Battleground Smackdown

We selected battleground districts identified by BallotPedia, specifically Kentucky’s 6th, New Jersey’s 2nd, 3rd, 7th, and 11th, North Carolina’s 2nd, 9th, and 13th, and Washington’s 3rd, 5th, 8th, and 9th because our group previously studied these states, and these are historically battleground districts. We compared our calculated vote shares to those of 538. To investigate differences, we compared our methodology to 538’s classic model using data collected on October 16th, 2018. We found that fundraising is the key difference between the models and will discuss the specific effect it has in our select districts.

As depicted in Figure 1, ORACLE, compared to 538, leans towards Republicans in all of these races. The reason for this can be seen in Figure 2, which shows that ORACLE’s fundamentals predicted margins strongly favor Republicans, while 538’s fundamental margins are relatively moderate, but still lean towards the Democrats. ORACLE’s fundamental margins favor Republicans because it only uses data from the past three elections, in which Republicans won the majority of these selected districts. There are many factors contributing to the differences in fundamental margins, but we concluded that fundraising has a major impact on the differences. For example, NJ-03’s Democratic candidate has raised $2.4 million more than the Republican candidate, and in WA-08, the Democratic candidate has raised $1.6 million more than the Republican candidate. Fundraising can greatly affect a campaign’s exposure, and therefore, chance of winning. As shown in Figure 2, fundraising shifted most of 538’s predictions to favor the Democrats, with the exceptions of NC-02.

Figure 1. FiveThirtyEight and ORACLE Democratic Vote Share, Win Chance, and Estimations. Predictions favoring the Democratic candidate are blue, those favoring the Republican candidate are red.

Figure 2. FiveThirtyEight and ORACLE Democratic Margin Comparison and Fundraising Effects.

To test this claim, we removed the estimated weighted fundraising shift from 538’s predicted Democratic vote share. Results are shown in Figure 1. Fundraising shift was calculated by multiplying 538’s calculated shift by the weight for each district’s fundamentals. The fundamental weight was estimated from the figures that were found on 538’s district pages. Since the calculated fundraising shift is for the margin, we multiplied this by 0.5 and subtracted the shift to 538’s original predicted Democratic vote share. For example, in KY-06, we multiplied the shift, 5.6, by 0.4, the estimated weight. This number, 2.24, was multiplied by 0.5 and added to the original vote share, 50.1%. The adjusted 538 vote share prediction in KY-06 is 48.9%. For the most part, 538’s adjusted Democratic vote share match ORACLE’s predicted vote share in terms of in which party holds majority vote share, with the exception of NJ-03. On average, 538’s adjusted fundamental margins shifted by 0.57% in favor of the Republicans, but it still gives 1.16% more to the Democrats than ORACLE does. The other factor that 538 takes into account when calculating fundamentals that ORACLE does not is the previous incumbent’s margin in the last election. However, in all of the selected districts, this factor shifts the fundamental margin in favor of the Republicans.

Another discrepancy between the two models is highlighted in the comparison WA-03 and WA-05. Unlike in WA-03 where ORACLE predicted the chance of Democrats winning as 17.6% less than 538, our model predicts a higher win percentage for Democrats than 538 does in WA-05 by 4.8%. A closer investigation of the methodologies shows that this is a result of the differences in how we calculate partisanship. In WA-03 and WA-05, both 538 and ORACLE calculate the district partisanship to be leaning towards the Republicans. However, in WA-5, 538 says the partisanship is almost twice that of WA-03 (2.0 in WA-03 vs 3.4 in WA-05), but our calculated partisanship says there is not much difference at all, (5.72% in WA-03 vs 6.07% in WA-05). Since 538 calculates that partisanship gives the Republicans much more of an advantage than our partisanship does in WA-05, 538 predicts the Democratic party with a lesser chance to win, 4.8% lower than we do in WA-05.

After comparing win percentages and vote shares between ORACLE and 538, there were clear trends to look at. We found answers by comparing the methodology of the two models, and what 538 included that ORACLE didn’t. The primary differing factor was 538’s inclusion of fundraising which generally gives Democrats advantages in these districts. Another factor would be the difference in how the two models calculate district partisanship because this affects select districts like WA-03 and sways these districts’ elections. For future models, ORACLE should take fundraising into account, as it has a big impact on elections, but only after the election can we tell if these district partisanship calculations were accurate.