There are a multitude of factors that dictate how someone votes in an election, and for a large share of individuals, whether they even vote at all. Political Science has a long history of attempting to define which characteristics influence the electorate to vote for a particular party; sometimes for academic purposes, but in the election industry, these defining characteristics are very valuable information.
There are internal factors and external factors that decide an individual’s vote. Internal factors are characteristics inherent to the person that increases their predisposition to vote for a particular political party, e.g. race, sex, or age. External factors are things that occur out in society that influence an individual’s political leanings, e.g. the salience of a social issue at the time of the election, economic growth or shrinkage, or the response of an administration to a national security threat. The effects of these events on the electorate are more difficult to quantify, but that is the aim of this paper, at least for one specific type of external factor: economic growth.
Do voters hold government officials, more specifically executives, accountable to economic performance during their tenure? What about the party of the executive; do voters punish the political party of the executive when economic performance is poor and reward it when it exceeds expectations? At first the answer seems like an obvious one. In fact, if there were any factor that did dictate the electorate’s vote, it would seem that most would say economic performance was it. But, the interesting part of this lies in the specifics, and implications of the question. If it is found that economic performance has little to no bearing on the electorate’s aggregate vote for the political party that is in power at the executive level, then this would indicate that voters do not make their decision in the context of political parties. Perhaps it would mean that they place more emphasis on the personal characteristics of the candidate. Alternatively, it could indicate that the partisan divide is so strong that voters, on the aggregate level, vote on party lines despite how good or bad the perceived economic leadership of the president or governor is during their tenure. And lastly, it could even indicate that campaigns are so effective at pushing a specific narrative that voters forget the reality of economic conditions.
But what results do you get when you examine the relationship between state economic conditions and presidential vote share? Do voters hold the president’s party accountable, and furthermore do they hold the president’s party accountable for state economic conditions that are outside of the national government’s control? In this paper we will discover whether or not a relationship exists, and determine whether or not accountability stays in its appropriate sphere. Building on this as well, we can make some determinations of other factors that affect presidential and gubernatorial success in elections.
Nearly every publication in political science regarding this subject looks at election results through the lens of two different voting behavior models. The first is the national referendum model, which “suggests that voters in subpresidential elections express their approval or disapproval of the sitting president and his policies with their vote.” The other model is the economic voting hypothesis, which “suggests that voters in these elections express support or dissent for the performance of the incumbent based upon how well the economy is doing,” (Atkeson and Partin, 1995). At its core, this last model indicates that voters who are financially better off than they were before the candidate took office will reward the incumbent, and conversely will punish them if that does not hold true. The economic voting hypothesis is what we are primarily taking a look at here, with a couple of variables that would shed some meaningful light on the applicability of the national referendum model in the gubernatorial analysis.
Institutions, The Economy, and the Dynamics of State Elections takes a deep look into the aforementioned topic at the state legislative and gubernatorial levels (Chubb, 1984). Chubb considers the relative effects of presidential coattails, the common backlash against the party of the president during mid-term elections, and state and national economic conditions. He believes that “. . . when it comes to assigning responsibility for economic performance, state voters have generally and increasingly looked outside of the state to the national economy and the president’s imputed performance in managing it,” thus a poorly performing national economy would influence the outcome of a state election, despite a state’s hypothetical independently strong economic performance. The analysis of this paper has shown Chubb’s claim to be inconsistent with voting behavior in the last two decades, but perhaps there has been a genuine change in behavior between the time period he analyzed and now. Lastly, Chubb says that at the state level, “the factors that account for variations in normal partisan voting across the states include idiosyncrasies of culture and history that subvert general explanations,” which corroborates the findings of the analysis here. Gubernatorial electoral outcomes are the product of far more variables than presidential electoral outcomes.
James King, of the University of Wyoming, in 2001 performed a similar analysis in his publication titled Incumbent Popularity and Vote Choice in Gubernatorial Elections. Studying gubernatorial elections in four states, his results found that “. . . voters use the ballot for governor to express approval or disapproval of current economic conditions and the president’s job performance, or as an easy means of evaluating candidates.” This would square with Chubb’s publication that retrospective economic voting is a reality, and that voters hold the governor accountable to state economic performance.
Atkeson and Partin, in Economic and Referendum Voting: A Comparison of Gubernatorial and Senatorial Elections come to similar conclusions. They find that “. . . governors, as chief executives of their respective states, are held responsible for the health of their state economies and are not generally shown to be liable for fluctuations in presidential approval.” This further affirms the existence of economic retrospective voting at the gubernatorial level. Atkeson and Partin also “. . . find something of an in-party incumbency effect whereby incumbent governors of the president’s party suffer more from a perceived worsening of state economic conditions than incumbents of the out party,” and finally they claim, “. . . governors . . . escape from these national-level evaluations of presidential performance and are instead held liable for state economic conditions.” This would suggest that national economic performance has no bearing on governors, regardless of political affiliation.
In a very interesting article written by Robert M. Stein in 1990, the argument is made that economic conditions affect presidential vote share, but not gubernatorial, saying “. . . state and local incumbents are less likely to be affected by voters’ retrospective economic evaluations than their national counterparts.” Stein’s usage of the word “incumbents” would indicate that voters do not punish or reward the incumbent party when an incumbent is term limited and the election is open-seat. Stein’s findings do confirm his aforementioned hypothesis, and he goes on to state that “Voters hold their governor neither responsible nor accountable for the state’s economic conditions and their voting behavior reflects this perception . . . This evidence of economic voting for governor, however, varies with the partisan affiliation of the incumbent candidate,” thus when economic effects do matter in an election, voters will punish or reward each party by different amounts for the same economic performance. Stein’s research shows that voters differentiate between the impact state and federal policies can have on their personal finances, wrapping up his article by saying that “voters recognize that their personal financial condition is more closely tied to federal policies and actions than to the state’s,” (Stein, 1990).
A different approach to testing the executive economic accountability theory is taken by David Samuels and Timothy Hellwig. They cited the controversy of trying to define the dependent variable in the accountability model; whether to measure accountability as number of seats in a legislature swings in vote share of an incumbent, or in its most stringent sense, partisan control of the office in question. They found that accountability can be observed when measuring it as seats in a legislature or vote share of an incumbent, saying “When we conceive accountability in terms of sensitivity of swings in incumbent vote shares . . . and when we use seats as the dependent variable, we find that incumbent performance is sensitive to economic performance.” However, when measuring accountability as partisan control of the office, the results were inconclusive, saying “. . . when we conceive of accountability as partisan control over government . . . our findings temper Cheibub and Przeworski’s (1999) pessimism.” The authors ultimate conclude that “citizen control of politicians is . . . imperfect because particular political contexts limit voters’ ability to hold incumbents to account by obscuring responsibility for economic performance,” (Samuels & Timothy, 2010).
Johnson and Ryu look to other countries to test these models, but with the inclusion of broken campaign promises in the analysis. They sought to determine if economic performance, broken promises, or some combination of the two were what voters cared most about in executive elections. Their findings indicated that neither of the two factors had any substantive effect alone, but an interaction between both variables resulted in statistically significant results, going on to say that “the relationship between broken campaign promises and incumbent vote change is affected by economic conditions.” This means that regardless of economic performance, as long as no campaign promise was broken, the executive was not rewarded nor were they punished for it; and alternatively, when a campaign promise was broken, voters held the executive accountable for economic performance, likely due to increased scrutiny (Johnson & Ryu, 2007).
Data was gathered from official and private sources (deferring to official whenever possible), from the years 1987 to 2013.
The hypothesis for the state level is as follows:
State Level H1: People vote to retain the governor, or party of the governor when he/she is term limited, when the state in question experiences strong economic growth.
State Level H2: On average, a governor will be positively affected when the governor is the same political party as the sitting president and their elections coincide.
The overall theorized model of the analysis is specified by the following regression equation:
%VoteShareGovParty = β0 + β1 % StateGrowth + β2 % NationalGrowthSamePres – β3 %NationalGrowthOppPres + β4 PartisanControl + β5 GovIsIncumbent + β6 GovSamePartyElectionasPres + ε
The following are variable definitions:
%VoteShareGovParty: The percentage of vote share received by the party of the incumbent governor, whether the governor was running for reelection or not.
%StateGrowth: Measured as the percentage change in Per Capita Real Gross State Product over the tenure of the governor.
%NationalGrowthSamePres: An interaction between a dummy variable that turns on when a governor is the same party as the president, and the percentage change in Real Gross Domestic Product
%NationalGrowthOppPres: An interaction between a dummy variable that turns on when a governor is the opposite party as the president, and the percentage change in Real Gross Domestic Product
PartisanControl: Serves as a baseline for the predicted vote share of a governor. This is measured by taking the average of the vote share of the Democrat Party for every gubernatorial election between 1987 and 2013, and providing that number if the incumbent party is Democrat, or 100 minus that number if the Incumbent party is Republican. The coefficient of this variable will be difficult to interpret, and is not meaningful for our sake.
GovIsIncumbent: This is a dummy variable that turns on if the governor is an incumbent.
GovSamePartyElectionasPres: This is a dummy variable that turns on if the governor is the same party as the president, and their elections fall on the same year.
Gubernatorial election results and information on gubernatorial incumbency were aggregated from Wikipedia, which has curated lists that source the data from each state’s secretary of state or equivalent. I am aware that Wikipedia is frowned on to use as a source in an academic context, but there was simply no other source that had the data in a remotely useful format. National RDGP growth, as well as state per capita RGDP growth figures were obtained from the Bureau of Economic Analysis. The rest of the variables were computed through Excel functions to prepare the dataset. Observations were discarded if the incumbent party was not Democrat or Republican, and after the completion of the dataset, there were 327 total observations being analyzed. For the regression analysis, Stata 14 was used.
The national level model is much the same, with two hypotheses as well:
National H3: People vote to retain the president, or party of the president when he is term limited, when the country as a whole experiences strong economic growth.
National H4: People vote to retain the president, or party of the president when he is term limited, when their state experiences strong economic growth.
The overall theorized model of the analysis is specified by the following regression equation:
%VoteSharePresParty = β0 + β1 % StateGrowth + β2 % NationalGrowth + β3 PartisanControl + β4 PresIsIncumbent + ε
The following are variable definitions:
%VoteSharePresParty: The percentage of vote share received by the party of the incumbent president, whether the president was running for reelection or not.
%StateGrowth: Measured as the percentage change in Per Capita Real Gross State Product over the previous presidential term.
%NationalGrowth: Measured as the percentage change in Real Gross Domestic Product
PartisanControl: Serves as a baseline for the predicted vote share of a president. This is measured by taking the average of the vote share of the Democrat Party for every presidential election between 1987 and 2013 and for every state, and providing that number if the incumbent party is Democrat, or 100 minus that number if the Incumbent party is Republican. Like the gubernatorial model, the coefficient of this variable will be difficult to interpret, and is not meaningful for our sake.
PresIsIncumbent: This is a dummy variable that turns on if the president is an incumbent.
Presidential election results and incumbency information were aggregated from The American Presidency Project. National RDGP growth, as well as state per capita RGDP growth figures were obtained from the Bureau of Economic Analysis. The partisan control variable was computed through excel functions. Observations were adjusted if there was a third party candidate to exclude that impact on the results.
After running the regression analysis on the data and ensuring that there were no omitted variables, we are presented with some very interesting results. Three different models were created for the gubernatorial level, with each subsequent model adding controls in an attempt to isolate the real effects that state GDP growth and national GDP growth have on the vote share for the political party that is in power at the executive level in state government. Though no heteroscedastic problems were foreseen, every model was controlled for heteroscedasticity by using the robust command in Stata. Refer to Table 1 for the results of all three models.
After producing a correlation matrix, we find no possible issues with multicollinearity in the model. Notice the low correlation between the PartisanControl variable and %VoteShareGovParty. The PartisanControl variable takes historical voting averages for the Democrat Party in each state and uses that number as a baseline to predict vote share in the year in question. However, because aggregate state partisanship is much more volatile than the country as whole, and is susceptible to much quicker political ideology shift, historical voting averages are not nearly as good of a baseline at the state level as they are for the national level. In an ideal world, this correlation number would be equal to 1.000, which would allow us to perfectly predict the outcome of each election with only one piece of data, but unfortunately this is not the case. As you will soon see, the presidential model correlation between these two variables was much higher, leading to a much more accurate model.
Neither state nor national economic performance has any effect on gubernatorial vote, and we cannot accept H1. The closest either variable gets to any significance is with a p-value of 0.102, but even this is in Model 1 when the other controls are not active. This is a surprising result, and though the p-values for the state, same-party-national and different-party-national variables are 0.335, 0.995 and 0.736 respectively in the complete model, the R2 is 0.429. This means that there are other variables out there that will account for the other 57.1% of the variance in vote share for the political party in power. If those were controlled for, there is a chance that state and national economic performance could come back down into significance. However, the challenge with this is that there is no shortage of unobservable variables that have some bearing on the outcomes of elections.
To those familiar with the field of Political Science, it should come as no surprise that incumbents enjoy a tremendous advantage over their opponents. In this analysis, it was found that, when controlling for the other previously mentioned variables, simply being an incumbent gives a gubernatorial candidate a 10.646% (though we must consider the constant forces us to start below zero, at -3.343, so it’s not exactly 10.646 but rather a touch below) vote share boost; with a p-value of 0.000. This is also an underestimate, due to some states only requiring a plurality of the vote to win the governorship; e.g. some gubernatorial results in this data show a governor winning with as little as approximately 40% of the vote. These observations could have been removed, but they comprised a significant portion of the data, and incumbency advantage was not the aim of this research.
Presidential coattails were also confirmed by this analysis, for better or for worse. When a governor shared the same political party as the incumbent president and their elections coincided, the governor enjoyed a modest vote share boost; with a p-value of 0.003, allowing us to accept our H2. This could be attributed to the president running a hard fought campaign during his election year, in an attempt to increase his approval rating before Election Day; and when the incumbent president could not run for reelection, perhaps those conditions did not necessarily hurt a same-party governor either. Presidential campaign efforts, on average, apparently have a positive effect on not only his party members at the federal level but governors as well. This may no longer be the case today, but on average from 1987 to 2013, it was.
These results call into question the rationality of the votes cast in gubernatorial elections, and force us to ask ourselves, if these results are an accurate representation of reality, why voters do not hold governors and their political party accountable to state economic performance? Perhaps this is a result of state economic conditions being to some extent a product of the national economy, and also a result of state economic conditions being vaguely defined in the context of the bigger national picture. Rarely are voters exposed to economic performance data, and when they are it is highly unusual to be state data. However, even this does not square with voters not being mindful of national economic performance either. There is a chance that voters view the governor as isolated from any decisions that would affect the state economy and insulated from decisions that affect the national economy, and because of this, state and national economic conditions do not hold any bearing on their vote in gubernatorial elections.
The presidential model produces far more predictable results. Three different models were created for the presidential level as well, with each subsequent model adding controls in an attempt to isolate the real effects that state GDP growth and national GDP growth have on the vote share for the political party that is in power at the executive level in the federal government. State economic growth is not significant in the complete model, however, and has a p-value of 0.918; preventing us from accepting H4 that hypothesizes that state growth has a positive impact on presidential vote share. Unlike the gubernatorial results that showed that state economic performance had no effect on general election outcomes, the president’s party is indeed accountable to national economic performance, with a p-value of 0.000; allowing us to accept our H3. Like the gubernatorial model, incumbency matters. Simply being an incumbent resulted in a significant vote share boost for the incumbent president.
After producing a correlation matrix, we find no possible issues with multicollinearity in the model. The PartisanControl variable correlates highly with %VoteSharePresParty, 0.8909, but this is to be expected. In these models we are using voting averages for the Democrat Party in years past to estimate current Democrat vote share. What this means is that we seek a high correlation with the PartisanControl variable, and if it were not correlated highly this would mean that partisan trends and a state’s propensity to vote more for a specific political party were non-existent. We know this is false.
Perhaps most interesting is the performance of the included quadratic national GDP growth variable. The inclusion of this variable is only due to the very high level of significance it attained, and it admittedly makes little theoretical sense within the context of the model. Because OLS has estimated a negative coefficient, this indicates that GDP growth has a diminishing effect with respect to vote share of the incumbent party of the president, to the extent that at a certain point, higher growth actually reduces it. Theoretically, voters are eager to kick out the party in leadership when there is very low GDP growth. Voters are eager to retain the party in leadership when there is a satisfactory amount of growth. And lastly, perhaps voters are enjoying an economic prosperity so much with very high GDP growth that they don’t care to turnout for an election and keep their favored party in power.
The performance of the presidential model squares with the economic voting hypothesis. Voters notice economic growth (or the lack thereof) and attribute this to the performance of the president. Though the president has no constitutional basis to manage the national economy, voters have given this responsibility to him.
The gubernatorial model does not square with the economic voting hypothesis. Governors are not beholden to any economic performance. However, the model does square with the national referendum hypothesis, as evident in the statistical significance of the “Governor Same Party as President” variable. Voters express their discontent or satisfaction with the sitting president by voting against or for same-party governors.
Though these results indicate that economic performance holds little significance in the context of elections in modern times, this may not have always been the case. There is a stark juxtaposition between the strong anti-federalist sentiment of the 19th century, and today’s interconnected and unified country. Voters 150 years ago placed much more emphasis on state-level government and politics, and this is just simply not the case anymore. There is a (nearly) century long trend of giving more and more state power to the national government, and I believe that in this, the significance of state level economic performance has been largely lost. However, this is not to say that this trend will continue, or that the results of this analysis will even hold true in the future. Perhaps we are yet to witness the imminent state politics renaissance, but for now, economic performance doesn’t matter for governors.
Hope you all enjoyed,
Atkeson, L. R., & Partin, R. W. (1995). Economic and Referendum Voting: A Comparison of Gubernatorial and Senatorial Elections . American Political Science Review , 89 (1), 99-107.
Chubb, J. E. (1988). Institutions, the Economy, and the Dynamics of State Elections. American Political Science Review , 82 (1), 133-154.
Johnson, G. B., & Ryu, S.-R. (2007). Campaign Promises, Economic Performance, and Electoral Accountability in Latin America . Annual Meeting of the American Political Science Association, (pp. 1-23). Chicago.
King, J. D. (2001). Incumbent Popularity and Vote Choice in Gubernatorial Elections. The Journal of Politics , 63 (2), 585-597.
Samuels, D., & Timothy, H. (2010). Elections and Accountability for the Economy: A Conceptual and Empirical Reassessment. Journal of Elections, Public Opinion and Parties , 20 (4), 393-419.
Stein, R. M. (1990). Economic Voting for Governor. Journal of Politics , 52 (1), 30-53.
U.S. Department of Commerce. (2015, April). Regional Data. Retrieved April 20, 2015, from Bureau of Economic Analysis: http://www.bea.gov
Wikipedia. (2014, December 2). United States gubernatorial elections, 1990. Retrieved
April 15, 2015, from Wikipedia:http://en.wikipedia.org/wiki/United_States_gubernatorial_elections,_1990
Woolley, J. T., & Peters, G. (2015). Presidential Elections Data. Retrieved April 1, 2015, from The American Presidency Project:http://www.presidency.ucsb.edu/elections