A series of polls provides new tests for how weather influences public beliefs about climate change. Statewide data from 5000 random-sample telephone interviews conducted on 99 days over 2.5 yr (2010–12) are merged with temperature and precipitation indicators derived from U.S. Historical Climatology Network (USHCN) station records. The surveys carry a question designed around scientific consensus statements that climate change is happening now, caused mainly by human activities. Alternatively, respondents can state that climate change is not happening, or that it is happening but mainly for natural reasons. Belief that humans are changing the climate is predicted by temperature anomalies on the interview day and the previous day, controlling for season, survey, and individual characteristics. Temperature effects concentrate among one subgroup, however: individuals who identify themselves as independent, rather than aligned with a political party. Interviewed on unseasonably warm days, independents tend to agree with the scientific consensus regarding anthropogenic climate change. On unseasonably cool days, they tend not to agree. Although temperature effects are sharpest for just a 2-day window, positive effects are seen for longer windows as well. As future climate change shifts the distribution of anomalies and extremes, this will first affect beliefs among unaligned voters.
While the evidence-based scientific consensus that humans are changing Earth’s climate has strengthened in recent years, acceptance of this idea among the U.S. public has not followed suit. Instead, there have been fluctuations without clear trend in overall levels of acceptance. The surface calm masks deepening partisan division (McCright and Dunlap 2011). Survey analysis finds that political outlook dominates other characteristics in predicting individual beliefs about climate. Even education and science literacy have divergent effects, depending on politics (Hamilton 2008, 2011b, 2012; Hamilton et al. 2012; Kahan et al. 2011a,b; McCright 2011; McCright and Dunlap 2011). Among self-identified Democrats or liberals, higher education and science literacy are associated with greater concern regarding anthropogenic climate change. Among Republicans or conservatives, education and science literacy have weak or even negative effects. Among unaligned respondents, both climate change beliefs and the effects of education or science literacy fall between these partisan extremes.
Scientists, political figures, journalists, and bloggers offer the public competing interpretations of observed changes such as the decline of Arctic sea ice (Notz and Marotzke 2012), global temperature and sea level rise (Rahmstorf et al. 2012), and the frequency and severity of climate extremes such as droughts and heat waves (Trenberth and Fasullo 2012). These interpretations, including true and false facts, often are differentially accepted by people according to their preexisting beliefs—a process called biased assimilation (Lord et al. 1979; Munro and Ditto 1997; Corner et al. 2012; Hamilton 2012). Personal recollections of climate or weather events could be filtered in a similar fashion, and interpreted differently in accord with more general beliefs (Goebbert et al. 2012). Trivial personal experiences such as ambient temperature, thirst, and visual cues can affect responses to climate questions as well (Joireman et al. 2010; Lewandowski et al. 2012; Li et al. 2011; Risen and Critcher 2011), further evidence that scientific and political arguments are not the only forces affecting public beliefs.
Recently, survey researchers have tested whether objectively measured climate or weather influence climate beliefs. Analyzing data from diverse rural areas, Hamilton and Keim (2009) found that perceived local impacts of climate change were highest in snow-country regions with winter warming trends. Shao (2012) found associations between decadal temperature trends and public concern about global warming. Akerlof et al. (2013) report associations between personal experience (to some degree objectively confirmed) and perceptions of global warming risks. Goebbert et al. (2012) examined relationships between self-reports of temperature, drought, or flood experience and corresponding objective long-term indicators. Controlling for ideology and worldview, their temperature-comparison indicator had little effect on perceived temperature experience. Objective flood and drought-comparison indicators, however, exhibit stronger effects on perceptions. Egan and Mullin (2012) tested for effects of temperature anomalies on nationally representative surveys. They found a significant effect of daily temperature on expressed beliefs about the evidence for global warming (not necessarily anthropogenic). This effect appears short lived for normal temperature variations, being strongest when just a 2-day window before the interview is considered. Exceptional heat waves have longer-lasting effects. Political identity and education moderate the temperature effects.
In the analysis that follows, we test for weather effects on agreement with the scientific-consensus view of climate change. We find impacts from temperature generally consistent with earlier studies, but they place into sharp focus the question of whose beliefs change with the weather.
2. The Granite State Poll
Four times each year, the Granite State Poll conducts telephone interviews with random samples of about 500 New Hampshire (NH) residents. This poll achieves prominence during New Hampshire’s presidential primaries and other elections. It employs standard, well-validated techniques for obtaining representative samples of the state’s adult population. Probability weights permit minor adjustments for design and sampling bias—the latter, from comparison with census data. Typical Granite State Poll questions cover political and opinion topics; several climate questions were added to the mix beginning in April 2010 (Hamilton 2010). By July 2012, climate questions had been carried on 10 survey cycles, involving more than 5000 interviews on 99 separate dates. A national Carsey Institute survey in 2011 [National Community and Environment in Rural America (NCERA)] asked the same climate questions and obtained results similar to those from New Hampshire (Hamilton 2011a, 2012). The quarterly New Hampshire data, however, provide a high-quality and uniquely resolved time series.
Table 1 describes variables in our analysis. For comparison, we show both statewide and national results on the climate questions. One response to the personal belief question, “Climate change is happening now, caused mainly by human activities,” corresponds to the main point of statements and reports on climate change from leading science organizations (Oreskes 2004), national academies of science (G8+5 2009; NRC 2010), reviews of research results (Solomon et al. 2007; Richardson et al. 2009), and surveys of scientists (Doran and Zimmerman 2009). Similar statements containing the same elements as our survey question have been made by many science organizations (e.g., AMS 2012). No leading science organization takes a position of disagreement that human activities are now changing the Earth’s climate. Telephone interviewers rotated the order in which responses were read to avoid possible bias. Fifty-three percent of New Hampshire respondents, and 52% nationwide, chose the now/human response to this question.
Background characteristics in Table 1 are the chief predictors identified by previous research (McCright and Dunlap 2011; Hamilton 2012). Our respondents could describe their politics on a seven-point scale from 1 = “strong Democrat” to 7 = “strong Republican,” with 4 = “independent” being the middle category (about 2%, set aside here, chose “other” or declined to answer). In keeping with the common poll finding that people who say they are “independent but leaning” toward one party behave as partisans in practice, we grouped the 1–3 categories as Democrats and the 5–7 categories as Republicans; only the neutral 4 category is counted here as independents (about 900 people or 18% of the estimation sample).
On 10 surveys from April 2010 to July 2012, the now/human response to our belief question fluctuated from 49% to 57%, without trend. Some fluctuations no doubt reflect sampling variation. Some could reflect unmeasured factors such as current events. If we adjust for respondent characteristics, for season, and for survey-to-survey fluctuations caused by unmeasured factors, are there detectable weather effects in these data?
3. New Hampshire weather data
Our focus on one relatively small state simplifies the integration of survey with weather data. Daily temperature and precipitation observations representing the weather across New Hampshire were obtained from the U.S. Historical Climatology Network (USHCN, http://cdiac.ornl.gov/epubs/ndp/ushcn/ushcn.html). The USHCN archive contains quality controlled records from a select group of high-quality, long-term weather stations that are adjusted for systematic biases (Menne et al. 2009). New Hampshire has five geographically dispersed USHCN stations that represent daily weather characteristics at relevant scales for temperature and regional circulation (Legates and Willmott 1990; Vinnikov et al. 2011).
Although all weather variables were considered, daily precipitation is difficult to quantify on a statewide basis because of the influences of terrain and season on the distribution of liquid and frozen precipitation. In New Hampshire, precipitation varies greatly in amount and type over short distances within the same regional weather pattern due to factors such as elevation and proximity to the coast. This leads to large differences among daily precipitation totals across the five USHCN stations. Daily snowfall amount and the depth of snow on the ground are only relevant for surveys conducted during the snow season. Whether for measurement or for substantive reasons, the precipitation indicators we extracted (based on total precipitation or alternatively, occurrence of 1-inch events) show no relation to climate beliefs.
Spatial patterns in minimum and maximum temperature are more consistent at regional scales than any measure of precipitation (Vinnikov et al. 2011). Although absolute temperatures from individual New Hampshire stations differ, daily departures due to regional weather patterns strongly correlate between stations. One principal component captures 84% of the variance over our study period. Following Keim et al. (2003) we use the statewide mean of USHCN temperature anomalies as our regional index.
Table 1 describes two of the weather variables we extracted from station records and integrated with the dated survey data. Temp2 is average temperature anomaly for the interview and one day prior. This 2-day window gives the best overall fit in predicting climate beliefs, although we also tested many other windows. Fig. 1 visualizes our time series. The upper curve tracks statewide temperature anomaly. The lower spike plot counts interviews on each date, across 10 surveys and more than 5000 interviews on 99 separate days.
4. Climate beliefs and temperature
Does short-term weather influence acceptance of the central point found in climate change statements by scientists? Table 2 shows results from three logit regression models. The dependent variable for each is a (0,1) indicator of whether people believe that climate change is happening now, caused mainly by human activities. Previous studies establish that age, gender, education, self-assessed understanding, and political orientation predict a wide range of climate-related beliefs. These are included as possible predictors in all three models. Odds ratios describe the multiplicative effects of a one-unit increase in each predictor on the odds of expressing a now/human belief. All background variables show statistically significant effects, in directions expected from previous research. Odds of a now/human response decrease with respondent age, are higher for women than men, increase with education and with self-assessed understanding, and are lower among Republicans than among independents and Democrats.
Model 1 is the base version, with season, two-day temperature anomaly, and the main effects of each background variable as predictors. Season exhibits no net effects. Unseasonably warm or cool temperatures on the interview day and the previous day, however, significantly shift the odds of believing that humans are changing the climate (p = 0.023).
Model 2 adds two complications. First, it incorporates party–education interaction terms in keeping with earlier studies. We find significant effects (p < 0.0005) in the expected direction: education raises the odds of agreeing with the scientific consensus among Democrats, does so more weakly among independents, and not at all among Republicans. Second, model 2 incorporates analogous party–temperature interaction terms. For independents only, the interaction proves highly significant (p < 0.0005)—indicating that temperature has a substantial effect for independents but not for people aligned with either party. Other models, not shown, tested for similar interactions between temperature and other background variables: age, gender, education, and understanding. Unlike party–temperature, these interactions proved weak and not significant. For parsimony and better precision, we omit them from the models of Table 2.
Model 3, a mixed-effects version of model 2, allows random intercepts to capture possible survey-to-survey variation that is not explained by variables in the model. Survey-to-survey variation could reflect unmeasured events such as political or economic developments, scientific reports, or news coverage about weather in distant places. The mixed-effects model 3 offers no improvement over a simple logit model (p = 0.433), however. The standard deviation of random intercepts (0.029) is only a fraction of its own standard error (0.090), indicating that there is little survey-to-survey variation in responses beyond that accounted for by the predictors. Based on this null finding, we focus our interpretation on the more parsimonious but substantively similar model 2.
5. Interpreting the interaction effects
Temperature effects among independents, strongest over a 2-day window, remain statistically significant for 1–4-day, and for 30–210-day windows. They are positive, if not significant, for all windows. Our finding that a 2-day window shows the strongest effect presents an unexpectedly detailed agreement with Egan and Mullin’s (2012) independent analysis.
The party–temperature interaction is visualized as an adjusted marginal plot in Fig. 2. Temperature anomaly over the past two days has little effect on the beliefs expressed by Democrats (high) or Republicans (low). The beliefs of independents, on the other hand, change with the weather. Adjusted probabilities of believing that climate change is happening now, caused mainly by human activities, shift from below 40% to above 70% over the −6° to +12°C range of temperature anomalies experienced on our interview days. The probabilities and 95% confidence bands graphed in Fig. 2 are calculated from model 2, adjusting for other predictors in that model.
In a related finding, Egan and Mullin (2012) report that political identity moderates the effects of temperature on responses to the question, “From what you’ve read and heard, is there solid evidence that the average temperature on earth has been getting warmer over the past few decades, or not?” They also find that Democrat and Republican responses are least affected by temperature. In their study, however, individuals who said they were “leaning” toward one party or the other show stronger temperature effects than either partisans or independents. Egan and Mullin also report a temperature–education interaction that is weak or absent in our data. These contrasts might partly reflect different dependent variables, which in our case specify human causation. Attribution, more than the reality of warming, is now the main point of public contention.
The party–education interaction effect is graphed in Fig. 3. Consistent with previous studies, we see a strong positive effect of education among Democrats, a weaker positive effect among independents, and a near-zero effect among Republicans. This result is no longer surprising but widely enough replicated to merit inclusion with most analyses of climate change views.
Distinctive features of our analysis include the following:
Statewide data from 5000 random-sample telephone interviews conducted on 99 days over 2.5 yr (2010–12) are merged with temperature and precipitation indicators derived from USHCN station records.
The dependent variable reflects agreement with scientific consensus statements that climate change is happening now, caused mainly by human activities.
Individual-level independent variables are age, gender, education, political identity, and self-assessed understanding.
Logit regression models include seasonal indicators, the political party–education interaction noted by other studies, and a party–temperature interaction.
Random intercepts in mixed-effects logit models test for variation between surveys that is not explained by temperature, season, or individual respondent characteristics.
Robustness of conclusions is supported by tests of alternative temperature windows, precipitation indicators, interaction effects, and model specifications.
We find that over 10 surveys, Republicans and Democrats remain far apart and firm in their beliefs about climate change. Independents average in between these extremes, but their beliefs appear weakly held—literally blowing in the wind. Interviewed on unseasonably warm days, independents tend to with the scientific consensus on anthropogenic climate change. On unseasonably cool days, they tend not to. These findings establish a climate science counterpart to poll-based and popular depictions of unaligned voters as a low-information group, who indecisively choose candidates for last-minute or superficial reasons (Galupo 2013; Kazin 2013).
Although the political classification of independents may vary from state to state, our self-identified New Hampshire independents generally resemble their counterparts on a nationwide survey that asked the same questions (Hamilton 2012). Independents comprise 18% of our New Hampshire estimation sample, compared with 17% nationally. They are similar with respect to education: 48% (NH) compared with 49% (U.S.) college graduates. The New Hampshire independent sample is slightly older (53 yr vs 46 yr) and more balanced with respect to gender (53% vs 57% male), but these differences fall within sampling uncertainty. Together with the similar NH and U.S. climate beliefs noted earlier, the comparisons suggest that our New Hampshire respondents are not wildly untypical.
Answering another recent national survey, many undecided voters said that they “need more information” on the issue and could “easily change their mind” (Yale Project on Climate Change Communication 2012). Such responses, like the volatility seen in our analysis, should encourage efforts at science communication to reach less partisan and committed segments of the population. Although undecided voters might be open to new information, however, they also fit a less sanguine image of people waiting for a change in the weather—that is, for personal experiences that feel consistent with “global warming.”
Hansen et al. (2012) note that in a warming climate the distribution of temperature anomalies will shift upward, with positive anomalies and heat events becoming more frequent. If so, we could expect the average views of independents to shift toward the scientific consensus, becoming more like the average views of Democrats. Although temperature effects in our data are sharpest for just a 2-day window, we see positive and often significant effects for longer windows as well, suggesting at least some degree of persistence from mundane weather experience. Experience with extreme events probably has longer-term impacts (Egan and Mullin 2012; Goebbert et al. 2012). Personal weather experience may thus represent a persuasive, although globally unfortunate, path toward greater concern about climate change.
Granite State Poll questions on environment and climate have been supported by the National Science Foundation (New Hampshire EPSCoR EPS-1101245), and by the Carsey Institute and the Sustainability Institute at the University of New Hampshire. Both the NCERA and Granite State surveys were conducted by the Survey Center at the University of New Hampshire. David Moore and Cameron Wake contributed to climate question design.