Publications & Resources

Modes and Misses

11/18/2022

Are the polls getting worse? 
 
In the U.K., the polls were, in the aggregate, pretty far off in predicting the outcome of the 2015 Parliamentary election.  A blue ribbon panel led by Patrick Sturgis concluded that “The opinion polls in the weeks and months leading up to the 2015 General Election substantially underestimated the lead of the Conservatives… In historical terms, the 2015 polls were some of the most inaccurate since election polling first began in the UK in 1945.”  The report by Sturgis and his colleagues also notes that the polls in the U.K. have often been off in their predictions of the vote shares for the various parties but those errors didn’t get so much publicity in the past because the polls correctly picked the winning party. 
 
More recently, the polls also seemed to miss the outcome of the Brexit vote in the U.K.  Most of the polls indicated that the outcome would be close and it was relatively close—52 percent voted to Leave versus 48 percent to Remain.  Still, most polls predicted a narrow win for Remain and thus got the outcome wrong.
 
By contrast, Nate Silver’s analysis suggests that there has been only a small downtick in the accuracy of polls in the U.S. in the last two election cycles.  For example, he shows the absolute difference between the projected margin for the winner (in percentage terms) and the actual margin was a little more the 9 percent in the 2016 Presidential primaries.  The average for Presidential primaries since 2000 was about 8 percent, so 2016 wasn’t that much worse than the typical year.  The average discrepancy in 2014 in the projected margin for House races in the general election was almost 8 percent, up from an average of about 6 percent for the period from 1998 to 2014.  For gubernatorial and Senate races, the discrepancies between the projected and actual margins was smaller—about 5 percent on average since 1998 and no worse in 2014 than in the previous years.   Still, as Silver points out, an average error of 9 percent is a big error; he calls it “whopping.”  One reason that people in the U.S. have not been so concerned about the accuracy of the polls is that, despite errors in the projected margin, the polls mostly picked the winner correctly—85 percent of the time in this year’s Presidential primaries, according to Silver.  That was not the case in the U.K. general election or with the Brexit results.
 
So what’s going on the U.K.?  Sturgis and his colleagues attribute the problems in 2015 to samples that “systematically over-represented Labour supporters and under-represented Conservative supporters.”  In her analysis of the Brexit results, Claire Durand points to mode differences as a possible factor.  The web surveys came closer to the actual results than the telephone surveys did.  Silver’s assessment indicates that the reverse may be true in the U. S.
 
The accuracy of the polls is clearly a big concern for everyone who does surveys.  Recognizing this, AAPOR has appointed an Ad Hoc Committee on 2016 Election Polling, headed by Courtney Kennedy, to assess the performance of the U.S. polls in the 2016 election.  Stay tuned to see what kind of marks the AAPOR committee gives to the polls in the current election.
  
References

Durand, Claire.  Brexit: why and how were we misled? The modes again.  Retrieved from http://ahlessondages.blogspot.com/

Silver, Nate (2016).   The State of the Polls, 2016.  Retrived from http://fivethirtyeight.com/features/the-state-of-the-polls-2016/
 
Sturgis, Patrick, Nick, Baker, Mario, Callegaro, Stephen, Fisher, Jane, Green, Jennings, Will, Jouni, Kuha, Ben, Lauderdale and Patten, Smith (2016) Report of the Inquiry into the 2015 British general election opinion polls.  London, GB, Market Research Society and British Polling Council.