Researchers use machine learning to reveal how gasoline prices drive presidential approval ratings

The cost of filling up a vehicle with gasoline plays a major role in how American voters view their commander in chief. A recent study published in American Politics Research reveals that gasoline prices operate as one of the strongest predictors of presidential approval ratings, acting in an uneven pattern where initial price spikes cause the most political damage. The research indicates that voters judge presidents based primarily on the direct financial pain felt at the pump, rather than using fuel costs as a broader economic warning sign.

Political commentators and campaign strategists frequently debate the electoral impact of energy prices. Public opinion polls predictably fluctuate alongside changing economic conditions, often leading to intense political messaging surrounding the cost of living. Rangan Gupta, an economist at the University of Pretoria in South Africa, alongside colleagues Christian Pierdzioch and Aviral Kumar Tiwari, wanted to formally test this relationship. They sought to determine exactly how the cost of fuel moves the American public’s approval of the president.

Two main academic theories exist to explain why voters might punish or reward a president based on the cost of driving. The first is the pocketbook mechanism, which suggests that voters react to the immediate, personal financial losses they experience when fuel becomes expensive. Under this theory, voters judge the government based on their own shrinking bank accounts.

The second theory is the sociotropic mechanism. This idea proposes that citizens view energy costs as an easily visible informational source about the general health of the national economy. Because people see gas prices advertised on large signs during their daily commute, they might use those numbers to guess whether the larger financial system is thriving or failing.

Previous investigations into this public opinion dynamic often produced conflicting results. Many early mathematical models assumed a straight line of cause and effect, where every dollar increase in fuel costs would logically produce an equal drop in presidential popularity. The research team suspected that human behavior and media coverage do not operate in such simple straight lines. They wanted to test for irregular thresholds where voter attitudes suddenly shift.

To analyze these shifting public attitudes, the researchers used an advanced computational technique called random forests. A random forest is a machine learning algorithm that builds hundreds of individual decision trees. These decision trees act like complex flowcharts that test different paths of data to categorize information and predict an outcome.

By averaging the results of hundreds of these individual decision trees, the algorithm avoids placing too much weight on any single piece of misleading information. This allows the computer to find hidden, irregular patterns in massive sets of data. The researchers chose this method because it can identify uneven behavioral shifts without requiring humans to guess the shape of those patterns beforehand.

The researchers gathered monthly data spanning fifty years, beginning in October 1973 and ending in December 2023. They compiled numbers on retail gasoline prices adjusted for inflation, global crude oil costs, and historical presidential approval ratings gathered closely by the Gallup polling organization. Gallup surveys provide a highly consistent measure of public opinion because they have asked the exact same question regarding presidential job performance for decades.

The researchers did not pull these broad economic indicators out of thin air. They utilized a massive economic database comprising hundreds of financial and macroeconomic time series. This extensive collection included hard data on national employment hours, manufacturing output, real estate building permits, and international trade volumes. By condensing this massive array of information into eight core factors, the researchers ensured the algorithm had a complete picture of the American economy.

Beyond standard economic performance, the algorithm also accounted for broader societal anxieties. The researchers incorporated formal metrics of macroeconomic uncertainty and financial market volatility. These uncertainty indexes capture the unpredictable shocks to the economy, reflecting the general unease consumers and businesses feel about future financial conditions. Including these measures was incredibly important for testing whether gasoline prices simply act as a proxy for general economic dread.

To seal any remaining gaps in predicting public opinion, the team also included databases tracking geopolitical acts or threats pulled from historical newspaper archives. These metrics tally the number of news articles focused on international conflict, nuclear threats, and terror acts. Before running the models, the researchers specifically separated the cost of unrefined global crude oil from the retail price of gasoline.

While international crude oil serves as the primary ingredient for motor fuel, retail prices also reflect domestic shocks. Local refining capacity, regional taxes, and seasonal driving demands all alter the final price consumers pay at the local station. Isolating the retail cost allowed the researchers to zero in on the precise numbers that voters see and pay.

The machine learning algorithm revealed that the link between retail gasoline prices and public approval is highly nonlinear. When gasoline prices sit at a low level and first begin to rise, the negative impact on the president’s popularity is exceptionally strong. The researchers attribute this sharp drop to an initial surge in news media coverage, which suddenly alerts previously inattentive voters to their shrinking purchasing power.

As energy prices continue to climb past a certain threshold, the penalty to the president becomes less severe. The public’s approval rating bottoms out and remains at a relatively low, flat level despite additional cost increases. At this advanced stage, dissatisfied citizens likely become accustomed to the high costs, and the ongoing negative press coverage offers no fresh shocks to their political opinions.

Putting past presidential approval to the side, inflation-adjusted gasoline prices emerged as an incredibly powerful predictor of future approval ratings. The algorithm successfully used historical fuel prices to forecast out-of-sample shifts in presidential popularity. This indicates that fuel prices carry actual forecasting value, passing a rigorous test of predictive accuracy that goes beyond simply matching past trends.

Because fuel costs proved to be a leading predictor independently of other broad economic health metrics, the researchers argue their results heavily favor the pocketbook mechanism. If voters were simply using gas prices as a sociotropic substitute to guess the overall state of the economy, the algorithm would have relied more heavily on the broader financial data provided to it. Instead, the direct price of filling a tank retained its unique forecasting power. This implies that voters hold political leaders accountable for direct personal financial strain.

While the statistical link between the gas pump and the Oval Office is incredibly strong, American presidents usually possess very little authority over global energy markets. The worldwide supply and demand balance dictates the cost of energy, leaving political leaders with limited options to alter prices quickly.

A president might temporarily suspend domestic taxes or adjust import tariffs, but making vast systemic changes is generally beyond executive power alone. Because of this dynamic, presidents will likely continue to face political backlash for global energy fluctuations that exist outside of their direct control.

The research team suggests that these political realities make a strong case for shifting toward renewable energy sources. Reducing a nation’s reliance on fossil fuels could eventually insulate domestic consumers from sudden global supply shocks. In turn, achieving this sort of green energy transition might shield future political leaders from the unpredictable voter backlash caused by international energy markets.

Future investigations will need to explore how energy costs predict political outcomes across longer timeframes. The current analysis focused on forecasting changes just one month ahead. Exploring alternative forecasting models and extending the prediction window will help political scientists better understand exactly how long it takes for a price shock at the pump to register in a national election booth.

The study, “Gasoline Prices and Presidential Approval Ratings of the United States,” was authored by Rangan Gupta, Christian Pierdzioch, and Aviral Kumar Tiwari.