The Presidential-Economic Dance: Are New Economic Variables In Rhythm with Traditional Economic Indicators and Presidential Approval?

Transcription

1 The Presidential-Economic Dance: Are New Economic Variables In Rhythm with Traditional Economic Indicators and Presidential Approval? Sara Margaret Gubala University of South Carolina Nathan Dietz Research Associate Department of Research and Policy Development Corporation for National and Community Service September 20, 2002 DRAFT ABSTRACT: This paper examines the relationship between the state of the economy and the popular approval of the president. Although inflation and unemployment should impact the population in the aggregate, it is unclear exactly why individuals should attribute blame to the president when inflation and/or unemployment are high. We hypothesize that while aggregate economic indicators are important determinants of presidential popularity, variables that have a more direct link or personal link to individuals will have the most influence. Therefore, individuals will condition their opinion of the president s job performance primarily on these personal economic variables. While expanding upon the work of other scholars, our paper provides a new assessment of presidential approval from a macroeconomic perspective. In addition, our theory also suggests that fiscal variations and financial changes (measured through loan rates) will have a greater individual impact on consumer spending and thus will affect consumer opinions of the president s job performance. We also consider non-economic variables such as partisanship and temporal variables (such as rallies) that may help explain and predict the dynamics of postwar presidential approval. We hypothesize that our improved measurement of macroeconomic performance, will improve the performance of our model, and our appropriate treatment of the dynamics of presidential popularity, will enable us to more accurately explain short-term and long-term trends in presidential approval. Additionally, our improved model should provide insight into the basis of the long boom periods of widespread presidential approval during the Eisenhower and Clinton administrations. An earlier version of this paper was first presented at the Annual Meeting of the Midwest Political Science Association, Chicago, Illinois, April 25-28, The authors would like to thank Joe BaFumi, Roger Carlsson, Michael Colaresci, Christina Corduneanu-Hucci, Tom Durkin, Susan Hammond, William Jacoby, Susan Johnson, Bill Keech, Stephen Weatherford, and John Williams for helpful comments and advice on this paper. The authors especially thank William Jacoby for his thoughtful comments, assistance with conversion of the data and last but not least, his advice and encouragement throughout the writing of this paper. The paper could not have been completed without his help! The analyses presented within this paper are based on presidential approval data compiled by George C. Edwards III and Alec M. Gallup (1990), and was supplemented by data archived in the Roper Polling Organization

2 ( The economic indicator data was assembled from data archived at the Bureau of Labor and Statistics and Economy.com ( The data were analyzed using E-VIEWS

3 1 Introduction The relationship or careful dance that occurs between the state of the economy and popular approval of the president has interested scholars over time. Many empirical studies have demonstrated that macroeconomic performance has an important effect on political support for elected officials. 1 Even though the president does not have complete control over economic performance, the president, as leader of our nation, is ultimately held responsible for the nation s economic health. Declining economic conditions can lead to the demise of a president s leadership and at the very least can diminish his approval ratings, which can seriously damage a presidents prestige and ultimately will affect a president s power to govern. Presidential approval (as a measure of political support) is an important political resource that influences presidential power. Presidents gain or lose power due to changes within the economy and due to events or shocks that impact the political system, which have a tendency to mobilize support either for or against the president. The public must decide whether or not they approve of the president based upon what the current circumstances are. While economic conditions have been hypothesized to influence presidential approval, exactly how the economy impacts presidential support has yet to be concluded. For instance, the typical specification of a model of presidential approval includes inflation and unemployment as the only measures of economic performance. 2 While both inflation and unemployment are hypothesized to affect presidential support, they have vastly different economic and political implications. According to Kernell (1978) and Kenski (1977a), inflation causes only minimal change in approval, while Monroe (1979, 1981) suggests that popularity responds more precisely to present and lagged changes in inflation. Similarly, most models that include unemployment as a predictor of approval have typically found that unemployment does influence electoral outcomes and support for the president. However, Hibbs (1974) found that unemployment exhibits neither a statistically significant, nor a theoretically important influence on presidential approval. Trying to compare the impacts of these commonly used predictors, Lau and Sears (1981) argue that inflation has a greater political impact than unemployment and that the tradeoff 1 See Monroe (1979) and (1984) and Paldam (1981). 2 Relevant studies that look at the influence that inflation and unemployment has on presidential popularity include: Mueller (1970); Hibbs (1974); Stimson (1976); Kenski (1977abc); Frey and Schneider (1978); Kernell (1978); Monroe (1978); Shapiro and Conforto (1980ab); Golden and Poterba (1980); Kenski (1980); Hibbs and Vasilatos

4 2 between the two indicators must be considered. As Norpoth and Yantek summarize: A president typically presents the public with an economy in which an improvement in one of those macroeconomic conditions is brought at the expense of a deterioration of the other. How then does the public balance the economic seesaw between the two conflicting goals of full employment and stable prices? (Norpoth and Yantek 1993, 803). While the economy does influence presidential approval, conflicting results in the literature make it more difficult to determine where the influence occurs. One thing is certain: economic change does influence politics especially through elections and evaluations of politicians. Contrary to Stimson (1976), who suggested that there are not significant economic influences on popularity, numerous studies have shown that there is a relationship between the economy and approval. However, methodological differences make if more difficult to draw general conclusions about the nature of this relationship. For instance, most empirical studies model presidential approval as a function of leading, lagged and current economic variables, as well as items from public opinion surveys that tap citizen evaluations of the economy. These studies are also theoretically divided on whether or not voters are interested in national (sociotropic 3 ) or personal (pocketbook) economic conditions, or whether they are motivated primarily by retrospective or prospective concerns. 4 The result is that although the perceived state of the economy is clearly an important indicator of presidential approval, it is not clear whether factors outside inflation and unemployment matter to the public. Support for the president fluctuates with changes in general economic conditions. A president skilled enough, or fortunate enough, to preside over a healthy economy is rewarded with public popularity (Kinder 1981) 5. Furthermore, although inflation and unemployment should impact the population in the aggregate, it is unclear exactly why and when individuals should attribute blame to the president as these economic indicators change over time. The underlying assumption of this blame-game implies that the electorate perceives and understands economic trends and that the electorate is (1981); Monroe (1981); Chappell (1983); Monroe and Levi (1983); Norpoth and Yantek (1983); Monroe (1984); Chappell and Keech (1985ab); Ostrom and Simon (1985); and Michales (1986). 3 See Kinder and Kiewiet 1979, 1981; Feldman 1982; Weatherford See MacKuen, Erikson and Stimson (1992) for the stark conclusion that citizens appear to use only prospective evaluations, and Clarke and Stewart (1994) for a re-analysis that shows that other variables also affect popularity. 5 Despite the fact that Kinder (1981) finds support for the idea that a healthy economy leads to rewards in terms of approval, Mueller finds quite the opposite result. Presidents are only punished when the economy slumps and are not credited when the economy improves.

5 3 able to factor this economic information into their assessment of the president. This assumption is rather naïve because it implies that there is homogeneity in information acquisition within the electorate. 6 People differ in both the amount and type of economic information that they have and these differences contribute to variation in economic judgments. How people evaluate economic performance does differ but we can assume that the evaluations are often based upon similar criteria. The conditions under which an individual might approve of the president may be based upon personal conditions, both satisfactions and dissatisfactions. If one is sufficiently pleased or displeased with a situation in which one finds oneself, if one s pleasure or displeasure is sufficiently salient, and if one sees an outside agent as having contributed to that pleasure or displeasure, then feelings about one s personal situation may be carried over to the outside agent (Sigelman and Tsai 1981, 372). Model Specification In our model, this agent is presumed to be the president and support for the president should rise and fall with changes in the economy. However, we improve on the simple specifications of earlier models of approval in several ways. First, we explore the possibility that citizens respond to other indicators of economic activity besides unemployment and inflation. By adding variables that measure changes in the macro economy, such as short-term and longterm consumption indicators, productivity, and monetary policy, we explore the possibility that citizens are more attuned to less familiar changes in the national economy than earlier models have presumed. Second, we unpack the commonly used summary measure of inflation to see what types of price increases exert the most impact on presidential approval. Finally, we allow for the possibility that the inflation and unemployment variables do not exert a constant impact on the popularity of every president, but rather that the popularity of Democratic presidents may depend more on unemployment than on inflation, for instance, while Republican presidents may worry more about inflation. We estimate this improved model using a technique first used by Hibbs (1974) to control for serial correlation in the unmeasured shocks to the president s approval rating. As we argue below, this model is appropriate even though more sophisticated 6 The naïve assumption is important to consider but is really not a problem for our model. The real problem is that we do not know why these economic indicators tend to affect presidential approval. The conflicting results found in previous studies do not help understand exactly why citizens respond the way that they do to changes in inflation and unemployment.

6 4 models that assume a more complicated dynamic structure to the approval time series have come into vogue in political science in the last decade 7. The Personal Link Economic conditions have consistently been found to be related to presidential popularity, especially when the public evaluates the performance of the national economy. Economics is the fate of politicians. It is an article of faith that success or failure of governments in dealing with the economy decides whether or not they survive politically (Norpoth 1984, 253). Within the literature, there are ongoing debates over the extent to which macroeconomic indicators influence presidential approval and vote choice. This literature assumes that people evaluate presidents based upon their expectations about what the president should do and what he should be like. The president should embody the spirit of the nation and be a symbol for all people. Symbolically, the president is held responsible for all of the problems and conflicts within our nation. Poor economic conditions, as exhibited through periods of high inflation and unemployment, indicate that the president is not fulfilling his duty to maintain prosperity and consequently should have disparate effects on presidential approval (Brace and Hinckley 1991). Previous studies have concluded that national economic perceptions affect the vote/approval of the president more than personal economic circumstances (see Kinder and Kiewiet 1979; Sears et al, 1980). Weatherford (1983b) suggests one reason for this: that personal economic circumstances are somewhat removed from the overall economic policy judgment. Declines in personal financial condition do not automatically acquire political meaning. Even if the individual suffers an income decline, he might remain generally satisfied with his standard of living and income relative to that of others similarly situated (Weatherford 1983a). In the aggregate, citizens appear to be sensitive to the possibility that their own 7 We tested one of these sophisticated models found within the approval literature. Williams (1990) uses a vector auto-regressive (VAR) model, where the lagged values of monetary factors, interest rate, unemployment, CPI, federal expenditures and GNP are all included in the approval equation and similar equations are estimated where the dependent variable is each of the economic series. Williams equation (using our data): Approval = β 1 + β 2 (GDP) + β 3 (CPI) + β 4 (Unemployment) + β 5 (bank prime rate) + β 6 (currency), AR(1) model. We tested for redundant variables and differenced all of the variables and re-tested again for redundant variables. The results show that the AR(1) specification explains everything until the political factors are controlled for. By not controlling for autocorrelation, the model is jeopardized and it is untrustworthy. Our model and philosophy are completely different from Williams so it is not surprising that our conclusions are significantly different. Additionally, Ostrom and Smith argue that Williams does not take the possible cointegrating relationships into account and that his results are suspect until someone does that.

7 5 individual economic circumstance might not be generalizable to the nation as a whole, even though an individual unemployed worker might erroneously judge the nation s unemployment rate to be rising or a housewife strapped for grocery money might mistakenly infer that inflation is going up. Most empirical findings imply that blame for economic troubles is based on a concern for national economic problems rather than personal experiences. However, the use of individual-level survey data to test these hypotheses makes it difficult to understand what citizens consider when evaluating the economy. Feldman (1982) asserts that attribution of blame for personal economic circumstances depends upon where responsibility is assigned for political problems. Thus, unless there is a perception of governmental responsibility for economic well being then the public may not attribute blame to the president. While studies that find support for the sociotropic voting hypothesis may be correct, the question of what types of aggregate economic variables drive citizen evaluations remains open. We hypothesize that in the aggregate, the blame that citizens assign to the president will stem from measures of macroeconomic variables that have personal consequences. These include, but are not limited to, commonly used measures of economic health such as inflation and unemployment, variables whose use in past models has shed little light on the reasons why they are significantly associated with presidential approval. We believe that aggregate measures of personal economic conditions, such as the price of clothing and food and other necessities, will be most closely linked to presidential approval, since they are most closely linked to the everyday life of citizens. Modeling approval as a function of aggregate changes in the economy is sensible, but we must consider aggregate measures that are presumed to be most closely linked to individual attitudes and beliefs. In this study, we approach the question of whether personal economic variables matter from a different perspective. Generally speaking, we define personal economic variables to be those which individuals have the most exposure to, those that are the most salient and are often those which affect individual pocketbooks. The idea that people act on the basis of changes in their personal well-being has been a widespread conclusion since Downs s seminal work (1957). Tufte s (1978) evidence also suggests that not only do people act on the basis of personal conditions but that politicians also believe that people vote their pocketbooks. Presumably, the idea that people vote their pocketbooks can be extended to approval, such that individuals who approve of the president do so with their pocketbooks in mind. From a theoretical perspective,

8 6 this argument is substantiated by the idea that individual evaluations of the performance of the government/ president should depend upon the economic context in which they live (Palmer & Whitten 1999). Individuals should condition their opinion of the president s job performance primarily as a result of personal economic variables that may specify individual memory patterns through which the economy influences presidential approval. These memory patterns are thought to be influenced by experience and knowledge. People s knowledge of how they are being affected by economic conditions is acquired on the basis of daily experiences: buying food, clothing, and other goods; keeping track of budgets and savings; shopping for a new house, and so forth (Feldman 1984, also see Popkin et al., 1976). We argue that these individual indicators will have a greater effect on an individual s pocketbook immediately than will broadly defined economic indicators such as unemployment and the inflation rate, which represent much more extensive and less salient concepts that are harder to define and recognize. There is some evidence within the literature supporting our initial hypothesis that personal economic indicators influences public opinion 8. Changes in prices, wage rates and hours worked, which are all components of short-run changes in real income, influence voting behavior (and possibly public opinions). For example, a brief increase in electronic prices may not be as noticeable as long-run changes in food costs, which increase the cost of living and the flow of money out of Americans wallets each day. While the public may be responsive to changes in inflation, Conover, Feldman and Knight (1986) argue that the public is more responsive to changes in unemployment than they are to changes in inflation. This is simply due to the salience of the indicator in the media and among family and friends (Hibbs 1979). Although unemployment is understood by most Americans to be something that is bad, the measure is imperfect because it includes people who have just left the job force and those who have been out of a job for longer periods of time. While it may be true that inflation is less salient than unemployment 9, inflation has a far greater meaning than is often conceptualized by most people. According to Nordhaus (1975), inflation can have a domino effect because inflation may lead to balance of payments problems, inefficient resource allocation, and an arbitrary redistribution of income. Individuals in society do not typically notice these effects but the 8 See Campbell et. al (1960) suggests that there are political consequences to economic changes that effect personal circumstances, the result is that there is increased criticism of the administration.

9 7 outcomes of these effects can have a direct effect on individual circumstances. As a result, the component factors that go into the determination of inflation may paint a better picture of national conditions than the aggregated measure. This is hypothesized to be the case because the public is most likely to respond to their own personal economic situations. The sorts of economic conditions and events the individual knows and understands best are those closest to home. There is no question that citizens have realistic, firmly grounded impressions of their own personal-financial condition, and surely we are justified in having faith in their ability to sacrifice in maintaining or improving it (Weatherford, 1983a). Conover, Feldman and Knight (1986) provide some evidence for our theoretical argument by stating that inflation is an ambiguous measure. Most people do not understand the nature of inflation (Peretz 1983). According to Keech (1995), inflation is a general increase in the money prices of goods and services and a decline in the purchasing power of currency (114). The consumer price index (CPI) is the most common measure of inflation. CPI measures the retail price of a fixed basket of goods and services typically purchased by households. CPI is an explicit price index in the sense that it directly measures the movements in the weighted average of the prices of the goods and services in the market basket through time (Froyen 1999, 25). Per Capita CPI represents another major index of consumer prices. Slightly different goods go into this measure which weights the current cost relative to the costs in the reference period The problem with inflation is that it is often misrepresented because it is often thought to mean higher prices rather than increasing prices. This misunderstanding of inflation makes the inclusion of this variable within a model of explaining presidential approval somewhat problematic because most people are making assessments of inflation based upon experience with items that are not reflected in the government s package of indicators for inflation. The inherent misunderstanding surrounding inflation is troublesome because many people do not see the link between increasing real incomes and increasing prices. Increased wages are symbols of hard work not hard times for most people but the reality is that these two indicators are tied together. 9 Unemployment is included within the model because it is an important indicator that is most often used to measure presidential economic performance. The unemployment rate represents the percentage of workers in the labor force (those that are working and seeking work) who are unemployed.

10 8 By unpacking the traditional CPI measure and using its components food, education, medical, housing, etc. as predictors of presidential popularity, our study provides a test of this hypothesis. For example, one hypothesis would suggest that increased housing costs will affect individuals more than national unemployment because most individuals would feel the effects of these changes within their own pocketbooks, whereas aggregate unemployment changes may or may not affect individual levels of consumption and expenditure. Housing starts 10 are also included as a personal economic variable that are presumed have an impact on presidential approval. This variable is a measure of consumer confidence, which should provide important cues about the economy. The number of housing starts may be an indication that the economy is good and that the general public is relatively satisfied with the current conditions such that the number of loans or proposals to build homes increases. Housing starts, as a measure of consumer confidence, should be related to presidential approval. As housing starts increase, approval for the president should also increase. The Fiscal/Monetary Link Real Dollar Value of Approval While expanding upon the work of other scholars, our paper provides a new assessment of presidential approval from a macroeconomic perspective. Unlike Feldman s (1984) work, which measures the change in financial well-being through a survey question, we operationalize the fiscal/monetary link through financial variables. We argue that variables measuring fiscal or monetary change may trigger aggregate changes in presidential approval because of the impact that they have on American s pocketbooks. As stated previously, individuals should be influenced by variables that capture their individual spending and financial situations 11. Variations in fiscal policy, measured through the bank prime rate, will affect consumer spending and thus will affect consumer opinions of the president s job performance. Bank prime rates are the percent of interest charged by major banks to their largest customers. It also can be seen as a 10 This indicator reports the number of housing units on which construction has started. This indicator implies consumer confidence that the economy will be strong, that money will be available for construction financing and housing purchases. 11 Monetary variables, such as currency in circulation, may contribute to economic change. For example, if people are holding onto more money, meaning here that the flow of currency is decreasing within the economic system, we might assume that individuals are either worried about the economic situation (if they are thinking about the future) or they could be responding to current conditions. Money supply, overall, is an important indicator for the economy because changes in this supply influence interest rates, which can in turn stimulate housing purchases and investments in capital goods..models that included a measure of currency and replaced Per Capita CPI with GDP had no more explanatory power than models that excluded this variable. Thus, we excluded the currency measure.

11 9 measure of perceived risk in the economy. Interest rates are presumed to be influential because when interest rates are lower, people are more likely to spend money, which in turn stimulates the economy. Although general economic well-being is considered to be the determining factor for predicting whether or not expenditures will occur, interest rates are also a measure of economic health. Bank prime rates should be correlated with presidential approval. As prime rates increase, presidential approval should decrease due to the burden that increasing interest rates have on the population at large. General Measures of Economic Well-Being We also consider variables that are often characterized as general/ typical indicators of economic health or well-being. In this category, we include Gross Domestic Product (GDP) as a measure of the nation s economic situation, unemployment as a measure of economic health, and growth of the stock market as measured through the Dow Jones Industrials 30 industries average. The Dow Jones Industrials Average is the most reported stock index whose values have been compiled daily since the late nineteenth century. The stock market is an important indicator for economic health. ). Volatility within the stock market, while influencing personal economic circumstances, also plays a role in the stability and health of the national economy. The public, through the stock market, registers its optimism or pessimism about the future performance of companies through the quantities of stock it supplies and demands (Carroll 1995, 120). Additionally, the stock index measures expected profitability since markets price stocks on profit potential rather than solely on past earnings (Rogers 1994, 248). We include an independent variable capturing the stock prices because the stock market tends to discount in advance the ups and downs in the economy (Lewis 1962, 49). Stock market values tend to be salient within the media and the general public. As a result, it seems probable that that significance of the indicator will influence presidential approval because as stock prices increase, this is symbolic that the economy is improving and is expanding, which will have a positive effect on approval for the president. It is presumed that the president may be rewarded for positive changes within the economy and the stock market price increases are presumed to have positive effect on the economy. GDP is included within this model because it is a measure of all final goods and services in the economy. It is important to consider because approximately 70% of GDP represents

12 10 consumption and consumer expenditures (Froyen 1999). GDP is the measure of all currently produced final goods and services evaluated at market prices (Froyen 1999, 14). GDP accounts for all goods and services produces in the United States in a given year and is probably the most reported economic statistic (Carroll 1995, 39). GDP is the best-known standard of living indicator and to truly understand presidential economic performance GDP really needs to be considered. GDP may be seen by some as the bottom line as far as economic performance is concerned, the components of GDP and other driving economic indicators (e.g., employment rate, productivity) that are the building blocks of economic growth and more closely measure the impact of economic policies (Carroll 39). GDP has two components, consumption and investment. For our purposes, consumption is the most important because it describes the component of GDP that is explained by the household sector s purchases of goods and services (that are currently produced). Consumption is really durable and non-durable goods and consumer services and this accounts for approximately 70% of GDP. Therefore, the fact that GDP has a direct link to the consumer makes this variable a likely candidate for inclusion in our model (however though we must consider the fact that GDP is highly sensitive to changes in the average price level and thus probably correlated with CPI). Finally, GDP is a standard measure used to assess the economy and consequently it may have a direct influence on approval of the president. We expect that increases in GDP to be positively associated with the aggregate support for the president because an increase in GDP symbolizes that the economy is growing, which should have a positive impact on support for the president. Other studies have used the misery index, M, as a measure of extremely harsh economic times: M = U + π U = unemployment and π = inflation. However, according to Keech (1995), the fact that the misery index is an weighted sum of the rates of inflation and unemployment [it] is an arbitrary simplification. It seems that Democrats should value unemployment differently than Republicans such that 1 additional point of unemployment may be twice as painful as one point of inflation. In order to capture this relationship, we could weight the variables in the misery index so that Democrats are more susceptible to unemployment and Republicans are more susceptible to

13 11 inflation. One potential bias in doing so is that we do not know the extent of the bias and by arbitrarily placing a weight on this value, we may be misrepresenting the true but unknown underlying relationship between the economic variables. Thus, we test the hypothesis that Democratic and Republican presidents are affected differently by inflation and unemployment by including two interaction terms: one each to capture the effects of inflation and unemployment during a Democratic administration. The complete effect of a one-point increase in inflation and/or unemployment for a Democrat is thus the sum of the coefficient of the interaction term and the coefficient of the relevant indicator. We expect that inflation and unemployment will both be negative and significant in the equation, but that the interaction term for unemployment under Democrats will be negative, while the interaction term for inflation under Republicans will be positive. Finally, to capture the interactive effects of truly miserable economic circumstances, when inflation and unemployment are both high, we include a third interaction term, misery, which is simply the product of inflation and unemployment, which we also expect to be negative. Thus, the full impact of inflation (for instance) under Democrats will be the sum of the coefficient for inflation, plus the interaction term, plus the product of the misery coefficient and the current unemployment rate. Exogenous Links Finally, we also consider salient non-economic variables such as rallies and partisanship that may help explain and predict the dynamics of postwar presidential approval. Economic conditions are not the sole determinant of presidential approval variations. However, people deal with economics every day, and the economy has always been a concern in national politics, while events and issues, those that bring good news and bad news, may be ephemeral or depend on particular actions or conditions by public officials. Consequently, the economy may determine the baseline for explaining popular evaluation of presidential performance and certain election outcomes (Shapiro and Conforto 1980, 64). While the economy provides baseline approval or disapproval of the president, other factors such as rallies also play an important role in influencing approval. John Mueller (1970) defined a rally point as an incident that is international; involves the United States, and particularly the President, directly; and is specific, dramatic, and sharply focused. We use an expanded definition of rally events, which also includes significant domestic events as well as presidential honeymoons. Rallies may influence the approval rating of the

14 12 president through shocks, which reverberate through the political system and temporarily affect the approval dynamics. Like most shocks, rapid approval increases typically correspond to adjoining decreases in approval as a sense of normalcy is restored in the political system. We control for rally effects with a dummy variable, RALLY 12, where 1 = rally occurred during the month and 0 = no rally occurred. We use this dummy variable to construct a variable that represents the impact of the rally on presidential approval by estimating the number of months since a rally has occurred. The rally variable is important for our analysis because while an increase in unemployment or an increase in prices may cause a president s approval ratings to decline, these fluctuations may be related to other factors such as international crises that also may influence presidential approval. As the number of months between a rally event increases, we expect the impact of a new rally event to be greater than say if a rally occurred in January and then another occurred in March of the same year. Partisanship is thought to be an important determinant of policy outcomes. Scholars such as Alesina and Hibbs have argued that partisanship is the most fundamental basis for political influence over macroeconomic policy and outcomes (Keech 66). Hence, partisan strategies are often contingent on economic conditions. Because partisanship may influence the type of policies passed during a president s administration, it may be necessary to include a measure of partisan ties in the House and Senate that could help us explain and predict why and when policies that could cause long run shifts in the natural rate of unemployment or the natural rate of consumption will be proposed. Divided government should affect policy outcomes because of the competitiveness that exists between the two political parties, which may lead incumbents to seek policies with short-run advantages at the cost of long-run problems. Additionally, partisan differences stem from the dissimilar outlooks that each has toward the economy social class, prosperity, distributional issues, etc. Instead of simply measuring the president s party, we include a dummy variable for each presidential administration except the Kennedy administration (the reference category). We hypothesize (as do others such as Golden & Poterba) that popularity is a function of the 12 The RALLY variable data from was obtained by coding rally events from the Gallup Opinion Index chronological list of dates. According to Gallup this chronology is provided to enable the reader to relate poll results to specific events, that may have influenced public opinion (See any addition of the Gallup Opinion Index). Rally events prior to November 1971 were coded from various other sources and events after January 1993 events coded using Gallup s list of chronological events influencing public opinion during the year. The decision as to

15 13 individual characteristics of each president, the non- economic events that occur during the presidency as well as the economic events. Beyond these individual-specific administration effects, the partisanship of the president may also influence approval through the interaction between economic variables such as unemployment and inflation. This is hypothesized to be true because of the partisan theory of economics, which suggests that Democrats/Liberals are more susceptible to unemployment than are Republicans who are more susceptible to inflation. However, the tradeoff between unemployment and inflation is difficult to explain. On one hand, the costs of unemployment are so great in terms of loss of output and lower incomes, but on the other hand, inflation costs are just as troubling in terms of the effect that inflation has on the economic system. Models of presidential support should include both economic and political factors in order to avoid misspecification (Frey and Schneider 1978). According to Kenski (1977), those approval models that fail to include noneconomic variables may artificially inflate the salience of economic factors. However, it may also be the case that these non-economic variables act as suppressors attenuating the effect of economic conditions on presidential approval (Shapiro and Conforto 1980). The Dynamics of Presidential Approval: Modeling Issues A. Unit roots, Cointegration, and Error-correction models There has been significant evidence that the economic conditions contribute to government popularity and electoral choice (see Arcelus and Meltzer 1975; Bloom and Price 1975; Goodman and Kramer 1975; Kenski 1977abc; Frey and Schneider 1978; Kernell 1978; Hibbing and Alford 1981). In their survey of the time-series models of presidential approval, Ostrom and Smith (1994, Table 1, ) show that the most of these studies have relied upon standard assumptions about the lasting effects of economic change, political events, and unmeasured shocks to presidential popularity. As in these studies, the dependent variable in our analysis is which rally events were to be included is rather arbitrary, which runs the risk of biasing the estimates for the economic variables.

16 14 presidential approval 13 : the percentage of those surveyed who approve of the job that the president is doing averaged within monthly responses to the Gallup poll question, Do you approve of the way in which Mr. is handling his job as president?. Ostrom and Smith s survey indicates the substantive conclusions drawn by these studies depend critically on the independent variables included in the model. Earlier studies closely follow Mueller s (1970) specification, relying on a now-canonical list of independent variables: unemployment, inflation, an indicator for the cumulative effects of war, indicators for rally events, and a variable that measures the deterministic trend that Mueller observed for every president except Eisenhower: as time goes by, controlling for all other factors, the president experiences a secular decline in his popularity. Mueller argues that this trend reflects the cumulative effects of the coalition of minorities that forms during every president s term: the unmeasured reasons why an ever-growing segment of the population disapproves of the president s job performance. Mueller s treatment of these unmeasured sources of discontent, along with various other unmeasured determinants of presidential approval, has driven many of the recent approval models (Hibbs 1974; MacKuen, Erikson and Stimson 1992; Beck 1991; Clarke and Stewart 1994; Ostrom and Smith 1994; etc.). Several of these models hypothesize, and confirm, that changes in economic indicators like inflation and unemployment have both immediate and lagged effects on the president s popularity. Modeling these lagged effects of these variables is straightforward if the effects of economic change are assumed to wear out after a specific number of months. In contrast, Monroe (1981) argues that these changes can have a persistent impact on approval well into the future, with the effect of a change eventually approaching insignificance many months later. Her (Almon) distributed-lag model relies on the assumption that every variable has effects that can be immediate and/or long-lasting, theoretically taking a very long time to diminish. The most sophisticated approval models (Ostrom and Smith 1994; Clarke and Stewart 1994) pay particular attention to the long-term effects of the unmeasured variables in the statistical model, which are contained in the error term. Drawing on econometric time-series 13 The presidential approval data was obtained from the Edwards and Gallup (1992), Roper Center Presidential Archive at and from Gallup Polls published in the Gallup Monthly Index. Approval data points that were missing (all except those from the first month of a presidents term) were linearly

17 15 studies of macroeconomic change, they begin by testing to see whether the dependent variable is stationary, or whether it is integrated (whether it has a unit root, in other words). 14 The distinction is not innocuous: in a time series that has a unit root, the effects of a shock never diminish, causing unpredictable aggregate behavior as the cumulative effects of past shocks keep reverberating forever at full strength. If true, the presence of a unit root would seriously hamper inference about the effects of changes in the individual economic variables measured in the model. In the long run, the variance of a variable with a unit root will explode, theoretically becoming infinitely large. Thus, it is imperative that time-series models of presidential approval accurately determine the degree to which shocks to the dependent variable persist over time. If approval has a unit root, inferences drawn from t-t-statistics and F-tests are potentially spurious, especially when the independent variables appear to share a common trend (Maddala and Kim 1998, 28-29). Even if approval is stationary, modeling its dynamic properties is crucial, since a failure to account for autocorrelation in the disturbances biases t-statistics and goodness-of-fit measures by inflating them. However, the presence of economic indicators, many of which have been demonstrated to have unit roots, may actually alleviate the problems caused by the unit root in the approval time series, if one actually exists. If two variables are cointegrated, they each have a unit root, but the relationship between them is not spurious: they travel together over time in an established equilibrium relationship, and an OLS regression will give valid estimates of the causal impact of one variable on the other. The key to estimating a relationship between two cointegrated variables is to account for the reequilibration mechanism that keeps the two variables moving together through time. This is the motivation behind the error correction model (ECM) estimated by several scholars (Beck 1991; Ostrom and Smith 1994; Clarke and Stewart 1994; Durr 1994; Zorn and Caldeira 2000). The intuition behind the ECM should be appealing to those who study the links between citizen attitudes, governmental actions, and economic changes. In an ECM, there is a interpolated to fill in the gaps. The missing data points from the first month of a president s term were acquired through backward extrapolation of the data. 14 The term unit root refers to the solution to the dynamic equation for a variable like y t, where y t = ρy t-1 + ε t, some random disturbance. In the event that ρ=1, the value of the dependent variable y at time t is a function of the last realized value (y t-1 ) plus some normally distributed disturbance with mean zero (ε t ). A random shock ε t felt at time t will never dissipate over time, whereas in a stationary series, when ρ < 1, the time series is mean-reverting, since the effects of ε t eventually wear off.

18 16 mechanism that determines the correct level of approval for a given set of economic circumstances: Implicitly, we are assuming that for every mix of observed outcomes, there is a corresponding level of approval [ ]. The level of approval associated with a given distribution of environmental outcomes reflects rewards and punishments levied on the president by the public in response to deviations of observed from institutionally expected outcomes. (Ostrom and Smith 1994, 131) The question ECM s can answer is how quickly the public reacts to these deviations by establishing a new set of expectations and rewarding or punishing the president as the situation warrants. It should be noted that theory can guide our expectations about why such a long-term equilibrium pattern might exist, but that cointegration is purely a statistical concept, and cointegrating relationships between variables in a statistical model can be shown to exist whether or not they have any theoretical interpretation. The most common ECMs that have been estimated start from the assumption that the dependent variable in question is presidential approval, and that economic variables may or may not be in a long-term equilibrium relationship that changes over time, but that there is no feedback so that, say, the economy responds to unexpected changes in presidential approval by increasing or decreasing unemployment (Beck 1994). 15 B. Is there a unit root in presidential approval? As appealing as the ECM is for theoretical reasons, there is no need to estimate it if the approval time series is stationary that is, if it does not have a unit root. The most common tests for unit roots the augmented Dickey-Fuller (ADF) test and the Phillips-Perron (PP) test share a common flaw: that the null hypothesis is that the series has a unit root. Critical values for both these tests must be specially generated since the test statistics do not have standard distributions (Maddala and Kim, Chapter 3) but nevertheless, the power of each test can be limited by small sample sizes, and the statistical burden is to demonstrate that the time series is stationary that 15 Beck argues that such a feedback mechanism is theoretically implausible, but Williams (1990) finds support for the idea of a political business cycle by demonstrating that changes in approval tend to predict changes in macroeconomic policy. Testing for several cointegrating relationships among, for instance, various interest rates makes sense if the analyst has no reason to suspect that one rate is certainly the cause of reequilibration in another rate. However, a true test of a cointegrating relationship between macroeconomic indicators and presidential

19 17 is, that it is relatively well-behaved in the long run. Clarke and Stewart (1994) also estimate an ECM after using the ADF test as a pretest for the existence of a unit root, but this model may be inappropriate if approval is in fact stationary, and the presumption of the test is that a unit root exists. Ostrom and Smith use the ADF test along with another test, the KPSS test, in which the null hypothesis is stationarity, but cannot confidently conclude that, for the Reagan administration, the approval series has a unit root. However, as Beck (1994) points out, the small sample size in Ostrom and Smith s model may be driving this conclusion; considering the longer series starting with the Eisenhower administration, his ADF tests reject the null hypothesis that there is a unit root in the approval time series. He concludes that the use of an ECM when the dependent variable is near-integrated may be appropriate, but that the issue is unresolved (1994, 24). More recent work suggests plausible reasons why the approval time series should be stationary and why tests for a unit root in presidential approval may be biased toward the conclusion that a unit root exists. Alvarez and Katz (2000) present two arguments for why a unit root is not present in presidential approval. One is technical: that since the approval variable, as usually measured, is bounded by 0 and 1 below and above, it is impossible for the variance to explode as an integrated time series must do eventually. They argue that if the time series has a unit root, we should expect to see approval hit either the lower or upper boundary and stay there for an appreciable period of time. The criticism is similar to that posed by John Cochrane (quoted in Maddala and Kim, 127), who wondered: Interest rates are the same now as in Babylonian days. How can there be a unit root in interest rates? One general answer is that the variance of a time series with a unit root will eventually explode, but that its failure to do so in a finite time frame is not in and of itself evidence that the variable is stationary. Another is given by Ostrom and Smith, who argue that the real approval variable is latent, since citizens are constrained by the possible survey responses of yes/no/don t know. Thus, the aggregate series, in which the observed proportion of supporters is measured, may have a unit root if the latent approval variable does, since the measured approval variable is a nonlinear transformation of the approval which has not been attempted, as Ostrom and Smith (1994) point out in their critique of Williams study (1994, ) should probably remain agnostic about the direction of causality, as Williams does.