by Arne Bathke

When elections are thrust into the media spotlight we are inundated by numbers, graphs, and figures. Every day, someone presents new poll results and predictions, statistical analyses, or interpretations. 

Candidates are trying to find out which target group to focus on with their efforts to sway voters. Journalists want to be the first in forecasting the outcomes of important races. Political analysts and scientists strive to comprehend the underlying political and sociological movements that determine election results.

Neither group can succeed without a good understanding of statistics. In fact, knowing your stats well gives you the extra edge over your competitors.

Statistics and politics have long historical ties. The term “Statistics” derives from the Italian “statista” which means “Statesman”, and it is commonly attributed to the German historian, jurist, and political scientist Gottfried Achenwall (1719-1772). He was one of the first who regarded it a science of its own to collect data about state, society, politics, and economics. 

Obviously, the field of statistics has grown much beyond the mere collecting of data. But it is still true that judicious data collection is one of the foundations of good statistical analysis and prediction. 

When looking at poll results, size (of the sample) does matter to a certain extent, but much more important is that the data are collected intelligently, using common sense – and the principles of randomized and unbiased sampling. Only if these principles are applied, you can quantify the uncertainty of the prediction, represented by a number called the margin of error. If two candidates are within the margin of error, then the small differences in their percentages could just as well be due to chance. A new poll might easily reverse the candidates’ order.

Also, remember to look who sponsored and who conducted the poll, and how the questions are worded – sometimes its only purpose is to persuade voters.

More sophisticated statistical methods combine election and poll results to investigate potential voting fraud, or include other demographic data to analyze complex associations between variables, and societal trends.

Motivated by the embarrassment of the 2000 US presidential election, a whole issue of the journal Statistical Science was devoted to voting and elections (Vol.17,4,2002). It not only demonstrates some of the things that went wrong in that election, but also how the statisticians found out about it.

Know your stats well. If you understand the basics of statistical inference, you will enjoy election seasons a lot better. And you won’t be fooled by the figures when looking at poll results.

Arne Bathke is a professor in the Department of Statistics.