« James Fallows on the effects of debates | Main | Limbaugh: Obama "little black man-child" »

August 21, 2008


Has Andrew Gelman considered the variability within his model? The magnitude of the change was only around 0.2%.

The scale makes it look as if there was a dramatic change in the last 100 days. However, another way to view that graph would be that there was essentially no change during that period.

I agree with David. I think the graph is a bit difficult to interpret. The number of data points before day 120 are too sparse to be statistically significant and in the final 20 days the variance goes through the roof. Throw out the last 3 data points and it looks like the trend would remain nearly flat.

Couldn't we as easily interpret this as meaning that something significant happened in the last few days of the election?

The comments to this entry are closed.