# A Comparison Between the Oracle of Blair and FiveThirtyEight's House Models in Ohio and West Virginia

In order to compare our ORACLE of Blair to FiveThirtyEight House Forecast to look for any noteworthy differences between the two models, we examined Congressional districts in Ohio and West Virginia where either the Democratic voter share or Democratic chance of winning was drastically different between the two models. For our purposes, we define drastically different to be a 5% difference in voter share or a 10% difference in chance of winning.

Under these criteria, West Virginia districts 1 and 3 and Ohio districts 2, 3, 6, 7, 12, 13, 15, and 16 show a drastic difference. This means that ten districts in the two states, just over half, indicate noteworthy differences between our model and that of FiveThirtyEight. However, it is important to note that while on average our predictions and FiveThirtyEight’s differ by 3.6%, both models expect that the average race in one of these two states to be decided by over 20%. So, in not a single race does one model predict a different candidate to win than the other. Only Ohio’s 16th district is predicted as a close race in one model and a safe race in another model. The discrepancy between the results of the two models is likely due to differences in the weighting of fundraising and partisanship, among other factors.

DistrictOur Dem %538 Dem %DiffOur Dem %538 Dem %Diff
WV-0130.0%35.7%-5.7%0.0%0.0%0.0%
WV-0242.8%43.1%-0.3%5.8%5.0%0.8%
WV-0347.1%44.0%3.1%20.0%9.5%10.5%
OH-0146.4%47.5%-1.1%13.0%22.1%-9.1%
OH-0245.4%42.7%2.7%17.8%4.1%13.7%
OH-0366.9%72.9%-6.0%100.0%100.0%0.0%
OH-0437.6%42.0%-4.4%0.7%2.8%-2.1%
OH-0537.8%36.9%0.9%1.3%0.2%1.1%
OH-0638.4%31.5%6.9%1.7%0.0%1.7%
OH-0738.3%44.4%-6.1%1.6%8.3%-6.7%
OH-0835.2%33.0%2.2%0.3%0.0%0.3%
OH-0971.9%69.4%2.5%99.9%99.9%0.0%
OH-1040.5%43.7%-3.2%3.9%5.3%-1.4%
OH-1182.1%80.5%1.6%100.0%100.0%0.0%
OH-1245.0%48.4%-3.4%17.8%31.3%-13.5%
OH-1360.8%68.0%-7.2%98.4%100.0%-1.6%
OH-1445.5%46.3%-0.8%18.3%16.4%1.9%
OH-1545.8%41.5%4.3%20.2%2.0%18.2%
OH-1649.1%42.9%6.2%43.3%4.1%39.2%

Our model does not take fundraising into account at all, so in districts where there is a major distinction between the two candidates in terms of fundraising, our model predicts drastically different outcomes than that of FiveThirtyEight. One example of a major difference between candidates in fundraising exists in Ohio’s 16th district. The ORACLE of Blair predicts that Susan Moran Palmer (D) will have a 49.1% vote share and a 43.3% chance of winning. In contrast, 538 only predicts a 42.9% vote share and a 4.1% chance of winning. This results in a staggering disparity of 39.2% between the two models in terms of Palmer’s probability of winning. However, such a change in the chances of winning is to be expected with the comparatively small vote-share disparity, as even a slight reduction in the number of votes a candidate receives in any remotely close race will have drastic impacts on a candidate’s ability to carry the district. The significant difference between our two models results from, again, 538’s incorporation of campaign fundraising. While Palmer’s campaign has only raised around $100,000, her opponent Anthony Gonzalez (R) has raised well over$1,000,000.

DistrictDem FundraisingRep Fundraising
WV-01$208,000.00$386,000.00
WV-02$532,000.00$952,000.00
WV-03$1,472,000.00$682,000.00
OH-01$2,692,000.00$749,000.00
OH-02$416,000.00$375,000.00
OH-03$230,000.00$0.00
OH-04$484,000.00$423,000.00
OH-05$173,000.00$466,000.00
OH-06$16,000.00$742,000.00
OH-07$2,062,000.00$306,000.00
OH-08$48,000.00$277,000.00
OH-09$158,000.00$0.00
OH-10$742,000.00$556,000.00
OH-11$147,000.00$16,000.00
OH-12$2,563,000.00$891,000.00
OH-13$817,000.00$76,000.00
OH-14$1,617,000.00$919,000.00
OH-15$825,000.00$1,068,000.00
OH-16$124,000.00$1,207,000.00

Our model also uses a different formula for district partisanship than FiveThirtyEight’s does. Our Blair Partisanship Index (BPI) takes a weighted average of the Democratic percentage relative to the national average in the 2012 and 2016 presidential elections, and the 2014 and 2016 congressional elections. It weights recent elections and congressional elections higher than earlier elections and presidential elections respectively. FiveThirtyEight’s partisan lean calculation does not include past congressional elections and does include state legislative elections. They give the 2016 election a 50% weight, the 2012 election a 25% weight, and recent state legislative elections a 25% weight. The effect of not including congressional races in the partisanship metric is shown in West Virginia’s third district. FiveThirtyEight’s model considers this district 37.3% more Republican than the nation as a whole while The ORACLE of Blair only rates them as 17.6% more Republican than the nation. In 2014, the district was only 5.81% more Republican than the country at large, which decreases the average partisan lean if it is included as all of the other elections had results that were at least 20% more Republican than the nation as a whole. Since BPI includes this election, and FiveThirtyEight’s partisan lean does not, FiveThirtyEight’s partisan lean is far more Republican than ours is. The result of this discrepancy is a jump in predicted Democratic percentage from 44% to 47.1% from FiveThirtyEight’s model to ours. The effect of this increase in Democratic voter share is an increase from 9.5% to 20% in the Democrat’s chance of winning in West Virginia’s third district.

DistrictPresidential 2012House 2014Presidential 2016House 2016
WV-01-15.63%-6.88%-23.14%-15.45%
WV-02-13.19%1.22%-20.23%-4.66%
WV-03-18.43%-5.81%-26.79%-20.38%
OH-01-5.05%-6.18%-4.55%-5.70%
OH-02-7.55%-8.93%-9.54%-12.93%
OH-0318.73%13.61%18.97%14.60%
OH-04-9.11%-10.63%-18.79%-14.47%
OH-05-6.96%-12.64%-14.42%-17.38%
OH-06-8.35%-3.11%-23.30%-17.15%
OH-07-6.82%-12.05%-16.69%-15.34%
OH-08-14.93%-14.02%-19.13%-22.39%
OH-0916.67%17.34%10.55%14.71%
OH-10-2.93%-10.36%-4.94%-12.71%
OH-1131.40%28.99%31.46%26.28%
OH-12-7.30%-14.01%-7.05%-15.53%
OH-1312.02%18.06%2.29%13.03%
OH-14-3.64%-8.67%-7.12%-9.09%
OH-15-4.82%-8.99%-9.20%-12.68%
OH-16-6.12%-6.70%-9.83%-12.48%

When constructing our model, we believed that using fundraising was superfluous since the results of these factors should already be reflected in the polls. Additionally, our inclusion of past House races in BPI was done out of belief that house races would do a better job than Presidential races at predicting the current House election. However, it remains to be seen whether The ORACLE of Blair or FiveThirtyEight’s House Forecast fairs better on election day.

CL, JD, TO, KSZ