Causal Inference and the Election
Beware the grand narrative
Applied economics often deals with causal inference. What is causal inference? Wikipedia defines it as, “the process of determining the independent, actual effect of a particular phenomenon that is a component of a larger system.” That still sounds fancy, so let’s make it simple: causal inference is about finding cause and effect. This is very difficult to do in economics. In other fields, it’s relatively simple. If a crop biologist wants to identify the effect of a new fertilizer on corn, the process is straightforward. Plant two fields with the same seed, spray one field with the new fertilizer, and leave the other one alone. Then see if there is a difference in crop yields. Any difference can be attributed to the new fertilizer. Medical studies follow a similar, albeit far more expensive path. To test the efficacy of a new drug, test subjects are separated into a control and treatment group. The treatment group is given the drug, the control group a placebo. Neither the patient nor medical professional knows who is being given which. The health of both groups is then tracked over several months.
Economists don’t have that luxury. Either because of complexity or ethics, randomized controlled trials are usually not feasible. In environmental economics, researchers have tried to determine the effects of air pollution on worker productivity and cognitive function. With unlimited resources and no pesky internal review board, setting up an experiment would be straightforward. Find two nearly identical towns or neighborhoods. Put a coal power plant next to one, and start burning away. Track both towns and see what happens.
Since this is impossible, as well as evil, it can’t be done. Instead, economists rely on what are known as “natural experiments” to determine the causal effect of air pollution. Coal power plants are not randomly sited, but sometimes they close randomly. What happens to the people who live nearby? A temporary closure can be used to mimic a randomized controlled trial. Natural experiments aren’t as foolproof, but they can be fairly accurate. Today, a robust literature shows that air pollution, even in small amounts, quickly has negative impacts on human health, cognition, and productivity. Over the 30 years, environmental economists have built a conclusive narrative that air quality should be closely monitored, and that while acute pollution events and “bad air days” are particularly harmful, any reduction in air quality will have harmful effects. This conclusion is the result of dozens of social scientists spending thousands of hours identifying and studying different natural experiments, working alongside hard scientists from other fields, and building a large literature that mostly shares the same results. This is how we gain knowledge.
Politics does not work this way.
Over the past few weeks, pundits from across the political spectrum have created a bevy of grand narratives to explain Donald Trump’s and the Republican Party’s definitive win in the 2024 elections. Some narratives include Harris’s ability as a candidate, Biden’s age and late withdrawal, racism, Trump’s personality, the Democrat’s move to the left on cultural issues, inflation in 2021, misogyny, lack of trust in the media, and the Trump assassination attempt, just to name a few.
These attempts at crafting a grand narrative overlook a few things.
First, the rise of Donald Trump changed US history. While he has dominated US politics for the last nine years and won a decisive victory in 2024, this was all downstream of his 2016 election win. That win also prompted a lot of the same grand narrative talk as 2024. Almost all the articles about “why Trump won” from 2016, overlook a key concept: luck.
Trump won in 2016 by the smallest of margins. Less than 100,000 voters perfectly located in Wisconsin, Michigan, and Pennsylvania delivered him the presidency. There were a lot of issues that made the election close. Hillary Clinton had the lowest favorability ratings of any politician in America. The Democrats had been in the White House for eight years. Trump found and tapped into a real anger. Ultimately, however, it was a knife-edge election. If the weather had been different on election day, if the Comey letter had not been sent, Clinton would have won. Then all the think pieces about Trump would not have been written. Instead, if Clinton wins by 100,000 votes across three key states she is being credited with running a brilliant campaign against the biggest upstart politician the US has ever seen.
It reminds me of a close sports game. Many football games are neck-and-neck. The two teams go back and forth, with exciting plays and lead changes throughout. Then, towards the end of the game, the team that is losing will have only a few minutes to drive down the field and attempt a game-winning field goal. If that field goal attempt goes in, they will win. If it misses, they will lose. With only a few seconds left, the team will line up to kick the 50-yard game-deciding field goal. If the kick goes in the team storms the field. The winning coach will say they won the game because they believed in each other or dug deeper than the other team. The winning QB will be credited with leading the team to victory.
What a bunch of bunk.
That team won because the kicker made a last-second field goal! If he had missed, then the other coach would be saying the exact same thing about digging deep and conviction. Belief and effort matter in sports, but ultimately that game was decided by the foot of a single player. If that kick is lined up ten times, the kicker would have made some and missed some. The winner of the game was essentially a coin flip. Reporters like to assign causality to sports and elections, but they forget that, at least in elections, ties aren’t possible. If an election is essentially a tie, then it’s random chance that produces the winner.
This was most obvious during the 2016 election but has been true for most of the most recent elections. Over the last several cycles, the only time the winning presidential candidate had a 10 percent margin over the losing candidate was in 2008. Just as there is chance in any sporting game, there is chance in every election. People like to assign definitive causality and pretend like chance doesn’t play a role. It does.
Consider 2024. The election was between Trump and Harris, but only due to an almost surreal series of events. If Trump agrees to the standard presidential debate structure, Biden debates him for the first time in September. At this point it is too late to replace him as a candidate if he has a horrible performance, and the election is Trump vs. Biden instead of Trump vs. Harris. If Trump doesn’t move his head just right during his campaign rally in Butler, PA, he’s killed by Thomas Matthew Crooks and the election is Harris vs. a different Republican. There is a lot of luck in politics.
Second, there’s only a presidential election every four years. That doesn’t provide a large enough sample size to prove a narrative. Maybe Trump beat both Clinton and Harris and lost to Biden because of anti-woman gender bias. But maybe Trump lost to Biden and beat Clinton and Harris because of an age bias. Perhaps voters prefer older candidates. After all, Trump beat two younger candidates and lost to one older candidate.
I don’t think there was an age bias, but my point is you can’t look at such a small sample size and say definitively what happened. It’s long since been muted that there is a height bias in presidential elections. The taller candidate usually wins. That, however, conveniently ignores that George W. Bush was the shorter candidate in both 2000 and 2004 and relies on Bill Clinton beating two candidates who were less than an inch shorter than him. It’s easy to create stories like this because there are so few elections. Six of the last seven elections have been won by the candidate with less hair. Does that mean Americans have an anti-hair bias?
Finally, there’s no way to confidently declare cause and effect because every election has many different variables, not least of all the candidates themselves. Would any Democrat have lost to Trump in 2024? Maybe, but there’s no way we will ever know. Did inflation doom the Democrats from the get-go? Also possible, but without a time machine we can’t prove anything. If Trump was assassinated last summer, maybe the coalition of Trump fans combines with the never-Trump Republicans to give the Republican candidate a landslide victory. Maybe Trump fans stay home and refuse to rally around the Republican replacement and Harris wins. It’s impossible to say.
With so many variables changing every four years, cause and effect cannot be definitively stated. A better way to look at elections is to determine what shifted the needle. To return to the sports analogy, a close football game might come down to a last-second field goal, but many plays led up to that close finish. Some players played well, others poorly. Good plays were called and bad. All those decisions led to close score. It’s unfair for the kicker to be blamed for the loss or credited with the win because if the team had played better the entire game he wouldn’t have been needed and if they played worse he would have been irrelevant. The same is true for elections. Many factors, from economic indicators to candidate likability to weather decide who votes and for whom. There is no single narrative about what caused victory or defeat.
Instead, it would behoove both losing and winning parties to examine each part of their strategy and determine what increased their vote share and what decreased it. The same way a team should play to their strengths and avoid their weaknesses, a political party must constantly adapt and evolve. What have attracted votes and what has repelled them. What moves the needle in your favor. That’s how elections are won and lost.