The Importance of Salience: Public Opinion and State Policy Action on Climate Change

Transcription

1 The Importance of Salience: Public Opinion and State Policy Action on Climate Change Rebecca Bromley-Trujillo University of Kentucky John Poe University of Kentucky Jordan Leising University of Kentucky 1

2 Abstract: This paper examines the relationship between issue salience and climate change policy adoption in the states. We argue that existing measures of issue salience are limited because they do not adequately capture two components of issue salience: issue concern and attention. In order to capture these components of issue salience, we utilize three measures: public opinion data, Google Trends search records, and interest group membership. We find that the measure of issue salience that incorporates both components (concern and attention) demonstrates the greatest influence on the enactment of climate change policy in the states. As more people within a state show concern in, and attention to, climate change the state legislature is more likely to respond with climate change policies. While this study focuses on climate change policy specifically, our findings have broad implications for studies focusing on the important link between public opinion and public policy. 2

3 Democratic theory argues that there is a relationship between the importance of issues and how they are addressed in government when an issue is salient to the public, government responds to the wishes of its citizens (Dahl 1956). While scholars often argue that changes in issue salience matter greatly for the political process, it has yet to be adequately studied within the context of state politics. This paper examines this dynamic by studying the relationship between issue salience and climate change policy adoption in the states. Climate change policy is an ideal test case for this process, given the prominence of activity in the states and the shifting of this issue on and off governmental agendas. As the de facto leaders in American climate policy innovation, states play a critical role in the climate policy space by adopting the vast majority of policies enacted in the United States. This situation is not unique to climate change policy, as states are also quite active in many other policy areas. Understanding the relationship between issue salience and policy adoption in the states is crucial to understanding how public policy is made in the United States. Existing research shows that there is a relationship between public opinion about policies and their adoption by elites, yet the magnitude and precise nature of the relationship is still under debate (see Erikson, MacKuen and Stimson 2002; Erikson, Wright and McIver 1993; Monroe 1998; Stimson, MacKuen, and Erikson 1995; Page and Shapiro 1983). 1 Burstein (2003) finds that seventy-five percent of studies gauging the impact of public opinion find that opinion drives 1 Page and Shapiro (1983) urge caution in their discussion of the opinion-policy linkage, due to their results being drawn on highly salient issues. Stimson, MacKuen, and Erikson (1995) find that there is a dynamic response from policymakers, which varies in accordance with constitutional design and other factors. Monroe (1998) finds that 55 per cent of policy outcomes across a 13-year period were in line with preferences of the public. 3

4 a response to policy change. 2 Paradoxically, Lax and Phillips (2009a; 2012) conclude that policy congruence with public preferences within the states is about as good as a coin flip, with increased congruence being more likely as the prominence of the issue grows in the states. 3 We believe that issue salience is the key to explaining the relationship between public policy and public opinion in the states, and that the conceptualization and measurement of issue salience in this arena deserves more attention. Researchers have recognized the theoretical importance of issue salience for the relationship between opinion and policy for quite some time (Burstein 2003; Page and Shapiro 1983). Virtually no public opinion scholar would say that issue salience is unimportant. Unfortunately, current measures of salience are often imprecise and may not capture the concept fully. The two primary measures of issue salience in common use are the New York Times citation index and Gallup s Most Important Problem (MIP) question, both of which are theoretically problematic. The traditional New York Times measure does not directly reflect what the public believes to be important, but rather what newspaper editors, reporters, and editorial 2 Burstein (2003) reviews the top three journals in political science and sociology and finds 30 studies that examine the relationship between opinion and policy published between 1990 and Of the 52 effects detailed in these studies, 35 presented statistically significant effect of opinion on policy. When salience is taken into account, Burstein finds that opinion always has an effect, thereby justifying theoretical consideration. 3 Lax and Phillips utilize the New York Times as a measure of salience, taking the log of the number of stories related to a range of gay rights policies. Using this measure, they find a strong conditional linkage between issue salience and gay rights policy adoption. 4

5 boards choose to print. Alternatively, the MIP question is vague and overly limiting, as naming a problem as important does not preclude the importance of other issues to the respondent. We argue that issue salience is comprised of two related, but distinct concepts: issue concern and issue attention. As issue salience is considered an important component for virtually all work on public opinion, and especially the link between opinion and policy, failure to think critically about both aspects of salience can result in serious practical and theoretical problems. This research considers how these components of issue salience influence policy adoption in the states. In addition to expanding the theoretical understanding of issue salience, we also provide a clear test of the relationship between issue salience and state climate change policy enactment. In so doing, we provide improved measures of issue salience over that of past research. In addition, we contribute to work concerning the determinants of state climate change policy adoption. Previous work on this topic has largely ignored the impact of issue salience on adoption, given the lack of adequate measures at the state level. We find the influence of issue salience both issue concern and issue attention to be striking. States where salience (particularly issue attention) is high are much more likely to adopt new policies. Despite the somewhat unique dynamics present in climate change politics, we believe that these findings are applicable to a broad set of policies. The Importance of Salience Salience is generally defined as a level of concern, attention or importance placed on a given issue (Wlezien 2005). While the precise definition of salience varies within scholarly work, the basic expectation remains the same: the likelihood of government responding to an issue increases with that issue s salience (Burden and Sandberg 2003). Past research has provided evidence in support of this expectation (Burstein 2003; Lax and Phillips 2009a; Lax 5

6 and Phillips 2012). In a review of past literature on the effect of public opinion on public policy, Burstein finds that when salience is not taken into account, public opinion has no real influence. Most research on the connection between public opinion and public policy has relied on the New York Times citation index as a measure of issue salience (e.g. Epstein and Segal 2000; Haider-Markel and Meier 1996; Lax and Phillips 2012). We argue that newspaper citations do not work well as measures of either issue concern or issue attention because they are supplyrather than demand-driven. In other words, newspaper indices have no real mechanism to incorporate how the public feels about an issue and instead assume that public salience mirrors elite salience. 4 Measures of salience derived from newspaper indices capture a variety of things beyond public concern or attention. Mentions in the New York Times reflect the levels of concern and attention of reporters, editors, and editorial boards at the Times. They reflect what elites believe people should know and what a newspaper s staff and owners believe will sell papers. Editors and newspaper editorial boards seem to have a bias for novelty and conflict, which may increase sales, but does not necessarily reflect what readers believe is important (Bennett 2011). 4 While these measures are likely related to public attention either because newspapers try to provide what people want to read or because elites help shape public interests the connection is uncertain and likely inconsistent. Without some additional measure of public concern or public attention to a policy problem, it is quite difficult to determine how well newspaper citations work as a measure for a given issue. 6

7 Some researchers have also relied on responses to Gallup s Most Important Problem (MIP) question (e.g. Aldrich, et al 1989; Burden and Sandberg 2003; Smith 1980, 1985). 5 Gallup s MIP measure attempts to capture both issue concern and attention to differing degrees of success. Gallup presumes that if a respondent says that something is the Most Important Problem, they are both concerned about the answer and are highly attentive to the problem. We believe that MIP understates problem concern and inconsistently captures problem attention. The MIP question has been thoroughly critiqued by Wlezien (2005), who asserts that the measure does not fully capture what he considers are two characteristics of salience: importance and problem status. Wlezien argues that an issue may be deemed important by the mass public, without being considered an actual problem and vice versa. For example, individuals generally consider the economy to be important, but may not consider it to be a problem until certain contextual factors are present (such as high unemployment rates). Wlezien s analysis indicates that the MIP question primarily gets at problem status and does not appear to tap importance very well. There are other potential problems with use of the MIP to consider. While Gallup may ascertain what people believe to be the most important problem, this does not preclude items not mentioned by the respondent from being important to that individual. It is typical for an individual to see many issues as salient at a given moment, which cannot be elucidated from the MIP question. The frequency with which individuals mention the 5 While it is argued that using a most important problem question in isolation may be problematic, other scholars have moved beyond this by measuring a more sophisticated version of the most important problem. Rabinowitz, Prothro and Jacoby (1982) take a two-step approach by asking respondents to rate the importance of 15 issues, afterward asking them to select the most important problem among those they initially deemed personally important. 7

8 environment/pollution as the most important problem facing the nation is very low. We contend that this does not provide strong evidence that issues pertaining to the environment are not salient to some members of the public. The typical ways in which issue salience is operationalized do not adequately capture the concept because they conflate issue concern with attention or assume that elite concern and attention work well as a proxy for all people. Similar to Wlezien (2005), we argue that salience is actually characterized in two ways: the level of concern and level of attention over an issue. While these two concepts are interrelated they can have independent influences on policy adoption at low levels. Alternatively, they can have a combined and often intertwined influence on the policy process at higher levels of salience. Issue concern reflects the degree to which an issue is viewed as serious by voters or survey respondents while issue attention reflects how much voters think about a problem. Issue concern may exist without much active attention to that problem. 6 Many people are vaguely concerned about public policy issues without taking the time or effort to learn much about them. Even politically apathetic people will tend to give some response of concern over problems if asked for their opinion (Zaller and Feldman 1992). These attitudes exist, but are often fragile and easily manipulated by the introduction of new considerations (Bizer et al. 2004, Zaller 1992). 6 Issue concern is conceptualized similarly to Wlezien s (2005) discussion of problem status, though issue attention is somewhat distinct from what Wlezien identifies as importance. Attention signifies a range of active investment in a policy, varying from limited information seeking, to a higher level of engagement. While we believe attention includes importance, this conceptualization goes even further. 8

9 Issue attention is likewise possible without clear concern over a policy problem. People can be aware of a problem even taking minor steps to learn about it without being concerned about it, simply out of curiosity or a desire to be informed. In these cases, attention can actually lead to increased concern. Alternatively, other people can give a significant degree of attention to a problem. Given this possibility, it is important to note that attention should fall along a spectrum where individuals may demonstrate low to high attention for an issue, a distinction that is important in our measurement choices. While concern over an issue is critical to forming a policy attitude, it is not enough to form a strong or coherent attitude that is likely to shape behavioral intentions for most people. Individuals can form weak attitudes on the spot based on any number of considerations they may bring to bear (Zaller and Feldman 1992). But actively considering a policy problem because it interests you increases the likelihood that an attitude will crystalize over time (Abelson 1988). 7 Policy attitudes that are personally important to people are more likely to motivate political behavior (Boninger et al. 1995), which, in turn, is more likely to influence the policy process (Abramowitz 1995; Arceneaux 2002; Burstein 2003). Attitudes become important to people largely because of the attention that they receive. 8 While issue concern and attention are both important aspects of salience in their own right, understanding how they combine is critical to understanding how issue salience influences the policy process. At low levels of concern or attention an individual can have one without the other. However, it is unlikely that levels concern and attention would not rise together at the high 7 Abelson refers to this phenomenon as ego preoccupation. 8 The exact relationship between attention to an attitude and its importance is complex in its own right, but is beyond the scope of this study. 9

10 end. As previous research indicates, we expect that when high levels of issue concern and issue attention are both present, an attitude will be translated into active political behavior (Abelson 1988). When citizens are both concerned and attentive to a policy problem they tend to vote based on that issue (Abramowitz 1995). Individuals who are both highly concerned and highly attentive to an issue are more likely to join interest groups to help influence policy directly (Boninger et al. 1995). The practical implication our split conceptualization of salience has for policy-making is that issue salience works in the same way that individual voters aggregate to form an electorate that decides an election (Erikson et al. 2002). Different aspects of salience apply to different subpopulations within a state. Part of a state s electorate may be concerned about climate change, but not give the topic much attention, while another subset of voters may give the topic moderate attention. A much smaller segment of the electorate may give the issue outsized attention, which will lead to collective political activity either through protests, interest group activity, or voting. Each of these groups should have an independent or mutually reinforcing impact on the policy process. By operationalizing salience as a single measure, researchers assume that a state population is homogenous in terms of attention and concern. We argue that this is not the case. State Public Opinion and Climate Change Policy Adoption While a number of studies have looked at how attitudes on climate change are formed (e.g. Borick and Rabe 2010; Li, Johnson and Zeval 2011), the relationship between those attitudes and the adoption of climate change policy is left nearly untouched by the literature on state policy adoption. However, work on environmental policy generally does suggest that environmental attitudes have an influence on policy adoption (Brace et al. 2002; Johnson et al. 10

11 2005). While this research provides thorough tests of the relationship between public opinion and environmental policy adoption, there are no direct measures of salience in this work. Work on the determinants of state climate policy adoption has become plentiful in recent years, incorporating both internal determinants and diffusion to explain such adoptions (e.g. Bromley-Trujillo 2012; Carley and Miller 2012; Matisoff 2008; Lyon and Yin 2010; Stoutenborough and Beverlin 2008). Some scholars have utilized state policy liberalism as an indicator of public preferences. Findings indicate that states with a more liberal citizenry are more likely to adopt climate change policy (Chandler 2009; Huang et al. 2007; Lyon and Yin 2010; Matisoff 2008). This provides some evidence that policy makers are responsive to the public for this environmental issue. Despite the relative boom in research for this issue area, issue saliency, as it relates to climate policy adoption, has still not been explored. Climate change policy provides a useful case for studying the effects of issue salience because of the rare relationship that climate change concern has with opinion about climate change policies. Many policy issues are framed as essentially multi-directional, where increased issue salience can differentially mobilize supporters or opponents of particular policy approaches. This means that researchers need to carefully measure the characteristics of salience concern and attention in competing political and policy camps. This problem is made somewhat easier with the issue of climate change because of the close relationship between belief in climate change and support for climate change policies. Individuals who do not believe climate change is a problem are largely the same individuals who believe climate change policies should not be passed, though, climate change deniers may be attentive to this issue while also indicating concern over negative impacts of policy action. Once a person believes climate change 11

12 is a problem, they are typically much more likely to endorse policy prescriptions to solve that problem (Zahran et al. 2006). There are a number of powerful disincentives to adopt climate change policy at the state level. States are often likely to free ride on the actions of others, as is traditional for collective action problems (Ostrom 2010). In addition, climate change is an abstract issue in the sense that direct causality between climate change and negative consequences to the individual is difficult to demonstrate (Leiserowitz 2006). A high level of issue salience on climate change should lead to increased likelihood of adoption. This makes climate change policy an ideal test case for the influence of salience on adoption. 9 By studying a policy area where salience and attitudes toward policy action are linked we can to some degree hold constant the opposing forces of public opinion. As public concern or attention to climate change goes up, policy adoption is more likely because most opponents of policy adoption simply do not believe in climate change (Dunlap, Xiao and McCright 2001; Leiserowitz 2006). Based on our theory of issue salience accounting for both concern and attention we propose a set of hypotheses regarding the relative influence of salience on adoption of climate policies. These hypotheses pertain to each type of issue salience: first regarding the impact of concern over climate change to policy adoption (as measured by public opinion data) and then considering the relationship between issue attention and adoption. For issue attention we use two 9 This relationship is indicated in the Appendix, Figure 1 using Pew Global Attitudes data. The level of support for climate change policies increases rapidly once an individual reports that climate change is a problem. 12

13 measures to capture a spectrum ranging from shallow attention (as measured through Google Trends) to deep issue attention (as measured through Sierra Club membership). Hypothesis 1: As issue concern over climate change increases, so does the likelihood of policy adoption. Hypothesis 2a: As shallow issue attention over climate change increases, so does the likelihood of policy adoption. Hypothesis 2b: As deep issue attention over climate change increases so does the likelihood of policy adoption. We then offer a set of additional hypotheses based on the possibility of interactive relationships between our measures. We believe that deep and shallow issue attention, in particular, represent the possibility for an interactive relationship. We also believe that high levels of issue concern and issue attention within a state may combine to influence policy. Hypothesis 3: As both shallow and deep issue attention over climate change increase, so does the likelihood of policy adoption. Data Hypothesis 4a: As issue concern and shallow issue attention over climate change increase together, so does the likelihood of policy adoption. Hypothesis 4b: As issue concern and deep issue attention over climate change increase together, so does the likelihood of policy adoption. This study examines state adoption of climate change policy from using original data. These data consist of policy adoptions among a set of nine policies associated with climate change, which are described fully in the appendix, Table The policies were selected 10 These data were collected through the Center for Climate and Energy Sources. The center provides information on state climate policy adoptions that includes text describing the history or date of adoption (U.S. Climate Policy Maps). Dates were obtained from this text and supplemented through examination of state regulatory listings. We obtained early adoption of 13

14 in order to cover the range of policy choices available to the states to mitigate and adapt to climate change. 11 While the specific goals of these policies vary, the overarching goals aim to mitigate and adapt to climate change. 12 Figures 2 and 3 demonstrate the variation among these policies, both in frequency of adoption by policy and differences across states. [Figures 2 and 3 about here] These programs are ideal for our analysis given they incorporate key policy choices for the American states to address climate change. When considering issue salience, we argue that it is important to include a broad group of climate policies that are likely to be driven by salience level. States engaging in this policy area have many choices and selecting one policy option does not speak to a broader commitment by a state to climate change policy. This argument is similar climate action plans from work by Wheeler (2008), while early adoption of renewable portfolio standards was obtained from Rabe s Statehouse and Greenhouse (2004). 11 Work by Rabe (2010) notes a multitude of approaches states have taken with regard to climate change, including efforts to increase the states share of energy obtained from renewable sources and efforts to conserve energy or reduce consumption. Policies are aimed at improving technology to make other sources of energy more environmentally friendly and to formulate plans of action that can help states decarbonize while maintaining a robust economy. Finally, states consider plans for adaptation to curb the effects of climate change. 12 The policies vary in terms of scope and stringency; however, as clear criteria for weights in this case are not present, we anticipate that any weighting choices are likely to produce bias (Daley and Garand 2005). While these policies take different approaches, directional hypotheses concerning our independent variables would be the same across these programs individually. In addition, a Chronbach s alpha of.83 demonstrates reliability of the scale. 14

15 to one made in previous work on general environmental policy commitment (e.g. Hays et al. 1996). These programs are also ideal given enactments have occurred primarily during the timeframe of our analysis. 13 We focus on the time-period for several reasons: First, Figure 2: Total Number of Enactments by Policy as of 2010 Green Building Standard Renewable Portfolio Standard Climate Action Plan Regional Climate Initiative Climate Advisory Board Greenhouse Gas Targets Advanced Coal Technology Vehicle Emission Standards Adaptation Plan Policy Enactments 13 Some states enacted policies prior to the time period we are examining. An alternative specification of the model included dummy variables for states that previously enacted a policy. These dummies produced no substantive changes and, as such, we leave them out of the final specification. 15

16 16

17 while states began enacting policies pertaining to climate change in the 1990 s, this activity accelerated greatly during the time frame we consider, given the failures at the federal level to achieve policy action (Rabe 2008). In addition, this time period allows us to utilize Google Trends data, which begins in This data provides a novel way of comparing issue salience across states. Variables and Estimation Strategy In order to test our theory of issue salience, we examine state climate change policy enactments. Our dependent variable is a count of the number of policies a state enacts in a given year, from the set of nine climate change policies listed in Table 1. As we have no policy reversals within our sample, the data are count distributed; thus, we employ a generalized negative binomial regression as the primary statistical model. 14 Given our approach is to study state policy enactment over time, we used clustered standard errors to account for the repeated measurements of states. 15 The large literature on the determinants of public policy indicates that a variety of factors beyond issue salience influence climate change policy enactments. Given previous research, we expect state political characteristics to have an influence on climate change policy adoption. Several studies that consider energy and climate change policy find liberal citizen ideology to 14 The new policy enactment variable is treated as count distributed because of its low yearly average (.53) and the fact that our sample contains no policy reversals. The observed maximum number of policies enacted in any given year for a given state is five and the minimum of observed new enactments is zero. 15 Alternate estimators were used for robustness tests and are reported in Table 2a in the appendix. 17

18 increase the probability of adoption (Chandler 2009; Huang et al. 2007; Matisoff 2008; Lyon and Yin 2010). Citizen ideology has also been considered more broadly as a factor in state policy adoption, for example in morality policy and taxes (Mooney and Lee 1995; Berry and Berry 1992). The relationship between government ideology and environmental policy adoption appears more limited, with very few scholars finding such a relationship (though, see Stoutenborough and Beverlin 2008). We incorporate measures developed by Berry and his colleagues including the revised citizen ideology series (Berry et al. 1998) and the NOMINATE measure of state government ideology (Berry et al. 2010). 16 As with state ideology, we control for partisan control of state government. We include multiple measures of partisan control to capture the complexities inherent in split control. Our measure of state legislative control compares democratic control of both chambers and split control to a fully unified legislature (Klarner 2003). We also include a measure of executive control that is coded as 1 for when the governor s party is also in control of the legislature and 0 otherwise (Klarner 2003). We control for legislative professionalism using a measure developed by Squire (2007). As indicated in the literature on environmental policy adoption and state politics in general, the more professionalized a legislature is, the greater access to resources for policy innovation (e.g. Ringquist 1993; Berry 1994). Thus, we would expect climate policy adoptions to increase as professionalism does. 16 The citizen ideology measure is constructed using interest group scores for members of Congress, ideological estimates of electoral challengers and vote weights by district. The government ideology measure utilizes DW-Nominate scores. Both measures run from 0-100, conservative to liberal. 18

19 State economic factors are also likely to play a role in climate policy adoption. A variety of environmental policy studies find that greater state wealth increases the probability of policy adoption (Chandler 2009; Matisoff 2008; Ringquist and Garand 1999). Such is the case more generally in the state policy adoption literature: greater state wealth leads to more resources and an increased probability of policy adoption and innovation (Berry 1994; Walker 1969). Environmental policy is often viewed as a luxury item to be weighed against things like per capita income and unemployment. Policy diffusion has also been shown as critically important in understanding the adoption of a wide variety of policies in the states (e.g. Karch 2007). In order to account for horizontal policy diffusion as an additional factor impacting state adoption, we include a lagged measure of neighboring policy adoptions. We sum the number of policies adopted by neighbors in a given year and divide by the number of neighboring states. Finally, we control for a state s general propensity to adopt environmental policy by including the level of total policy enactments in the state in the previous year. This variable works to capture the effect of a state population s general propensity to be more concerned and attentive to climate change than another state. Unmeasured features of states (e.g. persistent drought conditions), could easily be expected to have an important influence on policy enactment over time. The number of relevant state-level features that could be estimated is extreme and would typically be solved by inclusion of state-fixed effects. However, negative binomial models have been shown to be inconsistent when used with fixed effects due to the incidental parameters 19

20 problem (see Allison and Waterman 2002; Hilbe 2011; Green 2004a and 2004b). 17 In addition to the potential weakness of the fixed effect estimator, most of the variation in our sample is across rather than within states. 18 This means that fixed effects estimation would ignore most of the variation in policy enactment. By accounting for the level of policy enactment through the prior year we account for otherwise unmeasured state characteristics. 19 Unlike previous research, we include three measures to capture the influence of issue salience: (1) state-level estimates of public belief in the seriousness of the threat of climate change, as a measure of issue concern; (2) an index of issue salience built from Google Trends search results; and (3) Sierra Club membership (per 1000 citizens). We argue that state estimates of the level of belief in the seriousness of climate change represent concern while our Google Trends and Sierra Club membership measures represent different degrees of attention. Our Google Trends index captures low effort or shallow issue attention, while Sierra Club membership captures high-effort or deep attention and concern. Finally, we test the interactive relationships between our different measures as well. We include interaction terms for both Google Trends/Sierra Club and Google Trends/public concern. We used national-level public opinion surveys and multilevel regression with poststratification to generate a measure of public concern over climate change. We began with a set 17 The incidental parameters problem occurs when including indicator variables in a nonlinear maximum likelihood model. It results in varying degrees of bias in either coefficients, standard errors, or both, that is difficult to practically assess % of the variation is across states and not within. 19 By including the state policy level we are also able to control for differences in rates of adoption due to a potential policy ceiling. 20

21 of public opinion polls with nearly identical questions about the threat of climate change: In your view, is global warming/climate change a very serious problem, somewhat serious, not too serious, or not a problem? 20 The responses were grouped into two categories where a zero is classified as a respondent not believing global warming/climate change is a threat and one is coded so that a respondent does believe it constitutes a threat. 21 Multilevel regression and poststratification was then used to predict the percentage of each state in a given year that believed climate change posed a threat or the percent concerned about climate change in a state (see Lax and Phillips 2009, 2012). 22 In years when no survey was available (i.e. 2005) linear interpolation was used to predict missing values. 23 In order to capture public attention we use a novel measure derived from Google Trends data for four search terms and build an index for comparative search frequency of those terms. In other words, we compare the frequency of searches for global warming, climate change, acid 20 See Table 4 in the appendix for detailed information about the sources of our survey data. 21 The use of dichotomous responses to produce MRP estimates is currently the standard practice in state politics as the goal of MRP is to produce state-level percentages. 22 Alternative operationalizations of public concern were also created by grouping the response categories in different ways. We viewed the best operationalization as largely an empirical question and thus reported the strongest result in our main findings. 23 Multiple imputation was also used as a robustness check for the linear interpolation. This resulted in almost no change in the final models or marginal effects predictions, so the simpler method was reported. 21

22 rain, and pollution across states and by year and gave each state year a score. 24 These terms were chosen because of their connection with climate change/global warming and their relative lack of missing data (e.g. states without enough searches to register a score) compared to other possible terms. 25 Google Trends provides a rank ordering of states for each year based on each search term so that the state with the highest percentage of searches for climate change would receive a 100 for that term. States at the bottom of the list, or without enough searches to register will typically receive a zero. Each state has a score made up of the averages of their scores for each search term and scaled to 100. We used data from the Current Population Survey to control for states level of internet access across our time period. While Google search data is still relatively new, researchers have demonstrated its validity for state politics (Reilly, Richey and Taylor 2012). While still quite useful as a measure of broad issue attention across states, it is important to understand that Google Trends does have a certain amount of sampling bias. Even controlling for state internet usage, Google Trends measures are likely to over-represent wealthier and more highly educated people as they are more likely to use the internet. As these individuals are also 24 The score is an additive index of each of four search terms that is rescaled to a maximum of 100 so that coefficients are comparable to the other independent variables. Cronbach s Alpha is.70. Approximately 70% of the variation in this measure is across states. 25 See Table 3 in the appendix for a more detailed explanation of the composition of our index measure. 22

23 more likely to be involved in politics, we do not find this particularly problematic for our study, but research concerning other issues should proceed with Google Trends data carefully. 26 It is also difficult to gauge the attitudes of searchers based on their search terms. Our measure would register climate change related searches even if the searcher were a climate change denier. To this end, it is necessary to include some measure related to policy concern when using Google Trends, as we have done with public opinion estimates of climate concern. We use annual Sierra Club membership per one thousand citizens in a state as a measure that captures deep attention fed by high levels of concern over climate change. 27 Unlike newspaper indices or the Most Important Problem measure of issue salience, Sierra Club membership works well to capture both concern and attention at high levels. We believe that while Google Trends captures attention with relatively little cost, Sierra Club membership acts as a proxy for high cost or high effort attention and concern. Sierra Club members, and the broader pool of potential members of which they are a sample, are also likely to be the most concerned citizens in regards to climate change. By including all three indicators of issue salience we evaluate how low versus high levels of concern and attention work differently. 26 It is also important to understand that it is impossible to differentiate fifty searches by fifty people versus fifty searches by one person. Given that we are looking at state-level comparisons we assume that roughly the same distribution among searchers exists across states when controlling for internet access and state income per capita. In other words, we expect the ratio of searches to searchers to be similar across states even as the relative frequency of climate related searches changes across states. 27 Sierra Club membership data was obtained directly from the Sierra Club on May 6,

24 It should be noted that Sierra Club membership is plausible as a measure of interest group activity or lobbying efforts, as opposed to a measure of public support. While it is possible that this measure might additionally capture some of the influence of interest group activity, we believe that measurement of group membership as opposed to group activity minimizes this confounding problem. At best, Sierra Club membership acts independently of group activity for our purposes. At worst, it is a weak measure of group activity compared to lobbying efforts, fundraising totals, or campaign spending and works as a somewhat biased measure of salience Additional empirical tests indicate that Sierra Club Membership acts differently from interest group behavior, according to the underlying logic of the sub constituency theory of politics (Bishin 2009). Sierra club is arguably a Democratic interest group and increased group membership, if a sign of increased interest group strength, would mean that Sierra club membership is mediated through Democratic control of the legislature and governorship. In other words, there should be an interaction between Sierra club membership and legislative control. Given additional tests, we did not find any sign of an interaction between legislative control and Sierra Club membership. This seems to imply that whatever underlying mechanism high issue concern and attention uses, is likely not working through legislative partisanship. Sierra Club membership is influencing the state government regardless of partisan control. 24

25 Findings [Table 2 about here] Table 2: Policy Enactment Base Model Interactions Across Salience Types State Citizen Ideology 0.021* 0.017* 0.020* 0.021* (0.008) (0.007) (0.008) (0.008) State Government Ideology (0.009) (0.009) (0.010) (0.010) Legislative Professionalism 0.799* * 0.814* (0.402) (0.491) (0.384) (0.404) Per Capita Income (0.019) (0.017) (0.019) (0.019) Unemployment (lagged) (0.074) (0.073) (0.074) (0.075) Legislative Control Democrat 0.476* 0.608* 0.478* 0.500* (0.239) (0.224) (0.239) (0.246) Split Control (0.213) (0.209) (0.208) (0.223) Governor's Party in Legislative Control 0.758* 0.764* 0.752* 0.768* (0.303) (0.290) (0.303) (0.305) Neighboring State Adoption (lagged) * (0.126) (0.121) (0.127) (0.125) Past Policy Adoption * * * * (0.038) (0.037) (0.037) (0.039) Sierra Club Membership (per 1000) 0.132* 0.144* 0.128* (0.061) (0.055) (0.061) (1.022) Internet Usage * * (0.016) (0.016) (0.016) (0.016) Google Trends Index 0.013* 0.034* * (0.004) (0.008) (0.047) (0.004) % Who Believe Climate Change is a Problem 0.035* 0.043* 0.033* 0.062* (0.014) (0.015) (0.014) (0.031) Sierra Club/ Google Trends Interaction * (0.002) Google Trends/Public Concern Interaction (0.001) Sierra Club/Public Concern Interaction (0.011) Constant * * * * (1.771) (1.764) (1.767) (3.128) Observations All models use generalized negative binomial regressions with cluster robust standard errors at the state-level. Depenedent variable is the count of policies for each state year with an observed range of 0-5. Alternative model specifications using standard negative binomial approaches as well as multilevel approaches are included in the appendix as Table 2a. * p<

26 In column 1 of Table 2 we begin by estimating our basic model that includes internal determinants, diffusion from neighboring states, the existing level of environmental policy in a given state, and our measures of issue salience. States with more liberal electorates, Democratic control of state legislatures, unified party control between the Governor and legislature, and more professional legislatures tend to enact a larger number of environmental policies in a given year. Economic factors like income per capita and unemployment seem to matter less than would be expected. Neighboring state adoption of policies is borderline statistically significant where a state s neighbors can have a large substantive impact on policy adoption with a large amount of variance. When looking at our primary measures in Column 2 it becomes clear that issue salience does indeed matter. Sierra Club membership, Google Trends, and public concern for climate change are all positive and statistically significant, confirming our first three hypotheses. As these measures are on different scales it is difficult to compare their substantive impact from coefficients. Below, Figure 4 displays the marginal effect of each measure of issue salience on the predicted policy enactments in state years. The fourth plot in Figure 4 shows the marginal effect of the Sierra Club/Google Trends interaction shown in Table 2. The relative influence of our salience measures on predicted policy enactment becomes clearer with Figure 4 as each variable was rescaled to fit on a axis so that their influence could be directly compared. This means that the scales from the figures and those from Table 2 are different, but the relative effects of a maximum to minimum shift are unchanged. Of the three measures, the effect of public concern over climate change seems to be the least important. [Figure 4 about here] 26

27 27

28 Without issue attention to activate concern it appears to have a somewhat smaller influence on the policy process. Public concern alone accounts for less than half of an additional expected policy enactment in a given state-year. The results indicate that issue attention accounts for a greater level of policy change. Attention, as captured by our Google Trends search index, represents an additional 1.5 policies enacted in a given state-year when going from the least to most-search prone state. 29 It is possible that this effect is actually endogenous with the policy process that higher search rates for climate change reflect legislation that is already being debated. The influence of increasing Sierra Club membership going from its lowest levels of about 1 member for every 2,300 people in Mississippi to its highest levels of about 15 members for every 2,300 people in Oregon appears on par with issue attention captured by Google Trends. Almost two additional policy enactments are expected from the increase in issue concern and attention represented by the Sierra Club members. While these figures clearly confirm our main hypotheses, they say nothing about the relationship between the different indicators of salience. Columns 2-4 in Table 2 test interactive hypotheses where we combine the different measures of salience to capture a possibly interactive relationship. Only column 2 the relationship between Google Trends and Sierra Club 29 Figure 6 shows that the influence of issue attention, as measured by Google Trends, is not primarily driven by news coverage of legislation. By comparing the relationship between our Google Trends index and a 1-year lag of the index on the predicted enactment count, we demonstrate that issue attention leads legislative action. The predicted effect from the lagged Google Trends measure is reduced to an additional 1 policy per state-year as opposed to the 1.5 that current Google searches predict. 28

29 membership shows a significant effect. Unlike the main salience terms the sign of the coefficient is negative. Given the similar level and quality of influence between these measures on policy enactment that is evident from Figure 4, it is logical that they would have a combined effect. Furthermore, from a theoretical standpoint it makes sense that our two measures designed to capture issue attention with varying levels of issue concern would be the ones to have a more nuanced relationship. The final plot in Figure 4 shows that the combined relationship is fairly strong. The magnitude of the interaction between Sierra Club membership and Google Trends is such that the predicted range from low to high is about 2 additional policies per state-year. Issue attention acts as a catalyst to transform concern into political behavior. People who are actively attentive to politics are more likely to vote, protest, or join an interest group. This means that states with higher levels of attention are likely to have higher levels of political activity agitating for policy solutions. Discussion and Conclusion This study conducts an important empirical test for a new conception of salience. Our study provides evidence that issue salience plays a more nuanced role in influencing policymaking than has previously been discussed in the discipline. We find support for the argument that as issue salience increases in the form of concern and attention, states are more likely to adopt climate change policies. State legislatures are responsive to public opinion when people care about an issue. Issue concern as measured by the percentage of a state that believes climate change is a problem has a strong, independent impact on policy enactment. However, we find that issue concern has a weaker influence on policy enactment than salience measures that include issue 29

30 attention. This largely fits with our theory as we have argued that policymakers are less likely to respond to public opinion when the public lacks strong feelings about an issue. We argue that issue attention falls along a spectrum from shallow to deep. We find that Google Trends works as a measure of shallow attention. This level of attention requires relatively little effort, but still seems to have an influence on the policy-making process. Sierra Club membership represents a level of issue attention and concern in a state that has a significant influence on the policy process. Combining both measures of issue attention so to detect when states show high levels of low-effort and high effort-attention indicates a stronger relationship still. From a normative perspective, these findings are excellent news for those concerned with the influence of public will on policymaking. When individuals care about an issue, state policymakers are much more likely to be responsive. These findings provide important new insights for state politics generally and state-level policymaking on climate issues more specifically. From our analysis it is clear that issue salience is one of the driving forces behind climate policy adoption. Increasing public awareness and public attention to climate change seems to have a sizable influence on real policy outcomes within the states. Consider California (mixed control), Vermont (Democratic), and Utah (Republican). What we see from Figure 5 is that more liberal states like Vermont, that were also controlled by Democrats, tended to be early adopters, but; without a high degree of salient climate conditions, their adoption rate drops off. California had a similar early adoption peak despite their mixedcontrol; however, their subsequent adoption rate remains steady because climate issues are more salient. During this time period California has experienced natural disasters, including droughts 30

31 and wildfires, often associated with climate change by both citizens and policy makers. These events have likely prompted concern and attention from California citizens and policy makers. Utah passed no climate policies in the early part of our time period. However, as the state began to experience more severe climate events, their rate of policy enactment increased. For instance, the number of acres burned in Utah due to wildfires in 2004 was just over 76,000. In 2005 and 2006 it was just over 300,000 acres. In 2007 the total number of acres burned was 629,000. Salient events like severe droughts and wildfires motivate people to think about climate change and clearly correspond to policy adoptions in the following year. By 2008 the number of wildfires had bottomed out and in the following year, so did the number of new policy enactments. As suggested by our findings, the increase in salience results in state legislators being signaled by voters to act. This translates to a temporary reordering of legislative priorities. For climate change, states have numerous policy options to consider. The data indicates that states select from these options in a fairly unsystematic and probably localized way, but adopt more policies at times when the issue is most salient. 31

32 32