The causal impact of local weather anomalies on beliefs about the occurrence of climate change

Research has demonstrated that members of the public recognize anomalous weather patterns, and that subjective perceptions of the weather are related to beliefs about the occurrence of climate change. Yet despite two decades of scholarship and dozens of studies, inconsistent and insufficient data have made it difficult to credibly identify the causal impact of objective experiences on perceptions, and the impact of perceptions on beliefs regarding climate change occurrence. Here, we overcome these limitations by collecting and analyzing data from a 5-y panel survey of 2,500 individuals in Oklahoma, a US state that is highly divided on questions about climate change. Our findings indicate that the relationship between local weather anomalies and climate change beliefs is heavily dependent on baseline beliefs about whether climate change was occurring. For people who did not believe in climate change in the initial survey in our series, perceptions of anomalously hot and dry seasons shifted their beliefs towards the occurrence of anthropogenic climate change, whereas their perceptions of anomalously cool and wet seasons shifted their beliefs away from anthropogenic climate change. This relationship was not present among people who believed that climate change was occurring at the beginning of the study; their perceptions of seasonal temperature and precipitation anomalies had no effect on their beliefs about climate change. These patterns have substantial implications for the evolution of public beliefs about climate change.

Resolving the inconsistencies in this area of research has obvious implications for scientific knowledge and society.As climate change intensifies, an increasing portion of the population will experience anomalous weather patterns such as heat waves and drought.At the same time, people will continue to experience weather and climate extremes that have little or no relationship to climate change, such as an unusually cold day in February 2015 that allowed a Senator to bring a snowball to floor as 'evidence' climate change is not happening.This tension makes the connection between local weather conditions and climate change beliefs especially important.If there is a connection, the prevailing forces of climate change will generate weather patterns that push people towards recognition, but idiosyncratic weather extremes have the potential to induce a backslide.Therefore, scientists and policymakers need to know if and how weather experiences will shape the future of climate change beliefs, behaviors, and policy attitudes.
There are two primary factors that have limited scientific progress in this domain.The first is inconsistency in the questions studies address.Some studies focus on the relationship between objective and subjective experience (a in figure 1 Deryugina 2013, Donner andMcDaniels 2013).While some studies address more than one of these questions at once, the inconsistency from study to study causes confusion, fragmentation of knowledge, and an incomplete understanding of the underlying cognitive mechanisms that relate experiences to beliefs.
Insufficient data and inadequate causal identification also limit scientific progress.Questions about the extent to which experiences shape beliefs are inherently dynamic.They imply that individuals modify or change beliefs when they experience different conditions.Despite the dynamic nature of this process, most studies use static data from a single point in time to explore the dynamic relationships in figure 1 (e.g., Egan and Mullin 2012, Goebbert et al 2012, Hamilton and Stampone 2013, Howe et al 2015, Bergquist and Warshaw 2019, Lujala and Lein 2020, Hughes et al 2020).Although static data provide important evidence about the association between experiences and beliefs, they do not allow for a direct test of the extent to which individuals update or change their beliefs after a given experience.A few recent studies use dynamic data, and while they shed important light on the on the relationship between objective experience and climate change beliefs, they generally follow path c in figure 1, sidestepping questions about the underlying cognitive mechanisms that link objective experiences to subjective perceptions and subjective perceptions to climate change beliefs (Palm et al 2017, Kim et al 2021).This is particularly problematic for building knowledge about path b in figure 1, the relationship between subjective experience and climate change beliefs, because the two variables are likely endogenous.In some cases, subjective experiences shape climate change beliefs.In others, baseline beliefs about climate change may shape the conditions that individuals perceive (d in figure 1).
This study addresses these two challenges.We use data from a 5-y quarterly panel survey of 2,500 individuals to wholistically and causally identify the relationships in figure 1.We begin by examining path a, the impact of objective experience (deviations from average temperature and precipitation) on subjective experience (perceptions of deviations).We then examine path b, the relationship between subjective experience and climate change beliefs (certainty that climate change is occurring).In doing so, we provide a complete assessment of the underlying cognitive mechanisms that connect objective experiences to subjective perceptions and climate change beliefs.Likewise, the data and identification strategy we employ removes the endogeneity driven by path d in figure 1 to provide a credible estimate of the extent to which an individual's subjective experiences cause them to update their beliefs about climate change.These advances allow us to address some of the most perplexing aspects of the public's evolving understanding of climate change, and therefore of public support for policies and candidates that combat climate change.Our results provide strong evidence that individuals recognize deviations from average temperatures and precipitation in their area.Our results also indicate that, on average, subjective perceptions causally affect beliefs about the occurrence of climate change.When survey respondents perceived a deviation from average, they adjusted the amount of certainty they had that climate was changing.When they perceived that a season was unusually warm (cool), they become more (less) certain that anthropogenic climate change was occurring.Perceptions of anomalously wet or dry seasons also caused respondents to modify their beliefs, though these effects were more modest in magnitude.
Though the average effects are significant, our results also indicate that the impact of perceptions on beliefs is substantively different for climate change skeptics and believers.For respondents who began the survey series believing that climate change was occurring (believers), perceptions of anomalous weather did not cause them to change their climate beliefs from survey to survey.For skeptics, on the other hand, perceptions had a statistically and substantively significant effect on their beliefs about climate change.While this finding suggests that local weather pattens may ultimately sway skeptics towards believing in climate change, it is a double-edged sword in that cooler, wetter weather anomalies can undermine acceptance that the climate is changing.We discuss the implications of this finding in the conclusion.

Measures, methods, and models
We used data from a 5-y quarterly panel survey to investigate the effect of weather perceptions on climate change beliefs (Jenkins-Smith et al 2017). 3The survey respondents reside in Oklahoma, a state in the South Central US.Oklahoma is an ideal testbed for this analysis because it offers significant variation on the core variables of interest.Residents experience high sub-seasonal-to-seasonal variability in weather patterns and there is significant disparity in beliefs about climate change.According to recent estimates, 64% of Oklahoma residents believe global warming is happening; the others are not sure or believe that it is not happening (Marlon et al 2022).
The surveys were administered at the conclusion of each 3-mo season (winter, spring, summer, fall) from 2014 to 2020.Shown in table 1, each survey included questions that measured seasonal temperature perceptions, precipitation perceptions, and climate change beliefs.We combined the panel survey data with information on observed weather conditions collected from the Oklahoma Mesonet, a dense network of environmental monitoring stations that that began providing and archiving high-resolution data on weather and climate conditions in Oklahoma in 1994 (Brock et al 1995, McPherson et al 2007).There are 120 separate monitoring stations across Oklahoma-each of the state's 77 counties contains at least one station-and each one measures and records a wide variety of weather conditions in five-minute increments, including temperature and precipitation.We used these data to calculate seasonal deviations from daily average temperatures and precipitation amounts at each survey respondent's location following each 3-mo season (see table 1).
This combination of the panel survey and Mesonet data allowed us to systematically analyze the relationships between objective experience, subjective perceptions, and climate change beliefs in Oklahoma.The repeated measures of these constructs render our data uniquely well-suited for providing insight into the dynamics of the relationships between them.Moreover, the panel data structure allowed us to explore these relationships using within-respondent variation, which mitigates concerns that the relationships we observe are driven by geographic sorting or any other time-invariant characteristics of the respondents.These characteristics alone allowed us to credibly identify path a in figure 1, the impact of objective experience on subjective perceptions.Credible identification of path b, the impact of perceptions on beliefs, required an explicit strategy to remove the endogeneity driven by path d, the reverse impact of beliefs on perceptions.Given the structure of our data, we adopted an instrumental variable (IV) strategy, where objective experiences (temperature/precipitations deviations) serve as instruments that exogenously affect perceptions but have no independent impact on climate beliefs.
Valid instruments must satisfy two main criteria, the relevance assumption and exclusion restriction.The relevance assumption holds that there must be a strong correlation between the instrument and the potentially endogenous variable.Our exploration of path a (shown below) indicated that this was clearly the case.There was a strong positive correlation between temperature and precipitation deviations and perceptions of those deviations.The exclusion restriction requires that the instrument has no direct effect on the outcome of interest; it only affects the outcome of interest through the potentially endogenous variable.In this context, the assumption holds that objective experiences (temperature/precipitation deviations) only influence climate change beliefs by affecting perceptions.The exclusion restriction is fundamentally untestable, but in this case a strong case could be made for its validity.It is extremely unlikely that a respondent would update their beliefs about climate in response to seasonal weather deviation if they did not perceive that deviation.
Having satisfied these assumptions, we leveraged these instruments in a regression framework, estimating models that employ techniques appropriate for the panel structure of our data.Specifically, we implemented the IV strategy using a two-stage least squares estimator.The stage-one models took this form: where perceptions, P, of individual i at time t are a function of observed local weather deviations, D, a respondent fixed effect, g , i and an error term, e .it In addition to providing inputs into the stage-two models, the stage-one models identify the effect of objective experience on subjective perceptions (path a in figure 1).They also provide evidence to support the relevance assumption.
After estimating equation (1), we generated predicted values of P, denoted by P in equation (2) below, and used them in stage-two models of the form: where the climate change beliefs, C, of respondent i at time t are a function of the predicted value of their perceptions, P, and, as in the stage-one models, a respondent fixed effect and an error term.We clustered standard errors by respondent.By only using variation in respondents' weather perceptions attributable to plausibly exogenous observed weather deviations when estimating the relationship between perceptions and climate change beliefs, we can ascribe a causal interpretation to d; it represents the causal effect of an individual's weather perceptions on their climate change beliefs (path b in figure 1).We operationalized P in four different ways.The first two operationalizations focused on temperature perceptions, with the first specifying P as an indicator for the respondent perceiving the prior three months to be warmer than average and the second specifying P as an indicator for the respondent perceiving the previous three months as colder than average.For these specifications of P, we operationalized D as the deviation from average temperature at the respondent's location in the prior three months.Our third and fourth operationalizations of P focused on precipitation, with indicators for perceiving the past three months to be drier than average and wetter than average, respectively.In these models, we operationalized D as the deviation in total precipitation at the respondent's location in the season of interest.In all four models, we operationalized C using a scale that ranged from −10 to 10, where −10 indicated complete certainty that anthropogenic climate change was not causing average global temperatures to rise, 10 indicated complete certainty that it was, and 0 indicated complete uncertainty (see Jenkins-Smith et al 2020).Lastly, to account for possible heterogeneity in effects, we estimated these models using the full sample of respondents, then re-estimated them using subsamples of climate change believers and skeptics.We operationalized these groups by identifying the baseline view of each respondent when they began the survey. 4

Results
Table 2 presents the results from estimating equations (1) and (2).The top row presents the stage-one results.It presents the estimated effect of observed temperature and precipitation deviations on respondents' weather perceptions, the parameter represented by b in equation (1) and path a in figure 1.The results indicate that observed temperature and precipitation anomalies are robust predictors of respondents' weather perceptions.The first column indicates that each one-degree increase in average daily temperature in a season (relative to average) increased the probability that a respondent perceived the season as abnormally warm by 0.13.Conversely, the second column shows that each one-degree decline in average daily temperature (relative to average) increased the probability of a respondent perceiving the season as abnormally cool by 0.10.Precipitation deviations also had a strong effect on perceptions.As the fourth column table 2 indicates, each one-4 Specifically, we began by identifying each respondent's initial appearance in our dataset and recorded their answer to the following question: 'As you may know, the issue of global climate change has been the subject of public discussion over the last few years.In your view, are greenhouse gases, such as those resulting from the combustion of coal, oil, natural gas, and other materials, causing average global temperatures to rise?' We assigned individuals who responded 'Yes' to the group of baseline climate change believers-we assigned 2,103 respondents to this group.We assigned individuals who responded 'No' to the previous question to the group of baseline climate change skeptics-we assigned 1,760 respondents to this group.Then, separately for each group, we applied the panel instrumental variables approach described above to estimate the effects of weather perceptions on individuals' certainty with which climate change is occurring.inch increase in precipitation (relative to average) increased the probability that a respondent perceived the season as abnormally wet by 0.04.Each one-inch decrease in precipitation (relative to average) increased the probability that a respondent perceived the season as abnormally dry by 0.03.
Figure 2 illustrates these relationships using scatterplots that group respondents into 20 bins that represent the range of deviations respondents experienced during the survey.The points on the plot indicate the proportion of respondents in each bin who said that the season was unusually warm, cool, dry, or wet.As the top two panels illustrate, the proportion of respondents who perceived a season to be abnormally warm (cold) significantly increases (decreases) as the deviation from average temperature increases.The bottom two panels illustrate the same pattern for precipitation, but the effect is somewhat less striking.Together, these results clearly show that respondents were adjusting their perceptions in response to deviations from average temperatures and precipitation.Path a in figure 1 was noticeably present.In demonstrating this relationship, the stage-one results also indicated that both variables (precipitation and temperature deviations) clearly satisfied  the relevance assumption necessary to credibly estimate the causal effect of perceptions on beliefs using the IV strategy we outline above.Row 2 in table 2 presents the stage-two results.It presents the estimated causal effect of respondents' weather perceptions on their climate change beliefs.The estimates correspond to the parameter represented by d in equation (2) and path b in figure 1.On average, the results indicate that variation in perceptions caused respondents to update their beliefs about climate change.For temperature, perception that a season was abnormally warm increased certainty that climate change was occurring by roughly half a point on the 21-point scale.On the flip side, perception that a season was abnormally cool reduced certainty that climate change was occurring by about half a point.The impact of precipitation perceptions on climate change beliefs was statistically significant as well, but the effect was more modest in magnitude.Perception that a season was abnormally dry increased certainty in climate change beliefs by about one-third of a point.Perception that it was abnormally wet decreased certainty by one-quarter of a point.While the magnitude of the effect is somewhat small, these results provide causal evidence that, on average, respondents used their perceptions of the weather, particularly temperature, as the basis for updating their climate change beliefs.They indicate that path b in figure 1 is also present.
While the results in table 2 provide clear evidence that, on average, objective experiences affect subjective perceptions, and subjective perceptions cause individuals to update climate change beliefs, they provide little insight into potential heterogeneity in these effects across different groups of respondents.There are many groups in society and many possible sources of heterogeneity, but the most obvious cleavage in debates about climate change is the distinction between skeptics and believers.Past research indicates that climate skeptics exhibit significantly more variation in climate change beliefs than believers (Jenkins-Smith et al 2018).It is possible that this asymmetric variation in beliefs is driven, in part, by heterogeneity in the relationships between objective conditions, subjective perceptions, and beliefs.Climate change beliefs among skeptics, for example, may be more sensitive to variation in perceptions than beliefs among believers, because believers recognize the disjoint relationship between local-scale weather conditions and global-scale climate change.To empirically assess this possibility, we re-estimated the models discussed above using sub-samples of climate change believers and skeptics.For context, 2,027 respondents stated that they believed global climate change was not occurring the first time they took the survey (skeptics); 2,568 stated that they believed it was occurring (believers).
Table 3 presents the results of this analysis.The stage-one estimates were remarkably consistent across the two groups, indicating that baseline beliefs about climate change had little or no effect on the relationship between objective conditions and subjective perceptions, path a in figure 1.The stage-two results, by contrast, indicated significant heterogeneity across the sub-samples.There was a substantively strong and statistically significant relationship between perceptions and beliefs among respondents who did not believe that climate change the first time they took the survey (skeptics).In this group of respondents, perception that a season was abnormally warm increased certainty that climate change was occurring by nearly one point on the 21-point scale.Perception that a season was cooler than average reduced certainty that climate change was occurring by a little more than one point.The effect of precipitation perceptions was also substantial.Perception that a season was abnormally dry increased certainty in climate change beliefs by a little more than one point and perception that a season was abnormally wet decreased certainty by three-quarters of a point.Among believers, those who began the survey series stating that climate change was occurring, there was no relationship between perceptions and beliefs.These respondents perceived anomalous weather patterns, but those perceptions did not cause them to update their beliefs about climate change.
Figure 3 illustrates these findings by recreating figure 1 for the two sub-populations.On average, both groups noticed when temperatures and precipitation amounts departed from average, but only skeptics used this information to update their beliefs about climate change.
Our analyses to this point provide strong evidence that individuals' weather perceptions impact their climate change beliefs, with those effects concentrated among individuals who did not believe in climate change at the time they completed their initial survey.In additional analysis, which we present in much more detail in appendix A, we explore how cumulative weather conditions over a multi-year period affect climate change beliefs.Specifically, we address this issue using the repeated observations of our survey respondents to first perform three calculations: (1) The change in climate change beliefs-using the measure of climate change beliefs we use throughout the paper-from the respondent's first survey wave to their final survey wave; (2) The cumulative temperature departure from the 15-year averages experienced by each respondent (using all available observations for each respondent; (3) The total number of periods the respondent perceived as abnormally warm (again using all available observations).Then, using these three measures and an instrumental variables (IV) strategy, we estimate the effect of the cumulative number of periods a respondent perceived as abnormally warm on their change in climate change beliefs, instrumenting the number of periods the respondent perceived as abnormally warm with the measure of the cumulative temperature departure.Our results show that (instrumented) cumulative perceptions of abnormally warm periods strongly predict changes in respondents' climate change beliefs, implying that our main findings can generalize across time.
A second additional analysis, which we again present in much more detail in appendix A, assesses whether a respondent's participation in the survey makes them increasingly likely to perceive changing weather conditions.To examine this possibility empirically, we estimate a regression model that predicts a respondent perceiving a season as abnormally warm with temperature deviation from the 15-year average, respondent wave, and an interaction between the measures of temperature deviation and respondent wave.Our results illustrate that the interaction is not statistically significant, indicating that respondents' ability to perceive actual weather conditions did not meaningfully change throughout their time as a survey respondent.
A final analysis, again detailed in appendix A, examines whether our results are sensitive to estimating the empirical model over a balanced sample that only includes respondents who participated in every wave of the survey.The results of this analysis are remarkably consistent with the primary results in table 2-the estimates just have somewhat less precision given the reduced sample size.Overall, though, these results provide strong evidence that our results are not a function of differential sample composition across time.And, together, the results of these three supplementary analyses instill additional confidence in our primary results.

Implications
The evidence in this study suggests that, on average, members of the public perceive anomalous weather patterns and these perceptions cause them to update their beliefs about the occurrence of climate change.But the average effects mask significant heterogeneity in the population.The connection between objective conditions and beliefs appears to be quite different for climate change believers and climate change skeptics.Believers, who accept that climate is changing, notice when a season departs from average, but this does not cause them to become more/less certain that climate is changing.Skeptics, who initially doubted the occurrence of climate change, also recognize departures from average, but they (unlike believers) use this information to update their beliefs about climate change.They notice when a season is usually hot or dry, and noticing this causes them to become a bit more certain that climate change is occurring.They also notice when a season is unusually cool or wet, and this causes them to be a bit less certain that climate change is occurring.
Is this good news?On one hand, weather is not climate and weather anomalies have been used by both climate change advocates and deniers to buttress their positions.The evidence in this study suggests that these tactics are chiefly effective among climate skeptics.Pointing to unseasonably cold temperatures, for example, is likely to reinforce skeptic's beliefs that climate change cannot be occurring.But the opposite is also true; drawing attention to unusually hot or dry conditions may make them a little less certain of their underlying beliefs about climate change.This dynamic suggests that individuals who doubt the occurrence of climate are unlikely to settle on an underlying opinion.Like local weather patterns, their beliefs will continue to vary from season to season.On the other hand, there is an underlying current in the variability of local weather patterns; on average, temperatures are rising and, despite variation from season to season, individuals in most parts of the world are going to experience warmer than average temperatures more frequently than they experience cooler than average temperatures in the future.As this current continues, we expect that skepticism will gradually erode.
An important caveat is required here: the data for this study were collected over a period that was anomalously wet and cool in Oklahoma, and among a population (Oklahoma residents) especially prone to skepticism about climate change (Howe et al 2015).Further, while the data captured significant variation in temperature anomalies, there was less variation in precipitation (see figure 2).For that reason, our data may be limited in the degree to which they can inform inferences about the magnitudes of the causal linkages between weather, weather perceptions, and climate beliefs during periods of persistently dry and hot weather.It is possible that our estimates are generalizable to such a period, leading to a rapid shift in public opinion about the certainty with which climate change is occurring.Alternatively, it is plausible that such a scenario will generate cognitive dissonance for skeptics, and lead some to de-link weather perceptions from beliefs about anthropogenic effects on climate, generating adaptive narratives designed to preserve deeply held beliefs.Some of these narratives already circulate widely, such as the argument that humans have little demonstrable impact on changing weather and climate (Koonin 2014(Koonin , 2021)).For these reasons, it is important that scholars of climate change opinion expand the available data, both to other populations and over different patterns of weather anomalies.
respondent (using all available observations for each respondent; (3) The total number of periods the respondent perceived as abnormally warm (again using all available observations).Then, using these three measures and an instrumental variables (IV) strategy, we estimate the effect of the cumulative number of periods a respondent perceived as abnormally warm on their change in climate change beliefs, instrumenting the number of periods the respondent perceived as abnormally warm with the measure of the cumulative temperature departure.Specifically, the first-stage model predicts the total number of periods a respondent perceived as abnormally warm with the cumulative temperature departure from the 15-year averages and based on model results, generates the predicted number of periods perceived as abnormally warm.The second-stage regression predicts change in climate change beliefs using the predicted number of periods the respondent perceived as abnormally warm.Essentially, this analysis sheds light on how, using data from each respondent's full set of survey waves, respondent perceptions over a long period of time caused their climate change beliefs to change.We performed this analysis separately for two analytic samples: (1) All respondents, and (2) Only respondents who were in the survey for at least 12 waves.
Table A1 below provides the results of this analysis.The first stage shows that, as expected, cumulative temperature departures strongly predict the cumulative number of seasons the respondent perceives as abnormally warm.More importantly, though, the second-stage results show that cumulative perceptions of abnormally warm periods strongly predict changes in respondents' climate change beliefs.Substantively, the results imply that the findings can generalize across time-at least across a three-year period-and are not just a wave-to-wave phenomenon.
Does survey participation sensitize respondents to perceiving weather conditions?It is possible that, over time, respondent's participation in the survey could makes them increasingly likely to perceive changing weather conditions.To examine this possibility empirically, we estimate a regression model that predicts a respondent perceiving a season as abnormally warm with temperature deviation from the 15-year average, respondent wave, and an interaction between the measures of temperature deviation and respondent wave.If survey participation were serving as an 'information treatment,' then we would expect a significant, positive coefficient on the interaction between temperature deviation and respondent wave-a significant positive coefficient would indicate that the relationship between actual weather conditions and respondents' perceptions of those conditions strengthened over time.
The results presented in table A2 below, however, illustrate that the interaction is not statistically significant, indicating that respondents' ability to perceive actual weather conditions did not meaningfully change throughout their time as a survey respondent.As such, the analysis provides evidence that our results are not attributable to survey participation per se, and rather represent a true effect of perceptions on climate change beliefs.
Panel structure and implications for empirical results.In this section, we provide additional detail on the structure of the data resulting from the five-year, quarterly panel survey we leverage in our analyses.Like any survey with 20 waves, there was some respondent attrition throughout its conduct, which we backfilled to meet our target sample size for each wave.Even with this attrition and backfill, there were, on average, there were 10 observations per respondent.In most cases, people responded to consecutive waves (with no gaps between waves), but there were some gaps in responses from a few respondents.To illustrate this point, we calculated the average gap per person, with one begin the baseline (one increment of time between each observation).For this sample, the mean gap was 1.19 with a standard deviation of 0.49.The gaps were accounted for using panel regression models that account for unbalanced data.Most respondents entered at the beginning of the survey (in 2016), but recruitment for attrition continued throughout the survey.The most significant recruitment effort was in 2019 (wave 16), when approximately 1,000 new respondents were added to the panel.This batch of respondents was, on average, slightly more likely to believe that climate change is occurring; 63% of people that joined in wave 16 believed that climate change was occurring when they joined the panel; only 55% of people who joined the panel before wave 16 believed that climate change was occurring.Although unlikely, it is possible that gaps in the data and changes in the underlying sample could lead to changes in results.This possibility prompted us to re-run our models over a balanced sample that only includes respondents who participated in every wave of the survey.As shown in table A3 below, the point estimates are remarkably consistent with our primary results presented in the body of the paper-the estimates just have somewhat less precision given the reduced sample size.Overall, though, these results provide strong evidence that our results are not a function of differential sample composition across time.

Figure 1 .
Figure 1.Hypothesized causal pathways from experienced weather to perceived weather patterns and beliefs about climate change.

Figure 2 .
Figure 2. Binned scatterplots showing the proportion of respondents who indicated that a given season was (A) warm, (B) cool, (C) dry, (D) wet in relation to the temperature (A), (B) and precipitation (C), (D) deviations in the given season.

Figure 3 .
Figure 3. Empirically observed causal pathways from experienced weather to perceived weather patterns and beliefs about climate change.
Precipitation DeviationsSubtract mean total rainfall at the respondent's location in the 3-mo season before survey completion from the 15-yr average total rainfall for that season at that location.Temperature Perceptions Response to survey question: Would you say that this [season] has been warmer, cooler, or about the same as previous [seasons]?-Abnormally warm (1) or not abnormally warm (0) 0.31 0.46 -Abnormally cool (1) or not abnormally cool (0) Precipitation Perceptions Response to survey question each time a respondent took the survey: Would you say that the amount of precipitation that fell this [season] was more, less, or about the same amount as in previous [seasons]?-In your view, are greenhouse gases, such as those resulting from the combustion of coal, oil, natural gas, and other materials, causing average global temperatures to rise? -Complete certainty that climate change is not occurring (-10) -On a scale from zero to ten, where zero means not at all certain and ten means completely certain, how certain are you that greenhouse gases are/are not causing average global temperatures to rise? -Complete uncertainty (0) Complete certainty that -Complete certainty that climate change is occurring (10) Baseline Climate Change Beliefs Response to survey question the first time a respondent took the survey: In your view, are greenhouse gases, such as those resulting from the combustion of coal, oil, natural gas, and other materials, causing average global temperatures to rise?

Table 2 .
Coefficients and standard errors from panel instrumental variables regressions.Each column presents results from a panel instrumental variables regression.The first-stage model predicts an individual's weather perceptions as a function of observed deviations in average temperature (left-hand panel) or total seasonal precipitation (right-hand panel) from the 15-year climatological average and a respondent fixed effect.The second-stage model predicts an individual's certainty about the occurrence of climate change as a function of the predicted values of a respondent's weather perceptions generated from the first-stage regression and a respondent fixed effect.

Table 3 .
Coefficients and standard errors from panel instrumental variables regressions, by baseline climate change beliefs.p< 0.10; ** p < 0.05; *** p < 0.01.Robust standard error clustered by respondent in parentheses below coefficient estimate.Separately for the top and bottom panels of the table, each column presents results from a panel instrumental variables regression.The first-stage model predicts an individual's weather perceptions as a function of observed deviations in average temperature (left-hand panel) or total seasonal precipitation (right-hand panel) from the 15-year climatological average and a respondent fixed effect.The second-stage model predicts an individual's certainty about the occurrence of climate change as a function of the predicted values of a respondent's weather perceptions generated from the first-stage regression and a respondent fixed effect. *

Table A1 .
Coefficients and standard errors from instrumental variables regressions.

Table A3 .
Coefficients and standard errors from panel instrumental variables regressions.

Table A2 .
Coefficients and standard errors from panel regression with individual fixed effect.* p < 0.05; *** p < 0.01.Robust standard error clustered by respondent in parentheses below coefficient estimate.