If you haven’t checked out Reuters’ American Mosaic Polling Explorer, please do. It’s a nicely put-together visualization (if I say so myself) that Mo Tamman and others at Reuters put together to show off the rich trove of US election polling data we’ve gathered over the course of this year.
The elections site has an explainer on how to use the explorer, but the gist is that we’ve been polling about 2,500 people a week online since January, which means that by the time the elections hit, we’ll have reached 150,000 people. That sample size, and the continuous and consistent nature of the polling, means we can drill down into ever-smaller communities and see what they think about any given issue with confidence that the sample is statistically valid. Or, as the explainer asks:
Want to know what white men in the South think about immigration policy? How African-American women older than 50 view gay marriage? Whether veterans and their families approve of the country’s foreign policy?
Check it out. It’s simple to select one of the dozens of questions we’ve posed to respondents over the year, then filter it by a huge set of possible demographic factors (gender, income, residence, sexual preference, religion – the list goes on). You can even run a cross-tab on a second question (ie, people who responded this way on this question had this response on this other question.)
At some points there won’t be enough respondents to give a statistically valid answer, and the system will tell you so. Otherwise it defaults showing the results in the most detailed way it has data for – weekly, month, or over the entire period.
It’s true that not everyone is a fan of online polling, and part of that antipathy is based on the fact that not everyone (or every American, in this case) is on line. But there are issues in more traditional telephone polling as well, not least the fact that fewer and fewer people have land lines. Ultimately, it’s all about how the people conducting the survey – in this case, research company Ipsos and us – have weighted the samples we have to make them as demographically representative as possible.
(As a not-entirely-complete aside, anyone interested in polling and the US elections should definitely check out Nate Silver’s 538 site, now a part of the New York Times. It’s a fantastic resource. His book, The Signal and the Noise is also excellent, and I plan to post on that soon as well.)
When done well, what online polling really offers is this kind of depth and breadth of coverage. Size does matter when it comes to figuring out valid samples, and having more respondents means we – and hence you – can drill deeper into more questions. Kudos to Mo for driving this project – and other, equally innovative ones – through the year.
It’s really immersive. Check it out.