Donald Trump won the presidency in the wee hours of the morning of Nov. 9 to the shock of many people, including election forecasters. The poor quality of the big data used to create their forecasts might account for why so many people did not anticipate the election results.
Statistician extraordinaire Nate Silver stated that Trump would have a better chance at winning the presidency than the Chicago Cubs had at winning the World Series. Yet Trump was announced the winner, just one week after the baseball team broke a 108-year curse. Hillary Clinton was projected, by many forecasters, to win both the popular vote and the Electoral College vote. She won the popular vote 47.7 percent to 47.5 percent, CNN reported Nov. 10. CNN also reported that she lost the Electoral College vote 290-228.
The upset of Trump’s victory prompted many people who tracked the polls long before Election Day to wonder how several Electoral College forecasts could have been so wrong. According to Meta Brown, a professional statistician and business analyst, forecasters have a difficult job in analyzing state-level data to divine how the Electoral College may vote.
“The forecasters have some real challenges. The quality of state-level data is poor. And the Electoral College is volatile, meaning that a small change in popular vote within a few states can dramatically change the Electoral College outcome. Public opinion has also been volatile, with dramatic reactions to debates and news reports,” Brown said in an exclusive conversation with 21st Century State & Local. “Polls and forecasters correctly gauged the popular vote in Clinton’s favor. Forecasters told us directly that Trump had a chance to win in the Electoral College. It’s just that many people were shutting out that message.”
State polls frequently yield poor data because they are hard to fund and hard to standardize between states, Brown said. Various reports can occur if different people with different interests are spending varying amounts of money to run polls across all 50 states.
“A state poll requires about the same effort as a national poll. That means a good state poll costs about the same as a good national poll, and 50 good state polls cost about 50 times as much as a good national poll,” Brown said. “Pollsters have to eat, so they are limited to research that somebody with cash will sponsor. That’s not to say that there are no state polls, but that what remains is not necessarily up to the highest research standards, or consistent in methods from state to state.”
