¹ See, e.g., Weissberg, Robert, Public Opinion and Popular Government (Englewood Cliffs, NJ: Prentice Hall, 1976)Google Scholar; Monroe, Alan, ‘Consistency Between Constituency Preferences and National Policy Decisions’, American Politics Quarterly, 12 (1979), 3–19CrossRefGoogle Scholar; Page, Benjamin I. and Shapiro, Robert Y., ‘Effects of Public Opinion on Policy’, American Political Science Review, 76 (1983), 175–90CrossRefGoogle Scholar; Page, Benjamin I. and Shapiro, Robert Y., The Rational Public (Chicago: University of Chicago Press, 1991)Google Scholar; Erikson, Robert S., Wright, Gerald C. Jr and McIver, John P., Statehouse Democracy: Public Opinion and Policy in the American States (New York: Cambridge University Press, 1993)Google Scholar; Marshall, Thomas R., Public Opinion and the Supreme Court (Boston, Mass.: Unwin Hyman, 1988)Google Scholar; Hartley, Thomas and Russett, Bruce, ‘Public Opinion and the Common Defense: Who Governs Military Spending in the United States?’ American Political Science Review, 86 (1992), 905–15.CrossRefGoogle Scholar

² Note that this conceptualization of the opinion-policy connection effectively integrates the ‘covariation’ and ‘satisfying’ models of congruence offered by Weissberg, Public Opinion and Popular Government. The conceptualization actually has deep roots in political science, including Deutsch, Karl W., The Nerves of Government (New York: The Free Press, 1964)Google Scholar and Easton, David, A Framework for Political Analysis (Englewood Cliffs, NJ: Prentice-Hall, 1965)Google Scholar. For a more complete development of the thermostatic model, see Wlezien, Christopher, ‘The Public as Thermostat: Dynamics of Preferences for Spending’, American Journal of Political Science, 39 (1995), forthcoming.CrossRefGoogle Scholar

³ See, e.g., Converse, Philip E., ‘The Nature of Belief Systems in Mass Publics’, in Greenstein, Fred I. and Polsby, Nelson W., eds, Handbook of Social Science, vol. 4 (Reading, Mass.: Addison-Wesley, 1964)Google Scholar; Kinder, Donald E., ‘Diversity and Complexity in American Public Opinion’, in Finifter, Ada W., ed., Political Science: The State of the Discipline (Washington, DC: American Political Science Association, 1983).Google Scholar

⁴ Also see Sniderman, Paul, ‘The New Look in Public Opinion Research’, in Finifter, Ada W., ed., Political Science: The State of the Discipline II (Washington, DC: American Political Science Association, 1993).Google Scholar

⁵ Page, and Shapiro, , The Rational PublicGoogle Scholar; Stimson, James, Public Opinion in America: Moods, Cycles, and Swings (Boulder, Colo.: Westview Press, 1991)Google Scholar; Stoll, Richard, US National Security Policy and the Soviet Union: Persistent Regularities and Extreme Contingencies (Columbia: University of South Carolina Press, 1990)Google Scholar; Durr, Robert H., ‘What Moves Policy Sentiment?’ American Political Science Review, 87 (1993), 158–70CrossRefGoogle Scholar; Wlezien, Christopher and Goggin, Malcolm, ‘The Courts, Interest Groups, and Public Opinion about Abortion’, Political Behavior, 15 (1993), 381–405.CrossRefGoogle Scholar

⁶ In particular, see Durr, , ‘What Moves Policy Sentiment?’Google Scholar but also see Stoll, , US National Security PolicyGoogle Scholar, and Wlezien, and Goggin, , ‘The Courts, Interest Groups, and Public Opinion about Abortion’.Google Scholar

⁷ See, e.g., Abramowitz, Alan I., ‘Issue Evolution Reconsidered: Racial Attitudes and Partisanship in the US Electorate’, American Journal of Political Science, 38 (1994), 1–24CrossRefGoogle Scholar; Jacoby, William G., ‘Public Attitudes toward Government Spending’, American Journal of Political Science, 38 (1994), 336–61.CrossRefGoogle Scholar

⁸ Downs, Anthony, An Economic Theory of Democracy (New York: Harper, 1957).Google Scholar

⁹ Page, and Shapiro, , The Rational Public.Google Scholar

¹⁰ Also see Bartels, Larry, ‘Constituency Opinion and Congressional Policy Making: The Reagan Defense Buildup,’ American Political Science Review, 85 (1991), 429–56.CrossRefGoogle Scholar

¹¹ The model of preferences and policy developed here implies equilibration in both preferences and policy, i.e., that when the two variables are out of equilibrium they adjust to each other over time. In effect, the model implies a ‘cointegrating’ relationship between preferences and policy, which is explicitly considered below.

¹² Hartley, and Russett, , ‘Public Opinion and the Common Defense’.Google Scholar

¹³ Roper also asked the same question in the December of every year since 1973 and this data is used in analysis that follows to explore public responsiveness to policy.

¹⁴ Note that the percentage of people who think spending is ‘about right’ varies within a fairly narrow range over time and the percentages of people who think we are spending ‘too little’ and ‘too much’ over time are virtual mirror images of each other. Not surprisingly, analyses using the percentage of the public that responds ‘too little’ or ‘too much’ separately produce results that differ only marginally from those using the measure of Net Support.

¹⁵ See Ellwood, John W., ‘Budget Authority vs. Outlays as Measures of Budget Policy,’ paper presented at the Annual Meeting of the American Political Science Association (Washington, DC, 1986).Google Scholar

¹⁶ Although regular appropriations account for most funding on an annual basis, additional funding often has been provided in supplemental appropriations; see Wlezien, Christopher, ‘The Political Economy of Supplemental Appropriations’, Legislative Studies Quarterly, 18 (1993), 51–76CrossRefGoogle Scholar. Note, however, that combining the appropriations into a single measure is complicated by the fact that supplementals are decided well after regular appropriations, and, thus, may reflect more recent information about public preferences. Regular appropriations data were distilled from the Annual Senate Document, Appropriations, Budget Estimates, Etc.Google Scholar Real dollar values of appropriations were calculated by dividing current dollar values into the gross national product price deflator (1982 = 1.00), from The National Income and Product Accounts.

¹⁷ The economic and political measures are based on Kiewiet, D. Roderick and McCubbins, Mathew D., ‘Presidential Influence on Congressional Appropriations Decisions’, American Journal of Political Science, 32 (1988), 713–36CrossRefGoogle Scholar; Soviet spending measures were drawn from Hartley, and Russett, , ‘Public Opinion and the Common Defense’.Google Scholar

¹⁸ In order to preserve a precious degree offreedom, the value ofNet Support in 1971 (from Roper) was used as the measure of Net Support_t–2 for fiscal year 1974, i.e., as the measure of Net Support in 1972. This coding decision poses only negligible consequences for the estimated parameters. Note also that when a variable that captures the change in the federal deficit is included in the model, following Hartley, and Russett, , ‘Public Opinion and the Common Defense’Google Scholar, its estimated effect is negative, implying that as the deficit increases defence appropriations also increase. This finding contrasts with Hartley and Russett's result, but the coefficient is far too unreliable to credit.

¹⁹ Of course, as the results in Table 1 make clear, policy makers do respond to the change in Net Support_t–1, but the pattern is not distinguishable from the responsiveness to the level of Net Support_t–2 (recall that the level of Net Support_t–1 represents the sum of the change in Net Support_t–1 and the level of Net Support_t–2).

²⁰ As is suggested in the text and is evident from the analysis, presidential requests capture factors other than Net Support and the Party of the President that ultimately find expression in appropriations. Regarding the possibility that correlated errors produce biased estimates in the analyses of appropriations, note that the results do not differ substantially when the predicted percentage change in appropriations requests (based on the estimated model in Table 2) is used in place of the actual percentage change in requests.

²¹ Although the Durbin–Watson statistics suggest some measure of positive autocorrelation in the models of requests and appropriations, the estimates are not statistically significant (p < 0.10, two-tailed). Moreover, explicitly accounting for the existing autocorrelation does not meaningfully alter the results. Note also that the estimated effects of Net Support on requests and appropriations are slightly larger when the years 1981–83 are excluded from the analysis, apparently reflecting some level of non-linearity in the responsiveness of both the president and Congress. In effect, institutional actors become slightly less responsive to Net Support as the measure tends toward its upper (or lower) bounds – technically, the measure is bounded by – 100 and 100. Since the estimated non-linearity in institutional responsiveness is minor, the results based on linear specification are reported here.

²² The effect may seem counterintuitive, for it implies that defence appropriations increase in each year under Republican presidents. Note, however, that the effect of the party of the president on defence appropriations is not isolated to transitions in administrations; see Wlezien, Christopher, ‘The President, Congress, and Appropriations, 1951–1985,’ American Politics Quarterly, 24 (1996), forthcomingCrossRefGoogle Scholar. Moreover, the effect holds only when die level of Net Support_t–1 is included in the model, which is exactly what one might expect if appropriations feed back on Net Support.

²³ See, e.g., Brody, Richard A., Assessing the President: The Media, Elite Opinion, and Public Support (Stanford, Calif.: Stanford University Press, 1991)Google Scholar; MacKuen, Michael B., Erikson, Robert S. and Stimson, James A., ‘Peasants or Bankers? The American Electorate and the US Economy’, American Political Science Review, 86 (1992), 597–611CrossRefGoogle Scholar. But also see Sanders, David, Marsh, David and Ward, Hugh, ‘The Electoral Impact of Press Coverage of the British Economy’, British Journal of Political Science, 23 (1993), 175–210CrossRefGoogle Scholar, for a different formulation.

²⁴ See Nisbett, Richard and Ross, Lee, Human Inference: Strategies and Shortcomings of Social Judgment (Englewood Cliffs, NJ: Prentice-Hall, 1980).Google Scholar

²⁵ Thus, accounting for any positive simultaneity between changes in appropriations and Net Support, which is not straightforward, can only strengthen the negative, feedback relationship between them.

²⁶ See Abolfathi, Farid, ‘Threat, Public Opinion, and Military Spending in the United States, 1930–1990’, in McGowan, Patrick and Kegley, Charles W. Jr, eds., Threats, Weapons, and Foreign Policy (Beverly Hills, Calif.: Sage, 1980)Google Scholar. Also see Kamlet, Mark S. and Mowery, David C., ‘Influences on Executive and Congressional Budgetary Priorities’, American Political Science Review, 81 (1987), 155–78.CrossRefGoogle Scholar

²⁷ Niemi, Richard G., Mueller, John and Smith, Tom, Trends in Public Opinion: A Compendium of Survey Data (Westport, Conn.: Greenwood Press, 1989).Google Scholar

²⁸ Neither the GSS nor AIPO asked the question in 1978, so the value in that year represents the average of the values in 1977 and 1979. Excluding the year from the analysis that follows makes virtually no difference to the estimated parameters.

²⁹ Kegley, Charles W. Jr and Wittkopf, Eugene R., American Foreign Policy: Pattern and Process (New York: St Martin's Press, 1989)Google Scholar. The potential limitation of the measure is most notable in 1991, when Net Support increased in apparent correspondence with American involvement in Desert Storm. Note, however, that the measure of Net Dislike outperforms the more common spending-based indicators of need for defence spending (see fn. 17) in the models of Net Support. Thus, it appears that the measure captures much more information that is relevant to the public demand for spending, such as relates to important real world events that have no basis in spending, e.g., the Soviet invasion of Afghanistan.

³⁰ Note, especially, that the estimated responsiveness to appropriations change is not the result of the correspondence between the large increase in appropriations and the dramatic drop in Net Support in 1982 (see Figure 1). When a dichotomous variable for 1982 is included in the model, the effect of Appropriations only drops slightly and remains highly reliable (b = −1.57; s.e. = 0.51). Other analyses show that the public is not solely responsive to the (dichotomous) direction of appropriations change and is not independently responsive to the party of the president.

³¹ The positive, statistically significant intercept in Table 3 suggests that the public's preferred level of defence spending tends to increase over time, ceteris paribus. The intercept (7.07), taken together with the coefficient for Appropriations, (1.90), implies that the amount of appropriations the public wants increases by about 3.72 per cent (in real dollars) on an annual basis. In other words, assuming Net Dislike remains unchanged, Net Support will tend to drift upward unless appropriations increase by just less than 4 per cent, which exceeds the average annual growth in the real gross national product. Although the tendency may largely be due to ‘income effects’, the possibility is not considered here; for a straightforward discussion of income effects, see Kreps, David, A Course in Microeconomic Theory (Princeton, NJ: Princeton University Press, 1990).Google Scholar

³² Outlays data were drawn from the Budget of the United States Government, Fiscal Year 1993 (Supplement, 02 1992).Google Scholar

³³ Moreover, separate analyses show that the pattern holds across subcategories of education and party identification, implying that the evident responsiveness to appropriations is not driven by a particular segment of the American public.

³⁴ Analysis using cointegration methodology produces results that do not differ meaningfully from those presented herein – see the Appendix. Such public responsiveness must reflect information communicated by the mass media (see fn. 23) or in other ways. For a basic discussion of two-stage information flows, see Stimson, James A., ‘The Paradox of Ignorant Voters, but Competent Electorate’, in Lutz, Donald S. and Tedin, Kent L., eds, Perspectives on American and Texas Politics (Dubuque, Iowa: Kendall/Hunt, 1989)Google Scholar. For an interesting assessment of the mediating role of political elites, see Brody, , Assessing the PresidentGoogle Scholar, especially the discussion of how ‘rally’ events influence presidential approval in the United States.

³⁵ It is apparent in Table 4 that the two (March and December) measures of Net Support differ consistently, which finds expression in the intercepts and error terms. This pattern apparently is due to ‘house effects’; see Smith, Tom W., ‘In Search of House Effects: A Comparison of Responses to Various Questions by Different Survey Organizations’, Public Opinion Quarterly, 42 (1978), 443–63.CrossRefGoogle Scholar

³⁶ Since most of the effects on March Net Support_t–1 already are reflected in December Net Support_t–2, it appears that presidents are not responding to the most recent information available to them when appropriations are requested (recall that presidents are not responsive to March Net Support_t–1 – see Table 2). This conjecture is supported by analysis of requests that includes the measure of December Net Support_t–2.

³⁷ The estimate is based on the following model of appropriations:

Appropriations_t = −3.15 + 10.26 Party of the President_t–1 + 0.35 Net Support_t–1.

³⁸ The number is calculated by dividing the estimated intercept (7.07) by the absolute value of the coefficient for Appropriations (1.90). Also see fn. 31.

³⁹ The numbers were generated by inserting the value of appropriations change (3.72) into the equation described in fn. 37 and then solving the equation for the values of Net Support.

⁴⁰ Also see Wlezien, , ‘The Public as Thermostat’.Google Scholar

⁴¹ See, e.g., Beck, Nathaniel, ‘The Methodology of Cointegration’, Political Analysis, 4 (1992), 237–48CrossRefGoogle Scholar. To determine whether preferences and policy are cointegrated, it is first necessary to assess whether the separate series are integrated. Intuitively, an integrated series is one that is quite long-memoried, where the effects on the variable at a particular point in time become incorporated into future values. Technically, an integrated series is non-stationary, meaning the current change in the variable is unrelated to its lagged level. Dicky–Fuller tests indicate that Net Support and defence appropriations (and Net Dislike of the Soviet Union) are integrated; indeed, the series appear to be first-order integrated. Such series also are referred to as a ‘random walk’ or ‘unit root’ process. If Net Support and defence appropriations are cointegrated, then a linear combination of the series (in levels) produces a stationary series, one where the current change is (negatively) related to the lagged level.

⁴² The procedure used here relies on the discussion in Durr, Robert H., ‘An Essay on Cointegration and Error Correction Models’, Political Analysis, 4 (1992), 185–228.CrossRefGoogle Scholar

⁴³ Although this specification is theoretically satisfying, the implied specification of the model of appropriations change is not. That is, policy makers would be expected to adjust appropriations for year t in response to disequilibria between Net Support and Net Dislike in year t – 1 and appropriations for year t itself, an obvious impossibility. This specification ‘problem’ actually is much more apparent than real, because it is not necessary to estimate an error correction model of appropriations change: (1) policy makers are expected to (and do) respond positively to the level of, not the change in, Net Support in year t – 1; and (2) as is implied by the foregoing analyses, and confirmed by the analysis that follows, the public adjusts Net Support immediately in response to appropriations change