Via email, Alan Abramowitz sends a plot showing that President Obama's approval levels with various demographic and political groups in the January 17-23 Gallup poll are almost perfectly correlated with his 2008 vote share among those groups:
Abramowitz's finding is consistent with exit polls which showed that the swing against the Democrats in 2010 was relatively uniform across demographic groups. The implication is that politicians and the media spend too much time worrying about appeals targeting specific groups and too little on what moves support up and down across groups (principally, the economy).
To show that the swing from 2008 to 2010 was uniform, ISTM you'd want to show that support among each of the groups had dropped by the same %. Then the line of best fit would go through the origin. That is, deviations from the line of best fit going through the origin would be a measure of uniformity of swing.
However, perfect correlation using the unrestricted line of best fit for the points shown would not prove a uniform swing. Imagine that Reps and Cons had substantially increased their approval, while Dems, Libs, and Blacks had substantially decreased their approval. The line of best fit would meet the y-axis at a point well above zero. The slope would be well below 45 degrees. R-squared might still be near 1, even though the President had made substantial gains among Reps and Cons, while taking substantial losses among Dems, Libs, and Blacks.
Posted by: David in Cal | January 24, 2011 at 04:44 PM
All true, but see the link above for the '08-'10 swing, not this post.
Posted by: bnyhan | January 24, 2011 at 04:49 PM
The lack of a gridlines on that graph made it difficult to read. However, the actual points on the graph don't all support the uniform swing.
It looks to me as if about 10% of Reps voted for Obama, but 14% now approve of him; 20% or 21% of cons voted for him, but 28% of cons now approve of him. On the other hand, 97% of blacks voted for Obama, but only 89% now approve of him; 90% of libs voted for Obama, but only 75% now support him.
I'm surprised that Obama gained ground with Reps and cons, but that's what the figures seem to show. Anyhow, if Obama went from 10% to 14% among cons, and from 90% to 75% among libs, that's not a uniform swing.
Posted by: David in Cal | January 24, 2011 at 08:50 PM
I'm confused by this talk of the intercept. If R^2 = 0.98 wouldn't 98% of the variability in approval be explained by 2008 vote?
It seems there are two parameters to the linreg. 1) Increasing the slope shows voters are more polarized towards their views. 2) Shifting the line up/down would be nationwide trends of Obama.
I like the lack of gridlines. It's chartjunk.
Posted by: JP | January 25, 2011 at 12:32 PM