7/30/19 The Hill:
While looking into this phenomenon in 2016, I contacted a number of polling groups. I knew that they “weight” and adjust their samples to make them reflect certain demographics of the U.S. population. In simple terms, if their sample ends up with too many young people, they assign greater weight to responses from older people. The methods they use to do so vary and are arcane, to say the least. For example, ABC states that it “adopted iterative weighting, also known as raking or rim weighting, in which the sample is weighted sequentially to Census targets one variable at a time, continuing until the optimum distribution is achieved.”
But one of the most interesting things I learned had to do with one big factor for which they typically don’t “weight” or adjust. It’s one that I think is arguably among the most important when it comes to polls measuring political issues: political affiliation. In other words, the national pollsters I spoke with told me that if they end up interviewing significantly more Democrats than Republicans — which is often the case — they don’t necessarily adjust the results to try to make the sample reflective of the U.S. voting population.
Wow! You could have knocked me over with a feather.
No comments:
Post a Comment