Doug Rivers on Polling

Margins of Error

Polling margins of error often mislead due to unaccounted weighting distortions. Doug Rivers highlights that many polls report margins of error based on simple random samples, ignoring the complexities of weighted samples. This oversight can significantly inflate the actual margin of error, leading to unreliable predictions. Russ Roberts finds this practice troubling, noting that it undermines the credibility of polling data 1. Rivers explains that the actual margins of error are much larger than stated, especially in state polls with small sample sizes 2.

   

Weighting Issues

Weighting in polling attempts to correct sample biases but introduces its own challenges. Doug Rivers explains that automatic dialers often over-select certain demographics, requiring adjustments to achieve a representative sample. However, these adjustments can add variability, complicating the accuracy of the results 3. Rivers notes that even with weighting, systematic errors persist, as some groups are underrepresented by factors of five to ten, necessitating significant weighting that can skew results 4.

   

Systematic Errors

Systematic errors in polling often stem from uncorrected skews in sample representation. Doug Rivers observes that these skews account for the majority of polling errors, as seen in the New Hampshire primary where no poll accurately predicted the outcome 5. He explains that low response rates and outdated weighting methods exacerbate these issues, as media polls often rely on antiquated statistical techniques 6. Rivers emphasizes the need for modern statistical methods to improve polling accuracy.