Journal of Student Research 2021

Time-Series Analysis of Wave Elections

83

We then applied the methodology used to find a threshold for the House on the data from the Senate. This resulted in a threshold of 2%, pictured in Figure 7. When applied to this data set, that threshold gives us 31 segments which, like the above threshold, runs into the same problems as the Rothenberg definition, which had a requirement that 20 seats must be gained in the House for an election to be a wave. Thus, like the above threshold, this implies that waves are less significant than this study assumes and is not a usable threshold for a definition given our assumptions. Thus, we will not be able to make a definition that includes the Senate. However, further examining the data does provide interesting results. Figure 6: The election data for the Senate after the Bottom-Up algorithm is used with a threshold of 1.7% .

Figure 7: The election data for the Senate after the Bottom-Up algorithm is used with a threshold of 2%.

We then sought to instead find a threshold that matches the 16 waves found by the threshold that met our criteria for the House. We found that a threshold of 2.5% gives a matching number of waves when used on the Senate data. This is pictured in Figure 8. Some interesting things to note are the period of stability from 1984 to 2014 and the fact that 2016 is the most recent wave in memory according to this threshold. When we compare our method’s results to Ballotpedia’s [2], we catch every election that they consider a Tsunami election. We also catch every election that