We’re smack in the middle of election season, and we’ve seen it a dozen times: The polls predict the outcome of a primary, the votes are cast, and the poll results are, once again, proven wrong. It’s an exhausting and exasperating cycle. As a data journalist, Maarten Lambrechts is more concerned with how these numbers are reported on and analyzed down to the tiniest percentages.
In an act of goodwill toward the less, shall we say, statistically inclined portion of the population, Lambrechts created Rock ‘n Poll, a colorful data visualization that demonstrates—both thoughtfully and whimsically—how the discrepancies between poll results and election results actually come about. The short answer? Chance plays a major role in even the most carefully conducted surveys with large sample sizes.
In the interactive visual, Lambrechts turns the ultra-detailed polling process into a narrative, with an election in a candy-colored fictional country, PollLand, at its center. Each PollLand voter is represented by a colorful ball that falls from the sky to visualize a particular result as the poll unfolds. It’s easy to follow because the interactive graphics are paired with a clear, engaging text that swiftly takes even the dimmest PollLandians through the polling process with ease. And in PollLand, there’s not a single statistical formula—an enormous plus in my book.
By simulating the surveys that specialized polling companies conduct in attempt to predict election outcomes, Lambrechts demonstrates the outsized role of sheer chance. The sample size starts at 10, then 100, then 1,000. Finally, eight polls of 1,000 participants each are generated at the same time and compared side by side. Of the 64 polling results, only four were correct up to one decimal and several had deviations of greater than two percentage points. The margin of error is due to the randomness of the polling participants, which leaves a lot up to happenstance.
Lambrechts argues that scrutinizing poll results down to the decimal and not giving an expected error is misleading: It sets up unrealistic expectations for polling accuracy that just can’t be met. And it’s leading us to poll-result fatigue.