资源环境科技发展态势分析平台

In prior installments in this report series, we showed that majorities of Americans believe that the earth has been warming, that this warming is caused by humans, and that Americans largely support public carbon pricing policies and tax incentives to address issues related to climate change.

Do those preferences shape public policy? According to some political theorists, democracies only function effectively if elected representatives enact the policies that their polities support (Dahl 1989). This is thought to lead to popular support of government and confidence in the democratic process. Public opinion can shape policy-making through at least two mechanisms: (1) elected representatives can learn the policy preferences of all of their constituents through representative sample surveys, and (2) constituents with passionate opinions on an issue can shape their representatives’ policy pursuits by voting, sending letters and making telephone calls to their representatives, attending protests, and more.

Public opinion surveys can transmit information to elected officials about their constituents’ wishes, and evidence suggests that representatives have at least sometimes been responsive to such data (Hulland, Baumgartner, and Smith 2018; Jacobs and Shapiro 2000). Yet it is hard to blame elected officials if they do not follow the will of the public, because useful survey data are rarely available. Government officials deal with numerous issues at any one time, and public opinion surveys rarely document opinions on all of those issues. More importantly, most surveys are of the nation as a whole, whereas no one in the US Congress represents the entire country. Instead, each representative is sent to the nation’s capital to represent a state or congressional district. Surveys measuring the opinions of residents of such limited geographic areas are even more rare and limited in scope than are national surveys. So to allow representatives to consider the views of all of their constituents, a new type of data must be provided to them.

In this report, we propose a new technique for generating such data, using high quality national surveys of representative samples of American adults to yield accurate assessments of opinions in the US states. We refer to this method as the aggregation-disaggregation technique (ADT). To implement the ADT, an investigator must have a large set of national surveys for which respondents were selected using unclustered sampling methods. That requirement rules out the use of the vast majority of what are called “household surveys,” which involve multistage random sampling of homes in clusters located near one another, thereby minimizing travel for interviewers. Instead, population surveys must involve either random digit dialing to landlines and cell phones by human interviewers or mailing paper questionnaires to households selected via simple random sampling from all those in the country.

To implement the ADT, one must first collect a set of conventionally sized surveys of representative samples of a single population (e.g., all American adults) that have asked the same question with the same wording in each survey. Normally, the surveys would have been conducted repeatedly over a sustained period of years. Having collected these data, the next step of the ADT is aggregation–combining all the data into a single dataset. Next, the disaggregation phase involves producing separate estimates of public opinion in each state. To do so involves multivariate estimation that statistically adjusts for changes in opinions over time to yield estimates of public opinion in each state in each year when a survey was conducted.

The ADT is built on two important assumptions that have foundations in the literatures on American public opinion and survey methodology. First, the ADT assumes that public opinion changes slowly over time, and that shifts at the national level are likely to occur roughly evenly across each of the states. Support for this presumption comes from research by Page and Shapiro (1992), who demonstrated first that public opinion is rarely subject to dramatic shifts over time. Furthermore, they found that different subgroups of the American population (as defined by demographics, for example) do not drift apart on issues of public policy and instead changed in sync with one another. Page and Shapiro (1992) referred to this phenomenon as “parallel publics,” whereby a change in public opinion in one subgroup was found in all other subgroups. Of special interest for the present inquiry, Page and Shapiro (1992) examined differences between geographic subgroups (Northerners vs. Southerners, Urban vs. Rural) on a multitude of issues and rarely found different trend lines in such groups. Their overwhelming body of evidence shows that public opinion shifts in parallel fashions across geographic subgroups.

The second assumption of the ADT is that random samples, when sufficiently powered, should not only provide accurate estimates of national public opinion but should also provide accurate estimates within states. True random sampling requires that all individuals in a population have a known, non-zero chance of being selected, and so-called “base weights” must be used to equate those probabilities of selection across sample members. The most common form of this involves random digit dialing (RDD) of landlines and cellphones. Research has found that samples recruited using these methods continue to yield highly accurate estimates of the population under study (Callegaro et al. 2014; Chang and Krosnick 2009; Cornesse et al. 2020; MacInnis, Krosnick, and Cho 2018; Yeager et al. 2011). Although an individual national probability sample may not have a sufficient number of observations in each of the 50 states to provide precise estimates of public opinion, combining data from many surveys conducted over time using probability samples can produce sufficient numbers of observations. For detailed information on the methodology of the survey and data included in this report, please see the Climate Insights 2020: Opinion in the States Technical Report, available under “Methodology and Data” here. For state-by-state public opinion data, please see the Appendix of this report.