How weather forecasting can help political pollsters

Sunday, November 13, 2016, 3:53 PM - The U.S. presidential election captivated the world, and, for months, had the eyes of the world glued to the polls to see which candidate was in the lead.

Most political pollsters predicted a Hillary Clinton victory; the likelihood she would prevail sitting between a 65 to 95 per cent probability. With those confident odds, it came as a colossal shock when Donald Trump was announced the next President of the United States, and the public quickly turned to scrutinizing political pollsters and their inaccuracy.

Predicting the future, as you can imagine, is a tough business with complicated factors, and if there is one group of professionals who understands these challenges, it's weather forecasters.

Similarities and Differences between Weather Forecasts and Pollsters

Interestingly enough, political pollsters and weather forecasters have some things in common: they both observe the present and predict the future.

Weather models rely on sources such as satellites, weather balloons, buoys, and aircraft to measure multiple parameters (like temperature, air pressure, and wind) throughout the lower levels of the atmosphere. The data obtained is irregularly spaced, so it must be processed through data assimilation to perform quality control before it can be useful. Computer models ingest the data as their starting point and the complex equations of fluid dynamics and thermodynamics are applied to predict the future state of the atmosphere: a forecast.

Image credit: Wikimedia Commons

Pollsters rely heavily on observation, primarily through telephone polling. They attempt to get an accurate picture of the voters' political views and extrapolate a result from there. You can already see a major flaw in this method: landline telephones are a thing of the past and many models use automatic dialing, which does not contact cellphones. On top of that, the number of successful calls is small compared to the hundreds of millions of voters. Therefore, they are using a small sample size to predict an outcome, which affects the accuracy of the prediction.

Original, via https://www.flickr.com/photos/85635025@N04/12894913705 under cc-by-2.0

It's clear a main issue for pollsters is the shortage of initial information, or 'observations.' So, how do they improve?

Applying Weather Forecasting Issues to Pollsters

Lack of observations isn't just a pollster issue; weather forecasts face the same hurdle. With the entire world to map and minimal observation points, models are forced to interpolate between these observations to help fill in the grid.

There are also certain parameters that are more difficult to measure than others, such as humidity (amount of water vapour in the atmosphere). Fortunately, by using relatively well-sampled parameters (like wind) and finding correlations between those variables, we can better model humidity.

This approach can also be applied to pollsters. Clifford Mass, a professor of Atmospheric Sciences from the University of Washington, suggests:

“Pollsters could select from a long list of potential 'predictors' - or parameters that could be combined statistically to estimate who would win the election. Examples might include:

• Demographics (ancestry, age, etc.)
• Unemployment rate (long term, short term, trend)
• Education level
• Facebook activity for each candidate
• Trend in economic activity
• Crime rate and trend in crime rate

...and many more.”

It's a fairly straightforward idea: Use well-known and easy-to-obtain information, then apply statistics to determine a relationship between these predictors, and calculate who is likely to win. In practice, it's significantly more complicated.

Would this really work?

Meet Allan Lichtman, a professor at the American University in Washington, D.C. He claims he has successfully predicted the past 30 years of presidential elections (apart from Al Gore). He even predicted a Trump win back in September.

Professor Lichtman uses a series of parameters or "keys" to decide the winner and does not depend on polling information at all. He has written a new book describing his "keys" titled Predicting the Next President: The Keys to the White House 2016, which you can refer to if you'd like more information.

In the end, whether political polling is deemed helpful or not, it appears the accuracy of their probabilities can be improved upon, and weather forecasting techniques could play a major role.

Sources: Cliff Mass Weather Blog

Header image source: U.S. Army