Do increases in petroleum product prices put the incumbent party at risk in US presidential elections?

Abstract

This article investigates the impact of petroleum product prices on recent United States’ presidential elections by modelling the probability of the incumbent presidential party losing a state (under the United States’ electoral college system) it had carried the previous presidential election. The main finding is that the probability of the incumbent party losing a state previously carried increases with petroleum product prices but only in those states that have primarily energy consuming economies. We also find that increases in the number of international conflicts, increases in real state per-capita income growth, and increases in state per capita grants-in-aid all reduce the likelihood of losing previously carried states while higher taxation growth increases this likelihood.

Notes

¹ For example, a recent Newsweek issue printed a story titled ‘Paying at the Pump: Record-high gas prices are putting pressure on politicians to come up with a solution, and making energy a hot topic in this year's presidential election.’ See http://www.msnbc.com/id/4610936/site/newsweek/

² We do not include the 2004 presidential election primarily due to data limitations. As we will see, our analysis is a US state-level analysis and petroleum product price data at the state level was available only through 2002 at the time that this analysis was conducted.

³ For instance, the state of California, by virtue of the fact that it is the most populated state in the country, has 55 electoral votes. Were a presidential candidate to win even a small majority of popular votes in that state, that candidate would ‘win’ California by receiving all 55 electoral votes (there are states that can and often do split their electoral votes, but these are few in number with relatively small populations). This system can generate some peculiar outcomes because the dispersion of electoral votes across states is much smaller than the dispersion of population in the country. For instance, the state of Maine has a population of 1.3 million with 4 electoral votes. California has a population of about 36 million people, over 27 times that of Maine's. Yet, California's electoral count is only about 14 times that of Maine's electoral count. As a consequence, it is possible that a candidate could win a large number of smaller states but lose one or two relatively larger states and thereby gain enough electoral votes to win the election yet still not obtain a majority of popular votes. While rare, this did occur recently in the highly contested 2000 presidential election.

⁴ Our focus on modelling the probability of losing a state previously carried is reasonable and appropriate given the objective of our study. At least three other possible models could have been considered. Perhaps the most obvious would have been to consider modelling the probability of the incumbent winning a state previously lost. This is intriguing but there are two issues that arise here. First, it's not obvious as to why we should expect this to be a better modelling approach nor is it obvious why we might expect results that differ substantially from what is presented in this article. Secondly, from a practical perspective, the number of times such an outcome occurred in our set of presidential elections is less that 10% of the total 357 observations which calls into question the statistical reliability of such a model. That said, perhaps with a few more elections and more such observations, this may be a reasonable avenue for future research. A second alternative may have been to model the probability of retaining a state previously won. This too has a number of difficulties associated with it. There are many states, for instance, that appear to vote for a particular party's candidate, such as Rhode Island for the Democrats and Indiana for the Republicans, irrespective of economic climate. If the desire is to explain such behaviour, which is not the purpose of this article, the political dynamics at play in such states may require a different modelling approach than the one applied here. Third, one could model the percentage of votes cast against the incumbent party in a state, irrespective of whether or not the incumbent wins the state. However, from the candidate's perspective, in the US's presidential voting system, popular votes are not nearly as important as the number of electoral votes since it's the winning of states that ultimately determines the election's winner. Moreover, such data at the state level is not readily available over time. Regardless, it is not clear that such a model would provide substantially different results from the dichotomous modelling structure adopted here. That said, however, it may be a potentially interesting avenue for future research.

⁵ It is important to note at this point that there is no indication that there is always a select few states that comprise these 89 observations. Indeed, a total of 39 states recorded at least one instance of voting against the incumbent party during the period covered and 16 states recorded three or more such instances. This strongly suggests that our results are not being driven by a few select states and considering the entire set of 50 states is appropriate.

⁶ In addition to Chamberlain (1980), Greene (1993, pp. 655–57) provides a detailed discussion of this technique. LIMDEP 7.0 was used to estimate this equation.

⁷ This data can be obtained at http://www.eia.doe.gov/emeu/states/_price_multistate.html. This composite price is constructed by EIA as a weighted average of a number of petroleum products prices. Products include distillate fuel, jet fuel, liquefied petroleum gasses, residual fuel and motor gasoline. The weights are based largely on each fuel's share of total petroleum expenditures, the largest of which is motor gasoline. Hence, there is a very high correlation between the composite petroleum price used here and the price of motor gasoline. Indeed, for 40 of the 50 states, this price correlation is in excess of 98% over the period 1970–1999. The lowest correlation is Alaska's, where the price correlation is 96%.

⁸ As with all log-transformed variables, we multiply by 100 to put this into percentage terms.

⁹ For instance, ln(ΔPETPR _{i , t} ) for election t = 1980 represents the growth in petroleum prices between 1977 and 1980.

¹⁰ For additional details, interested readers are referred to Decker and Wohar (2005).

¹¹ The focus on basic industries comes from theoretical and empirical support for Economic Base Theory. This theory asserts that a regional economy consists of sectors that are exogenous (or ‘basic’ sectors in that they define a regional economy's economic base) and sectors that are endogenous (i.e. ‘derivative’ or ‘nonbasic’) to that economy. Exogenous sectors represent the specialized production of goods and services for a given region whose primary market lies outside regional boundaries, the sales from which inject income into the regional economy. Derivative industries are those that are dependent on the basic industries and whose purpose is to support the local economy. While challenged many times in its rather lengthy existence, Economic Base Theory remains one of the main theories of regional economic analysis for several reasons. First, empirical support for the model remains reasonably strong. Second, it is the central structure employed in almost every regional forecasting model today. And third, it is easy to implement relative to other models (Polzin, 2001). Moreover, the overall performance of a state's economy appears to be driven by the performance of its basic industries. For instance, Polzin (2001) finds, consistent with earlier studies, the primary determinant of a state's nonbasic employment growth is growth in a state's basic industries.

¹² For manufacturing industries, we utilized the US Department of Energy's EIA's 1998 report on fuel consumption ratios by industry for the US as a whole. This ratio, calculated by EIA, measures a given industry's energy usage, in thousands of BTU's, per dollar of total value added generated in that given industry. The fuel consumption ratio for total US manufacturing was 9.4 BTUs per dollar of industrial value added. We classified those industries with ratios in excess of this average as ‘energy intensive.’ These industries are textiles, wood products, article, petroleum and coal products, chemicals, nonmetallic mineral products (stone, clay and glass) and primary metals (primary iron and steel manufacturing). For nonmanufacturing industries, we consulted the BEA's 1998 ‘USE’ table from their input-output matrix for the US economy. This provided us with information on the inputs to each industry's total output. If an industry realized a significant amount of its total value came from oil and gas production and/or petroleum refining, we considered it an energy consuming industry. This analysis revealed that nonmetallic mineral mining, railroad, trucking and other service transportation sectors including local and interurban transportation (buses, taxis, etc.) should be added to our set of energy consuming industries.

¹³ This employment data, as well as data on personal income, tax revenues, and population, comes from the US Bureau of Economic Analysis’ Regional Economic Information Service (REIS) and can be obtained at http://www.bea.gov/bea/regional/data.htm

¹⁴ The data was adjusted for inflation using the US gross domestic price deflator.

¹⁵ State unemployment data comes from the US Bureau of Labour Statistics, www.bls.gov.

¹⁶ This variable was calculated using REIS data.

¹⁷ While our hypothesis suggests that voters in the United States are largely driven by self-interest rather than some broader social welfare calculus, it should be noted that what motivates voters when it comes to such policies is open to debate. See Gibson (1994) for a discussion of this debate.

¹⁸ If true, an implication of this hypothesis would be that incumbents, feeling their future electoral success is at risk, may promote budgets that inject sizable grants into key electoral states. While this is likely, investigating the determinants of variations in grants-in-aid to states is beyond the scope of this study. However, interested readers are referred to Grossman (1996) for research on this issue.

¹⁹ Prior to 1983, this data was reported by the US Treasury Department. Since then, this data has been reported by the US Bureau of the Census. Fortunately, the data is regularly published in issues of the Statistical Abstract of the United States. We compiled this data from various issues of these abstracts.

²⁰ This data comes from Center for the Study of Civil War at the International Peace Research Institute, Oslo (PRIO) and the Department of Peace and Conflict Research, Uppsala University. We did not include a dummy variable indicating if the US was at war since for the elections we are considering, from 1976 through 2000, the US was not substantially engaged in what might be considered a highly active war with substantial US troops committed during the actual election year itself. Indeed the only election in close proximity to a US war was the 1992 election following the first Persian Golf War in 1990–91. In an earlier version of the model, we included a dummy variable for the 1992 election in an effort to see if this would pick up a ‘war’ effect but the variable was statistically insignificant, perhaps due to the fact that the 1992 election was too far removed from the war and concerns over the general economy (which we already control for) dominated.

²¹ These results highlight an interesting issue. Recall that ΔPETPR _{i , t} measures the growth in real petroleum prices over an entire presidential term, suggesting that voters in energy consuming states are tending to evaluate the incumbent party over the entire presidential term and react accordingly. One might wonder, however, if voters place more weight on more recent price movements since such information is closer in time to the voting decision. To check this possibility, we calculated ΔPETPR _{i , t} over both a 1-year time frame and a 2-year time frame. When these two variables were successively introduced into our model, we found that, while the coefficient on petroleum price growth continued to be positive, the coefficient itself was smaller and not statistically significant. This would seem to suggest that in energy consuming states, voters are inclined to evaluate the state of petroleum prices over the entire presidential term. In light of these results, one might then postulate that in addition to price growth, voters might also adversely respond to the volatility of such prices because greater volatility might signal a much less stable economic environment, particularly for those living in energy consuming states. While additional research on these subjects, perhaps for gubernatorial or congressional elections, is desirable, we also introduced a price volatility measure into our model and it too proved statistically insignificant. We thank an anonymous referee for these suggestions.

²² Strictly speaking, this variable is significant at the 10% level in Model 2 but not in Model 1. However, with a p-value of 0.1034 in Model 1 and the potential for collinearity between income growth and unemployment, we can be reasonably confident that income growth does indeed have a statistical impact on these election outcomes.

²³ It is worth reminding readers that these conflicts are not conflicts involving US military personnel. Therefore, this result should not be taken to mean that voters are unlikely to vote for the incumbent when the nation is at war. Indeed, since the US was not actively at war during any of the election years considered here, our model cannot speak to the likelihood of retaining a wartime president.

²⁴ Indeed, the DMY₈₀ variable is negative in all models, and quite surprisingly statistically significant at the 10% level in Model 3. It is difficult to ascertain why this result came about and it certainly merits further attention in future related research. However, it should be remembered that this election was a highly contested election (unlike the following 1984 election). Therefore, the insignificance of the dummy may be easier to accept.