News update via NF Bot

PubMed link

Publication date: Available online 30 October 2026 Source: Physiology & Behavior Author(s): Seungyul Lee, Teri Kim

Published 5 November 2026 • © 2026 The Author(s). Published by IOP Publishing Ltd Environmental Research Letters, Volume 21, Number 21Citation V Cologna et al 2026 Environ. Res. Lett. 21 219501DOI 10.1088/1748-9326/aea350 Original content from this work may be used under the terms of the Creative Commons Attribution 4.0 license. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. 1. Short statement The market research company that collected the survey data for this study made consequential mistakes when recruiting respondents, such that some participants were sampled twice. Specifically, 44 participants who had filled out the survey twice pre-vote had different IDs in our dataset but actually reflected the same people. Post-vote, we identified 57 participants who had completed the survey at least twice with some of these participants getting access to two separate survey links and viewing both conditions. Correcting these mistakes slightly changed the results of some of our analyses. The values in the original manuscript should therefore not be taken at face value; readers should instead consult the tables in this correction. Our substantive conclusions are largely unaffected. We continue to find no evidence that advocacy backfired among right-leaning respondents, no effect of advocacy on the voting intentions or self-reported voting behavior, a significant interaction between treatment and political orientation on trust pre-vote, and a significant positive effect of the treatment on trust post-vote. However, the evidence for a positive main effect of the treatment on trust in the pre-vote survey is now weaker and model-dependent: It becomes no longer significant in the model without socio-demographic covariates (p = .071) and only marginally significant in the model with covariates (p = .049). Thus, this pre-vote effect should be interpreted with caution. Finally, the coefficients in our original tables were labeled as standardized, but only the income variable was standardized. The tables now correctly indicate unstandardized coefficients (b). 2. Why do we suggest this correction? The market research company that collected the survey data underlying this study made a consequential mistake when recruiting the respondents: A few participants were sampled twice. Correcting this mistake changed the results of some analyses slightly, however, critically, the main results remained largely unaffected. In the following, we explain in detail what happened and how we corrected the mistake. 3. How did we collect data for this study? Data was collected by a market research company with a pool of Swiss panelists. For the first survey (pre-vote), we asked the market research company to resample participants from the trust in science and science-related populism (TISP) study, as well as collecting data from additional participants (see pre-registration). Pre-vote, we created one survey link—which was sent to the market research company—and participants were randomly assigned to either a control or a treatment group. Post-vote, we created two separate surveys: one for the control group and one for the treatment group. We sent the market research company the two survey links—one for the control group and one for the treatment group—with a list of the IDs that had to be invited to each of the surveys. This would ensure that participants who were in the control group pre-vote would be invited to participate in the control group post-vote, and vice versa. 4. What happened? The market research company opened two projects on their end: one to resample TISP participants and one to sample participants from a new pool of panelists. In doing so, the project manager at the company erroneously allowed panelists resampled from TISP to also participate in the second project, thus allowing some participants to participate in the study twice. Since these were separate projects, panelists who participated twice were assigned two different IDs in our survey. On our end, it was therefore impossible to see that the same participants had participated in the study twice albeit with different IDs. Only the market research company could have seen this, because their panelists have unique identifiers on their end that, for data privacy reasons, are not accessible to their clients (i.e. in this case us). As 44 participants were allowed to participate twice in the pre-vote survey, some were assigned to the control group twice, and some to the treatment group twice and in many cases, they were assigned to the control and the treatment group. Since these participants had completed both conditions, their IDs were added to both the control group and the treatment group lists that we sent to the market research company to resample participants post-vote. Therefore, these participants received two invitations in the post-vote survey: one to participate in the control group and one to participate in the treatment group. Some of these participants completed both surveys post-vote. An additional problem we observed, but for which the company has not been able to provide an explanation because the project manager responsible for this project has left the company, is that some participants who had only completed one condition pre-vote were able to access both surveys post-vote (control and treatment), i.e. got access to both survey links. 5. How did we notice this? When running further analyses for another paper using the same dataset, we noticed that some participants were not correctly reassigned to the same condition post-vote. For example, some participants were in the control condition pre-vote and in the treatment condition post-vote. We contacted the market research company to find out how this could have happened. While inspecting this, they noticed that the same participants had completed the pre-vote study twice with different IDs. However, the company could not give an explanation as to how other participants gained access to both survey links post-vote. 6. What did we do to correct the dataset? Pre-vote. We asked the market research company to inform us which IDs were linked to the same participants so that we could inspect duplicate responses from the same participants (we uploaded the raw dataset with this information on OSF). Overall, 44 unique participants filled out the survey twice (with different IDs on our end). For example, the same participant could have had ID = 12 345 when first completing the survey and ID = 67 891 when completing the survey a second time. We created new unique IDs by linking the two IDs and sorting them so that the order of survey completion did not matter (e.g. 12345_67891). We retained the first complete response and deleted the second response. This decreased our sample from n = 1622 to n = 1578. The remaining analyses pre-vote remained unchanged (see code with step-by-step description on OSF). Post-vote. We collected n = 940 complete responses. As in the pre-vote survey, we created unique IDs based on the new information provided by the market research company. In total, we identified 57 duplicate responses, of which 12 were from participants who completed both conditions pre-vote and one participant completed the survey three times, leading to 58 excluded responses. The remaining participants had somehow gained access to both survey links. (Note: In the original publication we removed 48 duplicate responses, namely, those who had already appeared as duplicates on our end). First, we deleted duplicate complete responses by retaining the first post-vote response, as done pre-vote (n = 58 responses removed). Second, we also deleted responses of participants who had been assigned to a different condition pre- and post-vote (n = 49 responses removed). For example, if participant 12345_67891 was in the control group pre-vote and in the treatment group post-vote. Lastly, we removed participants who had completed both conditions pre-vote and had therefore been exposed to both the control and the treatment materials (n = 19 responses removed). Of these n = 19 responses, n = 11 had already been removed because they were assigned to a different condition pre- and post-vote and n = 2 had already been removed during the first step of deleting duplicate responses. Effectively, we therefore removed n = 6 participants in this last step. Together, these steps decreased our final post-vote sample from n = 891 before correction to n = 827 (940 − 58 − 49 − 6 = 827) after correction. All of these steps are detailed in the code published on OSF. Additionally, we note that our tables previously indicated standardized coefficients, but we relabeled them to unstandardized coefficients, as only the income variable was standardized. 7. How did this affect our results? Our substantive conclusions largely remained unaffected by these changes, but the evidence for one result weakened. Pre-vote, the effect of the treatment (vs control) on trust in scientists changed from b = 0.07, p = .047 to b = 0.06, p = .071 in the model without socio-demographic covariates, and from b = 0.08, p = .035 to b = 0.07, p = .049 in the model with covariates. The pre-vote main effect of the treatment is thus no longer significant in the model without covariates and only marginally significant at the conventional p < .05 threshold in the model with covariates. Hypothesis H2a, which we described as partially supported in the original manuscript (the t-test was non-significant), therefore now has weaker partial support. Post-vote, the effect of the treatment on trust in scientists changed from b = 0.12, p = .018 to b = 0.13, p = .014. The interaction effect of the treatment (vs control) and political orientation on trust pre-vote remained significant, changing from b = − 0.10, p = .008 to b = − 0.09, p = .017 in the model with sociodemographic covariates. This interaction term remained non-significant post-vote, changing from b = − 0.03, p = .512 to b = − 0.05, p = .332. More generally, we note that the treatment effects on trust are small. Unstandardized coefficients of 0.06–0.13 on a 5-point scale, corresponding to standardized differences of ca. 0.1–0.2, at or below the smallest effect size (d = 0.20) the study was powered to detect. The fact that effects of this size move across the p = .05 threshold when duplicate participants are removed reflects their fragility. Readers should therefore give more weight to the effect sizes and confidence intervals than to whether individual models fall marginally above or below the significance threshold. For transparency, our OSF folder includes the corrected datasets and code, as well as the original datasets and code. 8. Corrected tables and figure in main manuscript Corrected tables 1–5 as well as corrected figure 1 can be found below. Table 1. Blockwise linear regression on trust in scientists (pre-vote). Trust in scientists (Intercept) 3.49 3.45–3.54 143.69 <.001 3.41 3.28–3.53 52.60 <.001 Advocacy letter treatment 0.06 −0.01–0.13 1.81 .071 0.07 0.00–0.14 1.97 .049 Political orientation (right) −0.06 −0.11—−0.01 −2.39 .017 −0.06 −0.11—−0.01 −2.29 .022 Treatment * Political orientation (right) −0.08 −0.15—−0.01 −2.20 .028 −0.09 −0.16—−0.02 −2.38 .017 Male 0.06 −0.02–0.13 1.55 .121 Language region: Italian −0.05 −0.18–0.08 −0.79 .430 Language region: French 0.03 −0.05–0.11 0.83 .407 Residence: urban 0.17 0.09–0.24 4.45 <.001 Age (in years) −0.00 −0.00–0.00 −0.57 .570 Income (in CHF) 0.03 −0.01–0.08 1.63 .103 Observations 1353 1207 R2/R2 adjusted 0.027/0.025 0.054/0.047 Note: Language regions are compared to German. Residence: 1 = Urban, 0 = Rural. Gender: 1 = Male, 0 = Female, other responses were coded as missing. Political orientation = 1 (strongly left-leaning), 5 (strongly right-leaning). Baseline is placebo letter treatment. Pre-vote survey wave. Political orientation was centered and income standardized. Values in bold are statistically significant at p < .05. Table 2. Linear regression predicting trust in scientists post-voting. Trust in scientists post-voting (Intercept) 3.36 3.16–3.55 34.15 <.001 Advocacy letter treatment 0.13 0.03–0.23 2.47 .014 Political orientation (right) −0.09 −0.16—−0.02 −2.47 .014 Gender: male 0.12 0.02–0.23 2.30 .022 Language region: Italian 0.04 −0.14–0.22 0.45 .651 Language region: French 0.10 −0.02–0.22 1.71 .088 Residence: urban 0.22 0.12–0.33 4.18 <.001 Age (in years) −0.00 −0.01–0.00 −1.74 .083 Income (in CHF) 0.02 −0.03–0.08 0.85 .393 Treatment * Political orientation (right) −0.05 −0.16–0.05 −0.97 .332 Observations 650 R2/R2 adjusted 0.083/0.071 Note: Language regions are compared to German. Residence: 1 = Urban, 0 = Rural. We included the non-preregistered interaction term treatment * political orientation to be able to compare results with the pre-vote survey. Political orientation was centered and income standardized. Values in bold are statistically significant at p < .05. Table 3. Blockwise linear regression on normative perceptions (pre-vote). Normative perceptions of scientists in society and policymaking (Intercept) 3.76 3.70–3.82 123.37 <.001 3.82 3.66–3.98 47.32 <.001 Advocacy letter treatment 0.06 −0.02–0.15 1.49 .137 0.05 −0.04–0.14 1.18 .237 Political orientation (right) −0.13 −0.19—−0.07 −4.19 <.001 −0.14 −0.20—−0.08 −4.26 <.001 Treatment * Political orientation (right) −0.10 −0.19—−0.01 −2.22 .026 −0.08 −0.18–0.01 −1.83 .067 Male −0.00 −0.09–0.09 −0.01 .991 Language region: Italian 0.11 −0.05–0.27 1.34 .181 Language region: French −0.05 −0.14–0.05 −0.89 .374 Residence: urban 0.17 0.08–0.27 3.78 <.001 Age (in years) −0.00 −0.01–0.00 −1.83 .067 Income (in CHF) 0.02 −0.04–0.07 0.58 .565 Observations 1353 1207 R2/R2 adjusted 0.050/0.048 0.072/0.065 Note: Language regions are compared to German. Residence: 1 = Urban, 0 = Rural. Gender: 1 = Male, 0 = Female, other responses were coded as missing. Political orientation = 1 (strongly left-leaning), 5 (strongly right-leaning). Baseline is placebo letter treatment. Pre-vote survey wave. Political orientation was centered and income standardized. Values in bold are statistically significant at p < .05. Table 4. Logistic regression predicting voting intentions pre-vote. Voting intention (yes vs no) (Intercept) 0.01 0.00–0.04 −4.86 <.001 Advocacy letter treatment 0.92 0.57–1.49 −0.33 .739 Trust in scientists 5.02 3.18–8.18 6.70 <.001 Political orientation (right) 0.77 0.58–1.03 −1.76 .078 Political party: SVP 0.21 0.11–0.40 −4.79 <.001 Gender: male 0.83 0.49–1.42 −0.69 .493 Language region: Italian 0.82 0.32–2.21 −0.41 .683 Language region: French 0.63 0.35–1.13 −1.55 .121 Residence: urban 1.84 1.11–3.07 2.36 .018 Income (in CHF) 1.00 0.80–1.30 −0.00 .999 Age (in years) 0.97 0.96–0.99 −3.36 .001 Climate change concern 2.26 1.75–2.97 6.07 <.001 Father of girl: Yes 1.90 0.82–4.66 1.45 .146 Observations 595 R2 Tjur 0.467 Note: Voting intentions: 1 = yes, 0 = no. Model excludes participants who selected the response options ‘I would abstain’ (n = 141) and ‘I do not know’ (n = 198). Language regions are compared to German. Residence: 1 = Urban, 0 = Rural. Income was standardized. Values in bold are statistically significant at p < .05. Table 5. Logistic regression predicting self-reported voting behavior post-vote. Self-reported voting behavior (yes vs no) (Intercept) 0.00 0.00–0.04 −4.60 <.001 Advocacy letter treatment 0.64 0.37–1.11 −1.57 .116 Trust in scientists 5.87 3.67–9.82 7.06 <.001 Political orientation (right) 0.57 0.41–0.78 −3.45 <.001 Political party: SVP 0.62 0.29–1.29 −1.29 .196 Gender: male 0.84 0.46–1.50 −0.58 .559 Language region: Italian 4.66 1.40–18.42 2.35 .019 Language region: French 1.52 0.81–2.92 1.28 .200 Residence: urban 0.79 0.45–1.38 −0.82 .412 Income (in CHF) 1.00 0.80–1.29 0.01 .996 Age (in years) 0.98 0.97–1.00 −1.93 .054 Climate change concern 2.35 1.76–3.19 5.67 <.001 Observations 428 Note: Self-reported voting behavior: 1 = yes, 0 = no. Model excludes participants who did not vote (n = 188) and selected the response options ‘Prefer not to say’ (n = 31). Language regions are compared to German. Residence: 1 = Urban, 0 = Rural. Income was standardized. Values in bold are statistically significant at p < .05. 9. Further corrections in the main manuscript The statement in the abstract that ‘scientists’ advocacy was associated with greater public trust, particularly among left-leaning individuals’ remains supported by the corrected analyses. However, we note that the pre-vote association is small and only marginally significant when controlling for socio-demographic variables. The association among left-leaning individuals (the treatment x political orientation interaction) and the post-vote association remain significant. Before correction: More than half (55.51%) of participants in the treatment condition believed the advocating scientists were objective (14.37% believed they were not objective; 30.11% were neutral). After correction: More than half (55.64%) of participants in the treatment condition believed the advocating scientists were objective (14.32% believed they were not objective; 30.04% were neutral). Pre-vote difference in trust in scientists between the treatment and control group before correction (Mtreatment = 3.51, SD = 0.65; Mcontrol = 3.46, SD = 0.64; t(1618) = − 1.580, p = .114) and after correction (Mtreatment = 3.51, SD = 0.65; Mcontrol = 3.47, SD = 0.63; t(1575) = − 1.425, p = .154). Post-vote difference in trust in scientists between the treatment and the control group before correction (Mtreatment = 3.49, SD = 0.68; Mcontrol = 3.42, SD = 0.68; t(857) = − 1.577, p = .115) and after correction (Mtreatment = 3.50, SD = 0.69; Mcontrol = 3.42, SD = 0.69; t(805) = − 1.654, p = .099). Exploratory paired t-tests showed no significant differences in trust in scientists pre- and post-vote before the correction within the control (tpaired(488) = 1.426, p = .154) and treatment (tpaired(390) = 1.952, p = .052) groups and after the correction within the control group (tpaired(446) = 1.333, p = .183) and treatment (tpaired(379) = 1.930, p = .054) groups. Pre-vote differences in normative perceptions between the treatment and control group before correction (Mtreatment = 3.79, SD = 0.84; Mcontrol = 3.74, SD = 0.80; t(1615) = − 1.280, p = .201) and after correction (Mtreatment = 3.79, SD = 0.84; (Mcontrol = 3.75, SD = 0.79; t(1571) = − 0.964, p = .335). Post-vote differences in normative perceptions between the treatment and control group before correction (Mtreatment= 3.80, SD = 0.85; Mcontrol = 3.62, SD = 0.87; t(863) = − 3.161, p = .002) and after correction (Mtreatment= 3.82, SD = 0.85; Mcontrol = 3.64, SD = 0.87; t(810) = − 2.971, p = .003). Paired t-test for normative perceptions between control groups pre and post before correction (tpaired(488) = 2.984, p = .003) and after correction (tpaired(446) = 2.588, p = .01). In the discussion section, ‘right-leaning’ was corrected to ‘left-leaning’ in the sentence below. This corrects a reporting error in the original text rather than a change resulting from the corrected data. The negative coefficient for political orientation (right) in table 3 of the original manuscript already indicated that more right-leaning participants held weaker, not stronger, normative perceptions, so the original sentence contradicted the reported model. The corrected sentence reads: ‘Furthermore, residents of urban (vs rural) regions as well as people with more right-leaning political views tended to have stronger normative perceptions regarding scientists’ involvement in society and policymaking (Cologna et al 2024).’ Before correction: About half (53.3%) of our participants indicated that they planned to vote in favor of the CPL in the pre-vote survey (25.2% intended to vote against, 8.8% to abstain, and 12.8% did not know). After correction: 53.6% planned to vote in favor (24.9% against, 8.9% abstain, 12.6% did not know). Before correction: 61.7% reported that they actually voted in favor of the law, which remained unchanged after correction.ants indicated that they planne Before correction: the final pre-vote sample was n = 1622 (n = 815 control, n = 807 treatment), 49.6% female, mean age 44. After correction: n = 1578 (n = 789 control, n = 789 treatment), 49.9% female, mean age 44. Before correction: the final post-vote sample was n = 891 (n = 489 control, n = 402 treatment), 46% female, mean age 47, resampling rate 59.25%. After correction: n = 827 (n = 447 control, n = 380 treatment), 45.9% female, mean age 48, resampling rate 52.41%. V.C. acknowledges support from the Swiss National Science Foundation Postdoc Mobility fellowship (P500PS_202935) and the Harvard University Faculty Development Fund. Conflict of interest R.K. and M.S.S. are signatories of the public statement in support of the Climate Protection Law.

Publication date: Available online 30 October 2026 Source: Physiology & Behavior Author(s): Seungyul Lee, Teri Kim

to RePEc:cup:bjposi:v:46:y:2016:i:02:p:479-480_00

Frontiers in Oncology has retracted a 2025 corrigendum on the anticancer effects of ABTL0812 in glioblastoma models after the authors failed to satisfactorily explain concerns about image integrity.