Polling Lessons from 1936 Presidential Election

Polling results vary depending on what sample is used — which is why in 1936, pollsters predicted Franklin Roosevelt would lose in a landslide.Political pollsters’ blood pressure always rises on Election Day as voters prove their surveys are accurate or not. For example, the Literary Digest predicted a landslide victory for the Republican presidential candidate based on its research, but the Democratic candidate won by a large margin. 

That was the 1936 race between GOP candidate Alf Landon and Democrat Franklin D. Roosevelt.

Not much has changed during the past 76 years. Numerous pollsters predicted a Mitt Romney win in 2012, but were off by wide margins. The outcome of both elections highlights how polling results can lead to incorrect projections if the survey methodology doesn’t adapt to change. 

Based on an analysis by NY Times polling guru Nate Silver, polling firms that used traditional telephone polls with landline-only samples had results skewed to the GOP. 

On the other hand, telephone polls that included cell phone numbers and random Internet-based surveys were more accurate in projecting the election’s outcome.

 Reasons why newer sampling techniques trump traditional methods include:

  • More than three in 10 households in the United States don’t have landlines. The demographics of cell-only households tend to be younger and urban and people of color — groups that voted heavily for President Obama, according to an MSNBC analysis.

  • Online survey techniques have improved considerably. Sample designs are representative. More than 70 percent of American households have Internet access.

  • According to Silver’s analysis of 90 polling firms, four of the 10 most-accurate used Internet-only survey techniques, five used telephone surveys that included cell phone numbers and one used a robo-call technique.

Samples used in the Roosevelt/Landon race created off-the-mark projections. The 1936 Literary Digest survey used a sample design that relied heavily on telephone book listings and motor vehicle registration. Remember, this was the middle of the Great Depression. People with telephones and cars tended to be Republican-leaning voters, thus results skewed to the GOP. George Gallup and others used a different survey sampling technique that proved to be more accurate and reliable.

Ironically, Gallup’s research for the 2012 election was among the least accurate in forecasting the outcome of Obama/Romney election, according to Silver.