Indiana Political Scientists: Better Polling Requires Better Outreach

Nov 14, 2016

Credit Judy Baxter / https://www.flickr.com/photos/judybaxter/

As the presidential election came to an end, a trending question among citizens went like this: how did the polls get it so wrong?

A pair of Indiana political scientists say some increasingly common errors led to this year’s wildly inaccurate political polling.

Butler University’s Gregory Shufeldt says groups such as minorities and disaffected white voters were consistently underrepresented.

“Certain groups have a harder time being represented in the polls, and because of that there are certain assumptions pollsters have to make when they’re speaking about certain demographics or specific subgroups of the population. So the margin of error around certain projections, around certain races, is going to be larger than it might otherwise be," Shufeldt says. 

Shufeldt says polling models underrepresent certain groups and because of that pollsters have had to make an increasing number of assumptions about those demographics and subgroups – which also increases the margin of error around those projections.

But Purdue political scientist Josh Scacco says it’s alarmist to say polling is broken.

“Whether or not this is actually a crisis in terms of public opinion, I think it overblown by individuals and pollsters are going to look and assess the extent to which they need to change their methods to reach people that they haven’t been able to reach in their polls," Scacco says.

Scacco says pollsters are attempting new ways to get in contact with minority voters, in particular, to get clearer samples. He cites offering survey questions in Spanish as a way of reaching a more diverse group of the electorate.

Shufeldt says because this year’s polling was so poor, pollsters will face pressure to look at how they model the assumptions they make about certain groups.

“I absolutely think there will be more pressure put on pollsters and I think pollsters will be looking internally on how they model in the assumptions that they make about turn out rates particular groups but also the voter breakdowns of those groups,” Shufeldt says.