¹ Andrew Rich, Think Tanks, Public Policy and the Politics of Expertise (New York: Cambridge University Press, 2004), p. 219.

² David B. Truman, The Governmental Process: Political Interests and Public Opinion (New York: Knopf, 1951), p. 57.

³ Mancur Olson, The Logic of Collective Action: Public Goods and the Theory of Groups (Cambridge, Mass.: Harvard University Press, 1965); Robert Salisbury, ‘An Exchange Theory of Interest Groups’, Midwest Journal of Political Science, 13 (1969), 1–32; Anthony J. Nownes and Grant Neeley, ‘Public Interest Group Entrepreneurship and Theories of Group Mobilization’, Political Research Quarterly, 49 (1996), 119–46.

⁴ Jack L. Walker Jr, ‘The Origins and Maintenance of Interest Groups in America’, American Political Science Review, 77 (1983), 390–406; Jack L. Walker Jr, Mobilizing Interest Groups in America: Patrons, Professions, and Social Movements (Ann Arbor: University of Michigan Press, 1991).

⁵ Terry M. Moe, The Organization of Interests: Incentives and the Internal Dynamics of Political Interest Groups (Chicago: University of Chicago Press, 1980), p. 36.

⁶ Virginia Gray and David Lowery, The Population Ecology of Interest Representation: Lobbying Communities in the American States (Ann Arbor: University of Michigan Press, 1996); for a complementary approach, see Anthony J. Nownes, ‘The Population Ecology of Interest Group Formation: Mobilizing for Gay and Lesbian Rights in the United States, 1950–98’, British Journal of Political Science, 34 (2004), 49–67.

⁷ Beth Leech, Frank Baumgartner, Timothy La Pira and Nicholas Semanko, ‘Drawing Lobbyists to Washington: Government Activity and Interest Group Formation’, Political Research Quarterly, 58 (2005), 19–30.

⁸ David S. Meyer and Douglas R. Imig, ‘Political Opportunity and the Rise and Decline of Interest Group Sectors’, Social Science Journal, 30 (1993), 253–70; Nownes, ‘The Population Ecology of Interest Group Formation’.

⁹ Virginia Gray, David Lowery, Matthew Fellowes and Jennifer L. Anderson, ‘Legislative Agendas and Interest Advocacy: Understanding the Demand Side of Lobbying’, American Politics Research, 33 (2005), 404–34, p. 413.

¹⁰ Moe, The Organization of Interests.

¹¹ Kevin M. Esterling, The Political Economy of Expertise: Information and Efficiency in American National Politics (Ann Arbor: University of Michigan Press, 2004); Gary S. Becker, ‘A Theory of Competition Among Pressure Groups for Political Influence’, Quarterly Journal of Economics, 98 (1983), 371–400; Tim Groseclose and Jeffrey Milyo, ‘A Measure of Media Bias’, Quarterly Journal of Economics, 120 (2005), 1191–237.

¹² Randall L. Calvert, ‘The Value of Biased Information: A Rational Choice Model of Political Advice’, Journal of Politics, 47 (1985), 530–55, employs a similar notion of biased information. In a simplified version of his model, a decision maker seeks advice regarding two alternatives, a ₁ and a ₂. The decision maker’s prior preference is for a ₁ which she believes will give her utility u ₁ > u ₂ (u ₂ is the utility drawn from a ₂) with u ₁,u ₂ ∈ [0,1]. The adviser can make errors in evaluating the policy, so, for example, the probability that a ₁ is recommended given that u ₁ = 1 (i.e., that a ₁ is truly better for the decision maker) is p ^1/b where b is the adviser’s bias. As b > 1 increases, the adviser is more likely to recommend a ₁ (is biased in favour of the decision maker’s preferred alternative). Calvert finds that, unless her services are too costly, the a ₁-biased adviser is preferable to a neutral or a ₂-biased colleague. The intuition is that a recommendation that a ₂ is better given that the adviser is biased towards a ₁ is rare and therefore carries more weight in the decision maker’s calculations. In this sense, the biased adviser reduces uncertainty. Our argument, by contrast, is that think-tank research is biased towards a ₁ because (a) like Calvert’s decision maker the legislator prefers a ₁ and (b) to meet this demand, an entrepreneur will supply a think tank with b sufficiently large to generate a study for which p ^1/b = 1. In our argument, uncertainty is increased by freely provided biased studies because studies provide support for legislators’ debate positions. For Calvert, uncertainty is reduced because the legislator seeks to inform her choice and pays for advice towards that end. Credible debate and informed choice are simply different, and a utility-maximizing decision maker demands biased information for different reasons in each situation.

¹³ Another way to think about this notion of bias is in terms of statistical inference. If ε is construed as the standard error of an estimate of a policy outcome, a confidence band can be constructed. Given that degree of uncertainty, bias in the sense used here comes from the side of the confidence interval being emphasized, i.e., ‘the resulting effect can be as low as’ or vice versa.

¹⁴ Cass R. Sunstein, Free Markets and Social Justice (New York: Oxford University Press, 1997), p. 367.

¹⁵ Richard L. Hall and Alan V. Deardorff, ‘Lobbying as Legislative Subsidy’, American Political Science Review, 100 (2006), 69–84, p. 76.

¹⁶ David Austen-Smith, ‘Credible Debate Equilibria’, Social Choice and Welfare, 7 (1990), 75–93; David Austen-Smith, ‘Information Transmission in Debate’, American Journal of Political Science, 34 (1990), 124–52.

¹⁷ Gray et al., ‘Legislative Agendas and Interest Advocacy’.

¹⁸ David C. Hammack and Stanton Wheeler, Social Science in the Making: Essays on the Russell Sage Foundation, 1907–1972 (New York: Russell Sage Foundation, 1994), p. 12.

¹⁹ Jane Dahlberg, The New York Bureau of Municipal Research: Pioneer in Government Administration (New York: New York University Press, 1966), p. 16.

²⁰ David H. Rosenbloom, Building a Legislative-Centered Public Administration: Congress and the Administrative State, 1946–1999 (Tuscaloosa: University of Alabama Press, 2000), p. 65.

²¹ David M. Ricci, The Transformation of American Politics: The New Washington and the Rise of Think Tanks (New Haven, Conn.: Yale University Press, 1993).

²² Joseph G. Peschek, Policy-Planning Organizations: Elite Agendas and America’s Rightward Turn (Philadelphia, Pa.: Temple University Press, 1987); G. William Domhoff, ‘Where Do Government Experts Come From?’ in G. William Domhoff and Thomas R. Dye, eds, Power Elites and Organizations (Beverly Hills, Calif.: Sage, 1987), pp. 189–203; James A. Smith, Brookings at Seventy-five (Washington, D.C.: The Brookings Institution, 1991).

²³ Rich, Think Tanks, Public Policy and the Politics of Expertise.

²⁴ Ricci, The Transformation of American Politics, p. 208.

²⁵ Donald A. Abelson, ‘Think Tanks in the United States’, in Diane Stone, Andrew Denham and Mark Garnett, eds, Think Tanks across Nations: A Comparative Approach (New York: Manchester University Press, 1998), pp. 107–27.

²⁶ Ricci, The Transformation of American Politics.

²⁷ For examples, see Benjamin I. Page and Robert Y. Shapiro, The Rational Public: Fifty Years of Trends in Americans’ Policy Preferences (Chicago: University of Chicago Press, 1992); John Zaller, The Nature and Origins of Mass Public Opinion (New York: Cambridge University Press, 1992); William G. Mayer, The Changing American Mind (Ann Arbor: University of Michigan Press. 1993); Ronald Ingelhart, Culture Shift in Advanced Industrial Society (New York: Cambridge University Press, 1990); Shmuel T. Lock, Robert Y. Shapiro and Lawrence R. Jacobs, ‘The Impact of Political Debate on Government Trust: Reminding the Public What the Federal Government Does’, Political Behavior, 21 (1999), 239–64.

²⁸ Austen-Smith, ‘Credible Debate Equilibria’; Austen-Smith, ‘Information Transmission in Debate’.

²⁹ The costs of information search are very small – asking a staff member to make a phone call to a think tank, for example – and are diminished by repeat play and reputation effects as described below.

³⁰ Section 501(h) of the US federal income tax code limits the amount of lobbying activity of non-profit organizations – the manner in which think tanks are overwhelmingly organized – to an insubstantial part of the organization’s activities or, more specifically, to 20 per cent of the first $500,000 of expenditures incurred annually. Hall and Deardorff, ‘Lobbying as Legislative Subsidy’, p. 74, discuss the way in which such information impacts the ‘budget’ or resource constraints of legislators allied with the ideological preferences of lobbyists, a position wholly consistent with the free provision to such legislators of information that is costly for the think tank to acquire. In essence, ally legislators ‘have a higher marginal willingness to pay for progress on [the lobbyist’s issue] A and thus will use more of the subsidized resources to expand their effort promoting A’ (p. 76). For the lobbyist or think tank, then, the cost of providing information to allied legislators is outweighed by the policy advocacy it purchases.

³¹ This is not to say, however, that the influence of advocacy-orientated policy analysis off the equilibrium path is not possible. Actions taken ‘off the equilibrium path’ are unexpected, and given that the perfect Bayesian equilibrium concept employed in the debate game does not restrict the beliefs players can form after such an action, the game can resolve itself in a variety of ways (the existence of multiple equilibria).

³² Return to the abstract scenario in the introduction to this article. Suppose that the policy ‘bias’ ε~N(0, σ). Suppose further that σ_C is the overall standard deviation on the policy ‘bias’ for the think-tank community and σ_i is the standard deviation of the bias of think tank i. Think tank i can commit resources to research quality – say, hiring high-quality post-doctoral social scientists or vetting studies with experts outside the think tank before reporting the final results – that bring σ_i toward σ_C. This convergence increases the reliability of biased information. Intuitively, a member of a congressional committee demands a policy prediction that is biased in favour of his or her ideal position for use in debate, but if σ_i diverges from σ_C the credibility added through research-based information procured from think tank i is diminished. Thus, in equilibrium, competition among think tanks leads to convergence in the variance of bias as we have defined it.

³³ The Austen-Smith models are ‘one-shot’ signalling games, and though we do not formally derive them here, some implications are straightforward. In the congressional debate (individual-level) context, the realizations of private information through signalling by members are not independent across periods. A legislator’s argument against one bill – because, say, he or she considers it to be associated with ‘big government’ – carries over to other bills at other times as in the notion of ideological branding (see, e.g., William R. Dougan and Michael C. Munger, ‘The Rationality of Ideology’, Journal of Law and Economics, 32 (1989), 119–42). Moreover, the history of earlier signals are clearly observable to a legislator’s colleagues after debate on one bill closes due to the memories of particular legislators who remain, transcription and archival, and so forth. Given the failure of those two conditions, the one-shot result is not the same as that in a repeated game (Ayça Kaya, ‘Repeated Signaling Games’ (unpublished manuscript, Department of Economics, University of Iowa, 2006), p. 2). In repeated signalling games, costly signalling can occur even after a separating equilibrium – the content of signals corresponds to the true policy preferences of the legislators who send them – is reached in an earlier iteration (see Kaya, ‘Repeated Signaling Games’, pp. 29–32). Since signalling does not stop with a separating result, the demand for research-based information continues, and a think tank’s reputation for providing such information that comports with research standards and is reliably ‘biased’ in favour of a legislator’s preferred policy becomes important in the competitive marketplace. Reputations cluster around reliably ‘biased’ policy positions.

³⁴ We use the term ‘founding’ throughout in reference to the initial establishment of a think tank.

³⁵ Rich, Think Tanks, Public Policy and the Politics of Expertise; Lynn Hellebust, Think Tank Directory: A Guide to Nonprofit Public Policy Research Organizations (Topeka, Kan.: Government Research Service, 1993).

³⁶ In 15 per cent of the sample (fifty-three think tanks), we were unable to locate tax returns. In such cases, we conducted a ‘hot deck’ multiple imputation to generate missing values. One important issue with the use of the Rich (Think Tanks, Public Policy and the Politics of Expertise) data is survivor bias. Since the data were collected at a single point in time, only those think tanks that survived management, economic and other changes were sampled. This means that it is much more difficult to understand the effect of the attrition of think tanks on our analysis. Because we focus on the factors that lead to the start-up or founding of a think tank, we believe that our data are preferred to net change as in the case of market entry.

³⁷ Zoltan J. Acs and David B. Audretsch, ‘Births and Firm Size’, Southern Economic Journal, 56 (1989), 467–75, p. 468.

³⁸ We are interested in understanding the factors that lead to the establishment of think tanks, and as a consequence, do not take into consideration organizational failures. Since our data are retrospective, they may include think tanks that ultimately failed. However, we believe that such failures represent only a small fraction of our sample since so few were created at the beginning of the period we study. There appears to be scholarly consensus about the population through the 1950s. Excellent sources of information underlie our data (see Rich, Think Tanks, Public Policy and the Politics of Expertise, pp. 221–3), and we have confidence that our sample correctly identifies nearly all think-tank formations during the period of study.

³⁹ Keith T. Poole and Howard Rosenthal, Congress: A Political-Economic History of Roll-Call Voting (New York: Cambridge University Press, 1997).

⁴⁰ We also examined US Senate polarization uncovering substantively similar unreported results. This is not surprising given that party polarization in both the House and Senate exhibit similar patterns throughout the time period of our dataset (see Nolan McCarty, Keith T. Poole and Howard Rosenthal, Polarized America: The Dance of Ideology and Unequal Riches (Cambridge, Mass.: MIT Press, 2006)).

⁴¹ J. Bradford DeLong, ‘A History of Bequests in the United States’, in Alicia H. Munnell and Annika Sunden, eds., Death and Dollars: The Role of Gifts and Bequests in America (Washington, D.C.: The Brookings Institution, 2003), pp. 33–52; Michael J. Graetz and Ian Shapiro, Death by a Thousand Cuts: The Fight Over Taxing Inherited Wealth (Princeton, N.J.: Princeton University Press, 2005).

⁴² David W. Rohde, Parties and Leaders in the Postreform House (Chicago: University of Chicago Press, 1991).

⁴³ See Richard F. Fenno Jr, Learning to Govern: An Institutional View of the 104th Congress (Washington, D.C.: The Brookings Institution, 1997).

⁴⁴ If a member first served in the 105th Congress, for example, and lost the next election, but returned to the House for the 108th Congress, that member is also coded as a freshman in both congresses. The member’s party may give some credit for seniority in such a scenario, yet being elected after such a gap in service is an empirical rarity in modern congressional history (see for example, Nelson Polsby, ‘The Institutionalization of the House of Representatives’, American Political Science Review, 62 (1968), 144–68).

⁴⁵ A. Colin Cameron and Pravin K. Trivedi, Regression Analysis of Count Data (Cambridge: Cambridge University Press, 1998).

⁴⁶ See Patrick Brandt, John T. Williams, Benjamin O. Fordham and Brian Pollins, ‘Dynamic Modeling for Persistent Event-Count Time Series’, American Journal of Political Science, 44 (2000), 823–43, for a discussion of issues estimating time-series event count models. We estimate numerous specifications to control for trend, linear and quadratic time trends, logarithmic and piecewise linear spline functions. None of the alternative specifications altered the statistical significance or substantive interpretation of our findings.

⁴⁷ W. K. Newey and K. D. West, ‘A Simple, Positive Semidefinite, Heteroskedasticity and Autocorrelation Consistent Covariance-Matrix’, Econometrica, 55 (1987), 703–8.

⁴⁸ Marginal effects are calculated as average marginal effects (as compared to the more common) marginal effect at the mean (see Tamás Bartus, ‘Estimation of Marginal Effects Using Margeff’, Stata Journal, 5 (2005), 309–29, for a complete discussion).

⁴⁹ We note that the shift in party polarization from the 103rd to 104th Congress was especially large and represents the single largest congressional shift in the data. However, the increase in polarization from the 102nd to 103rd Congress was 1.65 standard deviations and the increase from the 104th to 105th Congress was 1.07 standard deviations. The overall increase in Congressional polarization from 1991 to 1997 was 4.15 standard deviations.

⁵⁰ We have also estimated models that including a quadratic polarization term, producing marginal effects for polarization whilst remaining statistically significant in all the cases. For expository clarity, we focus on models with linear effects.

⁵¹ Clive W. J. Granger, ‘Investigating Causal Relations by Econometric Methods and Cross-Spectral Methods’, Econometrica, 34 (1969), 424–38.

⁵² Calvert, ‘The Value of Biased Information’.

⁵³ Esterling, The Political Economy of Expertise.

⁵⁴ Christopher H. Achen and W. Phillips Shively, Cross-Level Inference (Chicago: University of Chicago Press, 1995).

⁵⁵ Gray et al., ‘Legislative Agendas and Interest Advocacy’.

⁵⁶ Esterling, The Political Economy of Expertise.

⁵⁷ Mark A. Smith, American Business and Political Power: Public Opinion, Elections, and Democracy (Chicago: University of Chicago Press, 2000).

⁵⁸ Rich, Think Tanks, Public Policy and the Politics of Expertise.