Notwithstanding valid reasons for qualified trust in science and scientists, the aim of the present work is to understand factors that legitimately promote or hinder trust. Prior research makes the distinction between trust as a perception (for example, seeing a scientist as worthy of being trusted) and as a behaviour (for example, following a scientist's recommendations) 31 . In the current study, we were interested in perceived trustworthiness-conceptualized as having the qualities of expertise (that is, seeing the scientist as epistemically competent), benevolence (that is, seeing the scientist as concerned for the wellbeing of others) and integrity (that is, seeing the scientist as honest) 32 -as well as whether this perceived trust spills over into actual behavioural willingness to follow recommendations.

We argue that certain features of scientists might exert powerful influences on how trustworthy they appear to be. Recently, scholars 33,34 have called for scientists to act and communicate with greater intellectual humility (IH)-the awareness that one's knowledge and beliefs might be limited or wrong 35 . While IH is most often defined as an intrapersonal or meta-cognitive trait, other definitions include interpersonal and contextspecific manifestations 35,36 . In the current research, we similarly define IH as the intrapersonal awareness of the limitations of one's knowledge, which can in turn have interpersonal and perceivable aspects (for example, respectfulness). IH is distinct from other similar traits, such as general humility, in that it specifically focuses on the limitations of one's knowledge, as opposed to one's general limitations 35,37 . In this way, IH is uniquely relevant to the domain of science relative to general humility, as being aware of the limitations inherent to the scientific enterprise and communicating transparently about them is critical for improving the scientific process 33 . As such, IH may be perceived as the expression of following the norms of science 33 . As scientific norms so closely resemble what it means to be intellectually humble, when scientists fail to behave in ways that reflect IH, it might be especially detrimental and jarring as it goes against both the fundamental norms of science and people's expectations for how a responsible scientist should act. Just because scientists understand that the scientific process entails these features, they may not always behave in ways that reflect IH. We therefore ask: how do people's perceptions of scientists' IH influence their perceived trustworthiness of scientists and willingness to follow research-based recommendations?

To our knowledge, no one has yet examined the impact of perceptions of scientists' IH on perceived trustworthiness of them and their research. Prior work has examined how someone's trait levels of IH predict their trust in science 5,38 and how a scientist's communication of topic complexity can impact someone's trust and levels of IH about a scientific topic 39 . Other work has demonstrated that certain behavioural qualities of scientists that might map onto high IH are associated with positive outcomes. For example, scientists who admit to mistakes and support open science practices are perceived as more trustworthy 40,41 , and expressing uncertainty and complexity as part of scientific communication can sometimes foster trust 39,42,43 . While these studies focus on specific methods of conducting or communicating about science, they do not directly assess more global perceptions of scientists' IH. One study on physician humility demonstrated that participants reported greater trust and willingness to adhere to the recommendations of fictitious physicians who were ostensibly described by a prior patient as 'very humble (versus 'not humble') in (their) approach to patient care' 44 . These studies, however, focused on patients' perceptions of physicians' humility towards their patients and did not assess characteristics of IH or perceived trustworthiness of scientists and their research.

We argue that perceptions of scientists' IH may be an especially important predictor of people's perceived trustworthiness of scientists because IH uniquely focuses on admitting the limitations of one's knowledge and beliefs. As science is inherently uncertain and limited, being willing to recognize when one is wrong, rely on others' knowledge or update one's beliefs is critical for the scientific process 33 . We therefore hypothesized that these characteristics of IH render it particularly relevant to the perceived trustworthiness of scientists.

More specifically, we reasoned that perceiving a scientist as having low IH should feed into lower perceived trustworthiness because being unwilling to recognize one's fallibility can decrease the quality of one's research. That is, low IH may impede aspects of the scientific process that make it valuable and rigorous (for example, avoiding confirmation bias, separating one's ego from one's research programme, being willing to update beliefs based on data, recognizing when one has made an error, recognizing limitations of the science, continuously learning and leaning on the knowledge of others). Indeed, people who are lower in IH are less scrutinizing of information that confirms their worldviews 45 , are less effective at recognizing gaps in their knowledge 46 , are more overconfident in their knowledge 47 , are less motivated by learning 48 , use less effective learning strategies 49 and are less open to exposing themselves to different perspectives 50 . Although it may be difficult for participants to understand how low IH is related to specific questionable scientific practices, scientists who express an insight into the limits of their own knowledge and their continuous learning and updating of beliefs would be perceived as trustworthy in their general approach to research and science 40 .

By contrast, we reasoned that perceiving a scientist as having higher (versus lower) IH should foster perceived trustworthiness because being willing to recognize one's fallibility supports learning-oriented approaches and rigorous scientific practices such as those listed above, leading to perceptions of expertise. Moreover, because people higher (versus lower) in IH are more likely to endorse prosocial values 51 and are perceived as having more prosocial characteristics 52,53 , people might believe that scientists with higher IH have the public's best interests at heart, leading to perceptions of benevolence. Finally, IH may also imply that the scientist is less motivated by personal or political agendas because they are more focused on pursuing knowledge, leading to perceptions of integrity. We therefore predicted that people would report greater perceived trustworthiness of scientists-in the form of greater perceived expertise, benevolence and integrity-who they perceive as having high (versus low) IH. Further, we aimed to examine whether, in addition to perceiving scientists as trustworthy, participants would exhibit more behavioural trust-operationalized as seeing the scientist's research findings as trustworthy and being interested in receiving more information about how to follow their research-based recommendations.

In the current study, we assessed perceptions and outcomes of seeing scientists as intellectually humble in five pre-registered studies (N = 2,034).

In study 1, we examined the correlates of seeing scientists as intellectually humble and found that those who saw scientists as higher in IH indicated higher perceived trustworthiness of scientists and support for science-based beliefs. In study 2, we experimentally

The error bands represent the 95% CI. N = 298. Correlation tests are two sided. manipulated the IH of an ostensible scientist, demonstrating that describing a scientist as high (versus low) in IH increases their perceived trustworthiness and trust in their research findings. In studies 3 and 4, we replicated this finding while also testing for moderation by the gender (study 3) and race/ethnicity (study 4) of the scientist, and further demonstrated effects of a scientist's IH on intentions to follow their research-based recommendations (study 3) and information-seeking behaviour aimed at learning more about the scientist's recommendations (study 4). In a final study, we tested tangible communication approaches scientists might use to convey their IH to the public in a census-matched representative experiment. In all studies, we show the benefits of seeing scientists as intellectually humble across a variety of polarized scientific domains including medical science (studies 1 and 2), psychological science (studies 3 and 5) and climate science (studies 1 and 4). For all studies, we also present exploratory Bayes factors for null effects using JASP version 0.18.3 and JASP's default priors 54,55 . Bayesian results did not substantially differ when testing robustness with different priors (see OSF). All descriptive information and plots of posterior distributions are available on OSF.

While not pre-registered, we also explored the influence of perceived IH on each dimension of perceived trustworthiness (that is, expertise, benevolence and integrity), as measured by the Muenster Epistemic Trustworthiness Inventory 32 . Further, while there are many IH scales that conceptualize IH in slightly different ways, in the current work we focus on the conceptualization of IH captured by the Comprehensive Intellectual Humility Scale 36,56 . We chose this scale because it contains subscales that assess both the more intrapersonal dimensions of IH (lack of intellectual overconfidence and independence of intellect and ego) as well as subscales that have some interpersonal components and are therefore more observable (openness to revising one's views and respect for others' views). Prior work shows that this scale shows similar associations with various outcomes as other IH scales that focus on purely meta-cognitive aspects 57,58 . While this scale is not normally used to measure perceptions of someone else's IH, throughout all studies we found that the adapted version of the scale was affected by our manipulations of IH, suggesting that it was picking up on signals of scientists' IH as we had intended it to. We present exploratory analyses using the separate IH subscales here and in the Supplementary Information. Results

Study 1 was a correlational test of the association between seeing scientists as intellectually humble and trustworthy. We asked participants (N = 298) to think of a broad range of scientists and rate them on perceived IH 56 . Participants also provided ratings of their perceived trustworthiness of scientists 32 and indicated their belief in polarizing science topics (climate change, vaccinations and genetically modified foods) 21 .

Perceived IH of scientists was strongly associated with perceived trustworthiness of scientists (r(296) = 0.62, P < 0.001) and belief in anthropogenic climate change (r(296) = 0.43, P < 0.001), support for vaccinations (r(296) = 0.49, P < 0.001) and support for genetically modified food (r(296) = 0.43, P < 0.001; Fig. 1), with effect sizes ranging from medium-large to large 59 . While all subscales of IH of scientists were associated with perceived trustworthiness, they varied in magnitude (openness to revising one's views, r(296) = 0.54, P < 0.001; respect for other's views, r(296) = 0.62, P < 0.001; lack of intellectual overconfidence, r(296) = 0.46, P < 0.001; and independence of intellect and ego, r(296) = 0.43, P < 0.001). Similarly, while IH of scientists was associated with all three dimensions of perceived trustworthiness, they varied in magnitude (integrity, r(296) = 0.62, P < 0.001; benevolence, r(296) = 0.57, P < 0.001; and expertise, r(296) = 0.47, P < 0.001). Importantly, perceived IH of scientists did differ by political orientation with a large effect size, with liberal participants perceiving more IH in scientists (mean (M) = 3.74, s.d. of 0.51) than conservative participants (M = 3.15, s.d. of 0.70; F(1, 296) = 70.24, P < 0.001, partial eta squared (η 2 p) = 0.19, 95% confidence interval (CI) for η 2 p of 0.12-0.27; Fig. 2). However, associations between perceived IH of scientists and perceived trustworthiness, and between perceived IH of scientists and belief in polarizing science topics, were robust to including simultaneous covariates of political orientation, political conviction, interest in science, science knowledge and conspiracy mentality (all P < 0.021; Supplementary Table 3). Finally, perceived IH of scientists was still significant when also including cognitive reflection and participants' own level of IH to these models (all P < 0.001), with the exception of the belief in climate change model (P = 0.083).

Overall, study 1 provided correlational evidence that people who see scientists as intellectually humble also tend to see those scientists as more trustworthy and hold evidence-based scientific beliefs.

In study 2, we tested our hypothesis using an experimental design. We randomly assigned participants (N = 317) to read one of three articles about an ostensible scientist, Susan Moore, researching new treatments for long COVID-19 symptoms. Participants in the control condition read about Susan Moore's research and findings. Participants in the low-IH condition read this same article but with cues of low IH (for example, "… according to one colleague, 'Dr Moore is not afraid to assert what she knows'"). Finally, participants in the high-IH condition read the same control article but with cues of high IH (for example, "…according to one colleague, 'Dr Moore is not afraid to admit when she doesn't yet know something'", see OSF for manipulations).

We found a large effect of our manipulation on perceived IH of the scientist, F( 2 Participants saw the scientist as significantly less intellectually humble in the low-IH condition than in the control condition (P < 0.001). The strength of the effect of condition varied by subscale of IH (lack of intellectual overconfidence, η 2 p = 0.57, 95% CI for η 2 p of 0.51 to 0.63; openness to revising one's views, η 2 p = 0.51, 95% CI for η 2 p of 0.44 to 0.57; respect for other's views, η 2 p = 0.31, 95% CI for η 2 p of 0.23 to 0.38; and independence of intellect and ego, η 2 p = 0.25, 95% CI for η 2 p of 0.17 to 0.33).

We also found a large effect on perceived trustworthiness, F(2, 314) = 25.50, P < 0.001, η 2 p = 0.14, 95% CI for η 2 p of 0.07 to 0.21 (Fig. 3). Participants saw the scientist as significantly more trustworthy in the high-IH condition (M = 6.43, s.d. of 0.64) than in the low-IH condition (M = 5.62, s.d. of 1.10; P < 0.001). Participants saw the scientists as significantly less trustworthy in the low-IH condition than in the control condition (M = 6.17, s.d. of 0.76; P < 0.001). The difference between the high-IH and control conditions did not reach significance (P = 0.078), though a Bayes factor analysis indicated some evidence against the null (that group means do not differ, BF01 of 0.27). The strength of the effect of condition differed by perceived trustworthiness subscale (integrity, η 2 p = 0.13, 95% CI for η 2 p of 0.07 to 0.20; benevolence, η 2 p = 0.19, 95% CI for η 2 p of 0.11 to 0.26; and expertise, η 2 p = 0.06, 95% CI for η 2 p of 0.02 to 0.12).

Finally, we found a small-to-medium effect on belief in the scientist's research on the new long COVID-19 treatment, F( 2 All results remained significant after controlling for participants' baseline belief in vaccines (all P < 0.002; Supplementary Tables 161718).

Study 2 revealed that a description of a scientist as having characteristics reflective of higher IH (versus lower IH) led to greater perceived trustworthiness of the scientist and belief in her scientific research. A description of a scientist with lower IH also led to lower perceived trustworthiness and belief in research than the control scientist, implying that low IH in scientists might be especially detrimental. However, because we described a woman scientist, we were not able to examine whether the effects were in part driven by the scientist's gender. Prior work has argued that the positive outcomes of expressing humility might differ based on social power, such that expressing humility may be less beneficial or even harmful for people of lower social power [60][61][62] . In particular, others have argued and shown that humility may be harmful if its expression does not adhere to gender norms 60,63 . In addition, there is evidence that people tend to be biased against women in science. For example, faculty rate women candidates for a STEM position as less competent and less hirable than men candidates 64,65 . It is therefore possible that low IH was viewed less positively in this study simply because it was perceived as notches indicate 95% CIs around the median, upper and lower hinges indicate 75th and 25th percentiles, respectively, and whiskers indicate 1.5× the interquartile range. All significant differences are labelled. Nstudy 2 = 317, Nstudy 3 = 369 and Nstudy 4 = 371.

counterstereotypical for a woman scientist. We therefore examined the effect of scientist gender in study 3.

In study 3, we aimed to test the effect of scientist gender and replicate the results of study 2 in a new scientific context. We randomly assigned participants (N = 369) to read an article about an ostensible psychological scientist researching why people should talk across political divides. As in study 2, there was a control condition as well as conditions describing the scientist as high in IH or low in IH. We crossed this with the described gender of the scientist (woman: Sandra Wilson, she/her pronouns; man: Robert Wilson, he/him pronouns). We found a large effect of IH condition on perceived IH of the scientist, F(2, 363) = 167.11, P < 0.001, η 2 p = 0.48, 95% CI for η 2 p of 0.41 to 0.54. Using a Tukey test to adjust for multiple comparisons, participants saw the scientist as significantly more intellectually humble in the high-IH condition (M = 4.30, s.d. of 0.50), than in either of the other conditions (Mcontrol condition = 3.81, s.d.control condition of 0.55; Mlow IH condition = 2.96, s.d.low IH condition of 0.71; all P < 0.001). Participants saw the scientist as significantly less intellectually humble in the low-IH condition than in the control condition (P < 0.001). There was an absence of evidence for an effect of (P = 0.129) or interaction with (P = 0.548) the gender of the scientist. The Bayes factor indicated evidence in favour of the null model that only included IH condition (model with gender, BF01 of 3.46; model with gender and its interaction, BF01 of 38.02). Again, the strength of the effect of condition varied by IH subscale (lack of intellectual overconfidence, η 2 p = 0.53, 95% CI for η 2 p of 0.46 to 0.58; openness to revising one's views, η 2 p = 0.48, 95% CI for η 2 p of 0.41 to Study 2 0.54; respect for other's views, η 2 p = 0.28, 95% CI for η 2 p of 0.21 to 0.35; and independence of intellect and ego, η 2 p = 0.16, 95% CI for η 2 p of 0.09 to 0.22).

We found a large effect of IH condition on perceived trustworthiness, F(2, 363) = 52.70, P < 0.001, η 2 p = 0.23, 95% CI for η 2 p of 0.15 to 0.29. Participants saw the scientist as significantly more trustworthy in the high-IH condition (M = 6.46, s.d. of 0.69) than in either of the other conditions (Mcontrol condition = 5.98, s.d.control condition of 0.81; Mlow IH condition = 5.33, s.d.low IH condition of 1.13; all P < 0.001). Participants saw the scientist as significantly less trustworthy in the low-IH condition than in the control condition (P < 0.001). There was an absence of evidence for an effect of (P = 0.165) or interaction with (P = 0.957) the gender of the scientist. The Bayes factor indicated evidence in favour of the null model that only included IH condition (model with gender, BF01 of 3.22; model with gender and its interaction, BF01 of 57.45). Again, the strength of the effect of condition differed by perceived trustworthiness subscale (integrity, η 2 p = 0.23, 95% CI for η 2 p of 0.16 to 0.30; benevolence, η 2 p = 0.26, 95% CI for η 2 p of 0.19 to 0.33; and expertise, η 2 p = 0.14, 95% CI for η 2 p of 0.08 to 0.20).

We found a small-to-medium effect of IH condition on belief in the scientist's research on the benefits of talking across political divides, F(2, 363) = 3.63, P = 0.028, η 2 p = 0.02, 95% CI for η 2 p of 0.00 to 0.05. Participants believed the scientist's research more in the high-IH condition (M = 4.54, s.d. of 1.07), than in the low-IH condition (M = 4.17, s.d. of 1.16; P = 0.020).

There was an absence of evidence for the difference between the high-IH and control conditions (M = 4.36, s.d. of 0.97; P = 0.401). The Bayes factor also indicated some evidence in favour of the null (that group means do not differ; BF01 of 3.06). There was also an absence of evidence for the difference between the low-IH and control conditions (P = 0.392). The Bayes factor also indicated some evidence in favour of the null (BF01 of 3.04). Finally, there was an absence of evidence for an effect of (P = 0.414) or interaction with (P = 0.930) the gender of the scientist. The Bayes factor indicated evidence in favour of the null model that only included IH condition (model with gender, BF01 of 6.59; model with gender and its interaction, BF01 of 110.31).

Finally, we found a small-medium effect of IH condition on intention to follow the scientist's research, F(2, 363) = 5.71, P = 0.004, η 2 p = 0.03, 95% CI for η 2 p of 0.00 to 0.07. Participants indicated higher intention to follow the scientist's research in the high-IH condition (M = 5.03, s.d. of 1.69), than in the low-IH condition (M = 4.29, s.d. of 1.96; P = 0.003). There was an absence of evidence for a difference between the high-IH and control conditions (M = 4.50, s.d. of 1.83; P = 0.058), and the Bayes factor indicated weak evidence against the null (that group means do not differ; BF01 of 0.50). There was also an absence of evidence for a difference between the low-IH and control conditions (P = 0.647). The Bayes factor indicated evidence in favour of the null (BF01 of 4.85). There was an absence of evidence for an effect of (P = 0.701) or interaction with (P = 0.642) the gender of the scientist. The Bayes factor indicated evidence in favour of the null model that only included IH condition (model with gender, BF01 of 7.99; model with gender and its interaction, BF01 of 101.55).

Study 3 replicated the experimental effects of scientists expressing IH in a different science domain (psychological science) and did not detect a difference by scientist gender. However, in studies 1-3 we did not explicitly assess perceptions of the scientist's race/ethnicity. Prior research has shown that people tend to presume the racial identity of others based on their implicit stereotypes, and often assume that those in leadership positions are white 66 . It is therefore possible that participants in studies 2 and 3 presumed that the presented scientists were white, absent of explicit cues. While it is unknown whether there are differential benefits of perceived IH across racial identities, there is substantial evidence for bias against scientists of colour. For example, faculty rate Black and Latinx candidates as less competent and hirable for STEM positions 64 . It is therefore possible that scientists of colour who exhibit high (versus low) IH might be viewed as less confident in their knowledge or less competent. To ensure that the benefits of perceived IH generalize to scientists of colour, in study 4 we tested whether the effects of IH were affected by the racial identity of the scientist.

Study 4 aimed to replicate the results of studies 2 and 3 in a new scientific context, while also experimentally varying the described race/ ethnicity of the scientist. We randomly assigned participants (N = 371) to read an article about an ostensible climate scientist researching the benefits of a plant-rich diet for reducing global carbon emissions. We again described the scientist as either high or low in IH, dropping the neutral control condition to increase power. We crossed the manipulation of IH with the described race/ethnicity of the scientist using a name manipulation from prior work (white: Claire Miller; Black: Shanice Banks; Latinx: Maria Rodriguez) 64 .

We found a large effect of IH condition on perceived IH of the scientist, F(1, 365) = 468.63, P < 0.001, η 2 p = 0.56, 95% CI for η 2 p of 0.50 to 0.61. Participants saw the scientist as significantly more intellectually humble in the high-IH condition (M = 4.08, s.d. of 0.61), than in the low-IH condition (M = 2.64, s.d. of 0.69). There was an absence of evidence for an effect of (P = 0.594) or interaction with (P = 0.223) the race/ethnicity of the scientist. Further, the Bayes factor indicated strong evidence in favour of the null model that only included IH condition (model with race/ethnicity, BF01 of 19.88; model with race/ethnicity and its interaction, BF01 of 95.88). Again, the strength of the effect of condition varied by IH subscale (lack of intellectual overconfidence, η 2 p = 0.56, 95% CI for η 2 p of 0.50 to 0.61; openness to revising one's views, η 2 p = 0.51, 95% CI for η 2 p of 0.44 to 0.57; respect for other's views, η 2 p = 0.43, 95% CI for η 2 p of 0.35 to 0.49; and independence of intellect and ego, η 2 p = 0.30, 95% CI for η 2 p of 0.22 to 0.37).

We found a large effect of IH condition on perceived trustworthiness, F(1, 365) = 84.14, P < 0.001, η 2 p = 0.19, 95% CI for η 2 p of 0.12 to 0.26. Participants saw the scientist as significantly more trustworthy in the high-IH condition (M = 6.28, s.d. of 0.80), than in the low-IH condition (M = 5.30, s.d. of 1.24). There was an absence of evidence for an effect of (P = 0.539) or interaction with (P = 0.928) the race/ethnicity of the scientist. Further, the Bayes factor indicated strong evidence in favour of the null model that only included IH condition (model with race/ ethnicity, BF01 of 19.96; model with race/ethnicity and its interaction, BF01 of 337.36). Again, the strength of the effect of condition varied by perceived trustworthiness subscale (integrity, η 2 p = 0.20, 95% CI for η 2 p of 0.13 to 0.27; benevolence, η 2 p = 0.23, 95% CI for η 2 p of 0.16 to 0.30; and expertise, η 2 p = 0.10, 95% CI for η 2 p of 0.05 to 0.16).

We found a small-medium effect of IH condition on belief in the scientist's research on the benefits of plant-rich diets, F(1, 365) = 8.88, P = 0.003, η 2 p = 0.02, 95% CI for η 2 p of 0.00 to 0.06. Participants believed the scientist's research more in the high-IH condition (M = 4.62, s.d. of 1.35), than in the low-IH condition (M = 4.17, s.d. of 1.54). There was an absence of evidence for an effect of (P = 0.670) or interaction with (P = 0.722) the race/ethnicity of the scientist. The Bayes factor indicated evidence in favour of the null model that only included IH condition (model with race/ethnicity, BF01 of 23.47; model with race/ethnicity and its interaction, BF01 of 316.90).

Finally, we found a small-medium effect of IH condition on participants' information-seeking behaviour, odds ratio (OR) of 1.46, 95% CI for OR of 1.16 to 1.86; P = 0.002, such that a higher percentage of people chose to receive information about switching to a plant-rich diet in the high-IH condition (36%) than in the low condition (21%). There was an absence of evidence for an effect of or interaction with the race/ethnicity of the scientist (all P > 0.312).

Study 4 provided additional evidence of the benefits of expressed IH in a different (and politically polarized) science domain (climate science) and also across scientist races/ethnicities.

Studies 1-4 showed that seeing a scientist as more intellectually humble increases perceived trustworthiness of them and their findings compared with seeing them as less intellectually humble. These results raise an important question: how can scientists most effectively express their IH? To answer this question, we first reached out to experts in the field of IH to identify ways that scientists can communicate with IH to the public. We organized the most commonly reported communication approaches into three categories: personal IH (for example, updating beliefs and willingness to share credit with others), limits of methods (for example, describing specific methodological limitations and methodological constraints of the current research) and limits of results (for example, describing the nuances and weaknesses in the results of the current research or describing limits to the generalizability of the findings). In study 5, we randomly assigned participants from a census-matched sample (N = 679), to read one of four (three IH communication conditions and one neutral control) interviews with a scientist discussing the psychological benefits of taking a break from social media 67 . Critically, this was a subtle manipulation of communicating IH (versus a neutral control) about a non-polarizing topic.

We found a small-to-medium effect on perceived IH of the scientist, F(3, 675) = 8.93, P < 0.001, η 2 p = 0.04, 95% CI for η 2 p of 0.01 to 0.07 (Fig. 4). Using a Tukey test, participants saw the scientist as more intellectually humble in the personal IH condition (M = 3.97, s.d. of 0.47) and the limitations of results condition (M = 3.93, s.d. of 0.50) than the control condition (M = 3.70, s.d. of 0.53; all P < 0.001). Perceived IH was also significantly higher in the personal IH condition than the limitations of methods condition (M = 3.82, s.d. of 0.55; P = 0.033). There was an absence of evidence for a difference between the limitations of methods condition and the control (P = 0.173). The Bayes factor indicated that the data were insensitive to detect this difference (BF01 of 1.38). The effect of condition was very similar across all subscales (η 2 p = 0.02-0.03).

Counter to our expectations, there was no effect of condition on perceived trustworthiness, F(3, 675) = 2.59, P = 0.052, η 2 p = 0.01, 95% CI for η 2 p of 0.00 to 0.03. The Bayes factor also indicated evidence in support for the null hypothesis (BF01 of 5.45). This may be due to a ceiling effect, with all conditions having a mean of over six on a seven-point scale (Mcontrol = 6.05, s.d.control of 0.83; Mlimits of methods = 6.15, s.d.limits of methods of 0.79; Mlimits of results = 6.28, s.d.limits of results of 0.65; Mpersonal IH = 6.17, s.d.personal IH of 0.77). Interestingly, the effect of condition was significant when looking at the individual subscales of integrity, F(3, 675) = 2.81, P = 0.039, η 2 p = 0.01, 95% CI for η 2 p of 0.00 to 0.03, and benevolence, F(3, 675) = 3.27, P = 0.021, η 2 p = 0.01, 95% CI for η 2 p of 0.00 to 0.03, but not expertise, F(3, 675) = 1.30, P = 0.274, η 2 p = 0.01, 95% CI for η 2 p of 0.00 to 0.02. In a secondary mediation analysis, we found a significant indirect effect on perceived trustworthiness from both limits of results (b = 0.18, s.e.m. of 0.05, 95% CI for b of 0.09 to 0.28) and personal IH (b = 0.22, s.e.m. of 0.05, 95% CI for b of 0.13-0.31) conditions through perceived IH using 10,000 bootstrapped samples (compared with the control). There was no indirect effect from the limits of methods condition (b = 0.09, s.e.m. of 0.05, 95% CI for b of -0.00 to 0.19); Supplementary Fig. 9).

We found a small-to-medium effect of IH condition on belief in the scientist's research on taking a break from social media, F(3, 675) = 9.26, P < 0.001, η 2 p = 0.04, 95% CI for η 2 p of 0.01 to 0.07. Using a Tukey test, participants believed the research more in the personal IH condition (M = 5.32, s.d. of 0.97) than in the limits of results condition (M = 4.89, s.d. of 1.07; P < 0.001) and the limits of methods condition (M = 5.02, s.d. of 1.08; P = 0.033). Participants believed the research more in the control condition (M = 5.39, s.d. of 0.97) than in the limits of results condition (P < 0.001) and the limits of methods condition (P = 0.005).

The personal IH condition and the control condition did not differ (P = 0.927). For the latter contrast, the Bayes factor also indicated support for the null hypothesis (BF01 of 6.69). We ran the same mediation model on belief in the scientists' research and found significant indirect effects from both limits of results (b = 0.10, s.e.m. of 0.03, 95% CI for b of 0.05 to 0.17) and personal IH (b = 0.13, s.e.m. of 0.03, 95% CI for b of 0.07 to 0.20) conditions (compared with the control) through perceived IH. There was no indirect effect from the limits of methods condition (b = 0.05, s.e.m. of 0.03, 95% CI for b of -0.00 to 0.11).

Finally, we found non-significant effects on participants' informationseeking behaviour relative to the control (personal IH condition, OR of 0.65, 95% CI for b of 0.40 to 1.04; P = 0.073; limitations of methods condition, OR of 0.67, 95% CI for b of 0.42 to 1.07; P = 0.093; and limitations of results condition, OR of 0.82, 95% CI for b of 0.52 to 1.28; P = 0.383).

Study 5 provided mixed evidence for the effectiveness of the expertrecommended approaches to communicating IH. We found that two of the three approaches effectively increased perceptions of IH of the ostensible scientist (personal IH and limitations of results). Further, perceptions of IH significantly predicted perceived trustworthiness as in prior studies (r(677) = 0.56, P < 0.001). However, only personal IH successfully increased perceptions of IH without backfiring, and even this approach had only indirect effects on perceived trustworthiness and belief in the scientist's research.

In the current research, we aimed to test the effects of seeing scientists as intellectually humble on trust in scientists and intentions to follow scientific evidence. In line with our predictions and prior theorizing 33 , we found that perceiving scientists as high in IH is reliably associated with perceiving them as trustworthy and believing their research (study 1). We also tested this experimentally and found that describing a scientist as high (versus low) in IH increases perceptions of their trustworthiness and belief in their research (studies 2-4). Critically, this effect was consistent across scientists with different gender (study 2) and racial/ethnic (study 3) identities. We also found that perceptions of IH contribute to more perceived trustworthiness in scientists across different topics including medical, climate and psychological science. Finally, we found that perceptions of IH contribute to intentions to follow the scientist's research-based recommendations and information-seeking behaviour aimed at learning more about the scientist's recommendations. Together, these four studies offer compelling evidence that perceptions of scientists' IH play an important role in both trust in scientists and willingness to follow their research-based recommendations.

In study 5, we tested three communication approaches to increase perceptions of IH and perceived trustworthiness. Of the approaches tested, personal IH was the most successful. This is perhaps unsurprising as people might pay closest attention to expressions that reflect a scientist's personal character (rather than limitations of the research) when making judgements of their trustworthiness. However, even this approach was limited in its direct benefits on perceived trustworthiness and intentions to follow scientific evidence, perhaps due in part to its subtlety and its comparison to a neutral (rather than low IH) control. Future work is needed to provide a deeper understanding of how scientists can authentically express IH when communicating about their science with the public and whether such communication enhances perceived trustworthiness.

We would be remiss to not acknowledge that two of the IH communication approaches backfired in the domain of believing a scientist's research. This finding raises the question of how perceived trustworthiness in the scientist as a person might differ from trust in the scientist's results. It is possible that conveying limitations about a particular study might signal that the scientist is more honest and intellectually humble but might simultaneously reduce trust in the research or recommendations that follow from it 68 . This is perhaps why prior research has found mixed effects of expressing limitations of research 40,42,43,69 . Future research should therefore seek to understand other ways to convey IH, while paying close attention to potential backfire effects. If certain expressions of IH are found to be helpful in increasing trust without decreasing belief in the science, it may be important to consider how to enhance the culture of intellectually humble communication within scientific institutions as well as between scientists and the public.

This research was characterized by several strengths distinguishing it from other studies in this domain (for example, use of experimental design, crowdsourcing of IH researchers and behavioural information-seeking measures) and it is, to our knowledge, the first set of studies to examine how participants perceive IH in scientists. Nevertheless, this research is also characterized by several important limitations. First, all studies used convenience samples collected from online pools. While this does make them more diverse than typical university pools 70 , it also limits their generalizability (for example, all participants chose to opt in to academic research websites and therefore may have more positive attitudes towards science). Similarly, while we collected a census-matched sample in study 5, the samples for studies 1-4 were notably less diverse in terms of race/ethnicity. Second, this research did not test whether there are specific underlying mechanisms for why perceptions of IH increase perceived trustworthiness. It will be fruitful to identify what precisely it is about IH that leads to more trust and whether these effects are unique to IH as opposed to other correlated traits/processes (for example, prosociality and agreeableness; Supplementary Figs. 3, 5 and 8). Future research should also directly test whether perceived IH combats common drivers of lower perceived trustworthiness, such as conspiracy beliefs 17,21 or perceptions of ideological motives 19 . It would also be advantageous to examine whether seeing a scientist model IH boosts the perceiver's IH, thereby motivating people to seek out accurate scientific knowledge themselves. Third, while we chose scientific disciplines that vary in their methods and topics of study, the current research can only speak to the effect of IH in these specific disciplines (medical science, climate science and psychological science). Future research should replicate this work in other disciplines (for example, economics, engineering and political science), as such work would clarify the generalizability of the relations identified in the present investigation. Finally, we used samples from the United States and cannot speak to how these results generalize to other geographic or cultural contexts.

In conclusion, this research builds on an emerging literature revealing the importance of IH in helping to solve major societal problems. IH is undoubtedly critical to the scientific process 33 . The current studies are some of the first to investigate how perceptions of people's IH impact their judgements, and we examined this question in a context that has paramount consequences for the wellbeing of society and its members: perceived trustworthiness of scientists and willingness to follow their research-based recommendations. Our work suggests that participants understand that IH is a beneficial characteristic for scientists to possess in these disciplines. At a practical level, this work suggests that scientists embracing and conveying IH is beneficial. Across three studies in different scientific domains, we found that scientists who are described as possessing intellectually humble characteristics-such as being willing to admit gaps in their knowledge, listening to the input of others and revising their beliefs in the face of new evidence-were more trusted than scientists who were described as possessing characteristics consistent with lower IH. Moreover, in study 5, we found that expressions of personal IH (for example, admitting that one's original predictions were wrong and updating one's thinking) promote perceptions of scientist IH, with downstream benefits for perceived trustworthiness of the scientist and their research. Although more work is needed to identify specific communication strategies that scientists can use to effectively convey IH, the current set of studies speaks to the benefits of scientists approaching their science and their public communications in ways that reflect high rather than low IH. In particular, given that we saw few differences between the high-IH and neutral control conditions in studies 2 and 3, this may suggest that the default perception of scientists is that they possess relatively high levels of IH and may especially dislike it when scientists display low IH or even arrogance. While future research should continue to disentangle valuing IH from disvaluing intellectual overconfidence or arrogance, we encourage scientists to be particularly mindful of displaying low IH, such as by expressing overconfidence, being unwilling to course correct or disrespecting others' views. These results also suggest that the field of science might more deliberately discourage low IH practices and reward scientists who approach their research and scientific communication with IH (as suggested in ref. 33).

All studies were approved by the University of Pittsburgh Institutional Review Board and determined to be exempt from requiring consent. All participants for each study read an information script before participating that informed them of the procedure, risks/benefits and compensation.

The goal of study 1 was to examine the public's perception of scientists' IH, and to test what these perceptions predict. Study 1 was pre-registered on 5 June 2022 at OSF. This and all studies were conducted in Qualtrics.

Participants. An a priori power analysis in G*Power 71 recommended a sample of 134 to detect a medium effect (r = 0.30) 59 using a correlational design (95% power, α = 0.05, two tailed). We aimed to collect data from 300 participants to allow for tests for moderation by political orientation. We collected a sample of 300 participants from Prolific Academic. Of these, two failed our pre-registered attention check, leaving a sample of 298 (47.65% female, 49.66% male, 2.35% non-binary/third gender, 0.34% prefer not to say; 2.01% Native American, 9.73% Asian or Asian American, 1.68% Black or African American, 0.67% Native Hawaiian or other Pacific Islander, 10.07% Hispanic or Latina/Latino/Latinx, 84.90% white, 2.01% Middle Eastern or North African; Mage = 39.40, s.d.age of 14.78). We collected equal numbers of liberals and conservatives in the sample (both N = 149). Materials and procedure. Perceived IH of scientists. We instructed participants to 'consider a broad range of scientists, such as medical scientists (for example, scientists who study human health conditions and diseases and how to treat them), social scientists (for example, psychologists and archaeologists) and natural scientists (for example, physicists and biologists)'. Participants then completed an adapted version of the Comprehensive Intellectual Humility Scale 56 . This is a validated 22-item measure containing four subscales-independence of intellect and ego (for example, "When someone disagrees with ideas that are important to a scientist, it makes the scientist feel insignificant"), openness to revising one's viewpoint ("Scientists are open to revising their important beliefs in the face of new information"), respect for others' viewpoints ("Scientists respect that there are ways of making important decisions that are different from the way they make decisions") and lack of intellectual overconfidence ("Scientists believe that their ideas are usually better than other people's ideas")-on a scale from 1 (strongly disagree) to 5 (strongly agree; α = 0.94). See Supplementary Table 7 for the results broken up by subscales across all studies.

Perceived trustworthiness of scientists. Participants then indicated their perceived trustworthiness of scientists on the Muenster Epistemic Trustworthiness Inventory 32 . This is a validated measure containing three subscales of 14 bipolar items-expertise (for example, incompetent to competent), integrity (for example, unjust to just) and benevolence (for example, immoral to moral)-on seven-point scales (α = 0.95). See Supplementary Table 8 for the results broken up by subscales across all studies.

Belief in polarizing science topics. Participants next completed three scales assessing beliefs in polarizing science topics in a randomized order. Participants completed five-item measures assessing each belief in climate change (for example, "Human CO2 emissions cause climate change", α = 0.94), support for vaccinations (for example, "I believe that vaccines are a safe and reliable way to help avert the spread of preventable diseases", α = 0.91) and support for genetically modified foods (for example, "I believe that genetic modification is an important and viable contribution to help feed the world's rapidly growing population", α = 0.93) on scales from 1 (strongly disagree) to 7 (strongly agree) 21 .

Exploratory measures. Participants completed exploratory measures of conspiracist ideation 72 (α = 0.87), science knowledge 73 , cognitive reflection 74 and participants' own IH 56 (α = 0.88). Finally, participants completed political (political identity, political orientation, political conviction, political interest and moralization of politics), science-related (interest in science and views of scientists in politics) and demographic (race/ethnicity, age, gender, education, religion and religiosity) measures. See Supplementary Tables 2 and 3-6 for analyses with exploratory measures.

The goal of study 2 was to examine the experimental impacts of scientist IH on perceived trustworthiness. Study 2 was pre-registered on 8 August 2022 at OSF.

Participants. An a priori power analysis in G*Power 71 recommended a sample of 252 to detect a medium effect (f = 0.25) 59 using a three-condition design (95% power, α = 0.05). We aimed to double this and collect 520 to test for moderation by political orientation. We collected a sample of 517 from Prolific Academic. Of these, 152 failed the first attention check, 9 failed the second attention check and 76 failed the manipulation check, leaving a sample of 317 (47.95% female, 49.53% male, 2.21% nonbinary/third gender, 0.32% prefer to self-describe; 1.58% Native American, 8.20% Asian or Asian American, 6.31% Black or African American, 0.32% Native Hawaiian or other Pacific Islander, 10.41% Hispanic or Latina/Latino/Latinx, 78.55% white, 0.63% Middle Eastern or North African, 0.95% 'Other'; Mage = 40.05, s.d.age of 15.10). We collected roughly equal numbers of liberals (N = 160) and conservatives (N = 157). Materials and procedure. Manipulation. We randomly assigned participants to one of three conditions. All participants read an article about an ostensible scientist (Susan Moore) at the University of Michigan researching a new anti-viral treatment for long COVID-19 symptoms. We described Susan Moore as high in IH (for example, "…according to one colleague, 'Dr Moore is not afraid to admit when she doesn't yet know something'", N = 115), low in IH (for example, "…according to one colleague, 'Dr Moore is not afraid to assert what she knows'", N = 103) or no information about IH (N = 99). We modelled the manipulations using the IH behaviours expressed in the Comprehensive Intellectual Humility Scale.

Big five inventory of the scientist. As a filler scale, we asked participants to report on their perceptions of the scientist's personality 75 using a 15-item measure with subscales of extraversion (for example, "Dr Moore is someone who is dominant, acts as a leader", α = 0.57), openness (for example, "Dr Moore is someone who is original, comes up with new ideas", α = 0.61), neuroticism (for example, "Dr Moore is someone who worries a lot", α = 0.54), conscientiousness (for example, "Dr Moore is someone who is reliable, can always be counted on", α = 0.65) and agreeableness (for example, "Dr Moore is compassionate, has a soft heart", α = 0.81) on a scale from 1 (strongly disagree) to 5 (strongly agree; Supplementary Tables 9 and 19-21).

Perceived IH of the scientist. As a manipulation check, participants completed the same IH scale from study 1, now adapted to be about Dr Moore (for example, "Dr Moore believes that her ideas are usually better than other people's ideas", α = 0.97).

Likability of the scientist. Participants also completed an adapted four-item likability scale 76 (for example, "Dr Moore is friendly") on a scale from 1 (very strongly disagree) to 7 (very strongly agree, α = 0.94; Supplementary Tables 9 and 13 Belief in the scientist's research. Participants completed a five-item measure assessing belief in the scientist's anti-viral treatment (for example, "I believe that this new anti-viral treatment is a safe and reliable way to help avert long COVID-19 symptoms", adapted from ref. 21, α = 0.88).

Exploratory measures. As an exploratory covariate, participants completed the same support for vaccinations scale (α = 0.94) and interest in science scale (inter-item r(315) = 0.39) as in study 1. Participants completed the same demographic measures as in study 1. See Supplementary Tables 9 and 16-18 for analyses on exploratory measures.

The goal of study 3 was to replicate the results of study 2, while also testing the impacts of scientist gender on perceived IH. Study 3 was pre-registered on 8 March 2023 at OSF.

Participants. An a priori power analysis in G*Power 71 recommended a sample of 374 to detect a small-medium effect (f = 0.20) 59 using a 3 × 2 between-subjects design (95% power, α = 0.05). We collected a sample of 450 from Prolific Academic to account for potential low-quality data. Of these, 3 failed the first attention check and 79 failed the second attention check, leaving a sample of 369 (49.86% female, 48.24% male, 1.90% nonbinary/third gender; 2.17% Native American, 5.96% Asian or Asian American, 7.05% Black or African American, 0.54% Native Hawaiian or other Pacific Islander, 8.67% Hispanic or Latina/Latino/Latinx, 82.66% white, 0.27% Middle Eastern or North African, 0.27% 'Other'; Mage = 42.63, s.d.age of 14.40). We collected roughly equal numbers of liberals (N = 195) and conservatives (N = 174). Materials and procedure. Manipulation. As in study 2, we randomly assigned participants to one of three IH conditions (high, N = 146; low, N = 113; control, N = 110). The IH part of the manipulation was identical to study 2. Unlike study 2, all articles were about an ostensible psychological scientist at the University of Pittsburgh researching the benefits of talking to people across the political divide in the United States. We crossed IH condition with the gender of the scientist, who was either a woman (Sandra Wilson, N = 184) or a man (Robert Wilson, N = 185).

Perceived IH of the scientist. Participants completed the same IH scale from studies 1 and 2, now adapted to be about Robert Wilson (α = 0.96).

Perceived trustworthiness in the scientist. Participants completed the Muenster Epistemic Trustworthiness Inventory about the scientist (α = 0.97).

Belief in the scientist's research. Participants completed a five-item measure assessing belief in the scientist's research on the benefits of talking across the political divide (for example, "I believe that talking to people from the opposing political party makes people feel happier", adapted from ref. 21, α = 0.72).

Intention to follow research. Participants completed two items about their intentions to learn more about how to follow the scientist's research recommendations ("I would be interested in receiving information about how to have productive conversations with people from other political parties", "I would be interested in learning more about Dr Wilson's suggestions for productive conversations", inter-item r(366) = 0.86).

Stereotype content of the scientist. As an exploratory outcome, participants completed two items assessing the warmth ('warm', 'friendly', inter-item r(366) = 0.91) and two items assessing the competence ('competent', 'capable', inter-item r(365) = 0.88) of the scientist (adapted from ref. 77; see Supplementary Tables 36-50 for results). It should be noted that the 'competent' item from the competence scale overlaps with one item from the Muenster Epistemic Trustworthiness Inventory (incompetent to competent). We therefore recommend caution when interpreting models that include both these scales. Demographic measures. Participants completed the same demographic measures as in prior studies.

The goal of study 4 was to replicate the results of studies 2 and 3 and ensure the effects of perceived IH were consistent across scientists with different race/ethnic identities. Study 4 was pre-registered on 15 June 2023 at: OSF.

Participants. An a priori power analysis in G*Power 71 recommended a sample of 374 to detect a small-medium effect (f = 0.20) 59 using a 3 × 2 between participant design (95% power, α = 0.05). We collected a sample of 450 from Prolific Academic to account for potential low-quality data. Of these, 9 failed the first attention check and 74 failed the second attention check, leaving a sample of 371 (51.21% female, 47.17% male, 1.35% nonbinary/third gender, 0.27% prefer to self-describe; 0.54% Native American, 8.63% Asian or Asian American, 6.47% Black or African American, 8.63% Hispanic or Latina/Latino/Latinx, 83.02% white, 0.54% Middle Eastern or