How Whatsgoodly predicted the millennial vote
Exit polls match Whatsgoodly’s projection: millennials were split 60/40 between Hillary Clinton and Donald Trump
Trump’s victory embarrassed nearly every pre-election pollster. Even the British research company hired by Trump’s own campaign predicted his chances at less than 30% the day before the election.
Part of the upset was how many millennials (aged 18–29) voted for Trump when they were expected to vote in a landslide for Clinton. Compared to President Obama in 2008, she underperformed with millennials by 3 to 13 percent. SurveyMonkey even predicted that Clinton would sweep 88% of the electoral college if only millennials voted.
We at Whatsgoodly focus only on millennials. While 98% of our polls are created by users, we made election polls every few weeks up until election day. Our results were within negligible distance from exit polls conducted after the election (1). (Note that the exit polls surveyed 4,629 millennials; not as many as our 14,595, but still a very low margin of error, at 1.4%.)
It’s hard to know all the issues with the mainstream pollsters until the AAPOR completes its review in May 2017. Even then, there are too many variables to really isolate any of them. But aside from having a much larger sample, we postulate that our results were accurate for two reasons: content diversity and lower nonresponse bias. Let’s go through both of those now.
Dangers of the Echo Chamber
If you’re a liberal millennial, you probably saw this image shared all over your social media, claiming that millennials overwhelmingly voted for Hillary Clinton. It’s misleading. The data comes from the same SurveyMonkey pre-election survey mentioned earlier, and most shares were of a cut of the data dated October 25th. It doesn’t represent how millennials actually voted (the government doesn’t even release that data), and it isn’t from an exit poll asking voters to reveal their picks.
This is a great example of content homogeneity: a story that validates intense emotions will ricochet like a superball around newsfeeds that promote similar content. It’s a complex but engaging algorithm that powers most of the social media world today, including Facebook, Instagram, and Twitter.
Instead of feeding into these echo chambers, Whatsgoodly seeks to break them. Users connect to their school or community, and within those communities, they can anonymously poll each other and engage in open, honest dialogue without fear of backlash. Whatsgoodly is about understanding how your entire community feels; it’s about empowering the moderates to be heard over the fringes who yell the loudest.
Capturing More Perspectives
Pollsters are struggling to recruit panelists. Landline telephones — which are much quicker and cheaper to call than mobile phones — have been steadily disappearing over the last 20 years. Almost 50% of Americans only have mobile phones now. As a result, response rates on market research surveys have dropped from almost 40% in 1997 to just 9% today.
Millennials are the biggest generation ever, and they’re at the crux of this shift. But the survey industry has lingered in the landline world. Even the national exit poll used by all major media outlets this year was only able to survey early and absentee voters via the telephone!
When millennials in particular get excluded from polls, it causes nonresponse bias, the unfortunate situation in which people who aren’t being surveyed are systematically different from the people who are. Nonresponse bias may be the main reason for SurveyMonkey’s significant inaccuracy this year. Their Election Tracking poll is sent to a random sample of the people who’ve given SurveyMonkey an email address, thus excluding millennials who shun conventional messaging like email compared to in-app communication. If these millennials are systematically different citizens (and indeed they are), we have a problem.
This is where Whatsgoodly comes into play. Our userbase is almost 100% millennial, and response rates on most polls are 60–90%, depending on how relevant the poll is to the community. It’s the only truly mobile millennial panel and a fundamentally different way to collect public opinion. Most of our users don’t care about email, and polls at our high schools show that the upcoming wave of voters has almost completely abandoned it. We won’t know for sure until May, but our users in Rust Belt states may have had a critical role in our prediction’s accuracy. Many of our Trump-supporting users, for example, had reservations with revealing their opinions elsewhere:
The Largest Panel of Millennials in the Country
Our predictions come from our userbase as-is, so we’ll be transparent about what that looks like. Out of 38,602 U.S. millennials who saw our first pre-election poll (see map), 14,595 of them voted on it.
Our demographic makeup looks like this (2):
Full breakdowns by school and more than 400 demographic attributes are available for every poll on the Survey Dashboard. The language breakdown, for example, shows an interesting result: millennials who grew up speaking both English and Spanish were much more likely to prefer Clinton back in August, but other segments were much less decisive.
Given how much attention they get on the web, it’s hard to think of millennials as being systematically ignored… But this election was a gut-punch for the polling industry, and a fundamental restructuring is needed. Organic community insights could lead the way.
If you’re interested in learning more about this dataset or working with our panel, contact us: insights at whatsgoodly dot com.
(1) Exit polls came from Edison Research for the National Election Pool, a consortium of ABC News, The Associated Press, CBSNews, CNN, Fox News and NBC News. They are based on questionnaires completed by 24,537 voters leaving 350 voting places throughout the United States on Election Day, including 4,398 telephone interviews with early and absentee voters. According to Fox News, which subscribes to Edison, 4,629 of them were between the ages of 18 and 29.
(2) Respondents with a reported age of 30 or above included for completeness, but none of them were included in the pre-election poll results.