Election 2016 Social Desirability Bias / Clinton Affirming: Benefiting from Bias
Normally, Social Desirability Bias is associated with a negative bias – people claiming not to support something when they actually do or people claiming to not do something when they actually do. But, topics or people can also benefit. This is the case for Clinton.
She is the first female US presidential candidate from a major party and is also standard-bearer for the Democrats, a party that emphasizes Identity Politics focusing on minority groups. She also has one of the highest unfavorable ratings of any candidate in history. These facts create a situation where many might not want to support her but feel social pressure to do so especially if they belong to one of the demographic groups that have been the targets of the Clinton campaign and long-term supporters of the Democrats, namely women and minority voters.
Women strongly tend to vote Democrat and have much higher turnout ratios than men. This year, however, the hope of Democrats is that they can attract even more women to vote Democrat while creating an environment for higher female turnout as well. One of the strong undertones of the Clinton campaign is that, by becoming the first US woman president, the last glass ceiling will be smashed. All you really have to do is look at Clinton’s twitter stream to notice the emphasis on women’s firsts for this to become apparent.
Many agree that it would be a positive for the country to elect a women, just as many believed that electing Obama had fringe benefits of breaking past racial stereotypes. In fact, parallels are frequently drawn between these two (potential) firsts.
Women must feel social pressure under these circumstances to support Clinton. The questions are how much social pressure do they feel and is it enough to make them change their minds?
We can tease out the level of Social Desirability Bias in favor of Clinton by comparing poll numbers from live and anonymous polls. As explained in other posts, live polls depend on live interviewers using phone calls to collect data whereas anonymous polls use more anonymous methods like ‘robocalls’ / IVR and internet collection. People tend to answer more honestly in anonymous polls, so by taking the difference between live and anonymous polls we can estimate bias levels.
Taking an average of female support for Clinton from live polls (CNN, McClatchy-Marist, and Quinnipiac University) and comparing it to an average of anonymous polls (Reuters, PPP, and The Economist/YouGov) we come up with Clinton receiving approximately 6 percentage points more in the live polls. This implies that on average women tend to superficially support Clinton to a far greater extent in the presence of another human than when alone answering the same question. The swing of 6 percentage points is extremely significant and could easily change the course of the general election. For instance, such a swing in the female vote in 2012 would have made the election too close to call. Another interesting side note is that the average percent of the female vote from the anonymous polls shows Clinton receiving approximately the same support as Obama did in 2012 – in other words, more or less what would be expected if the populace were voting in a completely gender-neutral manner.
Though the size of the swing could be surprising, it should not surprise anyone that women feel social pressure to support Clinton. This trend did not begin with the general election against Trump, but in fact when she was battling against Sanders. A number of well-known female leaders and feminists including former Secretary of State Albright and Gloria Steinem rebuked younger women for supporting Sanders and not Clinton. Such calls for women to support Clinton have continued and were very apparent during the DNC with many calling for the nation to smash the last great glass ceiling of the US presidency.
Minority, especially African-American, voters have been extremely loyal to the Democrats. The extent of their loyalty is truly astounding. In 2008 and 2012, over 90% voted for Obama in comparison to his Republican rival. These numbers are unprecedented. Such high numbers imply that on some level there would be social pressure, perceived or real, to continue his legacy – that is, to vote for Clinton.
When comparing the live polls to anonymous polls for African-American support of Clinton, we come up with similar conclusions to those for women. On average, live polls were approximately 10 percentage points higher than anonymous polls for African-Americans supporting Clinton. This swing is extremely large and would imply that many of the current live polls are clearly biased in favor of Clinton. In reality, such a swing fits in quite well for forecasting. Live polls actually show increased support for Clinton in comparison to actual results for Obama in 2012. This really does not pass the common sense test. The 10 percentage point decrease brings the percent of African-Americans supporting Clinton to within a half a percentage point of what Kerry, the last Democratic nominee before Obama, received in 2004. In other words, it more or less makes sense for the African-American vote to jump higher for Obama, the first US African-American president, and then to slide back to levels considered the norm before his arrival – this is what the anonymous polls are implying.
Summarizing, anonymous polls imply that Clinton will gain approximately the same female support that Obama received in 2012 and approximately the same African-American support that the Democrats received before Obama. These polls point to voters analyzing the race in a gender-neutral fashion and the Democrats losing the Obama-bonus received when he was the active candidate. Such statements seem acceptable judging from the data available.
The big surprise is the size of the Social Desirability Bias bump that Clinton seems to be receiving in live polls. Such swings are so large that further study is needed. Future polls should focus on the topic of how much social pressure individuals feel to support or not support particular candidates. Without further data, we will have to make approximate adjustments to polls and wait for the election results to finally confirm or deny the size of these biases.