Legislative Production and Ideology Voter Scores: Three Essays by Shaun Michael Tanger A dissertation submitted to the Graduate Faculty of Auburn University in partial fulfillment of the requirements for the Degree of Doctor of Philosophy Auburn, Alabama December 18, 2009 League of conservation voters scores, interest group scores, environmental policy, Campaign contributions, bill co-sponsorship, interest group politics Copyright 2009 by Shaun Michael Tanger Approved by David Laband, Chair, Professor of Forest Economics and Policy Larry Teeter, Professor of Forest Economics and Policy Daowei Zhang, Professor of Forest Economics and Policy Yaoqi Zhang, Associate Professor of Forest Economics and Policy Wayde Morse, Assistant Professor of Human Dimensions ii Abstract Legislators possess supply-side characteristics in the production process of public sector outcomes. Constituents demand legislative services and choose those who most closely resemble their preferences. There are several organizations that rate the performance of congressmen and senators with respect to their votes on identified sets of bills they (the rating organizations) concern themselves with. It is well-known that legislative production is influenced significantly by the ideology of legislators (Kau and Rubin 1979 ; Dougan and Munger 1989; Poole & Rosenthal 1996;Levitt 1996). However, while these scores have been labeled ideology scores it is not clear whether or not they are merely responses to constituent interest (Downs 1957:Kalt & Zupan 1984; Strattmann, 1998;Strattmann 2000). While this question is not novel, with findings on both sides of the discussion containing scholastic merit (Bender and Lott 1996), we seek to analyze the import of these scores on legislative production; do they influence contributions? co- sponsorship behavior on certain bills? Do economic conditions influence the scores themselves? We find evidence of the affirmative in all three cases. iii Acknowledgments The principle author would like to thank the multitude of assistance he received while in pursuit of this dissertation. While the majority of credit goes to him, nothing it seems would have been possible without the contributions and efforts of many others. Notably, David Laband, Henry Thompson, John Sophecleus, Wayde Morse, Daowei Zhang, Larry Teeter, Yaoqi Zhang, and several others in both the Forestry School and Economics Department at Auburn University. Also, I?d like to thank Charles Tanger and Sandy Rowell. I love you very much. iv Table of Contents Abstract ......................................................................................................................................... ii Acknowledgments........................................................................................................................ iii List of Tables ................................................................................................................................ v List of Figures .............................................................................................................................. vi Chapter 1: Introduction ................................................................................................................. 1 Chapter 2: An Empirical Analysis of Bill Co-sponsorship in the U.S. Senate: The Tree Act of 2007............................................................................................................................................... 4 Introduction ................................................................................................................................ 4 Literature Review ....................................................................................................................... 6 Methods.................................................................................................................................... 10 Results ..................................................................................................................................... 15 Discussion ............................................................................................................................... 19 Chapter 3: Political Ramifications of Voting Green: Evidence from the 2002-2006 Congressional Elections .................................................................................................................................................... 23 Introduction ............................................................................................................................ 23 Literature Review................................................................................................................... 25 Methods.................................................................................................................................. 31 Results .................................................................................................................................... 38 Discussion .............................................................................................................................. 41 v Chapter 4: Political trade-offs between the economy and the environment: A time series analysis from 1970-2008 .......................................................................................................................... 43 Introduction ............................................................................................................................. 43 Literature Review .................................................................................................................... 44 Methods................................................................................................................................... 48 Results ..................................................................................................................................... 61 Discussion ............................................................................................................................... 73 Chapter 5: Closing comments and future research ..................................................................... 78 References ................................................................................................................................. 81 vi List of Tables Table 1: Descriptive Statistics for Chapter 2 .............................................................................. 16 Table 2: Logistic Regression Results for Chapter 2 .................................................................. 17 Table 3: Descriptive Statistics for Chapter 3 ........................................................................ 32-34 Table 4: Ordinary Least Squares Regression Results for Chapter 3 ...................................... 36-37 Table 5: Descriptive Statistics for Chapter 4 ............................................................................. 55 Table 6: Stationarity Results for Chapter 4 ........................................................................... 59-60 Table 7: Cochran-Orcutt Regression Results for Chapter 4 ................................................. 62-70 vii List of Figures Figure 1: Environmental Kuznets Curve .................................................................................. 45 Figure 2: Average Senate LCV scores for 1973-2008: Adjusted and Unadjusted ..................... 55 Figure 3: Average House LCV scores for 1970-2008: Adjusted and Unadjusted ..................... 56 1 Introduction In the last half-century, Public Choice scholars have revolutionized our understanding of public sector production (organization and behavior). Previously, researchers interested in political behavior had argued, in effect, that government officials seek to maximize something loosely referred to as the ?public interest.? This view was founded on the premise that individuals make different decisions in the political arena than they do in their private lives. That is, Homo economicus and Homo politicus are different species. But starting with Downs (1957) and Buchanan and Tullock (1962), researchers in the Public Choice tradition argued that individuals always maximize their own private interests, even in a public sector context. Thus, Homo politicus is Homo economicus and an understanding of political behavior necessarily requires us to examine and appreciate the incentives and constraints facing individuals engaged in public sector activity. Moreover, public sector production can be analyzed as markets that are closely analogous to markets for private goods and services. In a public sector context, citizens qua voters demand public goods and services and elect representatives to supply these legislative services. Because provision of improved environmental quality has characteristics of public goods (joint production, non-excludability), much production of environmental goods and services has taken place in the public sector (as legislation). But important aspects of 2 the production of environmental legislation have been neglected in the scientific literature. In this dissertation, I attempt to at least partially fill 3 gaps in the literature on environmental policy: (1) whether economic special interests are significant predictors of the pattern of political co-sponsorship of environmental legislation, (2) whether support for environmental legislation is privately-captured by politicians, in the form of campaign contributions and/or voter support, and (3) whether political support for environmental legislation is influenced by macroeconomic conditions. These 3 issues are bound together by a common thread: whether the private incentives and constraints facing politicians influence their production of environmental legislation. So the analyses contained in the following chapters reflect a Public Choice perspective. There has been relatively little analysis of bill co-sponsorship behavior generally speaking, and no analysis whatsoever of co-sponsorship of environmental legislation. In chapter 2, I examine whether the pattern of co-sponsorship in the U.S. Senate of a 2007 bill seeking to reduce taxation of harvested timber was related to the strength of forestry interests in each state. That is, was the likelihood that a Senator agreed to co-sponsor this legislation influenced by the economic importance of forestry in his/her state? In a nutshell, the answer is ?yes.? Increasingly, it seems, private firms are marketing their goods and services with environmental appeals. This suggests that private market producers anticipate higher profits from their pro-environment behavior than they expect in the absence of that pro- environment behavior. In chapter 3, I explore whether this is also happening in political 3 markets, by examining empirically whether politicians who support environmental legislation reap higher ?profits? than those who do not support environmental legislation, where profits in this political context refer to campaign contributions and votes made to individual legislators. Using data from the U.S. Senate and House of Representatives for the period 2002-2006, I find little evidence that either campaign contributions or votes for incumbent politicians of either party are influenced by their support for environmental legislation. However, I find that campaign contributions to Republican members of the House of Representatives prior to the 2006 general election were influenced significantly by their environmental voting record. Finally, in chapter 4, I investigate whether political support for environmental legislation is subject to the Law of Demand - - that more is demanded/supplied when price is low than when price is high. Presumably, we can afford more public sector production of environmental goods and services when economic times are good than when economic times are bad. If so, this implies that political support for environmental legislation will ebb and flow with conditions in the macro economy. I conduct a time- series analysis of the relationship between political support for environmental legislation in the U.S. from the early 1970s through 2008 and a measure of economic well-being: per capita income. 4 An Empirical Analysis of Bill Co-sponsorship in the U.S. Senate: The Tree Act of 2007 ?Ultimately I vote the economic interest of my district. That?s what I?m elected to do, and if it grows jobs and strengthens industry, that?s a positive thing.? Artur Davis (U.S. Congressman, District 7-AL) Introduction Much empirical analysis of legislative production focuses on determinants of roll call voting on specific bills (recent examples in a forestry context are Mehmood and Zhang, 2001; and Hussain and Laband, 2005). But final voting in legislatures occurs on only a small fraction of bills introduced. 1 This suggests that there is much about the legislative production process that has heretofore escaped the attention of scholars, a large part of which might aptly be referred to as legislative destruction. Indeed, relatively little attention has been paid to the quite visible starting point: bill introduction and co-sponsorship. Public choice economists view the legislative process as a political market, in which interest groups attempt to influence the production of legislation that has pecuniary and non-pecuniary consequences for them and politicians compete to provide these groups with relevant legislation. 1 In the 109th Congress (from 2005-2006) there were 6,436 bills introduced in the U.S. Senate and House of Representatives, but only 316 bills were passed by both chambers and signed by the President to become law. http://giffords.house.gov/services/government-information/congress-faq/index.shtml). In their analysis of bill co- sponsorship behavior in the 99 th Congress, Wilson and Young (1997) report a passage rate of less than 10% of all bills introduced into the U.S. Senate. 5 In this context, bill co-sponsorship acts as a signal to interest groups that a legislator is working to promote their interests and thereby maximize his/her rents from such groups. The introductory quote by Rep. Artur Davis is a reflection of this behavioral motivation relative to his public life. Rep. Davis?s sponsorship of the congressional version of Senate Bill 402 is one very visible indication to members of his district that he promotes their interests (in this particular instance, their forestry interests). This bill would allow taxpayers to deduct 60 percent of timber gains from the sale or exchange of timber held for more than one year. The change results in a reduction of the income tax rate on corporations from the regular rate of thirty-five percent to fourteen percent, which is competitive with the tax rates paid by foreign timber companies. According to Rep. Davis (quoted in Orndorff, 2007), ?If there are unfair or arbitrary tax burdens facing an industry and that industry is faced with unfair anti-competitive practices from foreign competition, one way we can help is through the tax code.? Two-thirds of the acreage in Alabama is forestland (twenty-two million acres) and 75 percent is owned by non-industrial private landowners. As well, the forest products industry in Alabama (defined broadly to include growers, loggers, mills, etc.) generates thirteen billion dollars a year in revenues and employs approximately 170,000 employees, directly or indirectly. 2 During the 2005-2006 legislative cycle, Rep. Davis accepted $14,500 dollars from PAC?s (political action committees) that represent these economically powerful forest interests in Alabama. While this is may seem consistent with his decision to co-sponsor the House companion bill to S402, a single observation does not provide compelling evidence in support of a broader conclusion that the bill co-sponsorship behavior of legislators is related to the strength of economic interests with a state. In this paper we examine empirically whether interest group politics significantly 2 Total GSP for Alabama in 2006 was $138,534,000,000, see http://www.statemaster.com/graph/eco_gdp-gross- state-product-current-dollars. 6 explain bill co-sponsorship behavior in the U.S. Senate. Specifically, we focus on co- sponsorship of Senate bill 402, a bill seeking to amend the Internal Revenue Code of 1986 to allow a deduction for qualified timber gains. Senate co-sponsorship decisions concerning S. 402 are assessed using a model that identifies various political and forest industry interests/characteristics. We demonstrate that a Senator?s co-sponsorship of this bill is correlated with his/her seniority, tax-cutting ideology, strength of electoral victory in his/her most recent election, campaign contributions received from forestry interests, the relative contribution of forestry to Gross State Product, and the percent of total land in his/her state that is privately owned. Literature Review The notion that elected legislators and other public officials maximize their self interest, not some supposed social welfare function, was put forward powerfully by Buchanan and Tullock (1962). Tullock (1965) further developed the argument with specific application to government bureaucrats. Olson (1965) was the first to argue that interest groups? pursuit of special favors granted by government officials leads to reduced economic performance. This was followed by a formal demonstration by Tullock (1967) that such behavior, while privately rational, was socially costly. Ensuing contributions by Stigler (1971), Kreuger (1974), Peltzman (1975) and Becker (1983) helped flesh out our understanding that interest groups demand, and government officials (at all levels) supply, special favors (or rents). Such favors almost never take the form of cash payments from the Treasury, as this is simply too obvious. Rather, they take a myriad of forms - - e.g., regulatory control of entry into a market by a firm?s potential competitors, special provisions in the tax code, environmental regulations that impose 7 differentially high costs on one?s competitors, price support programs, no-bid contracts awarded to select firms, and the like. This model with some nuanced elaboration now is widely accepted by both economists and political scientists as a more accurate description of the behavior of public officials than the so-called ?public interest? theory of government (Kalt & Zupan, 1984; Peltzman, 1984). In recent years, researchers have started applying this public choice model in a forest policy context, to explore empirically whether interest group activity appears to help explain forest and natural resources policy outcomes. For example, Mehmood & Zhang (2001) find that the relative importance of the forestry and forest products industries in each state correlate significantly with congressional voting on bills that have consequences for forest landowners. Likewise, Zhang and Laband (2004) report that the relative importance of the forestry and forest products industries in each state was a highly significant predictor of whether each state?s U.S. senators signed a letter sent to President G.W. Bush in 2000, privately urging him to act on behalf of U.S. timber interests against imports of Canadian softwood lumber (Zhang & Laband (2005). A completely unexplored aspect of the interest-group theory of politics is the extent to which interest group activities can be tied not only to final voting on bills, but also to bill introductions. It makes intuitive sense that interest groups that stand to gain or lose will lobby elected officials regarding their votes on a bill that comes up for a floor vote. What is not-so- obvious is how active interest groups are with respect to lining up co-sponsorship support for a bill that is being introduced for consideration by the Congress. Most introduced bills are never voted on - - they die in committee, are withdrawn, etc. So the prospective benefits to interest groups of lobbying elected legislators to co-sponsor a given bill appear to be much lower than if 8 they wait to see whether the bill in question even gets reported out of the (sub)committee it was assigned to before ramping up their lobbying efforts. But even this is speculative, as we simply have little understanding of, or appreciation for, the importance of bill co-sponsorship in the legislative production process. What little that is known has been aptly summed up by Campbell (1982, pp. 415-16): ?Though seldom mentioned in treatments of congressional procedures and practices, the cosponsoring of legislation has become an integral part of the legislative process in both houses of Congress. There are at least three reasons to suspect that the cosponsoring of proposed legislation is important in the legislative process. First, there is a significant effort to recruit members as cosponsors. Most members when introducing legislation routinely circulate "Dear Colleague" letters to the entire membership explaining the desirable features of their proposals and requesting support in the form of co-sponsorship. Second, both the number and the diversity of cosponsors (e.g., their diversity by party, or in ideology) are often cited by legislators during floor debate and in public discussions as evidence of a bill's support. For instance, one member prefaced his remarks in floor debate by noting that he was acting "in concert with the 100 members of this body who have cosponsored legislation of similar intent during this Congress." Third, the importance of cosponsoring legislation is indicated by how frequently members of each house decide to cosponsor. To the typical congressman, the decision to cosponsor a bill seems to be neither a rare nor a routine matter. During the 95th Congress, the typical member of the House cosponsored 147 bills and the typical senator cosponsored 131 bills.? According to Campbell, legislators expend resources and reputational capital both to co- sponsor legislation and to convince others to co-sponsor also. While there is some theoretical speculation as to why Congressman and Senators co-sponsor legislation, there is little empirical research that identifies factors that are significantly correlated with bill co-sponsorship. Adherents of electoral-connection theories of legislative politics view bill co-sponsorship as position-taking by rational legislators as a means of communicating cheaply but effectively with constituents (e.g., Campbell 1982). However, in their analysis of when a congressman co- 9 sponsors a bill, Kessler and Krehbiel (1996) find evidence that congressmen use co-sponsorship to communicate within the legislature. Extremists on either side of the political isle are early sponsors; they are followed later on by more moderate legislators. Fowler (2006) augments the intra-legislative signaling argument by proposing a model where the connectedness of members is examined. Indeed, he suggests that patterns of bill co-sponsorship may serve to measure those among Congress who are influential. Wilson and Young (1997) argue that bill co-sponsorship is an overrated cue. That is, within a legislative body co-sponsorship provides a signal concerning members? expertise and interest in a bill at the time of introduction, but it serves little real purpose from then on. Campbell (1982) examined the number of bills co-sponsored by members of the U.S. House of Representatives and Senate. He argues that the intensity/frequency of bill co- sponsorship by U.S. Senators is influenced by four factors: electoral margin of victory, party affiliation, ideology, and general legislative activity. Campbell?s work was done while the interest-group theory of legislative production was still being developed; therefore, he focused his analysis on legislator characteristics and did not even consider possible explanatory variables that might have served as proxies for interest group activity. Indeed, to our knowledge, no one has explored empirically whether the interest-group theory of politics is relevant at the bill introduction stage of the legislative production process. In this paper, we provide compelling evidence that demand-side factors (related specifically to the economic importance of forestry within a state) as well as supply-side factors (consistent with Campbell?s previous work) significantly influence the likelihood of bill co- sponsorship in the U.S. Senate. 10 Methods We are specifically interested in whether the presence and/or relative importance of certain interest groups located within a legislator?s political jurisdiction is statistically correlated with that legislator?s co-sponsorship of legislation favored by those interest groups. Because data that reflects the presence and/or relative importance of interest groups typically is aggregated at the state level and not available at the congressional district level, we focus our empirical analysis on co-sponsorship behavior of U.S. senators. We use a traditional roll call analysis model and logistic regression techniques to analyze the co-sponsorship decisions on Senate Bill 402. In general, we expect bill co-sponsorship to be influenced by demand-side considerations, reflecting constituent interests, and supply-side considerations, reflecting the willingness and ability of a Senator to ?go public? with his/her support for a given piece of legislation. 3 With respect to S402, it is important to specify believably the relevant constituent interests; our quote from Rep. Davis offers suggestions. We can proxy the ?economic interest? in S402 in a given state by measures of the relative contribution of private forestry (industrial and non-industrial) to the state economy. This might take the form of total employment (indicated by Rep. Davis) in the forestry sector, the forestry sector share of gross state product, campaign contributions by forestry interests to a Senator, the fraction of land in a state devoted to forestry, and so on. No doubt, some of these measures will be highly correlated with each other. With this in mind, the specific model we estimated is: (1) Cosponsor i = ? 0 + ? 1 Forestry$$ i + ? 2 %PrivateLand i + ? 3 %ForestGSP i + ? 4 NTUscore i + ? 5 %Vote i + ? 6 %Seniority i + ? i , 3 Grier and Munger (1986, p. 349) put this point rather succinctly: ?Legislators possess political assets that economic interest groups may find valuable in pursuing their goals.? 11 where Cosponsor i = 1 if Senator i cosponsored S402; 0 otherwise; Forestry$$ i = the amount of campaign contributions received by Senator i during the 2005-2006 and 2007-2008 4 election cycles combined from forestry interests, both individuals and Political Action Committees (PACs); 5 %ForestGSP i = the percentage of state i?s gross state product (GSP) derived from forestry and forest products 6 ; %Privateland i = acres in state i that are private as a percent of total acres in the state; NTUscore i = Senator i?s 2006 score assigned by the National Taxpayers Union. Scaled between 0 and 100, it reflects the percentage of bills identified by the NTU that each senator voted in accord with the NTU?s desired position (typically to lower taxes). %Vote i = the % of the vote Senator i obtained in the last election (s)he ran in. 7 Seniority = Senator i?s current ranking in the senate based on his/her time in office. ? i = the error term assumed to be logistically distributed. 4 We identified contributions from the 2007-2008 legislative cycle only through May 2008 because relevant aspects of Senate Bill 402 were included in the Farm Bill that was passed on May 15, 2008. While it is possible that contributors may indeed donate money after the fact in response to passage of favorable legislation, such an analysis is not possible within the scope of this paper. 5 We explicitly model the legislative production process as a temporal one in which prospective demanders of more favorable tax treatment of timber sales convey this desire to a Senator prior to and post introduction of a bill providing tax relief. That is, campaign contributions that reveal an interest group?s intensity of desire for a policy outcome precede and coincide with bill introduction. One might argue that interest groups will be loathe to pony up money unless and until they are assured that a desired bill actually will be passed, thus campaign contributions will post-date bill passage and, obviously, bill introduction as well. While we don?t confine our attention to the relationship between bill co-sponsorship and campaign contributions that precede bill introduction, we identify the possibility of a more time sensitive contribution strategy as a line of future research. See Stratmann (1998) for analysis in a Roll-Call voting context. 6 We use four categories to compile the contribution of the forest industry to the overall GSP of a state. These include Forestry, fishing, and related activities; wood product manufacturing, furniture and related manufacturing activities, and paper manufacturing. 7 Although Campbell (1982) found margin of victory to be a significant predictor of bill co-sponsorship behavior, we use percent of total vote and not margin of victory. However, the two metrics obviously are close substitutes. 12 In virtually every study of floor votes on particular bills, campaign contributions have been found to be a statistically important explanatory variable. As in private markets for goods and services, the ?price? paid for a legislative service (in the form of campaign contributions) reflects the demanders? intensity of desire. Prospective contributors with an interest in legislative favors donate more money to legislators who support desired legislation than to legislators who do not support desired legislation. This is true regardless of whether or not a politician does so as a quid pro quo transaction or believes in the position from an ideological standpoint. Drawing from this established result, we expect that the likelihood that a given senator co-sponsors legislation favorable to timber interests is positively related to the amount of campaign contributions received from those interests. 8 Thus, we expect the coefficient on Forest$$ to be positive. Zhang & Laband (2005) use several timber-related economic variables to explain whether or not U.S. Senators signed a private communication to President George W. Bush, urging him to restrict imports of Canadian softwood. Drawing from their work, we use the share of Gross State Product (GSP) derived from forestry and forest products as an indication of the relative importance of private forestry. The greater the forestry-related contribution to GSP in a congressional district or state, the more likely is the district/state to have: (a) strong economic interests tied to the success of the forest industry, and (b) lots of potential voters who own timberland (or other land that could be converted to timber production) and thus would benefit from passage of S.B.402. Following the lead provided by Rep. Davis, we expect the likelihood that a senator co-sponsored S.B.402 to increase as this percentage increases. Similarly, we expect a positive relationship between the likelihood that a senator co-sponsored S.B.402 and the 8 Because the tax benefits provided by S402 accrue to all private owners of timberland, whether private landowners or industrial timber companies, we identify all forestry-related campaign contributions received by each senator, without separating them by donor category. 13 percentage of privately held land in each state. As suggested above, as the percentage of privately-held land increases, so does the likelihood that land owners may benefit from (actually or potentially), and therefore support passage of, a bill lowering taxes on timber sales. Turning to supply-side characteristics, it is well-known that legislative production is influenced significantly by the ideology of legislators (Kau and Rubin 1982 ; Nelson 2002; Downs 1957; Dougan and Munger 1989; Kalt 1981; Peltzman 1984). There are several organizations that rate the performance of congressmen and senators with respect to their votes on identified sets of bills they (the rating organizations) concern themselves with. Since S.B.402 seeks to lower taxes on the sale of timber, it is possible, if not likely, that each senator?s ideological stance on taxes (favors lower taxes generally versus favors higher taxes generally) may have influenced the likelihood that (s)he co-sponsored the bill. With respect to the scoring by the National Taxpayer Union, a higher score indicates that an individual opposed higher taxes and/or spending and a lower score indicates that an individual voted in favor of higher taxes and/or spending. 9 Thus, we expect a positive relationship between a senator?s NTU score and the likelihood that (s)he co-sponsored S.B.402. 9 From the NTU website (http://www.ntu.org/main/misc.php?MiscID=13): Every year National Taxpayers Union (NTU) rates U.S. Representatives and Senators on their actual votes--every vote that affects taxes, spending, and debt. Unlike most organizations that publish ratings, we refuse to play the "rating game" of focusing on only a handful of congressional votes on selected issues. The NTU voting study is the fairest and most accurate guide available on congressional spending. It is a completely unbiased accounting of votes. NTU assigned weights to the votes, reflecting the importance of each vote's effect on federal spending. NTU has no partisan axe to grind. All members of Congress are treated the same regardless of political affiliation. Our only constituency is the overburdened American taxpayer. Grades are given impartially, based on the Taxpayer Score. The Taxpayer Score measures the strength of support for reducing spending and opposing higher taxes. In general, a higher score is better because it means a member of Congress voted to spend less money. The Taxpayer Score can range between zero and 100. We do not expect anyone to score a 100, nor has any legislator ever scored a perfect 100 in the multi-year history of the comprehensive NTU scoring system. A high score does not mean that the member of Congress was opposed to all spending or all programs. High-scoring members have indicated that they would vote for many programs if the amount of spending were lower or if the budget were balanced. A member who wants to increase spending on some programs can achieve a high score if he or she votes for offsetting cuts in other 14 We expect the likelihood of bill co-sponsorhip to decline with legislative seniority, for straightforward economic reasons. With increasing tenure in office, the sheer amount of information that constituents have been provided (by a variety of sources) about their legislator also increases. This implies that over time there is lower uncertainty among constituents about how their legislator will react to different policy proposals. This reduces the marginal return, to the legislator, of signaling his positions on bills via co-sponsorship which further implies that a senior legislator will be less likely than a junior legislator to co-sponsor proposed bills. Finally, other things held constant, we expect to observe a negative correlation between the percent of the vote won by a senator in his/her most recent election and the likelihood that (s)he co-sponsored S.B.402. Again, this is based on the empirical work of Campbell (1982), who reported that more secure members in both the House and Senate co-sponsor fewer pieces of legislation than do less secure members. Campbell?s explanation is that junior members of the senate enjoy closer contact with a larger portion of their chamber, receive a large amount of media attention, and also are less likely to have specialized in particular policy areas. Because the distribution of responses to our dependent variable, co-sponsorship of S.B.402, is limited to Yes (1) or No (0), logistic regression is an appropriate regression estimation technique (Woolridge 2006). In logistic regression, the probabilities of each outcome are: )( 1 )1( ? ? Xi Xi e e PiYi + ===? and )( 1 1 1)0( ?Xi e PiYi + =?==? , programs. A zero score would indicate that the member of Congress approved every spending proposal and opposed every pro-taxpayer reform. 15 the likelihood function for the model is: )1( 1 )1( yi n i yi i PiPL ? = ?= ? , and the marginal effects for each independent variable, at each observation can be calculated as: ?)1( PiPi Xi Pi ?= ? ? , where, Pi and (1-Pi) are both probabilities that the dependent variable takes the value ?1? and ?0?, respectively, and ? is the estimated coefficient for each variable or observation. Our data set is based on the 89 Senators for whom we had complete information on all variables included in our analysis. By virtue of deaths, retirements, and special elections for other reasons, several states had newly-elected senators; since NTU scores are not available for legislators with less than 1 year of service, this created several observational casualties. Sample statistics are reported in Table 1. Results Since our goal was to identify factors associated with co-sponsorship of S.B.402, as opposed to developing a predictive model, we estimated specific models drawn from equation (1), rather than using forward or backward-selection estimation procedures. Selected logistic regression estimation results are reported in Table 2 and discussed below. Analysis of Pearson correlation coefficients calculated with respect to our independent variables revealed significant correlation between specific pairs of variables. NTU score appears as an explanatory variable in all of the models. In unreported estimation results, we included a party dummy which we found to be a statistically significant predictor of co-sponsorship of S.B. 402. However, the Pearson correlation coefficient between Party and NTU score was 0.92 16 Table 1. Descriptive Statistics Variable Mean Std.Dev. Minimum Maximum Co-sponsor a 0.26 0.44 0 1 Forest$$ (x1,000) b 15.93 26.18 0 157.33 Vote% c 62.48 10.15 48.60 99.00 Seniority d 15.94 10.77 2.00 50.00 NTU score e 47.56 31.76 9.00 92.00 % Forest GSP f 1.18 0.95 0.14 3.93 % Privateland g 79.87 22.85 10.78 99.08 N= 89 a. http://thomas.loc.gov/cgi-bin/bdquery/z?d110:SN00402:@@@P Taken from the Library of congress bill record b. taken from the center for responsive politics. http://www.opensecrets.org/ c. taken from the National Journal website, which reports election outcomes. d & e. National taxpayers Union score. http://www.ntu.org/main/ f. Bureau of economic analysis. http://www.bea.gov/regional/gsp/ g. Devised by the national wilderness institute, which reports public land holdings. The authors subtracted the % figures and subtracted from 100 to get the private land holdings. These do not include land held by municipalities. http://www.nrcm.org/documents/publiclandownership.pdf (p<.0001). 10 In our opinion, the NTU score is a more appropriate indicator of ideology than party, with respect to a tax-related bill such as S.B. 402. With a dummy variable denoting party, the individual?s intensity of preference (for or against tax cutting) is subsumed beneath an all-or- 10 H 0 : rho = 0 (the null hypothesis); H a : rho <> 0 (the alternative hypothesis) If the p-value for the test is small (usually less than 0.05) then the conclusion is that rho is not 0, thus the relationship is statistically significant. The researcher will then have to make a professional judgment to determine if the association is important in terms of the context of the focus of the analysis. 17 nothing score that may reflect other, non-tax, ideology. Put differently, the variation among observations is much larger using the NTU score than when the party dummy limits the response to one of two possibilities; this variance is more likely to reflect each senator?s ideology with respect to a tax-cut bill than is the dummy variable for party. Table 2. Logistic Estimation Results Explanatory Coefficient Odds Coefficient Odds Coefficient Odds Coefficient Odds Variable Estimate Ratio Estimate Ratio Estimate Ratio Estimate Ratio Intercept -2.6539 -6.3737** -0.0705 -2.0211** (2.4525) (1.6762) (2.0038) (1.6366) Forest$$ 0.0281** 1.029 0.0328*** 1.033 (0..0113) (0.0125) %ForestGSP 1.0787*** 3.186 1.0916*** 2.979 (0.3197) (0.3090) %PrivateLand 0.0348** 1.035 0.0314** 1.031 (0.0168) (0.0163) NTU score 0.0228** 1.023 0.0182* 1.018 0.0255** 1.026 0.0190* 1.019 (0.0094) ( 0.0099) (0.0103) (0.0105) %Vote -0.0348* 0.952 -0.0647** 0.937 (0.0308) (0.0322) Seniority -0.1113** 0.895 -0.1227** 0.885 (0.0442) (0.0487) _____________________________________________________________________________________________ _______________ % concordant : 80.6 % concordant : 84.8 % concordant : 85.0 % concordant : 86.2 -2 Log Likelihood: 78.310 -2 Log Likelihood: 72.550 -2 Log Likelihood: 73.378 -2 Log Likelihood: 69.340 Likelihood ratio: 23.40*** Likelihood ratio: 29.16*** Likelihood ratio: 28.33*** Likelihood ratio: 32.37*** N = 89 standard errors in parentheses *** significant at 0.01 level ** significant at 0.05 level * significant at 0.10 level Other observed correlations between pairs of possible explanatory variables led us to estimate models with different combinations of explanatory variables. Our intent was two-fold: (1) to avoid models plagued by multi-collinearity, and (2) to reflect the statistical viability of alternative, albeit related, explanatory variables. For example, in models 1 and 2 we include forestry-related campaign contributions received by each senator (Forestry$$) as an explanatory variable. Since these contributions are, as might be expected, highly correlated with the 18 importance of forestry to GSP (Pearson correlation coefficient = 0.43, p <.0001), in models 3 and 4 we include %ForestGSP as an explanatory variable, without including Forestry$$ in the model. Similarly, Seniority and % of vote won were found to be correlated; the Pearson correlation coefficient between Seniority and % of vote won is .36 (p = 0.0004). We include % Vote won in models 1 and 3, and Seniority in models 2 and 4. As expected, we find that the likelihood that a specific senator co-sponsored S.B. 402 increased with: (1) Forestry$$ - - the total amount of contributions (s) he received from forestry- related sources (individuals and/or interest groups), (2) %PrivateLand - - the percentage of all land in the senator?s state that was privately-owned, and (3) %ForestGSP - - the percentage of Gross State Product in the senator?s state that was derived from forestry. However, as indicated in models 3 and 4, we consistently found that in the presence of GSP, other forestry-related explanatory variables such as Forestry$$ and %PrivateLand are statistically insignificant. This is plausible, as %ForestGSP is likely to reflect both the relative contribution of the forest sector to employment in a state, as well as the relative strength of forestry-related special interests compared to other economic interests. This finding is consistent with those of Mehmood and Zhang (2001) who argued that economic interests in their respective districts loom large with respect to Roll Call voting by members of Congress. Specifically, they found that House members from districts characterized by a significant construction sector were more likely to support the Tellico Dam legislation as it provided a large project for constituents employed in the construction industry. Inclusion of the percent of private land in a state in our model follows the same logic. The more private land in a state, the greater the number of land owners that would benefit from the proposed change in the tax code contained in S.B. 402. Because these individuals are direct beneficiaries of the proposed tax change, they may be more likely to vote 19 than other individuals who do not receive such politically-conferred benefits. More pointedly, they may be more likely to vote to re-elect the sponsors of such legislation and/or to contribute significant amounts of money to help re-elect politicians who can credibly claim - - through co- sponsorship - - to have at least tried to secure such benefits for their constituents. The odds ratio that reflects the estimated impact of %ForestGSP is quite sizable - - indicating that, for example, doubling a state?s %ForestGSP from the sample mean of 1.18 percent triples the likelihood that the senator(s) from that state co-sponsored S.B. 402. Likewise (from models 1 and 2), every additional $10,000 that a senator received in campaign contributions from forestry-related sources increased the likelihood that (s) he co-sponsored S.B. 402 to approximately 30 percent up from 25 percent. We are most interested in the impact of this variable on co-sponsorship decisions as the other demand side variables are unlikely to change as quickly as campaign contributions from the interested parties. In terms of supply-side factors, we find that the likelihood of co-sponsorship of S.B. 402 was: (1) positively correlated with a senator?s National Taxpayers Union score, as expected, and (2) correlated negatively with both the senator?s percent vote won in his/her most recent election and his/her seniority, also both expected. The sign and the estimated relationship between seniority and the likelihood of co-sponsorship are negative. This means that, other things equal, junior senators were more likely than senior senators to co-sponsor S.B. 402. Discussion The motivation for conducting this analysis was to determine whether there is statistically significant empirical evidence that is consistent with the hypothesis that economic interests in a legislator?s state or district influence his/her bill co-sponsorship behavior. Our analysis of the pattern of co-sponsorship of the 2007 Tree Act (S.B. 402) in the U.S. Senate suggests that co- 20 sponsorship indeed was affected by the forestry-related explanatory variables we examined. In addition, we find that legislator-specific characteristics (ideology, seniority, and strength of electoral victory in the most recent election) also are significant predictors of bill co-sponsorship. Our best model, while fairly parsimonious, correctly predicted 86 percent of the co-sponsorship decisions relative to S.B. 402. While our findings are highly consistent with the well-established result that the importance of economic interest groups in a state influences roll call voting on legislation and policy implementation, we regard them as preliminary rather than definitive. Interest groups of all persuasions are vitally concerned with legislative outcomes, taken broadly to include roll call (floor) votes, decisions made in committee or subcommittee, etc. Because of the relative visibility of roll call votes (i.e., data is readily available), the attention of empirical researchers has tended to focus more-or-less exclusively on this specific aspect of legislative production. But the proportion of bills that reaches a floor vote is a small fraction of those that are introduced. This suggests that there is a lot of potentially important, if not occasionally critical, legislative activity that has heretofore escaped scientific analysis. In particular, a bill must be introduced to have any possibility of eventually receiving a roll call vote. But the mere fact that most introduced bills never reach a floor vote clearly implies that not all bill introductions are created equal. To our knowledge, there has been no previous analysis of the relationship, if any, between patterns of roll call voting and patterns of bill co- sponsorship or, for that matter, of the possible relationship between bill co-sponsorship, roll call voting, and campaign contributions. There are many questions that command answers in order to develop a better understanding of this policy nexus. For example, does the likelihood and/or speed that a bill reaches a floor vote depend on the number of co-sponsors? Does either depend on specific 21 attributes of the co-sponsors, such as their seniority, presence on certain critical (sub)committees, etc.? Similarly, are the outcomes of floor votes influenced by these aspects of bill co-sponsors? S.B 402 died in committee the first time it was introduced, however The Tree Act was successfully tucked into the 2008 Farm Bill. Once a bill has been introduced and the pattern of co-sponsorship has been established (and is therefore well-known within the legislature), possibilities for the merger of bills and/or horse trading on bills may become apparent. Bills that may not stand much of a chance on their own may be merged into an omnibus bill where supporters each get a little something, even at the cost of agreeing to support specific provisions that they ordinarily would vote against. The answers to the questions posed above have powerful and comprehensive relevance to the forestry community writ large, since legislative decisions have the potential to radically alter the economic landscape within which companies, shareholders, NGOs, landowners, and demanders of forest products operate. There is a concomitant incentive for members of the forestry community to find ways to generate desired political outcomes generally and especially with respect to items of specific interest. But the critical question is, ?what are the best strategies to adopt in pursuit of this general goal?? An economist will argue that the best strategies are defined as those which yield the maximum benefit per dollar spent to influence political outcomes. We know, for example, that in terms of maximizing political influence generally a company like Weyerhaeuser may be much better served by contributing money to the chair of the Senate Agriculture committee than to either of the U.S. Senators from the state of Washington. But with respect to generating political support for specific forestry-related items, it may be quite critical to have one or both of the Senators from Washington co-sponsor desired legislation. We just do not know (yet) how 22 important this may be. But clearly, at least from our perspective, if members of the forestry community are not aware of important relationships in this regard, they inadvertently may be failing to take advantage of critical opportunities to influence legislative outcomes. In short, we believe that additional research is needed in order to develop a greater appreciation for the importance of bill co-sponsorship in the policy process and to identify potential opportunities for members of the forestry community to influence legislative outcomes. Economic considerations have been shown to exert significant impact on patterns of roll call voting by legislators. In this paper, we have demonstrated that economic considerations exert significant impact at an even earlier stage of the legislative production process - - bill co- sponsorship. One take-home message from this line of research is that readers should understand that public policy is not created in a vacuum or by legislators acting altruistically in the ?public interest.? Interested parties seek to influence legislative outcomes and their presence (and success) in the process is reflected by patterns of campaign contributions, the pattern of roll call votes, and patterns of bill co-sponsorship. A second take-home message for the forestry community specifically, is that defining success with respect to legislative outcomes narrowly in terms of roll call votes may cause interested parties to overlook strategic opportunities to influence legislative outcomes earlier in the legislative process. But these opportunities may not only be quite important, they may be quite cost-effective. 23 Political Ramifications of Voting Green: Evidence from the 2002-2006 Congressional Elections ?All I can say is when you ask politicians what subjects come up at town-hall meetings-which is something I do a lot - issues like global warming and environmentalism never come up.? TIME Magazine journalist David Brooks (2005) Introduction Corporate social responsibility implies a comprehensive mission/objective that does not begin and end with profit maximization. Pressured by non-governmental organizations (NGOs) and the like, companies voluntarily have begun implementing policies that go further than restrictive government regulations. This 21 st century business model may not be without merit; indeed, there has been a good bit of discussion recently about whether embracing environmentally ?friendly? policies and technologies pay off financially for firms. At least one corporate giant seems to think that ?going green? will generate more profits. In a recent Wall Street Journal article 11 , General Electric (the world?s largest publicly held company) CEO Jeff Immelt argued that ?new restrictions on green house gas emissions would have a de minimus impact on my investors.? As well, Immelt believes that additional profits will be generated by further carbon emission controls that allow the company to capitalize on its robust line of ?friendly? products?wind turbines, efficient jet engines, etc. ?Wind, water, lowering emissions, 11 Will Social Responsibility harm business, (May 18, 2005) 24 having an environmental service business?the economics of scarcity are going to drive lots of technological innovation over the next ten, 20, 30 years,? Immelt says. ?This is an approach to growing the company faster.? We find Mr. Immelt?s pronunciations to be intriguing, because they imply a perception on his part that there is now a sizable level of public support for firm-specific actions that are viewed as environmentally ?friendly.? Further, he evidently believes that his company is (or will be) positioned to benefit financially from this shift in consumer preferences. We wondered whether such a shift in consumer preferences also is reflected in production decisions made in the public sector. While production in a legislative context is not characterized overtly by profits, elected officials may benefit from the production decisions they make by trading in the coins of that realm: campaign contributions and votes. Elected politicians normally are quick to sense and exploit such ?profit? opportunities. Mr. Immelt?s sentiment about the possible profitability of corporate environmentalism notwithstanding, there is reason to believe that such a payoff, on balance, may not exist in the public sector. Following up the quote at the beginning of this paper, TIME Magazine journalist David Brooks added: ?In surveys too, when you ask people for the 10 issues that matter most to them, it?s always health care, jobs, education, gas prices. Environment is never there.? This is highly consistent with remarks made by Representative Artur Davis (AL-7) when co-sponsoring recent legislation to lower the effective income tax rate on timber sales: ?Ultimately I vote the economic interest of my district. That?s what I?m elected to do, and if it grows jobs and strengthens industry, that?s a positive thing? (Orndorff, 2007). With this in mind, the empirical question we explore in this paper is whether having a demonstrable record of support for environmental legislation has a significant impact on a 25 congressman?s campaign contributions and/or percent of vote won - - that is, is there a ?payoff? for producers who embrace ?environmentally friendly? production in politics? We empiricize this question by using League of Conservation Voter (LCV) scores as a barometer of a Congressman?s support for (co-production of) environmentally friendly legislation. Focusing on the 107 th- -109 th Congress, we estimate the impact of LCV scores on percent of vote won in the 2002, 2004 and 2006 general elections and campaign contributions received in the election cycle building leading up to each of these elections. Controlling for other factors that influence campaign contributions and votes received, we find that LCV scores were not important to voters in any election cycle and that contributors did not react to LCV scores until the 2006 election. In that election, they punished Republicans with differentially high LCV scores. Literature Review Our analysis starts from a different perspective than Riddel (2003), who examined the donation strategies of environmental political action committees (E-PACS). Rather, we adopt the perspective of a prospective producer of environmental legislation: the individual legislator. Presumably, his (her) behavior with respect to any specific bill introduced reflects careful weighing of the present value of a stream of anticipated costs and benefits tied, directly or indirectly, to that vote. In certain measure, these costs and benefits may be observable to a third party in the form of campaign contributions and/or votes. Indeed, we focus our analysis on both of these measures. 12 However, we assume that the marginal utility of a campaign contribution dollar received by the legislator is the same irrespective of the source donor. Therefore, unlike Riddel (2003), we do not restrict our empirical examination of campaign contributions to those 12 We acknowledge, of course, that the complete set of a legislator?s anticipated bill-specific costs and benefits presumably also includes highly personalized (ideological) components that may not readily be quantifiable. 26 that derive exclusively from E-PACS. That is, we wish to ascertain whether a legislator?s campaign contributions in aggregate are affected by his/her support for environmentally friendly legislation. Undoubtedly the most commonly-used indicator of political support for environmentally friendly legislation is the LCV?s Environmental Scorecard (e.g., Shipan and Lowery, 2001; Nelson, 2002; Riddel, 2003; Hussain and Laband, 2005). This rating, developed ex post for each legislative session, scores each representative and senator according to the percentage of the time they cast votes in accordance with the LCV-supported position on a selected set of environment- relevant bills that advanced to floor votes. Thus, the values range from 0 ? 100. The Environmental Scorecard represents ??the consensus of experts from 22 environmental and conservation organizations who selected the key votes on which members of congress were graded.? A particular drawback to LCV scores is that they are available only for individuals who are in a position to influence production of environmental legislation (i.e., incumbent congressmen, not challengers). A second drawback is that they reflect only legislators? behavior with respect to a select set of bills that received floor votes; they do not reflect a more comprehensive spectrum of ways in which an individual can advance (or obstruct) environment- relevant legislation. However, since the scores are constructed from votes on a number of different bills, they arguably reflect each individual?s breadth of support for the environment, which may, in turn, be taken as a sign of his/her degree of commitment to environmental protection 13 . Riddel (2003) used E-PAC contributions as an alternative measure of candidate eco-labeling. However, in her analysis of individual donations to U.S. Senate candidates, Riddel 13 Unfortunately, if a Congressman does not vote at all on a bill due to any number of reasons, he/she?s vote on that particular bill is counted as a vote against the stated position of the LCV. 27 looked only at whether or not each of 4 E-PACs donated to each candidate and did not examine the intensity of the signal (the amount donated). One might also use candidates? stated positions on environmental issues as an indicator (Project Vote Smart). But, of course, stated positions do not necessarily imply policy fidelity, so basing empirical analysis on such a variable may be problematic 14 (Ringquist 2004). We use LCV scores as our indicator of each legislator?s documented record of support for the environment and estimate the impact on both total campaign contributions received and on the percent of vote won in the general election. There is considerable previous research on the determinants of vote percent won and campaign contributions (e.g., Jacobson, 1978; Grier and Munger 1985; Bronars and Lott, 1997; Dix and Santore, 2003; Riddel, 2003). Following in this tradition, we characterize the percent of vote won by a legislator in any given election as reflecting his/her political vulnerability (as indicated, perhaps, by the percent of votes won in the previous general election and/or the level of campaign contributions received by his principal opponent in the current election cycle), how much money he has available to spend (indicated by contributions to his campaign), and his political party affiliation. Onto this general model, we graft a measure of the individual?s support for ?pro? environmental legislation, defined in terms of his/her LCV score. In general, campaign contributions received by a candidate in any given election are influenced by many of the same factors that influence votes received, with one important difference: the current political vulnerability of a legislator in a given election (and thus the incentive of especially-interested parties to invest money to help ensure that the legislator remains in office) can be proxied ex post by the actual closeness of that race. 14 The Project Vote Smart sample is truncated as not all incumbents and challengers participate in the political courage test. 28 Our empirical analysis focused on U.S House races in the 2002, 2004, and 2006 general elections; the specific models we estimated are: (1) Contributions i = ? 0 + ? 1 PoCC i + ? 2 Sen i + ? 3 Age i + ? 4 PrevVote i + ? 5 PrevVote? i + ? 6 CurrVote i + ? 7 CurrVote? i + ? 8 Par i + ? 9 LCV i + ? i (2) % of Vote Won i = ? 0 + ? 1 Contributions i + ? 2 PoCC i + ? 3 PoCC? i + ? 4 Sen i + ? 5 PreviousVote i + ? 6 PreviousVote? i + ? 7 Par i + ? 8 UnOpposed i + ? 9 LCV i + ? i Where Contributions i = total PAC plus individual contributions ($?s) received by each winning candidate during the 2001-2002, 2003-2004, 2005-2006 general election cycle for the U.S. House of Representatives, as reported by the Center for Responsive politics. 15 PrevVote i = the percentage of the vote that the winning candidate received in the previous general election; CurrVote i = the percentage of the vote that the winning candidate received in the current general election. PoCC i = the amount of money raised by the winning congressman?s principal opponent in the general election cycle; LCV i = the League of Conservation Voters environmental score for the legislator, in the year prior to the general election in question; Seniority i = the number of years the congressman had served in the House of 15 www.opensecrets.org 29 Representatives up to the general election; Age i = the congressman?s age as of January 1, 2002, 2004 and 2006, respectively; Unopposed i = 1 if legislator i was unopposed in his/her most recent election; 0 otherwise; ? i = the error term, assumed to be i.i.d. We expect campaign contributions received by individual i prior to the 2002 (2004, 2006) general election to be influenced negatively by how badly he beat his opponent in the 2000 (2002, 2004) general election. Other factors held constant, there is less pressing need to donate significant sums of money to a legislator who crushed his opponent in the most recent election, because barring an unforeseeable event that besmirches the individual, (s)he is not as likely to face serious opposition in the current election as an otherwise similar candidate whose most recent electoral victory was narrow. However, as the general election draws nearer, the results from the past election may be a less important predictor of the legislator?s political vulnerability than his actual opponent?s perceived strength. Since campaign contributions are more likely to influence the outcome of a close election than a non-close election, campaign contributions should be greater in hotly-contested elections (proxied ex ante by the level of campaign contributions raised by the legislator?s principal opponent and proxied ex post by the actual election results) than in blowout elections 16 . We expect the amount of money contributed to legislator i to be related positively to his seniority and negatively to his age. In general, more senior legislators are able to produce better political outputs than less senior legislators, ceteris paribus; they will, thus, command greater investment by parties with interests in legislative 16 In our analysis both unopposed and current vote variables were attempted for Campaign Contributions. According to our analysis current vote is a more accurate marginally as far as contributors are concerned. While both measure how much of the vote a candidate captured, current vote is much more sensitive to changes and therefore a more accurate measure of the impact on contributor?s donation strategies. 30 outcomes. However, holding seniority constant, donors have a greater expected time horizon for reaping the political returns to campaign contributions with younger legislators than older ones. We include a control for party (Republican = 1) to investigate whether there were significant party-based differences in campaign contributions during the 3 election cycles we examined. Finally, we look to the sign and significance of the estimated coefficient on LCV to reveal whether legislators? demonstrated support for pro-environment legislation affected the overall level of campaign contributions received. We expect the percent of vote won by legislator i to be positively related to the amount of money (s)he raises and inversely related to the level of campaign contributions raised by the incumbent?s principal opponent in the general election (Jacobson, 1978). As with the level of campaign contributions received, we expect a legislator?s electoral success in the 2002 (2004, 2006) general election to be influenced by how badly he beat his opponent in the 2000 (2002, 2004) general election. With respect to percent vote won, however, we expect a positive relationship. That is, a legislator who beat his opponent handily in the previous election is not as likely to face serious opposition in the current election as an otherwise similar legislator who narrowly beat his opponent in the most recent election. In the extreme, a sufficiently strong (i.e., well-liked) legislator will face no challenger; legislators who run unopposed surely will receive a significantly larger percentage of votes than those who are actively challenged. Two of the three election years we focus on were mid-term elections, during which the party of the president typically suffers losses according to Erickson (1988). Since Republican George W. Bush was President of the United States from 2001 ? 2008, this should be reflected in lower vote percentages for Republican legislators, especially in 2006. We expect vote percent to be related positively to seniority for the simplest of reasons: success breeds success. Long-term winners 31 have demonstrated consistently that they are well-liked by the voters in their districts. Finally, with respect to the impact of a legislator?s support for pro-environment legislation on the votes (s)he receives from the electorate, Riddel (2003) found that donations by environmental Political Action Committees (EPACs) have a positive impact on the likelihood of victory in U.S. Senate elections. In turn, E-PAC contributions, which serve an eco-labeling function in Riddel?s analysis, are impacted positively by the candidate?s LCV score. So the impact of an individual?s environmental voting record on the likelihood of electoral success was indirect, rather than direct. This gives us reason to expect a positive relationship between LCV scores and the percent of vote won. Riddel also indicates that environmental extremist candidates are more likely to be rejected by the voters on Election Day than candidates exhibiting moderate support for environmental issues. 17 This suggests that the relationship may be complex (e.g., non-linear). Methods Our sample necessarily was limited to those candidates who received environmental voting scores from the League of Conservation Voters, which required that they served during the 107 th - 109th Congresses. This caused us to delete from our sample those House members who did not serve that full term for any of a multitude of reasons (e.g., death, retirement, expulsion, and entry by means of a special election in mid-term). Since we were interested in campaign contributions received during the period 2002-2006, we included only those incumbents from the 2000, 2002, 2004 elections. Our sample did not include the Speaker of the House, since the League of Conservation Voters does not include a score for this individual, noting that the Speaker votes at his own discretion. This left us with a usable sample of 376, 17 For additional analysis of the impact of extremism in politics, see Westley (2000). 32 392, and 378 individuals, respectively. Sample statistics for Democrats, Republicans, and combined samples are reported for 2006, 2004, and 2002 in Tables a-c. Table 3a. Descriptive Statistics ? 2006 Variable Combined sample Democrats Republicans N = 378 N = 189 N = 189 Contributions 2006 a $1,161,647.70 $1,032,242.42 $1,290,371.89 (712,926.65) (588,829.64) (798,899.02) % of Vote won 2006 b 68.79 75.57 62.05 (13.26) (13.52) (8.84) % of Vote won 2004 b 70.14 71.47 68.81 (12.68) (13.06) (12.17) Contributions to $240,043.44 $141,448.08 $338,119.88 Principal Opponent a (557,531.91) (389,473.91) (671,867.09) LCV 49.93 82.76 17.28 (40.51) (24.39) (23.28) Party (Republican = 1) 50.13 (50.07) Seniority b 11.44 12.53 10.35 (7.88) (8.50) (7.05) Age b 56.16 57.28 55.05 (9.56) (9.39) (9.63) Unopposed (%) 7.12 13.23 1.05 (25.75) (33.97) (10.23) ______________________________________________________________________________ a from Federal Election Commission Campaign Finance Reports b from the National Journal (www. National Journal.com). 33 Table 3b. Descriptive Statistics ? 2004 Variable Combined sample Democrats Republicans N = 392 N = 185 N = 207 Contributions 2004 a $1,098,166.50 $987,923.04 $1,197,120.65 (870,511.59) (629,838.67) (1,029,454.18) % of Vote won 2004 b 70.51 72.64 68.60 (12.71) (12.94) (12.18) % of Vote won 2002 b 70.61 71.50 69.77 (13.09) (12.94) (12.11) Contributions to $204,638.82 $182,761.65 $223,434.50 Principal Opponent a (538,382.54) (413,033.27) (629,109.08) LCV 47.36 84.46 13.98 (40.75) (20.37) (20.64) Party (Republican = 1) 52.81 (49.98) Seniority b 10.92 11.45 10.41 (7.75) (8.06) (7.44) Age b 54.80 55.65 53.98 (9.40) (9.24) (9.52) Unopposed (%) 6.38 4.84 7.73 (24.47) (21.52) (26.77) ______________________________________________________________________________ a from Federal Election Commission Campaign Finance Reports b from the National Journal (www. National Journal.com) 34 Table 3c. Descriptive Statistics ? 2002 Variable Combined sample Democrats Republicans N = 376 N = 184 N = 192 Contributions 2002 a $864,092.24 $827,908.84 $898,767.99 (554,398.26) (569,843.83) (538,385.95) % of Vote won 2002 b 71.92 72.29 71.57 (12.85) (13.85) (11.85) % of Vote won 2000 b 68.40 70.93 65.91 (13.28) (14.32) (11.71) Contributions to $128,255.50 $107,778.52 $147,879.27 Principal Opponent a (313,772.45) (286,804.81) (337,187.40) LCV 47.34 79.45 16.57 (37.39) (21.48) (18.93) Party (Republican = 1) 51.06 (50.05) Seniority b 10.40 11.05 9.79 (7.38) (7.92) (6.78) Age b 53.93 54.76 53.14 (9.30) (9.03) (9.50) Unopposed (%) 6.91 8.70 5.21 (25.40) (28.25) (22.28) ______________________________________________________________________________ a from Federal Election Commission Campaign Finance Reports b from the National Journal (www. National Journal.com). 35 Average campaign contributions increased from $864,092 in 2002 to $1,161, 647 in 2006 - - a 34 percent increase. But over that same, relatively brief, time span average contributions to principal opponents rose 87 percent - - from $128,255 in 2002 to $240,043 in 2006. Party- specific differences are revealing. Both nominally and in percentages, Republicans benefitted more than Democrats from increases in campaign contributions. Campaign contributions to Republicans increased by 44 percent, versus an average increase of 25 percent for Democrats. But the growth in contributions to principal opponents was heavily skewed against Republicans also. The average percentage increase in principal opponent campaign contributions for Democrats was 31 percent, whereas for Republicans it was 129 percent, starting from a base level that was considerably larger. That is, Republican incumbents faced a much stronger increase in competition for their seats than did Democratic incumbents. This is confirmed by the sharp decrease (increase) in the percent of Republican (Democratic) incumbents that were unopposed in 2006, as well as the decline in the percent of vote won by Republican incumbents (while the percent vote by Democratic incumbents rose). A significant difference between the two major political parties, at least for our purposes, is that the average LCV score for Democrats consistently is 4-5 times higher than the average LCV score for Republicans. Notwithstanding this sharp division between the parties with respect to mean LCV scores, not all Democrats are rigidly pro-environment and not all Republicans are rigidly anti-environment, so there is considerable dispersion in the actual LCV scores. All equations were estimated using Ordinary Least Squares (OLS) regression. We estimated a number of different model formulations, including linear and squared terms for each variable as well as interaction terms, to investigate the structure of possible relationships. Because there is such a striking difference (on average) between Republicans and Democrats 36 with respect to LCV scores, we estimated separate models for Republicans and Democrats, to investigate whether the impact of LCV scores on campaign contributions and vote received differed by party, as such possible differences would not necessarily be revealed in models that included all House members. Our testing indicated the presence of heteroskedasticity; we corrected for it using the adjustment recommended by White (1980). The estimated structural models are reported in Tables 3 and 4 and discussed below. Table 4. Ordinary Least Squares Regression Results for Campaign Contributions Dependent variable = Campaign Contributions in 2006 Combined Sample Democrats Republicans Estimated Standard Estimated Standard Estimated Standard Variable Coefficient Error Coefficient Error Coefficient Error Intercept 5,932,037*** (1613058) 5,165,151*** (1803347) 7,380,025*** (2785589) Prin 0.663211*** (0.0939) 0.703416*** (0.1580) 0.668282*** (0.1227) Sen 11,333.42** (4180.5) 13,381.72** (5276.6) 10,010.98 (6536.5) Age -9,850.31** (3011.9) -9,285.72** (4084.1) -10,283.1** (4283.5) Vote04 -85,737.6** (25828.0) -92,812.3** (29128.7) -81,990.3* (47543.3) Vote04sq 520.42** (162.7) 540.1122* (180.5) 517.875* (298.4) Vote06 -23,534.4 (31080.3) -398.497 (46214.6) 72,557.9 (45842.8) Vote06sq 126.5867 (189.8) -10.9586 (275.5) 468.5049 (297.4) Republican -79,828.8 (86352.3) LCV -1,908.43 (954.7) 109.7318 (1388.7) -4,137.29** (1632.3) R-square adj, 0.47 0.43 0.47 N 366 180 178 ***, **, * - - significant at 0.01, 0.05, 0.10 levels, respectively. Dependent variable = Campaign Contributions in 2004 Combined Sample Democrats Republicans Estimated Standard Estimated Standard Estimated Standard Variable Coefficient Error Coefficient Error Coefficient Error Intercept 1,849,646 (1739478) 4,172,040 (1576044) 342,131.4 (2543268) Prin 0.90276*** (0.1600) 0.64748*** (0.1072) 0.968099* (0.1684) Sen 8,775.838* (4642.2) 13,339.26** (6183.2) 5,003.355 (6302.1) Age -7,965.68** (3616.0) -12,619.8** (5103.6) -4,823.93 (4699.6) Vote02 -18,106.4 (31361.1) -69,066.6** (34531.1) 27,375.54 (50511.7) Vote02sq 96.0354 (200.7) 409.9066* (223.3) -190.983 (323.3) Vote04 7,857.248 (36840.6) 6,819.036 (43124.2) 2,341.549 (52207.5) Vote04sq -73.8577 (226.9) -61.7073 (268.0) -53.2656 (323.1) Republican 98,494.7 (133943) LCV -784.904 (1625.8) -1,757.12 (2395) -322.651 (1,894.1) R-square adj 0.37 0.36 0.34 N 380 176 198 ***, **, * - - significant at 0.01, 0.05, 0.10 levels, respectively. 37 Dependent variable = Campaign Contributions in 2002 Combined Sample Democrats Republicans Estimated Standard Estimated Standard Estimated Standard Variable Coefficient Error Coefficient Error Coefficient Error Intercept 2,928,338*** (1067082) 3,895,825*** (1236340) 1,955,327 (2096242) Prin 0.66534*** (0.1250) 0.60438*** (0.1140) 0.753731*** (0.2143) Sen 13,861.81** (6677.9) 22,983.3** (10104.8) 1,888.38 (6917.6) Age -12,351.2*** (2636.9) -14,153.4*** (3986.7) -10,962.2*** (3307.6) Vote00 -11,251.6 (11396.0) -7720.91 (13247.3) -22493 (18199.2) Vote00sq 41.90861 (77.9562) 8.626317 (88.3795) 144.14 (127.6) Vote02 -24,476 (27107.2) -52,011.7** (30769.9) 11,043.4 (52519.1) Vote02sq 113.1973 (166.0) 284.802 (186.9) -112.745 (321.4) Republican 81,017.87 (9366739) LCV 962.4744 (1222.2) 1,236.465 (1838.3) 1,932.824 (1665.7) R-squared adj. 0.33 0.31 0.36 N 359 174 177 ***, **, * - - significant at 0.01, 0.05, 0.10 levels, respectively. Table 4. Ordinary Least Squares Regression Results for Percent of Vote won in the general election Dependent variable = % of Vote in 2006 Combined Sample Democrats Republicans Estimated Standard Estimated Standard Estimated Standard Variable Coefficient Error Coefficient Error Coefficient Error Intercept 12.33993 (15.9513) 7.627974 (21.4789) 21.29534 (20.9314) Contr 2.451E-7 (6.882E-7) 1.339E-6 (1.233E-6) -2.61E-7 (7.448E-7) Prin -0.00001*** (1.781E-6) -0.00003*** (4.816E-6) -0.00001*** (1.749E-6) Prinsq 3.13E-12*** (5.37E-13) 1.03E-11*** (2.37E-12) 2.44E-12*** (4.43E-13) Sen 0.02278 (0.0547) 0.017458 (0.0679) 0.064418 (0.0786) Vote04 1.337554*** (0.4151) 1.400941** (0.5498) 0.970173* (0.5607) Vote04sq -0.0068** (0.00265) -0.00666* (0.00348) -0.00502 (0.00367) Par -7.65351*** (1.4076) Unopposed 24.60188*** (1.0122 22.43652*** (1.0988) 34.7495*** (1.0356) LCV 0.011691 (0.0156) 0.004481 (0.0235) -0.00172 (0.0216) R-square adj. 0.76 0.75 0.52 N 368 180 180 ***, **, * - - significant at 0.01, 0.05, 0.10 levels, respectively. Dependent variable = % of Vote in 2004 Combined Sample Democrats Republicans Estimated Standard Estimated Standard Estimated Standard Variable Coefficient Error Coefficient Error Coefficient Error Intercept 21.08955 (15.2548) 8.588469 (23.8579) 31.19015* (18.7843) Contr. -1.21E-6* (6.223E-7) -1.07E-6 (1.223E-6) -1.08E-6 (6.847E-7) Prin -0.00001*** (1.749E-6) -0.00002*** (4.126E-6) -0.00001*** (1.832E-6) Prinsq 1.94E-12*** (4.29E-13) 9.34E-12*** (1.73E-12) 1.59E-12*** (3.74E-13) Sen -0.05495 (0.0635) -0.03124* (0.0412) 0.01654 (0.0929) Vote02 1.286189*** (0.4003) 1.734288*** (0.6319) 0.76598 (0.5079) Vote02sq -0.00671** (0.00263) -0.00997** (0.00412 -0.00289 (0.00341) Par -5.91958*** (1.9407) Unopposed 17.81552*** (3.0937) 18.51433*** (4.0084) 18.72261*** (4.1222) LCV -0.0244 (0.0230) -0.03124 (0.0412) -0.02132 (0.0201) R-square adj. 0.45 0.43 0.46 N 379 176 197 ***, **, * - - significant at 0.01, 0.05, 0.10 levels, respectively. 38 Dependent variable = % of Vote in 2002 Combined Sample Democrats Republicans Estimated Standard Estimated Standard Estimated Standard Variable Coefficient Error Coefficient Error Coefficient Error Intercept 57.90277*** (6.3985) 60.83184*** (7.0196) 55.74752*** (10.2422) Contr -1.67E-6** (7.9E-7) -1.7E-6* (9.581E-7) -2.12E-6 (1.326E-6) Prin 0.00003*** (3.709E-6) -0.00003*** (6.146E-6) -0.00003*** (5.062E-6) Prinsq 1.27E-11*** (2.2E-12) 1.3E-11*** (3.22E-12) 1.38E-11*** (3.45E-12) Sen -0.03494 (0.0530) 0.11242 (0.0718) -0.20095** (0.0823) Vote00 0.120966 (0.1813) -0.10862 (0.1987) 0.394227 (0.2906) Vote00sq 0.001161 (0.00134) 0.002968 (0.00143) -0.00121 (0.00214) Par 2.802772* (1.7177) Unopposed 23.64496*** (1.3848) 23.16868*** (2.1260) 24.44873*** (1.6835) LCV 0.0122 (0.0228) 0.041215 (0.0379) -0.01532 (0.0228) R-square adj. 0.60 0.63 0.57 N 359 174 177 ***, **, * - - significant at 0.01, 0.05, 0.10 levels, respectively. Results We focus first on the factors that influenced campaign contributions (Table 3). As expected, campaign contributions across all three election cycles examined were related positively to legislators? seniority and negatively to their age. However, there consistently appears to be a much stronger linkage between seniority and campaign contributions among Democrats than Republicans. The estimated coefficients for Seniority uniformly are much larger (and statistically significant) for Democrats as compared to Republicans, where the estimated coefficients are positive, as expected, but not statistically significant. In the combined sample for 2006, as expected, we find that campaign contributions decline with % of vote won in the previous election, but the relationship is curvilinear and turns positive at 82 percent of the vote. These findings hold for both Democrats and Republicans considered separately. While we see the same pattern of signs on the estimated coefficients in the combined models for 2004 and 2002, they are not statistically significant. However, in 2004 the estimated negative linear term and positive squared term on vote received in 2002 are statistically significant for Democrats but not Republicans; they are not significant for either party with respect to campaign contributions 39 leading up to the 2002 election. So there is some empirical support for the proposition that the strength of an incumbent?s most recent election victory impacts campaign contributions in the subsequent period. We also observe across all three elections that a congressman?s campaign contributions were strongly linked to contributions made to his/her principal opponent. The latter reflect the strength of the challenger in the coming election; on the margin a tighter race induces more campaign contributions for both challengers and incumbents because donor contributions are more likely to influence the outcome of the election. With the exception of Republicans in 2004, incumbents raised between 60 and 75 cents for every dollar raised by challengers, regardless of party. In 2004, however, Republican incumbents raised nearly $1 for every dollar raised by their principal opponents. Unexpectedly, in the three aggregate models we find that campaign contributions to incumbents were not sensitive to how close the current election actually was. The party-specific results are consistent, with the exception of Democrats in 2002, where we observe an estimated inverse (expected) relationship between the actual vote percent received and campaign contributions. The absence of a consistent result may reflect colinearity between this vote percent received by the incumbent and the level of campaign contributions received by his/her principal opponent. Finally, there appears to be no impact of LCV score on campaign contributions received during the campaign cycles, for the combined sample of Republicans and Democrats. This finding held regardless of whether the estimated relationship was structured as linear or nonlinear. Among Democrats, we observe no empirical linkage between LCV scores and campaign contributions received during any of the election cycles examined. However, among Republicans in the 2006 election cycle, those with higher LCV scores paid a price, in terms of 40 foregone campaign contributions. We estimate that for every additional LCV point, a Republican incumbent received over $4,000 less in campaign contributions from all sources. We turn now to the factors that influence the percent of vote won by incumbent congressmen (Table 4). Not surprisingly, incumbents who ran unopposed received significantly higher vote percentages in all three elections than incumbents who faced competition. The size of these effects was similar for Republican and Democratic incumbents in the 2002 and 2004 elections, but in 2006 unopposed Republican congressmen received 34 percentage points more vote than opposed Republican incumbents, whereas unopposed Democratic incumbents received 22 percentage points more votes than opposed Democratic incumbents. As expected, we find that an incumbent?s percent of vote won was negatively related (with a diminishing effect) to the level of campaign contributions received by his/her principal opponent. Controlling for the other factors in our model, Republican incumbents enjoyed a 2.8 percent higher vote percentage than Democrats in 2002, but received an estimated 5.9 (7.65) percent smaller vote percentage in 2004 (2006). The 2002 result is at odds with the historical pattern of gains by the opposition party in non-presidential election years. However, this was the first congressional election after the 9/11/01 World Trade Center attacks and Republicans may have benefited from a general perception that they are more trustworthy than Democrats with respect to dealing with terrorists. On the other hand, the 2004 and 2006 slide by Republican incumbents suggests that the substantial gains made by the Democratic Party in both the House of Representatives and the Senate in the 2008 general elections were part of a longer-run pattern of voter dissatisfaction with Republican incumbents. In general, there is little consistent evidence linking seniority and percent of vote won, although there is a small but significant impact on Democratic (Republican) incumbents in 2004 (2002). 41 Focusing on our motivation for this empirical investigation, we find no evidence in support of the proposition that how congressmen vote on environmental legislation affects, either positively or negatively, the level of voter support they enjoy in the next general election. This result is consistent across members of both of the major political parties, despite the large difference in average LCV score noted previously. Discussion The motivation for conducting this analysis was to determine whether there is real-world evidence of positive (negative) returns to public sector producers from supporting (failing to support) environmentally friendly legislation. Our analysis of data from winners of the 2002- 2006 elections to the U.S. House of Representatives suggests that the comment by TIME Magazine?s David Brooks that we used to lead off our presentation was directly on-target, at least for the particular half-decade period covered by our data. Using League of Conservation Voter Environmental Scorecard ratings for incumbent congressmen that ran for re-election in the 2002, 2004, and 2006 general elections, we find no evidence that a congressman?s support for environmentally-friendly legislation had a political payoff in terms of greater support by voters. Further, among Democrats we find no evidence that a congressman?s support (or lack thereof) for environmentally-friendly legislation had a political payoff (cost) in terms of greater increased (decreased) campaign contributions. Among Republican incumbents, we find evidence, for 2006 only, of a statistically significant inverse relationship between total campaign contributions and LCV scores - - that is, as LCV scores went up, campaign contributions declined and vice-versa. However, in both the 2002 and 2004 election cycles we find no evidence of a statistically significant effect. 42 Although we might speculate about the importance of and/or explanation for this one- time impact among Republicans, from our standpoint such discussion is not advisable at this point in time. Even with our findings now established, it would be premature to declare that there are no significant political returns to supporting environmentally friendly legislation. Our analysis is restricted to incumbent legislators, because only incumbents have an established record of voting on environmental legislation that is tracked by the League of Conservation Voters. This data availability issue imposes a limitation on the generality of our finding. Ideally, one would conduct an analysis of the likelihood of electoral victory by all candidates (both challengers and incumbents). Indeed, environmental groups claimed credit for ousting incumbent Representative Tom Pombo (R-CA) from his House seat in the 2006 general election, after declaring him environmental public enemy #1 (LCV.org 2008). The anti-Pombo tactics employed by these groups (e.g., attack ads) are not as easily measured as campaign contributions to an incumbent or his/her principal challenger. So the empirical challenge with respect to employing more encompassing metrics of impacts is rather more daunting than the one we tackled. A second limitation of our analysis is that it is static rather than dynamic. That is, a cross-section analysis like ours fails to identify possible impact(s) on campaign contributions or (re)election likelihood of an individual?s changing support for pro-environment legislation over time. Quite obviously, such an analysis would require the investigator to focus exclusively not only on incumbent legislators, but more specifically on those who repeatedly were successful, with whatever biases that might imply. Nonetheless, it would be interesting to see what the consequences might be for someone whose demonstrable record of support (or lack thereof) for pro-environment legislation changed appreciably over time. 43 Political trade-offs between the economy and the environment: A time series analysis from 1970-2008 ?Against the backdrop of an economic downturn?conservation scores dropped across the board? (Cindy Schwartz, League of Conservation Voters Director, 2008, Associated Press article) I. Introduction What has come to be known as the Environmental Kuznets Curve (EKC) reflects a simple, straightforward theoretical proposition: human demand for environmental quality depends in a particular way on their economic circumstances. Specifically, when people are economically destitute, they are willing to degrade the environment (or tolerate environmental degradation) in efforts to improve their economic well-being. However, as humans? economic well-being improves a threshold is reached where environmental quality becomes positively- valued. At this point, environmental quality becomes a ?normal good,? in which further increases in economic well-being lead to declines in environmental degradation. A substantial number of economic studies have provided evidence on the empirical validity of the EKC, with mixed results (see below). However, most of these studies employ a similar methodology - - cross-sectional analysis of the relationship between a measure of economic well-being in a sample of countries at a particular point in time and incidence of one or 44 more specific indicators of environmental degradation at approximately the same point in time. In this paper, I take an emphatically different empirical approach - - time-series analysis of the relationship between economic well-being and political support for pro-environment legislation in the U.S. Senate and House of Representatives. Political support for pro-environmental legislation is measured using the Environmental Scorecard ratings of Congressmen and Senators published annually by the League of Conservation Voters. In brief, the question of interest is whether these Environmental Scorecard ratings, aggregated across each legislative body, are sensitive over time to changing economic conditions - - that is, whether there is a political trade-off between economic conditions and the environment. Over nearly 40 years? worth of scorecard ratings, I find that, among the U.S. House of Representatives and Senate, political support for the environment (as measured by League of conservation voter scores) is significantly impacted by the per capita income of individuals in the United States. In section II, I review relevant literature on the EKC. My model of the relationship between political support for pro-environmental legislation and economic conditions is developed in section III. In this section I also introduce and discuss the data used in my analysis. This is followed, in section IV, by a brief discussion of methods. In Section V, I present my empirical results. Discussion and conclusions round out the presentation. II. Literature Review In theory, desperately poor people are willing to despoil their local environment in order to improve their economic circumstances. However, beyond some threshold level of economic well-being, environmental quality becomes a normal good - - i.e., demand for it increases with increasing income. This suggests an inverted U-shaped relationship (Figure 1) between 45 economic growth in terms of per capita income and various indicators of environmental degradation (Grossman and Krueger 1995; Barrett and Graddy 2000), although the relationship also has been described as N-shaped (Grossman and Krueger 1995; Torras and Boyce 1998). Figure 1. Environmental Kuznets Curve Using different measures of environmental degradation -- air pollution (Selden and Song 1994; Grossman and Krueger 1995; Torras and Boyce 1998), water pollution (Shafik 1994; Grossman and Krueger 1995; Torras and Boyce 1998), deforestation (Cropper and Griffiths 1994), and ecologically imperiled species (McPherson and Nieswiadomy 2005) -- researchers have presented empirical evidence in support of EKC relationships. However, EKC skeptics have raised questions about the existence of an EKC for specific forms of environmental degradation ( Stern 2004). More pointedly, it has been argued that the EKC is an artifact of: (1) technological/structural change in production and associated impacts on indicators of environmental degradation (de Bruyn et al. 1998) and (2) specialization in production and trade of goods and services across the nations that permits richer nations to ?export? environmental degradation to poorer countries (Arrow et al. 1995; Stern et al. 1996). In addition, the lack of 46 econometric rigor, particularly the tests for some statistical properties such as variable distribution, serial correlation (in time series data), model adequacy and specification tests has raised concerns about earlier EKC studies (Stern 2004). While examination of the empirical relationship between specific indicators of environmental degradation and economic well-being may provide useful, perhaps compelling, insights, such investigation does not exhaust the set of possibilities. Of interest in its own right is the question of whether political support for pro-environment legislation is sensitive to economic conditions, irrespective of whether the legislation actually is enacted. As reported by Baumol and Oates (1988), separate findings indicate that higher income individuals support environmental laws and regulations more than low income individuals. This supposition is grounded in the Hedonic Pricing literature--separate findings indicate that property values are affected by air pollution levels in metropolitan areas and that exposure to air pollution varies inversely with income 18 . Although environmental quality may be characterized as a normal good beyond some threshold level of economic well-being, not all normal goods are created equal. In good economic times, political (voter) support for the environment, among many voter wants, may wax strong. However, when economic times are tough, politicians necessarily make hard choices between competing programs/initiatives/regulations that not only affect human quality of life, they also have economic costs and consequences that weigh more heavily in the voters? minds than when economic times are good. Thus, even if voters? preferences for environmental quality stay constant over time, their willingness to support environmental programs and regulations may be impacted by changing economic conditions that reflect on the feasibility of enacting such programs and regulations. 18 To our knowledge no researchers have conducted an analysis of whether or not citizens pay more for home in districts or states that have, ceteris parabis, politicians who vote differentially higher LCV scores. 47 Indeed, as the quote that led off our paper reveals, the leadership of the League of Conservation Voters acknowledges that poor economic conditions act as a constraint on environmental legislation. To be clear, LCV Director Cindy Schwartz? comment was made with respect to changing LCV scores for Maryland state legislators, but the general thought expressed surely applies in the context of national politics as well. This same fiscal reality also has impeded passage of environmental legislation at the international level. For example, Germany?s chancellor, Angela Merkel, who originally supported the Kyoto Protocol?s requirement of drastic carbon dioxide emissions, now calls it ?ill-advised climate policy? (Michaels 2008). The question we?re raising is, ?how sensitive is support for the environment to economic conditions in a political context?? Because most of the research to-date on the EKC has employed cross-section data, there has been little opportunity to observe how the pace of environmental degradation varies with changing economic conditions within a country. The above-mentioned findings by Baumol and Oates suggest that, within the United States, support for pro-environment regulation will increase as per capita income increases. But aside from conjecture, we have no empirically-based understanding of what this relationship actually looks like. For example, because government-produced environmental regulations are permanent, unless revoked, it seems unlikely that environmental degradation actually will increase during economic downturns. However, we do not know how much, if at all, citizen/political support for government policies aimed at protecting the environment erodes during economic slowdowns or downturns. It is the structure of the relationship between political support for the environment and economic conditions implied by the EKC, that we focus our scientific lens on in this paper. 48 III. Methods The general model implied by the EKC literature is that political support for the environment is hypothesized to be related to economic conditions: (1) Political Support for the environment = f (economic conditions) Arguably, economic conditions in the U.S. imply that we are, and have been for some time, in the region of the EKC to the right of the turning point (i.e., where environmental protection is a normal good). Since our empirical analysis is restricted to the U.S., the assumed relationship specified in equation (1) is positive. There is plenty of anecdotal evidence to support this presumed relationship. Between 1970 and 2006, for example, the United States' inflation-adjusted GDP grew by 195%, the number of cars and trucks in the country more than doubled, and the total number of miles driven increased by 178%. However, during that same time period, annual emissions of carbon monoxide fell from 197 million tons to 89 million, nitrogen oxides emissions fell from 27 million tons to 19 million, sulfur dioxide emissions fell from 31 million tons to 15 million, particulate emissions fell by 80%, and lead emissions fell by more than 98% (Dupont 2006), despite a nearly 50 percent increase in the resident population over the same time period. Since the 1970s, at least, per capita income in the U.S. has been rising while a broad suite of indicators of environmental degradation have fallen. Identifying a metric that accurately reflects political support for the environment is challenging, because the production process in politics can be difficult to track accurately. Much of the real action, in terms of support for, or opposition to, a bill takes place behind the generally closed doors of (sub)committee meetings. Moreover, there is a myriad of things a politician can 49 do to benefit the environment - - adding funding riders for environmental interest groups to pursue projects, changes in the tax code that encourage recycling or other environmentally- friendly behaviors, and the like. Tracking all of these possible political activities is sufficiently daunting that, to our knowledge, no one has even attempted it. Moreover, aggregating these activities into a single metric would be additionally problematic, as there is no established methodology for weighting the impacts. These difficulties notwithstanding, a number of empirical researchers have used the Environmental Scorecard (ES) ratings of Senators and members of the House of Representatives, developed and published annually by the League of Conservation Voters (LCV), as a barometer of their support for the environment (e.g., Nelson 2002; Shipan and Lowery 2001; Riddel 2003; Ringqvist 2004). Scaled between 0 and 100, the score reported for each senator or congressman reflects the percentage time that each senator or congressman voted in accord with the LCV?s desired position on a set of bills selected by the LCV that were reported out of committee and came to floor votes. Following in this tradition, we use LCV scores as our measure of politicians? support for the environment. LCV?s ES ratings are available for members of the U.S. House of Representatives (U.S. Senate) starting in 1970 (1973). The ES ratings have been produced annually (i.e., for each legislative session), except in 1987/88 when a single score was reported by the LCV for both legislative sessions. We use the same scores for both years. Because economic statistics are calculated at the state and national level, but not available for congressional districts, our analysis focuses on ES ratings averaged across all members of the House and Senate. Economic well-being can be (and is) measured in a number of ways. The most direct tie to the EKC literature is exploration of possible empirical linkages between LCV?s ES ratings and 50 real (i.e., inflation-adjusted) per capita income over time. However, one also might explore possible linkages between LCV?s ES ratings and other conditions that affect socio/economic well-being, such as the unemployment and/or inflation rates. We investigated the empirical linkage between LCV scores and per capita income via estimation of the following specific time series model: (2) LCVscore t = ? 0 + ? 1 PCI t + ? 2 share t + ? 3 control t + ? 4 presidency t + ? t where, LCV Score t = the average LCV score in the U.S. House of Representatives/Senate in year t. PCI = Per capita income in chained (2000) dollars Share t = the share of the seats in the House/Senate that was held by Democrats in year t. ControlD t = 1 if Democrats held majority power in both the House and Senate in year t, otherwise 0. ControlR t = 1 if the Republicans held majority power in both the House and Senate in year t, otherwise 0. PresidencyD t = 1 if the Democrats held majority power in both the House and Senate and the Executive branch in year t, otherwise 0. PresidencyR t = 1 if the Republican held majority power in both the House and Senate and the Exective branch in year t, otherwise 0. 51 We expect the average LCV score of the House/Senate in any given year to be influenced positively by per capita income. This expectation results from a straightforward application of EKC theory to demand for publicly-provided goods/services: as real income rises (falls), our collective ability to afford consumption of those items we value positively increases (decreases). 19 Much, if not most, environmental legislation consists of regulatory restrictions on business practices that are deemed to have detrimental consequence for the environment - - e.g., legislation protecting species legislation, wetlands mitigation, air and water quality restrictions, and the like (Zhang and Mehmood 2001; Anderson and Mizak 2006). Because these regulations mitigate economic growth, they are less likely to be embraced when economic times are tough than when they are rosy. Economic conditions held constant, we assume that the larger the share of the House/Senate held by Democrats the higher the average LCV scores will be, since the aggregate scores are averages of the individual scores of members of both parties. The larger the Democrat presence in the legislative body the higher the aggregate score because Democrats have, on average, much higher scores than Republicans. While party affiliation has been found by an army of researchers to be important in predicting both environmental roll call votes and individual LCV scores (e.g., Anderson and Mizak 2006; Snyder and Groseclose 2000; Mehmood and Zhang 2001; Levitt 1996; 2002), only Shipan and Lowry (2001) have put forth a theory as to why Democrats are the ?party of the environment. 19 Relatedly, we expect the average LCV score of the House/Senate to be related negatively to more narrowly-focused indicators of economic well-being, such as the unemployment rate and the rate of inflation, and thus also the Misery Index. 52 Moreover, if Democrats held a majority in the House/Senate, the average LCV score of Democrats (Republicans) is likely to rise (fall) for a different reason - - Democrats control the committee process that likely determines the fate of introduced bills. This implies that environmental legislation introduced into the House/Senate is more likely to formulate policy in ways that Democrats will support and Republicans will oppose. This is consistent with the contention of Mehmood and Zhang (2001) that Republicans are the more conservative party on economic issues and are expected to vote more in favor of free market approaches to environmental legislation and to contest government intervention. Our sample necessarily was limited to those candidates who received environmental voting scores from the League of Conservation Voters, which required that they served during the 91 st -110 th Congresses. Although the E.S. ratings for the House/Senate were calculated as the average of the individual members? scores, in any given year there typically were fewer than 435 (100) observations from Congressmen (Senators). There were deletions from each year?s sample for a number of reasons (e.g., prolonged sickness, death, retirement/resignation, expulsion, and entry by means of a special election in mid-term) that resulted in members who only served a partial term. The minimum number of observations for any one year was 370 congressmen and 81 senators. Our annual samples did not include the Speaker of the House, since the LCV does not include a score for this individual, noting that the Speaker votes at his own discretion. The LCV Environmental Scorecard produces ratings from 0 (worst) to 100 (best) for Congressmen and Senators, detailing their support or opposition for what is deemed by the LCV as environmentally important legislation. A potential pitfall of using LCV scores as a measure of political support for the environment across time is that the metric itself does not stay constant over time. We know, for example, that the number and substance of bills voted on each year that 53 forms the basis for the E.S. ratings changes over time (Groseclose et al. 1999). If the number of bills used to generate the LCV?s E.S. differs from year to year, then the resulting scores are calculated using different weights assigned to the constituent bills that comprise each year?s score. As well, through judicious selection of bills to include when compiling the E.S. ratings, the LCV may change its scale over time to strategically move the average score in some preferred direction. This can serve to obfuscate the true preferences of constituencies expressed through the roll call mechanism. Finally, with respect to comparisons between the House and Senate, note that a congressman and senator may vote identically on the same set of bills in a given year, but their E.S. ratings may differ by virtue of differences across legislative bodies with respect to the set of bills used by LCV to create the E.S. To make the voter scores more comparable over time and across chamber, we employ a linear transformation introduced by Groseclose et al. (1999). If the i-th member in chamber k has LCV score y k,i,t in period t, then the transformed score is tik y ,, ^ = tk tk b a y tik , , ^ ^ ,, ? , where the values tk a , ^ and tk b , ^ for each chamber-year are maximum likelihood parameter estimates of movements in the policy space. This adjusted score isolates the individual legislator?s position from general trends in the congressional chamber, i.e. changes in majority party leadership that would alter the types or number of bills introduced (Lopez and Ramirez 2008). 20 20 Groseclose et al. (1999) argue that the adjusted scores more accurately represent a politician?s position on legislation the score presumes to reflect. Indeed, they find that the adjustment results in substantive changes to the conclusions of previous research (e.g., Levitt 1996, Shipan and Lowery 2001), reduces the standard error of the estimates in most cases, reduces the sensitivity of empirical findings to the particular modeling assumptions, and improves the performance of the model on specification tests. 54 With respect to our analysis of LCV scores, implementing the Groseclose adjustment had the particular effect of resulting in adjusted scores that exceeded 100 for a number of individuals, in all cases Democrats, especially in the last decade. 21 Figures 2 and 3 document the time-pattern of LCV scores in both nominal (i.e., published) and real terms. Figures 2a and 3a reveal the time pattern of nominal LCV scores in the Senate and House of Representatives, respectively. The mean scores for the both the Senate (as a whole) and the House (as a whole) have hovered, with some variation, around 50. However, in nominal terms at least, the mean scores for Democrats and Republicans have diverged dramatically over time, with this increasing divergence being driven by a more or less steady increase (decrease) in mean score for Democrats (Republicans) in both legislative bodies. Although these trends may suggest that Democrats are becoming more environmentally friendly while Republicans are becoming less environmentally friendly, the index-adjusted LCV scores (Figures 2b and 3b) indicate that this is an incorrect conclusion. In fact, Democrats are becoming more environmentally friendly, at least in terms of LCV scores, but Republican support for pro-environment policies is holding quite steady over time, in both legislative bodies. Further, it is quite apparent that the nominal LCV scores understate the support for pro- environment policy initiatives by members of both parties, as the adjusted scores, even with the truncation at 100 for Democrats, are higher for both Democrats and Republicans. Sample statistics for LCV score averages for the U.S. House of Representatives and U.S. Senate, as well as Economic and Political Indicators are reported in Tables 5a-5c. 22 21 Dr. Groseclose indicated (personal correspondence) that he and his coauthors found a similar result. Specifically they found that adjusted scores for several Senators and House members were above 100 for the ideological ratings (Americans for Democratic Action and Americans for Constitutional Action) they worked with. In our analysis this result was much more pronounced. 22 We only report the full sample of Economic indicator variables in Table 5c, but for our statistical analysis as the Senate LCV scoring did not start until 1973 we truncate the economic indicator variables to N=36 for 55 Table 5a: Average Adjusted LCV scores for the U.S. House of Representatives, 1970-2008 Full Senate Senate Percent Senate Democrats Republicans Democrats Mean 59.01 73.82 36.90 56.81 Standard Error 1.47 2.29 0.50 0.99 Median 60.09 73.94 37.38 58.20 Standard Deviation 9.24 14.33 3.17 6.18 Minimum 43.03 47.75 28.13 47.10 Maximum 75.95 96.88 44.07 67.10 # Years 39 39 39 39 _____________________________________________________________________ Table 5b: Average Adjusted LCV scores for the U.S. Senate, 1973-2008 Full Senate Senate Percent Senate Democrats Republicans Democrats Mean 61.77 79.15 39.92 51.71 Standard Error 1.24 2.32 0.65 0.96 Median 61.30 78.69 40.17 50.25 Standard Deviation 7.47 13.95 3.91 5.78 Minimum 47.69 52.26 33.09 45.00 Maximum 76.03 99.21 47.15 62.00 # Years 36 36 36 36 _____________________________________________________________________ Table 5c: Sample Statistics for Economic Indicators Per Capita Democrats Republicans Democrat Republican Income Full Control Full Control President President Mean 18913.43 0.49 0.23 0.15 0.13 ? Standard Error 754.52 0.08 0.07 0.06 0.05 Median 18848.00 NA NA NA NA Standard Deviation 4711.17 0.50 0.43 0.36 0.34 Minimum 11955.00 NA NA NA NA Maximum 27319.00 NA NA NA NA # Years 39 39 39 39 39 all models run with Senate samples so that the data stream for the dependent variable and independent variables is consistent. 56 Figure 2a. Unadjusted LCV scores 1973-2008 --Senate Figure 2b. Adjusted LCV scores 1973-2008 --Senate 57 Figure 3a. Unadjusted LCV scores 1970-2008 ?House of Representatives Figure 3b. Adjusted LCV scores 1970-2008 ?House of Representatives 58 Once the LCV scores were properly adjusted, we estimated time series models based on equation 2 to determine if indeed there are any statistically significant linkages between indicators of economic well-being and political support for pro-environment policy, as reflected in the adjusted LCV scores. Models were run for samples consisting of the full House and Senate, House Democrats, House Republicans, Senate Democrats, and Senate Republicans. The party-specific samples were used to explore the possibility that the response functions differ from one party to the other - - which might reasonably be expected in light of the large divergence in mean LCV scores, both nominal and real, between the two parties. We then tested the various models to assess whether they met the required stationarity and white noise requirements for time-series estimation. 23 While models estimated across Republicans in both the House and Senate were AR (1) stationary, none of the other model/sample possibilities (House Democrats, Senate Democrats, House ? combined sample, Senate ? combined sample) met the requirements of stationarity 24 and randomly-distributed residuals. Each variable was then first-differenced and a Dickey Fuller (DF) test performed on the differences. The explanatory variables were all Augmented Dickey-Fuller (ADF) stationary unless otherwise noted in Table 6, which reports the test results. 23 The typical issue in time series estimation is that variables are not normally distributed, thus the coefficient estimates generated by Ordinary Least Squares (OLS) regression are characterized by understated standard errors. If theory suggests x t should affect y t and both have trends, they will be correlated and coefficients will appear significant when in reality the explanatory power of x t is overstated. The residual error term ? t in an OLS model should be distributed randomly (colloquially described as white noise ?WN). An OLS regression with non-stationary variables is characterized by autocorrelation (? t , ? t-1 ) in the residual series ? t . Autocorrelation implies there is useful information in the residual ? t relative to predicting y t . Obtaining this information requires either a different model or transformed variables. 24 If series are not stationary they may be difference stationary random walks, and OLS regressions in differences are then reliable. If variables are difference stationary, the difference model ; ?y t = ? 0 + ? 1 ?x t + ? 2 ?z t + u t captures the dynamic adjustment process. 59 In all cases the explanatory variables were first-difference stationary and met the WN requirement in the ARCH tests 25 (except for House Republicans, which still showed signs of heteroskedasticity in the ARCH (1) test). We nonetheless report the results of House Republicans with this in mind. After using an OLS regression procedure to estimate each of the four models (with per capita income as an explanatory variable) for each of the six samples (House Democrats, House Republicans, House ? all members, Senate Democrats, Senate Republicans, Senate ? all members), we tested for cointegration using the Engle-Granger method in order to estimate an Error Correction model. In several cases the residual of the spurious regression is not difference stationary in the Engle-Granger test, so we estimated a lagged transformation model 26 y t = ? 0 + ? 1 y t-1 + ? 2 x t + ? 3 x t-1 + ? 4 z t + ? 5 z t-1 + ? t LTM (3) Table 6a: Stationarity Test Results - Augmented Dickey Fuller Tests for Dependent variables Variables ADF test ARCH (1) Result ? F ? (r ) ? (?) F-stat res?? Series LCV House -0.78 6.67 0.002 ~0 0.19 WN 27 RW? Repubs. -1.11 6.31 0.077 ~0 8.39 WN RW Democ. -0.68 6.89 -0.016 ~0 3.29 WN RW Senate -0.22 1.24 -0.011 ~0 0.19 WN RW Repubs. -0.71 5.22 -0.084 ~0 0.43 WN RW Democ. -0.43 3.00 0.060 ~0 0.04 WN RW Critical ? ?3 r DWh-stat F(1,30) r Value -3.60