I get the sense that the folks at 538 have an aversion to making manual adjustments to the data feeding into their models (or in the case of the recall election, their polling avg.) because it feels like they'd be introducing their own biases into the model. But to your point, not making these adjustments allows for biased data to sneak in & gives the casual reader/the 538 analysts this false sense of fairness, since they're not making any ad-hoc adjustments/exclusions. There's certainly the case to err on the side of inclusion for edge cases, but I don't think the SurveyUSA was an edge case; given their recant, that should have been removed.
You are +/- 100% correct in your analysis re some polls are better designed and conducted than others. A lesson should be that pollsters need to be better educated. (Are there distinctions among highly educated white male pollsters and less educated pollsters, any color, gender, class? ;))
I get the sense that the folks at 538 have an aversion to making manual adjustments to the data feeding into their models (or in the case of the recall election, their polling avg.) because it feels like they'd be introducing their own biases into the model. But to your point, not making these adjustments allows for biased data to sneak in & gives the casual reader/the 538 analysts this false sense of fairness, since they're not making any ad-hoc adjustments/exclusions. There's certainly the case to err on the side of inclusion for edge cases, but I don't think the SurveyUSA was an edge case; given their recant, that should have been removed.
You are +/- 100% correct in your analysis re some polls are better designed and conducted than others. A lesson should be that pollsters need to be better educated. (Are there distinctions among highly educated white male pollsters and less educated pollsters, any color, gender, class? ;))