General Discussion
In reply to the discussion: Can we please kick Nate Silver to the curb [View all]hlthe2b
(102,239 posts)As we see in all of science, whether in conducting clinical trials or trying to differentiate populations and predict outcomes in polling. And that is for multiple reasons, but especially in the fact that choice of inputs is everything. By that I mean the common refrain: "garbage in, garbage out" in terms of findings.
Inputs in polling are not unlike inputs in the methods governing epidemiological studies and clinical trials. If you do not adequately account for bias in your sampling then you cannot ever hope to achieve a valid outcome.
More to the point, however, polling like such clinical trials are statistic-dependent. By that, I mean that there is always going to be the limitation of being able to reliably differentiate outcomes when the comparison groups are close. By that, I mean that a clinical trial attempting to show very small differences in outcome between two drugs will likely fail because the "power" (ability) of any study to show close differences is insufficient without enormous sample sizes, while a much smaller sample of participants can readly detect LARGE differences. Similarly, a poll can readily and accurately predict differences between very disparate populations sampled, while when the percentage of people voting differently is very close can fail to be accurately quantified. So a population that is so close to 50:50 is nearly impossible to poll reliably.
I understand the frustration with the polls seemingly failing us. Just as I've bemoaned the failure of a clinical trial to accurately show what I thought would be a significant difference in a new medical treatment. But this is the limitations of statistical methods, not a given person nor necessarily a given pollster.