Evaluating Our Legislative Predictions
Predicting elections is far from an exact science, as Americans around the country learned on Tuesday night with Donald Trump’s surprising victory over Hillary Clinton. Now that the results are in, how did Carolina Data Desk’s predictions fare?
Where the Model Succeeded
Our model was successful nine times out of ten in predicting the correct party in districts we classified as either slight or strong leans toward one party.
In every district that our model projected as a “Strong Lean” to one party, that party’s candidate did, in fact, emerge victorious. This was the case for 25 House districts (5 won by Democrats, 20 by Republicans) and 16 Senate districts (8 won by each party). Of the 24 districts we pegged as “Slight Leans,” 79% were won by the predicted party.
Where the Model Failed
Republicans far outperformed our model in the “Toss-Up” districts, winning 27 of the 33 districts in this category. Only four out of 22 toss-up districts in the House and one out of 11 toss-up districts in the Senate switched parties. Of these five districts, four flipped from Democrats to Republicans and one from Republican to Democrat.
One major assumption in our model was that each party was equally likely to win toss-up districts. This turned out to be a key flaw since Republicans were defending nearly three-fourths of these districts. Incumbents typically have a slight inherent advantage, and the model would have proved more accurate if it had accounted for this advantage.
The model was limited slightly by the lack of historical data. Since this was only the third election since these districts came into effect, the predictions only utilized two election years. In the future, incorporating more election results could improve the model’s accuracy. Other possibilities for improvement include better accounting for the share and voting tendencies of unaffiliated voters and determining which registered voters are likely to actually vote. Both of these changes could be made using voter history data, which shows which elections registered voters participated in.
- House District 6: Our model predicted this seat as a slight Democratic lean, but Republican Beverly Boswell won in a tight race.
- House District 49: We saw this seat as the most likely to flip to the Democrats. This held true as Democrat Cynthia Ball defeated incumbent Gary Pendleton. The 2300 votes for Libertarian candidate David Ulmer proved crucial, as the margin between Ball and Pendleton was only 850 votes.
- House District 59: This was one of our model’s bigger misses, as the predicted toss-up was won comfortably by incumbent Jon Hardister.
- House District 35: Predicted as a toss-up, incumbent Chris Malone won re-election in a close contest.
- House District 8: As predicted, this was one of the closest races, as Republican incumbent Susan Martin defeated Democratic challenger Charlie Pat Farris by only 202 votes.
- House District 46: This district was not as close as we predicted, as Republican Brenden Jones received over 60% of the vote.
- House District 41: Incumbent Democrat Gale Adcock held onto this seat, which our model projected as a toss-up.
- Senate District 15: We predicted a slight Democratic lean, but Republican incumbent John Alexander won a three-candidate contest.
- Senate District 25: In another of our worst predictions, incumbent Republican Tom McInnis easily kept his seat.
- Senate District 17: As expected, this race was closely contested, with Republican incumbent Tamara Barringer winning re-election. In a race won by only 1000 votes out of over 100,000 total votes count, the Libertarian candidate received 5000 votes.
- Senate District 1: In this projected toss-up, Republican incumbent Bill Cook won with 59% of the vote.
- Senate District 13: In another projected toss-up, incumbent Democrat Jane Smith lost her seat to Republican challenger J.R. Britt.
- Senate District 30: In our other big miss, this projected toss-up turned out to be a runaway victory by Republican incumbent Shirley Blackburn Randleman.