Election predictors have been doing it all wrong

Polls are one of the most coveted things in politics, and for good reason. We are all anxious for months leading up to the elections, and polls offer us a guess at the results taken from a sample of the actual voters! Furthermore, the campaigns themselves need polls to determine what moves should be made to give them the best chance of winning their race. But inaccurate polls don’t do anyone any good, and in recent elections there has been a common narrative of polls missing their mark. Most notably in the 2016 election, where several midwestern states such as Wisconsin, had results 5-9 points off of the expectation. So of course, pollsters and news organizations attempted to fix these problems. Some changed their polling methods, and others combined poll averages with fundamentals. The fundamentals are the expectation for any given race based on party registration in that area. This has had a mix of positive and negative results.     However the 2022 midterms may have finally given us the key to providing accurate predictions in the future. In 2022, the polls were actually fairly accurate. In fact, the polls were much more accurate than all of the models created by combining polling averages and fundamentals. 538, the gold standard of election models, has two different models that they use each cycle. One, called the lite model, that bases its predictions mostly off of polling averages, and another, called the deluxe model, that combines polling averages with fundamentals, (weighting the latter more heavily). Now approaching the 2022 midterms, 538 considered their deluxe model to be the most accurate, however, on Election Week, we learned that the lite model was much more accurate. This has left many political analysts scratching their heads, and wondering what the next step is for the 2024 presidential election.     Here at TMax we believe that we may have the answer. The currently presented issue is that it seems like sometimes the fundamentals are more accurate, and sometimes the polls are more accurate. For example, in 2022 and 2018, the polls were much better, but in 2020 and 2016, the fundamentals were much better. Now the solution starts with the key voter issue. Do voters make their decision based on the quality of the candidate, or based on the party they want to be in power. And the data collection from the past 6 years has pointed towards an unexpected answer: it depends on the type of race. Gubernatorial races tend to be based off of candidate quality while US house races tend to be almost entirely based off of party. Presidential races seem to shade towards the side of voting-by-party, and senate races seem to be directly in the middle, sliding either way depending on the year.      So what does this mean about predicting future elections? Well there is a clear pattern we have noticed. In races based off of candidate quality, polls tend to be the better bet, because they gather information about voters opinions of each candidate, however, in elections decided by party affiliation, the fundamentals tend to be more accurate, because if 99% of voters vote for a member of their party no matter who it is, you can get a very good sense of the results by looking at party registration. So if we put these pieces together, we get our final answer. In presidential election years, the best model is one that combines fundamentals and polling, with a greater focus on fundamentals, but in a midterm year, looking at polls will be much more accurate. This finding lines up well with the piece of data seen earlier, in 2018 and 2022 (midterm years) the polls were very accurate, whereas in 2016 and 2020 (presidential cycles) the polls were way off.      How do we know if this is right? Well we don’t, but with all the data presented, it is certainly a good guess, And for the 2024 presidential election, we will be keeping a close eye on our fundamentals