Welcome!

Hello! My name is Alison Hu, and I’m a sophomore at Harvard University studying Applied Math and Economics. I am deeply interested in data science and data journalism, and this blog will explore those areas in the context of the United States presidential election system. As part of my coursework for Gov 1347: Election Analytics, my goal is to analyze possible predictors for the outcome of the 2020 US presidential election and to convey my insights in a clear and engaging way. I hope you’ll join me along this journey!

If you have any comments, questions, or suggestions, please feel free to email me at ahu@college.harvard.edu.

December 10: Election Narratives

After a presidential election, the media and campaigns speculate on possible narratives to explain why the election turned out the way it did. In our final blog post, we test a narrative on this year’s election supporting Biden’s win: that the Biden campaign’s message was both effective in its message centered around decency and consistent across time. To do so, we look to campaign speeches and Twitter data.

November 23: Post-Election Reflection

With the outcome of this year’s much-anticipated election finally solidified and a new president-elect for the nation, we take a look at how our model performed.

November 1: Final Prediction

Using everything we have learned and explored these past seven weeks, we present our final prediction for the 2020 U.S. presidential election.

October 26: Shocks

Can shocks, defined as any unexpected or unpredictable events, influence an election’s outcome? This question has been studied by scholars in many different contexts, including shark attacks, war casualties, and sports game results. Though the causal effects of shocks are difficult to isolate, findings point again to the theory of retroactive voting: that voters do tend to either reward or punish incumbents based on past events. Thus it is plausible that shocks, even ones seemingly apolitical, can alter voters’ behaviors. We explore shocks that may affect the 2020 presidential election, including the COVID-19 pandemic and racial justice movements of magnitude unique to this year.

October 19: The Ground Game

Even as the Air War becomes increasingly important to elections, on-the-ground contact with constituents still remains an essential component to campaigning. The Ground Game proves to be a potentially more effective method in persuading and mobilizing voters, with messages and interactions being more personalized and targeted in comparison to mass media efforts – but this also makes ground efforts more costly and difficult to carry out. This week we continue our descriptive analysis to determine strategies in field office locations and consider what the Ground Game looks like in the current political climate.

October 12: The Air War

From the invention of the television to the creation of the World Wide Web to today’s ubiquity of social media, the reach and extent to which advertising influences our everyday lives has been constantly evolving over time. Likewise, as technology changes, so too does the role of advertising in elections. Because complete data is difficult to obtain across advertising platforms and states, and the effects of advertising are extremely difficult to isolate from other factors, this week we will not update our predictive model but rather focus on some interesting descriptive statistics.

October 5: Incumbency

Incumbency is often cited as an influential factor in election outcomes. In fact, for the seventeen post-war presidential elections thus far (1952 to 2016), an incumbent president has run for re-election eight times and only lost twice. This week, we look at whether adding an incumbency indicator variable affects our predictions as well as whether the supposed incumbency advantage applies to this year’s presidential election.

September 28: Polling

As we saw last week, economic indicators alone may not be highly predictive of presidential election outcomes, especially during this time. One way to improve our model is to use polling data, since these ideally reflect which candidate poll respondents may actually vote for. Similar to economic indicators, the predictive power of polls goes up as we get closer to elections. Using aggregated polling data, we can test out how polling data alone and polling plus economic indicators data affect our model.

September 21: Economic Indicators

This week we begin building our predictive model using economic indicators. Voters often use past results to make inferences about the future, and measures surrounding the state of the economy are some of the most easily obtained, and thus most used, features to make these judgments. However, voters focus on the near past, not the entire four-year term of the sitting president. For instance, if unemployment rates are high at the beginning of a president’s term but low in the time period leading up to an election, voters may still view him favorably and be more likely to vote for the incumbent party in the next election. This has implications both on how we should set up our model but also on how candidates choose to portray themselves during campaign season.

September 14: Introduction

As we witnessed during the 2016 presidental election season, predicting a presidential election’s outcome is difficult. Scholars have put forth numerous theories, with many factors that may help to determine who will win, yet there is always room for surprises and uncertainty. To set the scene for our prediction model, we begin by exploring the system we have in place for selecting our country’s president — the electoral college system — and its consequences. Next week we will begin building our predictive model.