Quote:
Originally Posted by mcsluggo
soooo, all in all you have a mixed soup.... and HAVE to control for many other variables to examine the issue. And as I said, I have seen studies where authors attempted to do so and found some level of positive coefficient on an education variable regressed against party affiliation (higher ed correlates with dem). But in the end the only reliable study is the simple fact that people that agree with ME are both well educated and intelligent, and those that disagree with ME are ignorant bafoons.
|
no, you don't have to control for everything else. At least not as the problem has been laid out in this thread. As a matter of fact, you shouldn't. Here, there was a hypothesis that Republicans target the uneducated, and that that targeting is evidenced by education and voting polling. If the hypothesis was that Republicans target the uneducated
after they control for income and age and gender and ethnicity and race and hair length and whatever else you want to control in order to get your answer (but that would be unreasonable even as a post-hoc, descriptive hypothesis. How could they possibly control for all that, then practically target the remainder?) - then you'd have a point. But when the original includes terms like "redneck" and "white voting blocks" all those variables are being considered causally conflated already, in the hypothesis. The primary variable, education, is the variable to test, and without controlling for all the others. To partial out all those other variables - which are assumed to be causally related - would give you a problem similar to the examination of data at the state level. Only, its that when you drill down to far, you are examining variation at a level to small to address your initial question, rather than to large for your initial question.