On Oct 10, 2006, at 11:57 AM, Miles Jackson wrote:
> I use Doug's Bush approval-gas price association as an example of
> bivariate regression in my methods course. It's an uncanny, strong
> relationship. However, as gas prices fall, the trend doesn't seem
> to be following Doug's regression line. Bush's ratings are back
> down to the low 30s, when the regression model predicts a
> significantly higher approval rating. Residual error, or
> meaningful shift in the trend?
When a model stops working, maybe the numbers are telling us something (though I used Gallup's approval rating, and the latest is 37, not the low 30s, like the Times/CBS poll - for some reason, maybe Gallup's Republican bias vs. the liberal NYT/CBS bias, Gallup's numbers are consistently higher). Something - Foley et al - seems to have shifted against the Reps in the last week or two.
Doug