Biased Sampling in Predicting a Presidential Election

A sampling error led to an embarrasing and wrong prediction.

Tags: methodology/statistics, politics/election/voting, quantitative methodology, sampling bias, sampling frame, straw poll, survey research, 00 to 05 mins
Year: 1989
Length: 1:03
Access: YouTube

Summary: A fundamental part of survey research is collecting a representative sample. This video illustrates the importance of this by showing how a biased sampling procedure led to a huge prediction error in the 1936 presidential election between Franklin D. Roosevelt and Alf Landon. Even though one sample size was enormous (more than 2.3 million respondents), it could not fix a fundamentally flawed sampling procedure, and incorrectly predicted that Landon would win the election. The sampling frame (the list of people from whom the sample was chosen) was built from lists of automobile and telephone numbers, which many low and middle income voters could not afford in the 1930's. Therefore, its sample reflected people with higher class backgrounds and "missed the larger pro-Roosevelt sentiment among low-income voters." In short, it did not reflect the population, leading to a very embarrassing and fatal error for the polling company (especially since Roosevelt won with 62% of the vote). This is also known as a straw poll, or a poll used to gauge political opinions that is conducted poorly or without utilizing proper sampling methods. At the same time, Gallup conducted a poll and correctly predicted the winner, based on a fraction of the sample size (50,000 respondents). Note this clip is from Program 14 of the Against All Odds series, and even more can be found here under the tab "Select a Video Unit." 

Submitted By: Bhoomi K. Thakore

---