Final Peer Review (wmx2) #12
Description
This project attempts to predict the results of presidential elections by way of the votes gained in key swing states. It picks out information from key counties in previous years, fits models on these counties, and then attempts to use this model to help predict the results of the 2020 election. Because the 2020 election occurred somewhat before this report was due, the group was able to use the 2020 election as a baseline benchmark of how accurate their model was / how well it actually predicted trends.
Some things I enjoyed about this report are:
- It is very well laid out and organized. I had zero trouble at all following the flow of the content. The passages were broken up to show logical progression of the project's progress over time, the graphs were all generally well laid out and color-coded, and appropriate links were left in order to send the reader to more resources as necessary.
- I appreciated how the group was very upfront in acknowledging the limitations of their model, and where they could not be confident in their answers, rather than merely pretending their model was correct. For example, going into the section predicting the 2020 election, they clearly state the limitations of their data / model with regards to predicting this result.
- The group was also very comprehensive with their considerations and analysis, and considering multiple possible facets of their predictions. I especially appreciated how they dedicated a whole section to considering the ramifications of mail-in balloting on the results of the election, and whether their model would be able to handle such a large influx of mail-in ballots, since previous elections did not have this much data for the category.
Some things I think could be improved about this report are:
- The group uses Linear Regression for their models. While a good model, as much of their data is continuous, it would potentially have been interesting to consider a logistic regression model / classification-based model, binary or not, detailing whether one county would vote for one side or another.
- The one nitpick I have with the layout is that almost all of the figures are graphs. It would be nice to have a table detailing each of the individual models, and their performance / accuracy / etc, instead of having to read through the paper and scan for it (which makes it much easier to accidentally look over / make less of an impact). Bolding the most important words / models / etc and color coding them would also help.
- There were some minor typos here and there (e.g. "preformed" instead of "performed"), but not enough to be substantially detrimental to the experience of reading the paper.