The polls: the coalitional president from a public opinion perspective.

Cohen, Jeffrey E.

As Thomas Cronin (1980; also Cronin and Genovese 2004) insightfully pointed out thirty years ago, the presidency is beset with paradoxes, at times being tugged, pulled, or pushed in differing and often contradictory directions. For instance, on the one hand the president is a symbol, representative, and leader of the entire nation. But the president is also a partisan who seeks benefits for some sectors of the polity, such as his party and those who voted for him. Presidents seek these particularized group-specific benefits as they try to build coalitions in support of their electoral and policy goals. Thus we may also view the president from a coalitional perspective, in which he builds a support coalition composed of specific groups and population subsets of the nation (Seligman and Covington 1989; Edwards 2000; Mueller 1970, 1973).

The coalitional perspective hypothesizes that groups and/or population subsets will hold the president accountable for conditions and actions that specifically affect and/or target group members. In this article, I test this coalitional hypothesis using newly available monthly tracking polls for each of the fifty states. Do state mass publics hold the president accountable for national economic performance, as the national leader perspective hypothesizes, or do their evaluations of the president hinge on state-specific factors, such as the performance of the state economy and presidential attention to the states, as the coalitional hypothesis would predict?

In the next section, I review these two competing perspectives on the presidency, the national leader versus the coalition builder. Although more research exists on public approval of the president from the national leader perspective, a small but important strain is compatible with the coalitional viewpoint. Then I discuss the data used to test these two competing views, which consist of monthly tracking polls across all fifty states from SurveyUSA. Such data have certain advantages over the Official State Job Approval Ratings (JAR) data (Beyle, Niemi, and Sigelman 2002), the other major source of information on state public attitudes toward the president, as reviewed in more detail below. Succeeding sections present the specific hypotheses for testing and the data analysis. The conclusion discusses the relative importance of these two perspectives for understanding presidential approval and directions for future research.

Two Perspectives on the President: National Leader versus Coalition Builder

At one and the same time, the president is the nation's leader but also a partisan who seeks to benefit some sectors of the polity over others, for instance, his party versus the opposition or those who voted for him as opposed to those who did not. Much research on the president assumes the first perspective, with the corollary assumptions that the public will hold the president responsible for the overall state of the nation. Studies in this tradition have, for instance, demonstrated that the state of the national economy affects public evaluations of the president.

It makes much sense to view the president as a national leader. The president, along with the vice president, is the only elected official to have the entire nation as his constituency. Furthermore, presidents preside over national ceremonies and occasions, such as the inauguration and the annual State of the Union address, something no other elected official can claim as a fundamental part of their post. In foreign policy, the president speaks for the nation, again something denied to any other elected official. And while the secretary of state represents the nation to other nations, the secretary does so as the president's agent. Along with these national duties comes accountability, as the public holds the president accountable for the state of the nation, its economic and domestic health, and its policies toward other nations. A large literature has studied this public accountability function, the most impressive and repeated finding that the health of the national economy affects public approval of the president (the literature is huge; see the review in Lewis-Beck and Stegmeier, forthcoming).

But the president also promotes policies and actions for the benefit of some sectors of the polity and society, and sometimes those actions impose costs on other sectors. For instance, presidents routinely seek advantage for their party or for those who supported him in his quest for the election. Presidents are more apt to appoint or nominate candidates for government posts from their party; rarely are those from the opposition so named. Presidents follow this group-specific strategy in order to build a coalition of support large enough to ensure electoral victory and congressional support for his policy initiatives, two important presidential goals.

The Madisonian framework of our government explains this aspect of presidential behavior and policy making. Rarely can presidents command. To prevail on a roll call vote before Congress requires that presidents build a coalition of support usually among specific groups, interests, and/or subsets of the nation and polity (Seligman and Covington 1989; Edwards 2000). At times, presidential coalitions will be composed of specific interest groups. Sometimes the coalition will be narrowly party based. And because of the Electoral College, in which states serve as the fundamental unit, we can think of the president's electoral coalition as being composed of states. In our current polarized political setting we often describe states as being red, blue, or purple, a colorful shorthand for denoting whether a state belongs to the Republican or Democratic electoral coalition or is up for grabs.

Presidents may use all of the tools at their disposal to build winning support coalitions, such as persuasion, vote trading, vote buying, dispensing patronage, evoking partisan loyalties, taking issue stands, speaking before select audiences, naming people to office, and so on, but never can the president command a member of Congress to support him on a roll call and never can a president force a citizen to vote for him or a friendly candidate from his party.

To win on roll calls (or in elections) does not require unanimity, but only 50 percent plus 1 (or in the Senate 60 percent plus 1, to overcome any potential filibuster). As William Riker (1962) taught us long ago, there is no reason to build a coalition larger than necessary to win, the minimum winning coalition. Consequently, some groups and interests will be excluded from the winning coalition, and those excluded may suffer, sometimes paying the costs of policy that the winners get to implement, while groups included in the coalition will receive benefits from the president.

Does this coalition-building activity have consequences for presidential support among group members? Are groups that receive targeted benefits from the president more likely to approve of his performance as president? Do groups that pay the costs of presidential coalition-building efforts hold the president accountable for those costs by withdrawing their approval of the president? Or is it the case that, despite presidential coalition-building behavior, when it comes to general job approval, the president is only held accountable for the state of the nation? In other words, will groups that may be harmed by presidential coalition-building efforts, for instance, because presidential policies impose costs on them, still approve of the president if the nation overall is doing better? How much do sociotropic standards enter into public evaluations of the president (the national leader perspective) versus "pocketbook" or "personalistic" standards (the coalitional perspective)?

Research Using the Coalitional Perspective

Although most research on presidential approval utilizes the national leader perspective, several highly influential studies can be viewed from the coalitional standpoint. As reviewed here, none presents a clear test of the coalitional hypothesis, which states that presidential approval of members in a particular group will be a function of the benefits and costs of presidential actions and the consequences of those actions on group members.

The seminal study in the coalitional vein is Mueller's (1970, 1973). In his early study of the dynamics of public approval of the president, Mueller notices a general cycle across presidents--their approval levels decline over time. Mueller offers his concept of a "coalition of minorities" effect to explain this decline. According to the coalition of minorities thesis, presidential approval is an aggregation of the approval of many groups. Presidents, as policy makers, must make decisions, and groups will respond differentially to these decisions. Some groups will applaud the president's decision(s), while others will criticize the decision(s). Over time, as a group comes to disagree with the president over decisions that he makes, members of the group will withdraw their support of the president, and presidential approval will decline.

Mueller does not offer a direct test of his coalition of minorities thesis, instead only suggesting that patterns in presidential approval data are consistent with the thesis. For instance, he notes that Eisenhower's approval slides less than other presidents, which he attributes to Eisenhower's lower activity level: by making relatively fewer decisions compared to other presidents, Eisenhower alienated fewer groups. Stimson (1976) and Brace and Hinckley (1992) offer other theoretical mechanisms to account for this cycle in presidential approval. Time may mean or measure several things, the coalition of minorities effect being only one possibility.

Hibbs, Rivers, and Vasilatos (1982) take the coalitional perspective in another direction. They ask whether different groups in the population, specifically partisan and economic classes, respond differently to economic conditions. They find that these groups do respond differently as their political and class interests would predict. For instance, those in the lower class react to increases in unemployment with declines in presidential approval, while upper-class individuals move in an opposite direction. Taking this stream of research in another direction, Ragsdale (1987) finds differences in group reactions to presidential speeches. Although these studies can be viewed from a coalitional perspective, they do not test the specific implications of the coalitional hypothesis. That hypothesis asks whether population subgroups respond to presidential decisions and the implications of those decisions specific to the group's interests, and not whether groups respond to changes in national economic conditions or presidential behaviors, like speeches, designed for a national audience.

Cohen and Powell (2005) is the first study to offer a direct test of the coalitional hypothesis by asking whether presidential visits to a state improve presidential approval in the visited state. In that study, which uses the JAR data, a presidential visit to a state is a behavior designed to target a population subgroup, in this instance

the mass public within a state. It was found that presidential visits to a state result in about a two percentage point increase in presidential approval, a modest, albeit statistical significant increment.

Several problems plague that study's analysis, however. First, the JAR data that was relied on, despite offering a large number of cases for analysis, were spotty across the states, containing many gaps in approval. Even though large number of control variables were applied, it could only be assessed whether a presidential visit to a state is associated with a higher approval level. The possibility always exists that every rival explanation was not controlled for, and thus, that the effect that was uncovered is spurious. A stronger test of causality would show that a presidential visit to a state leads to a change in presidential approval in the state. Without presidential approval in the prior month, something quite rare in the JAR data, one cannot make this stronger causality test.

Second, that study did not take into account that a presidential visit may be endogenous, that is, some factor associated with the state that is also associated with presidential approval in the state leads presidents to visit the state. For instance, Ragsdale (1984) found that changes in national approval lead presidents to make primetime television addresses. Declines in approval at the state level, perhaps for states critical to the president's coalition, may similarly stimulate a presidential visit to that state. By not taking into account the simultaneity between presidential visits and state approval of the president, the model discussed above is misspecified. As a result, the impact of presidential visits on state-level presidential approval may have been either over- or understated.

Thus, although several important studies can be viewed from a coalitional perspective, all have problems. They either do not test the core hypothesis of the coalitional perspective or data limitations inhibit their ability to design a strong test of the coalitional hypothesis. The next section describes newly available data that allow a stronger test of the coalition perspective on presidential approval.

The SurveyUSA Monthly Tracking Polls

Since May 2005, SurveyUSA has been posting its monthly tracking polls on its Web site (http://www.surveyusa.com). SurveyUSA is a polling firm that conducts monthly polls on public attitudes toward political personalities, issues, and elections for newspapers, television stations, and other news outlets across all fifty states. From August 2005 through January 2006, SurveyUSA offers data on state-level presidential job approval continuously, without any monthly gaps.

The SurveyUSA data are superior to the JAR data for this article for several reasons. First, unlike the JAR data, at least for the August 2005-January 2006 time span, there are no monthly gaps. From these data, one can create monthly job approval change scores, which will allow a stronger test of causality than the JAR data permit. Second, data exist for all fifty states, unlike the JAR data, which are more frequent for some states than others. Third, SurveyUSA uses the same question across all states and months, unlike the JAR data, which is a compilation of different questions, across different survey houses, utilizing different methodology. Thus, there is a greater degree of comparability for the SurveyUSA data than the JAR data. The SurveyUSA question reads: "Do you approve or disapprove of the job George W. Bush is doing as President?" which is identical to the Gallup job approval question, the basis of so much research. Fourth, each SurveyUSA poll uses the same sample size per survey, 600, with a margin of error of +/- 4.1 percent, using random digit dialing. Again, this introduces a degree of comparability across the states absent with the JAR data.

Hypotheses and Analysis

The coalitional perspective hypothesizes that groups will hold the president accountable for conditions and actions that specifically affect and/or target group members. Analysis below considers two such hypotheses, one that relates to the condition of the state's economy and the other looks at the impact of a presidential visit to the state. More formally,

H1: As the level of unemployment in the state rises (falls) from one month to the next. presidential approval at the state level will fall (rise).

H2: A presidential visit to a state will lead to an increase in presidential approval in the state from the past month to the month of the president's visit.

To test Hypothesis 1, the unemployment change hypothesis, requires two variables, monthly change in state-level unemployment and monthly change in presidential approval. Change in state-level unemployment is formally defined as State Unemployment [%.sub.t]--State Unemployment [%.sub.t-1], where t indicates the current month and t-1 the past month. Change in presidential approval is defined as [Net Approval.sub.t]--[Net Approval.sub.t-1], where t and t-1 are defined as before and net approval is defined as Presidential Approval--Presidential Disapproval. Subtracting disapproval from approval helps control for variations in the percentage of "don't know" responses. Hypothesis 2 uses the same dependent variable, but includes a presidential visit variable, scored "1" if the president visited the state in month t and "0" otherwise. Finally, I add a control for change in the level of national unemployment, defined as National Unemployment [%.sub.t]--National Unemployment [%.sub.t-1]. The national unemployment variable will allow us to test the national accountability hypothesis at the state level and compare how much a state's mass public holds the president accountable for the state versus the national economy. The six-month period for which we have data produces 250 cases for analysis, based on five pairs of months (e.g., August 2005-September 2005, etc.). The equation to be estimated can be formally stated as:

Monthly Change in Net Approval = Constant + Monthly Percent Change in State Unemployment + Presidential Visit Month, + Monthly Percent Change in National Unemployment.

Before proceeding to the analysis, there are several estimation issues that must be addressed. First, the SurveyUSA data, although presented as point estimates, are actually poll estimates within a sampling error of +/- 4.1 percent. This induces a degree of measurement error into the dependent variable, although that measurement error is likely to be random. This sampling error property of the dependent variable will have at least one analytic consequence, namely that results should not be highly efficient and the estimation should not produce high [R.sup.2]'s. But this should not affect the tests of statistical significance for the relationships between the independent and dependent variables. To account for this measurement error, the analysis below uses Huber-White robust standard errors.

Second, the time series element of the data set violates the regression assumption of independence of observation. With such a short series, six months, we have little to fear from autocorrelation, plus the series is too short to estimate an autocorrelation function. Still, the lack of independence of observations (past month's value of presidential approval may affect the current month's value) requires correction, which is accomplished by using the cluster option in STATA 9.

Finally, it is likely that a presidential visit is endogenous, that is, change in presidential approval may affect the presidential decision to visit a state. Ragsdale (1987) found that changes in national-level presidential approval lead presidents to speak to the nation on prime-time television. Instrumental variable techniques, such as two-stage least squares (2SLS), can be used to correct for the simultaneity between presidential approval and visits. To implement 2SLS requires an instrumental variable, that is, a variable that predicts presidential visits in this case, but not change in presidential approval. The analysis below employs two instrumental variables: [Level of State Unemployment.sub.t-1] and a dummy variable for states directly affected by Hurricane Katrina (Mississippi and Louisiana). (1)

Table 1, Model 1 presents the results of the 2SLS analysis, clustered on state with Huber-White robust standard errors. First, the overall fit of the estimation ([R.sup.2]) is meager, an implication of the error variance in the dependent variable. Despite this issue, however, several independent variables emerge as strong predictors of monthly change in presidential approval at the state level and results indicate some support for the coalitional perspective, but also support for the national leader model.

Consistent with the national leader perspective, change in the national unemployment level leads to changes in presidential approval. Each one point rise (fall) in national unemployment is associated with a corresponding drop (increase) of 5.8 percent in presidential approval, which is statistically significant with a p value of less than .01. Although this suggests extreme sensitivity of the public to unemployment, recall that monthly change in unemployment at the national level rarely exceeds 0.2 percent, the maximum in these data. A 0.2 percent change in unemployment will translate into an approximately 1.2 percent shift in approval, while a 0.1 percent change will shift presidential approval by about 0.58 percentage points. In the short run these shifts in approval appear meager, but when national unemployment is trending either up or down across a period of months, the cumulative effect of trending unemployment can have considerable effects on state-level presidential approval.

There is some suggestion of coalitional effects in these data as well. Results indicate that presidential approval in the states responds to changes in state unemployment but not presidential visits. Each one percentage point change in state unemployment, controlling for change in national unemployment, leads to a 0.35 percent shift in presidential approval. This finding skirts the traditional .05 cutoff for statistical significance, at p = .055. However, as noted below, when presidential visits, which does not affect approval, is dropped from the analysis, the significance level of state unemployment change improves to .047.

State unemployment is more volatile in the short run than national unemployment. For the most past, monthly state unemployment change rarely exceeds 1 percent; when it does, presidents will realize the 0.35 percent gain or loss in approval. Again, if unemployment in a state trends above and beyond national trends, this can cumulate into large losses (improvements) in presidential approval. But sometimes states feel unemployment shocks, as was the case for Mississippi and Louisiana due to the effects of Hurricane Katrina, where the former registered a 3.3 percent drop in employment and Louisiana a whopping 7.3 percent loss. Adding a dummy variable for Katrina does not affect substantive results, however, but dropping the Katrina observations from the analysis increases the effect of state unemployment change on presidential approval. Without these states, each 1 percent shift in state unemployment leads to a corresponding 4.4 percent shift in presidential approval (p = .02), which is only slightly lower than the effect of national unemployment change, which budges little from the prior estimation (b = 5.6, p = .008).

In contrast, presidential visits to a state appear to have no effect on state-level presidential approval, contrary to the findings in Cohen and Powell (2005). The lack of impact of presidential visits persists even if we drop the Katrina-affected states from the analysis. President Bush visited Mississippi and Louisiana in these data at least once per month almost every month of and after the hurricane. Noting this lack of impact I reestimated the model, dropping the presidential visit variable and thus the 2SLS estimation, but retaining the clustered regression and robust standard errors. This produces the results in the second panel on Table 1 labeled Model 2.

Conclusion

The analysis in this article presents evidence in support of both the national leader and coalition builder accountability hypotheses. Using SurveyUSA monthly tracking polls from August 2005 through January 2006, across all fifty states, and without any gaps across months or states, analysis found that changes in the national unemployment rate affect changes in presidential approval. As national unemployment rises, presidential approval at the state level declines. From this result, we can infer that even population subgroups are likely to hold the president accountable for the state of the nation, a suggestion that groups, in this case state mass public, apply a sociotropic standard when evaluating presidential job performance.

But changes in the health of the state economy also affect presidential approval at the state level. The worse the state's job climate, holding national unemployment constant, the worse the president's approval at the state level. Not only do voters seem to hold the president accountable for the state of the nation but also for the economic health of their states.

This finding suggests that groups react to the group-specific implications of presidential policies and actions. When presidents pursue policy directions that harm or help particular groups in American society and polity, the president can expect to see approval declines among groups harmed and approval rises among groups helped. In a society as complex as the United States, presidents may often make decisions that lead to some segments of the nation being harmed while other segments benefit. Moreover, in an age of polarized politics, when compromise with the opposition party is often fruitless or impossible, presidents may be even more inclined to pursue policies that keep their base happy, to the displeasure of the opposition. Presidents in an age of polarized politics are basically coalitional presidents. This article demonstrates that coalition-building behavior on the part of presidents can have implications for his approval ratings.

At the same time, findings here indicate that presidential visits to a state do not affect state-level approval, contrary to the analysis in Cohen and Powell (2005) that used the JAR data, but even that study detected only small approval gains from visiting a state. Coupling this finding with the above findings about the impact of economics on approval at the state level suggests that symbolic activities have less ability to affect approval than real-world conditions, like the economy. Still, this article looked at only one type of group-specific impact on presidential approval--state unemployment. This leaves open the question of whether other types of presidential decisions and actions will similarly affect the targeted or affected group. The larger message of this article is that the coalitional perspective may be a fruitful line of inquiry, one which will deepen our understanding of the linkages between the president, the public, and subsets of the public.

AUTHOR'S NOTE: I would like to thank Robert Erikson for telling me about the SurveyUSA polls and Costas Panagopoulos for his comments on this article.

References

Beyle, Thad, Richard G. Niemi, and Lee Sigelman. 2002. Gubernatorial, senatorial, and state-level presidential job approval ratings: The U.S. official job approval ratings (JAR) collection. State Politics and Policy Quarterly 2(Fall): 215-29.

Brace, Paul, and Barbara Hinckley. 1992. Follow the leader: Opinion polls and the modern presidents. New York: Basic Books.

Cohen, Jeffrey E., and Richard J. Powell. 2005. Building public support from the grassroots up: The impact of presidential travel on state-level approval. Presidential Studies Quarterly 35(March): 11-27.

Cronin, Thomas E. 1980. The state of the presidency, 2nd ed. Boston: Little, Brown.

Cronin, Thomas E., and Michael A. Genovese. 2004. The paradoxes of the American presidency. New York: Oxford University Press.

Edwards, George C. III. 2000. Building coalitions. Presidential Studies Quarterly 30(March): 47-78.

Hibbs, Douglas A., Jr., R. Douglas Rivers, and Nicholas Vasilatos. 1982. The dynamics of political support for American presidents among occupational and partisan groups. American Journal of Political Science 26(May): 312-32.

Lewis-Beck, Michael S., and Mary Stegmeier. Forthcoming. Economic models of the vote. In The Oxford handbook of political behavior, edited by Russell Dalton and Hans-Dieter Klingeman. Oxford, UK: Oxford University Press.

Mueller, John E. 1970. Presidential popularity from Truman to Johnson. American Political Science Review 64(March): 18-34.

--. 1973. War, presidents, and public opinion. New York: Wiley. Ragsdale, Lyn. 1984. The politics of presidential speechmaking, 1949-1980. American Political Science Review 78(December): 971-84.

--. 1987. Presidential speechmaking and the public audience: Individual presidents and group attitudes. Journal of Politics 49(August): 704-36.

Riker, William H. 1962. The theory of political coalitions. New Haven, CT: Yale University Press.

Seligman, Lester G., and Cary R. Covington. 1989. The coalitional presidency. Chicago: Dorsey.

Stimson, James A. 1976. Public support for American presidents: A cyclical model. Public Opinion Quarterly 40(Spring): 1-21.

(1.) Both variables correlate at the zero order with presidential visits: Katrina dummy r=.31 (p = .000), lag of state unemployment r = .17 (p = .004).

Jeffrey E. Cohen is a professor of political science at Fordham University and the author of several books including Presidential Responsiveness and Public Policy-Making, as well as articles in numerous journals including the American Political Science Review, American Journal of Political Science, and Journal of Politics.

TABLE 1
Impact of 1% Change in State Unemployment, % Change in National
Unemployment, and Presidential Visits on % Change
in State-Level Presidential Approval (a)

 b SE t p
Model 1

Second-stage results (b)
Constant -0.35 0.18 -1.87 .07
% Change in state unemployment -0.35 0.22 -1.62 .055 *
% Change in national unemployment -5.83 2.11 -2.75 .004 *
Presidential visit 0.27 0.56 0.05 .31 *

Model 2 (c)
Constant -0.32 0.17 -1.94 .058
% Change in state unemployment -0.33 0.19 -1.71 .047 *
% Change in national unemployment -5.83 2.11 -2.77 .004 *

* One-tailed test of statistical significance because
the hypothesis predicts the direction of impact.

(a.) Two-stage least squares results, with Huber-White robust SEs
and clustered on state. n = 250. Lag of state
unemployment and a dummy variable for states directly affected
by Katrina (Mississippi, Louisiana) are used as instrument.

(b.) n = 250, F = 6.63, probability of F = .0003, [R.sup.2] = .03

(c.) n = 250, F = 5.39, probability of F = .008, [R.sup.2] = .03,
based upon a regression that clusters on state and uses
Huber-White robust SEs.

Source:SurveyUSA monthly tracking polls, August 2005-January 2006.