interval around the dashed blue regression line. A value of 0 (dashed black lines) represents indifference. b. Preference shift. Each colored line connects the conflict attitudes of a single participant in Task 1 (indirect comparison) and Task 2 (direct comparison). Green lines depict participants who were more averse to conflict in Task 2, compared to Task 1 (82.7% of the participants). Purple lines depict the remaining participants, who were more conflict-averse in Task 1 compared to Task 2. Colored area is a normalized, smoothed histogram of participants’ conflict attitudes. Black line connects the population mean. Dashed black line represents indifference between conflict and ambiguity. Error bars are standard errors.

Full size image

Task 2: Direct comparison of conflict and ambiguity

So far, both conflict and ambiguity have only been contrasted with certain options. Next, we compared the two uncertainty types directly. Participants were presented with a series of choices between conflicting and ambiguous options. Within each trial, rewards and probabilities were identical, with probabilities framed as conflict in one option, and as ambiguity in the other (Table S2).

Surprisingly, while in Task 1 we found a mild preference for conflict over ambiguity, here conflict was strongly aversive (Fig. 3). Overall, in Task 2, participants chose the conflicting alternative in only 22.7% ± 1.7% of the trials, significantly less than the ambiguous one (bootstrap test, p < 0.001). This shift in preference was expressed not only at the population level but at an individual level as well (Fig. 3.b). We found that 82.7% of the participants who were more averse to ambiguity in Task 1 (the indirect comparison) reversed their preference in Task 2 (the direct comparison), becoming more averse to conflict. Additionally, regardless of the choices in the first part of the experiment, in Task 2 only 16.0% of the participants chose the conflicting alternative more often than the ambiguous one.

One possibility is that the shift in preference between the two tasks resulted from the change in choice context. An alternative possibility is that preferences for conflict and ambiguity depend on the center of the probability range. Unlike the probability ranges in Task 1, which were all centered around 50%, in Task 2, participants encountered several probability centers: 15%, 50%, and 85%. To test this hypothesis, we repeated the analysis above but considered only the subset of trials of Task 2 with probabilities centered around 50% (Fig. S5). We found that the shift in conflict preference was even more pronounced in this case, with 85.1% of participants presenting a higher preference for conflict in Task 1 (2-tailed t-test, t(220) = 18.873, p < 0.001).

Finally, we tested the role of the midpoint of the probability range on conflict attitude. Task 2 included decisions with three midpoints—15%, 50%, and 85%—each combined with different probability spreads of ± 5%, ± 10%, and ± 15% (e.g., 15% ± 5% represents a conflict between 10 and 20% success probability). Theoretical considerations we derived from prospect theory (Fig. S4, a and b) suggested that preference for conflict should increase with the probability mid-point (i.e., C[15%]≼C[50%]≼C[85%], with C[P] being the mean conflict attitude for lotteries centered around P). Our data, however, did not support this hypothesis (Fig. S4.c). In an exploratory analysis, we tested the hypothesis that conflict attitudes in Task 2 would be higher in ‘extreme’ midpoints (15% and 85%), compared to centered midpoints (50%). We found that participants chose conflicting options in 25.3% of the extreme-midpoints trials, significantly more than in trials with centered midpoints, 18.2% (2-tailed t-test, t(236) = 5.211, p < 0.001).

Discussion

Conflicting information is inherent in our information-rich environment. On its own, informational conflict does not imply any inherent good or bad quality. Our results, however, suggest that, depending on context, conflicting information can induce substantially negative bias.

Context-dependent conflict preferences

Our primary objective was to compare conflict and ambiguity attitudes. We found that the relationship between the two is context-dependent. When considered on their own, conflict and ambiguity attitudes were, at the population level, indistinguishable. Such a finding suggests that a conflict scenario – e.g., one information source reporting a "75% chance of a successful investment" while another reports only a "25% chance" – would be interpreted as implicating a probability ranging between 25% and 75%. Surprisingly, however, when contrasting ambiguity and conflict directly, we observed an overwhelming aversion to conflict.

Our results are consistent with context-dependent preference reversal of conflict-related decision-making. Classical preference reversal occurs when a decision-maker chooses one option over another but places a lower price on the chosen option20,21. Our design does not allow us to demonstrate such a strong version of preference reversal, since the two tasks used different lotteries, with only a small subset overlapping. Both designs, however, used the same outcome range ($5-$120), and the aversion to conflict was consistent across the three central probabilities used in Task 2 (Fig. S4). Furthermore, when considering only the subset of trials centered around the same probability midpoint in both tasks, 50%, the preference shift became even more pronounced. We thus believe that the results are more consistent with preference reversal than with preference shift based on the ranges of amount or probability.

What can explain such a shift of preference? One potential explanation could be that decision-makers are more averse to conflict compared to ambiguity, but only marginally. For instance, to switch from a certain $5, a decision maker may require high compensation when faced with ambiguity (e.g., $40), and slightly higher compensation (e.g., $42), when faced with an equivalent conflicting scenario. Such an attitude structure would result in similar behaviors in conflicting and ambiguous scenarios when presented separately, but imply a clear preference for ambiguity when directly contrasted with conflict. This explanation is inconsistent, however, with the mild preference we found for conflict compared to ambiguity when the two were presented separately (Fig. 2.b). Instead, we propose a theoretical framework that accounts for the shift of preference observed between the two parts of the experiment.

We suggest that conflict is indeed highly aversive. We propose that, if given leeway for interpretation, decision-makers tend to interpret conflict as ambiguity. This policy would explain the similarity between conflict and ambiguity attitudes when presented separately—participants simply behaved consistently across conflict and ambiguity trials, which they interpreted as the same conditions. When conflict and ambiguity are directly contrasted, however, participants could no longer interpret conflict as ambiguity and were forced to recognize the conflict, which they were averse to. This may explain why strong conflict aversion emerged only when the incongruence between conflicting options was contrasted with the congruent, ambiguous alternative. This explanation is reminiscent of the “comparative ignorance” hypothesis, according to which ambiguity aversion is produced by comparison with less ambiguous, in particular risk-associated, scenarios22. In this perspective, our results may reflect a “comparative conflict”, where aversion to conflicting information is produced by its comparison with less conflicting, in particular ambiguous, scenarios.

Association with regret, subjective probability, and personality traits

In addition to comparing conflict attitudes measured by the two experimental approaches described above, we tested whether conflict attitudes were associated with additional factors. We found a weak correlation between conflict attitudes and conscientiousness (Supplementary Text 1) and self-reported anticipated regret (Supplementary Text 2). These null or insubstantial findings align with our interpretation of conflict as an independent and idiosyncratic trait not underlined by other factors. Future studies should expand on our individual differences approach to explore other personality factors that may correspond with conflict attitudes.

Due to the shape of the subjective probability function, we hypothesized that a higher conflict midpoint (centered around 15% < 50% < 85%) would be associated with greater conflict preference (C[15%]≼C[50%]≼C[85%]). Our results did not support this hypothesis. However, an exploratory analysis revealed that conflicting options with midpoint probabilities of 50% were significantly more aversive than conflicting options with extreme midpoint probabilities, 15% and 85%. These findings are consistent with previous work showing that ambiguity attitudes are determined both by the probability range and its midpoint23. To further study how conflict attitudes are shaped by probability-midpoints and probability-ranges, future studies could systematically manipulate both factors to replicate these findings.

Limitations, implications, and outlook

To compare ambiguity and conflict attitudes, we utilized a binary decision-making task that contrasted certain with uncertain outcomes, but other methods could have been used toward the same goal. For example, Abdellaoui et al.24 quantified conflict attitudes using model-based approaches, and estimation of the certainty-equivalence of conflicting options (the premium needed for decision-makers to become indifferent between conflicting and certain options). While the aversion to conflict we identified is consistent with previous literature, the measurement of uncertainty attitudes can vary depending on the method of elicitation21,25. Future research could strengthen and generalize our findings by replicating them using alternative approaches to measure conflict attitudes.

We have opened with a common scenario in which conflicting information is likely to arise: seeking a second opinion26. Our results suggest that, as a general rule, decision-makers are averse to such informational conflicts, and that the greater the conflict, the greater the aversion. While we explored several cognitive bases for this aversion, additional unexplored mechanisms may contribute as well. For example, conflicting information may be perceived as generally non-informative, non-credible, or untrustworthy27,28,29.

A broad literature investigates the perception and utilization of advice and the role of uncertainty representation in advice-giving and receiving30,31,32. While we did not directly aim to test the effects of conflicting information on advice-taking, our data bears directly on how conflicting advice may influence subsequent actions. Consider a decision-maker who is seeking a second opinion regarding some action they wish to take, and that the second opinion they receive conflicts with the first. Interestingly, our data suggest that while each of the opinions in isolation might have been sufficient to prompt the decision-makers to opt for the contemplated action, the combined effect of conflicting opinions may discourage them from taking it. Importantly, in our experimental paradigm, both opinions were presented at once, while in many real-world scenarios, conflict is induced sequentially. Future studies may examine how conflicting advice shapes decision-making and the potential order effects in conflict generation.

More broadly, our data suggest important considerations for the communication of conflicting information in light of how it is likely to be perceived. Accurate communication of a conflict can have substantial implications. However, our results suggest that human cognition is averse not only to conflicting prospects themselves but also to the mere perception of conflict. When given room for interpretation, our results suggest that decision-makers will inadvertently substitute conflict with ambiguity. As a result, if a perception of conflict is important, communicators should be very deliberate in representing the incompatibility of the given scenarios. In such cases, however, conflict communicators should also be aware that, irrespective of the objective value of the different propositions, the mere presence of conflict is likely to evoke strong subjective distaste.

Methods

We pre-registered the task and its analysis at: https://aspredicted.org/Q2X_Y9D

Participants, exclusion criteria, and compensation

Participants were recruited through the online crowd-sourcing platform Prolific (https://www.prolific.co/) to complete a 30-min survey. All participants provided informed consent before beginning the survey, in accordance with the ethical guidelines approved by the Yale University Human Investigation Committee. The experimental task was preceded by an instructional tutorial that included simple directives (e.g., "The task presents two options. To test the certain outcome option, please choose the ‘certain outcome’ button"). The instructions included both a presentation and an explanation of the types of uncertainty used in the experiment. For conflicting lotteries, participants were shown an image of two individuals, one stating there is a 10% chance of receiving the outcome, and the other stating a 90% chance. The instructions explained that, at the end of the task, one trial would be randomly selected for a potential bonus payment. If the conflicting option was chosen in that trial, the outcome would be provided with either a 10% or a 90% chance. Similarly, for ambiguity presented with 25% and 75%, the instructions stated that the reward would be delivered with probabilities ranging between 25 and 75% (See Supplementary Text 4).

Participants who did not successfully comply with these instructions were not presented with the following experiment. These participants and all additional participants who started but did not complete the survey were not included in the analysis.

We recruited 253 participants who completed the survey. From this sample, we excluded participants who failed our attention checks. The task included 3 catch trials that could have been answered correctly by attending to their simple instructions. For example, one catch trial presented participants with the question “What is the product of 7 multiplied by 2?” followed by the instruction “Please pick the third option (regardless of the content)”. Participants who failed to respond correctly to more than one of the catch trials were excluded from the analysis. An additional criterion for exclusion, which we did not anticipate and did not include in our pre-registration, was prolonged overall participation time. The average task completion time was 34.6 min (standard deviation 32.3 min). Upon examining the data, we observed that a small proportion of the participants took substantially more time to complete the experiment. Therefore, we excluded participants whose completion time exceeded two standard deviations above the mean (99.0 min). Based on this criterion, we excluded 16 participants (6%). Including these participants in the analysis does not qualitatively change the results. After exclusion, the ultimate participant sample consisted of 237 valid participants: 109 self-identified as males, 125 as females, and 3 did not specify their gender.

Participants were provided with a fixed USD 5 compensation for completing the study. Additionally, we informed participants in advance that one of their experimental trials will be randomly selected to provide an additional bonus equivalent to the realized outcome of that trial through a conversion rate of 10%. For example, participants who chose the $120 option in risky trials, and for whom that trial was randomly selected, were granted a bonus of $12 with a 50% chance.

Experimental task

The task was conducted using the online Qualtrics survey platform. In the first part of the task, participants were presented with a binary choice between two alternatives. In these trials, one alternative offered a guaranteed low outcome (100% chance of receiving $5), and the other alternative offered a greater outcome ($6 to $120) that would be provided with some uncertainty. Uncertainty in the task was presented in one of three frames (Fig. 1, Table S1). In risky trials, participants could win a higher outcome (e.g., $8) with a 50% chance or, with a complementary 50% chance receive no outcome ($0). Conflicting trials consisted of uncertainty that was presented as two incompatible probabilities (e.g., either 25% or 75% chance of winning $8, Fig. 1). Conflicting trials included three ranges of probabilities, all centered around 50%: [40% or 60%], [25% or 75%], and [10% or 90%]. Finally, ambiguous trials were identical to conflicting trials, but rather than presenting two contrasting probabilities, presented a range of probabilities (e.g., 25%-75% of winning $8). Each combination of uncertainty frame and probability range was presented with the full set of the following 21 reward amounts: 5, 6, 7, 8, 10, 12, 14, 16, 19, 23, 27, 31, 37, 44, 52, 61, 73, 86, 101, 110, and 120 U.S. dollars (for similar experimental approaches see:33.

This resulted in a total of 147 trials: 21 risky trials, 63 conflicting trials, and 63 ambiguous trials. Each uncertainty type (risk, conflict, and ambiguity) was presented as a single block. Our main goal was to compare conflict and ambiguity attitudes. Since the conflict and ambiguity blocks featured the same range of outcomes and uncertainties and differed only in the framing of uncertainty, we designed the experiment to include an intermediate decision task separating the two. Specifically, the risk block was always presented between the conflict and ambiguity blocks, with the order of conflict and ambiguity randomized: with a 50% chance, the block order was Conflict, Risk, and Ambiguity; or, alternatively, Ambiguity, Risk, and Conflict. Within each block, the order of trials was randomized.

Similar to the method of Holt and Laury34, this experimental setup allows for the evaluation of uncertainty attitudes by estimating participants’ indifference point—the magnitude of the uncertain outcome that is subjectively equivalent to the certain outcome (e.g., that a certain $5 is subjectively equivalent to a 50% chance of receiving $23). Since participants’ choices may be stochastic, a useful quantitative approximation of uncertainty attitudes is the proportion of choices favoring the uncertain option, with more uncertainty-seeking participants exhibiting higher proportions of such choices.

The second part of the task included 45 trials of direct comparisons between conflicting and ambiguous alternatives. Each such trial contrasted conflicting and ambiguous options presented with the same rewards and probability ranges (e.g., [25% or 75% of winning $8], vs. [25%-75% of winning $8]). Unlike the lotteries in the first part of the experiment, whose probabilities were centered solely around 50%, lotteries in this block were centered around 15%, 50%, or 85%. Each of these probability centers was presented several times with three probability spreads: ± 5%, ± 10%, and ± 15% (Table S2).

Following these decision-making trials, participants rated their anticipated regret in response to hypothetical scenarios. This assessment was designed to evaluate both general regret as a trait35, and regret specifically associated with conflict. We presented participants with scenarios involving certain, ambiguous and conflicting options and asked them to rate on a scale of 1–100 the level of regret they anticipate experiencing in response to a specific choice and outcome (Fig. S3). For example, we asked participants to rate the level of regret they anticipate experiencing if rather than a certain $5, they chose a conflicting option indicating 10% and 90% chance of winning $27, and subsequently found out that they had won $0.

Monotonous preference in probability spread

In conflicting (and ambiguous) trials, we observed that a greater probability spread was chosen less frequently compared to a narrower probability spread. Namely, as reflected in the blue and red lines of Fig. 2.b, that p10..90 < p25..75 < p40..60, where pi..j is the choice probability of conflicting or ambiguous lotteries with probabilities i and j. To estimate the significance of this finding we performed a bootstrap analysis. Our null hypothesis was that each participant has a specific conflict attitude that does not increase with a narrowing probability thread (i.e., p10..90 = p25..75 = p40..60). To generate potential realizations of this hypothesis in a data-driven fashion, we considered for each participant their overall proportion of choices in the conflicting (ambiguous) alternative across all probability spread, poverall = (p10..90 + p25..75 + p40..60)/3. We then simulated for each participant a specific proportion of choices in each conflict (ambiguity) level by drawing several choices from a binomial distribution (with 21 trials, and each participant’s empirical poverall). Then, for each participant i, we considered a “monotonous score” (msi) which reflects the extent to which increasing probability ranges induce lower preference: msi = (p40..60 – p25..75) + (p25..75 -p10..90). Finally, we averaged msi across all participants to a single value of MSsimulated, and repeated the process to generate a distribution of MSsimulated values. Under the null hypothesis, MSsimulated distribution is centered around 0, and we report as p-value the proportion of simulated MS scores that are greater than the empirical MS score (the true sum of the difference in neighboring bars in Fig. 2.b).

Conflict and ambiguity attitudes

Corrected conflict (ambiguity) attitudes are defined as the proportion of choices in the conflicting (ambiguous) alternative, from which the proportion of choices in the risky options is subtracted. Consider a trial with a 25% vs. 75% chance of winning $10. A conflict-averse (“pessimistic”) participant may view this trial as providing a mere 25% chance of winning the $10. The participant may still elect to choose it if 25% of winning $10 has a high enough subjective value for that risk-seeking participant. Vice-versa, a conflict-seeking (“optimistic”) participant may subjectively interpret the lottery as offering an attractive 75% chance of winning the $10 but still choose to forgo it if the participant is risk averse.

The corrected conflict (ambiguity) attitude accounts for individual risk attitude. Consider the participant who interpreted conflict and ambiguity trials as simply representing the mean of the two presented probabilities, 50%. For such a participant, these trials were equivalent to the risky trials, and thus the proportion of choices in the conflicting (ambiguous) option is expected to be as frequent as in the risky trials, resulting in a corrected score of 0, reflecting conflict (ambiguity) neutrality.

To quantify participants’ attitudes towards conflict when directly contrasted with ambiguity (Fig. 3) we simply consider the proportion of choices in the conflicting option in these trials. To compare conflict and ambiguity indirectly, through the comparison with a certain option, we calculate for each participant an indirect comparison score using the proportion: \(\frac{{P}_{conflict}}{{P}_{conflict}+{P}_{ambiguity}}\), where \({P}_{conflict}\) and \({P}_{ambiguity}\) are the proportion of choices in the conflicting and ambiguous alternatives when contrasted with a certain alternative. A score of 0 or 1 indicates a low and high preference for conflict over ambiguity, and a score of 0.5 implies indifference. If both \({P}_{conflict}\) and \({P}_{ambiguity}\) are equal to 0, we replace the term with 0.5.

To test whether the population of participants significantly preferred ambiguous lotteries over conflicting ones in the direct comparison, we utilized a bootstrap test. We simulated a single population of unbiased decision-makers, matching the size of our empirical sample. For each simulated participant, we drew a random number from a binomial distribution assuming an equal probability (50%) for choosing conflict and ambiguity. We repeated this process 10,000 times and for each population we calculated the population mean choice of conflicting options. We report as p-values, the proportions of simulations whose population mean was smaller than that of the empirical mean.