A month after Biden’s historic presidential victory, various media outlets and pundits have begun to push narratives about why Biden won and how he could have done better. Some argue that Biden’s win was inevitable because of Trump’s unconventional character, others argue that Biden would have performed better if he had proposed more liberal policies, and yet more say that Biden’s high-tech campaign was essential to his victory. In this blog post I will be exploring one specific narrative, that Joe Biden won the election because of his appeals to working class white voters that voted for Trump in 2016. Some examples of this narrative in popular media include:

In this week’s blog post I will be exploring the language used by presidential candidates, especially Trump and Biden. I will examine what words each candidate used the most and how Trump’s and Biden’s language usage changed over the course of their campaigns. To conduct this exploration I used campaign speech transcripts from seven recent candidates. I used the quanteda r package to examine the specific words in each speech.

In this blog, almost three weeks after election night, I will reflect on my election model. Specifically, I will give a brief recap of my model, evaluate its performance compared to actual election results, and explore ways that my model could have been improved.

In this blog post, with only two days left until election night, I will make my final prediction for the outcome of the presidential race. Over the past few weeks I have made a variety of models and considered a wide range of variables. For my final election model, I have narrowed these variables down to polling averages, presidential approval, incumbency, and demographic variables. I have also decided to use a probabilistic model, specifically a binomial logistic regression, so that I can incorporate some of the randomness and uncertainty that is inherent in the election.

With less than two weeks until election night, I will finally discuss the effects COVID-19 might have on this year’s presidential election outcome. COVID-19 is at the front of every voter’s mind; it has changed our economy, our voting, and the way we live our lives. Many blame the severity of the pandemic on Donald Trump’s response, while others argue that the over 200,000 American deaths due to the virus were inevitable. This week I will discuss the potential impact this unprecedented shock might have on voting patterns around the US.

One of the most frequently discussed factors in voting is demographics. In this week’s blog post I will explore the effect that demographic change has had on historical voting behavior, create an updated prediction model for the 2020 election, and then predict the election outcome accounting for potential surges in certain demographics.

In this week’s blog post I will examine the most public facing aspect of an election, the campaigns themselves. One factor of a campaign that voters interact with frequently is advertisements. Politicians and their parties care immensely about advertisements, going so far as to spend hundreds of millions of dollars on advertising each election cycle. In this blog I will explore if campaign spending can be used to predict election outcomes, what the best time to spend money on ads is, and then use campaign advertising in a model to predict the 2020 election outcome.

One key aspect of elections, especially presidential elections, are the advantages afforded to the incumbent. Presidential incumbents are more nationally recognized, receive more press coverage, and have the power to allocate federal funds. In fact, in their 2012 paper “The Influence of Federal Spending on Presidential Elections,” Douglas Kriner and Andrew Reeves find that an increase in federal spending increases the incumbent’s vote share at the state level. In this post, I will examine the effects of incumbency and then explain how these effects apply to Donald Trump.

As the 2020 election draws nearer, national and state polls will gain increasing attention. Frequently, these polls seek to estimate the popular vote share of the presidential election by asking U.S. citizens their voting preference. This week, I will examine if polls can be used as a good predictor for the actual election outcome. I will do this by analyzing historical polling and voting data to see if polling has been an effective predictive model in the past. I will then see if polling data has been an effective predictor of historical elections at the state level. Finally, I will use the models I create and 2020 polling data to predict this year’s election.

One prominent method for predicting election outcomes is using economic variables as predictive models. This model, as proposed and discussed by Christopher H. Achen and Larry M. Bartels in “Democracy for Realists” and Andrew Healy and Gabriel S. Lenz in “Substituting the End for the Whole: Why Voters Respond Primarily to the Election-Year Economy”, is based on a theory of retrospective voting behavior. This theory argues that voters consider the past performance of the incumbent party through variables such as gross domestic product (GDP) growth and real disposable income (RDI) growth.

As the contentious 2020 election looms nearer, countless Americans will begin to make their predictions for election night. In this blog I will discuss insights from previous elections as well as offer my own predictions for the 2020 presidential election. This introductory post will focus on observational analysis of previous elections. Specifically, this post will examine historical trends in the two-party popular vote, two-party popular vote share margin by state, and swing states.