Back in August we conducted a poll of likely New Hampshire primary voters for the gubernatorial races. Instead of polling by phone we targeted our subjects via social media by scrutinizing their activity in politically oriented Facebook groups or their Twitter activity. This polling method has been essentially written off as biased towards those people who own and frequently make use of computers or smartphones. However, in an age where the rate of ownership and use of these devices is ever rising does it not make sense that our methods of polling should reflect these changes? The New Hampshire state primaries were held last week on September 13th and the results surprised quite a few party insiders who severely underestimated the support of supposedly lesser-known candidates. So the question is: how accurate were the results of our polling method compared to how people actually voted?
Our poll predicted (albeit exaggeratedly) Edelblut’s strong showing
To answer this question we need to revisit the poll data. In the Republican race our poll indicated that Frank Edelblut was favored by more than half of Republicans. Rep. Edelblut lost the primary by less than 1,000 votes and received ~30% of the vote. No other news poll listing all of the candidates ever reported Rep. Edelblut to have more than 5% of the vote. Our poll predicted that Frank Edelblut had ~57.3% of the Republican vote; far higher than what he actually received. However our poll also indicated that Rep. Edelblut had ~30.3% of the overall vote. This may indicate that our polling accurately reflected the fact that many self-professed Independents drew Republican ballots and cast for Edelblut in the primary. The Republican gubernatorial primary was one of the closest in the state’s history and essentially proved that the newspaper polls are woefully inadequate at gauging grassroots support.
Online poll much more predictive of the Democrat primary
In the Democratic primary our poll predicted that Van Ostern would have ~42% of the vote among Democrats. Van Ostern ended up winning the primary with ~51% of the vote. This likely indicates that most Independent voters that drew Democratic ballots cast for Van Ostern. Marchand was predicted by our poll to receive 24.7% of the vote among Democrats. Steve Marchand ended up receiving 25.3% of the vote, very close to our predictions and certainly well within the margin of error. Likely what hurt Marchand most was that more Independents turned out and drew Republican ballots than Democratic ones.
Even though the figures we reported did not exactly reflect the attitudes of voters, it certainly came much closer proportionally to the share of the vote each candidate ultimately received. In the future pollsters will have to recognize that targeted, online polling has it’s place in their repertoire. Traditional phone-based polling simply cannot accurately reflect the attitudes of the grassroots of either party or even swing voters. If there is one thing to be learned from our methods it is that it represents a much more diverse spectrum of partisan voters than phone polling. Whether or not this intra-partisan spectrum is relevant in predicting general election outcomes is yet to be seen however it certainly has shed much needed light on how loose groups of people on that spectrum will cast in their primaries.