First Panera post of the summer!

Hi everyone! This is my first blog post for the summer and I’m excited to discuss some of my project’s progress.

As I originally stated in my abstract (which can be handily located at, the primary objective of my Freshman Monroe grant is to contribute to a much larger national dialogue about the future of the Democratic Party. Following the stunning presidential defeat of Hillary Clinton two years ago, Democrats have been bogged down by an identificational crisis. Many of the party’s guiding principles, including support for ethnic and religious pluralism and an emphasis on economic equality, appear threatened by Donald Trump’s presidency.

These concerns have led to different theories of how to best improve Democratic performances in the 2018 midterms. Brian Fallon, a former strategist for the Clinton campaign, publicly theorized last year that attracting “Panera voters”, i.e. well-educated, affluent, suburban voters, would be the most beneficial strategy for Democrats moving forward. I aim to examine if that theory holds any weight by testing if vote swing from 2012 to 2016 has any correlative relationship with the amount of Paneras in a given local jurisdiction.

Unfortunately, the first phase of decoding Fallon’s theory proved to be the most time-consuming one. The first step was to develop a measurement of partisanship that (1) could be applied to every local jurisdiction in the country and (2) can be easily tracked over time. I needed to determine the most effective way of gauging shifts in partisan affiliation, as doing so would help me determine if there is a relationship between Panera prevalence and increased Democratic vote share/decreased Republican vote share. I came to the conclusion that I could reasonably do this by obtaining three data points from every county, independent city, or parish in the United States:

  1. Republican presidential candidate Mitt Romney’s performance in 2012 (expressed as a percentage)
  2. Donald Trump’s performance in 2016 (expressed as a percentage)
  3. The difference between these two values, also known as the political swing of the county from 2012 to 2016 (ex: If Trump won 40% of a county’s vote share whereas Romney won only 35%, the county ‘swung’ 5% more Republican)

These three data points would allow me to track if support for Republicans did actually decline in local jurisdictions with more Panera Bread locations, as I could investigate if counties with a Democratic swing from 2012 to 2016–i.e., counties where Trump actually under-performed Romney–are the ones with a higher prevalence of Panera franchises.

I quickly ran into a roadblock. After a few hours of desperately scouring the Internet for helpful data files, I had no problem obtaining data points 1 and 2. Every local jurisdiction (except for pesky Alaska, which continues to throw wrenches into my research) has readily accessible election records for the 2012 and 2016 presidential elections through their respective state boards of elections. However, obtaining pre-done calculations of vote swing from 2012 to 2016 seemed impossible. I eventually decided to give up on finding a data file and resolved to calculate the vote swing by subtracting Romney’s vote percentage from Trump’s vote percentage for each county by hand.

This process was extraordinarily taxing. I spent about five hours going through making individual calculations, and slowly but surely, I trudged through each state’s counties alphabetically (I had no idea how many counties are in Arkansas!) By the time I made it to Florida though, I decided to research if there were any softwares online that could do these calculations for me, since staring at Excel spreadsheets and typing endless percentages into my iPhone calculator was growing tedious.

Obviously, a software does exist for this purpose, eliminating my need to complete any mundane calculations by hand. Stata, a software which I also intended on using for determining a statistical relationship between Panera prevalence and vote swing later in my research, is also capable of merging data files and extracting calculations between them. Moving forward, I’ll teach myself Stata, merge the 2012 and 2016 results data files, and obtain the vote swing that way.

Thanks for reading!



  1. iancropley says:

    I love this concept! It seems like an interesting and creative way of looking at shifts in voting demographics. One question I have is whether you looked at changes in the number of Panera locations between the 2012 and 2016 elections. Whether you did or did not could have different implications for the study. If you are looking at the number of Panera locations in 2016, my interpretation of your research is that Panera is a “state of mind” reflected in a certain demographic (well-educated, affluent, suburban voters) that remained relatively static during the four year period between elections. In this case, your study just looks at the change in voting within this particular group. However, if your study involves looking at the number of Paneras in 2012 and 2016, my interpretation from your experimental design would be that the size of the affluent, suburban voter group has changed and that this is reflected in the voting results. I imagine you are doing the first because I saw no mention of tracking the number of Panera locations in your blog post (and also because it would be difficult and time consuming to get exact data on the change in number of Paneras), but I wasn’t quite sure.