Tuesday, September 28, 2010

Speaking of Kevin Drum

This post is brilliant too. I hadn't fully grasped the hisorical pattern of appeals to the constitution by right wing groups which grow when a Democrat is President (the latest of which supported George W Bush when he declared the constitution null and void).

On the other hand, this post includes an arithmetic mistake. Drum notes that 30% of people in a poll support the ACA and an additional 16% either oppose it or are undecided because it didn't go far enough (0% said they opposed the law and 30% said they didn't know) That means 46% either support it or don't support it because it didn't go far enough. Notably 40% oppose it. Even assuming that 0% opposed it because it didn't go far enough that means support or think didn't go far enough beats oppose 46% to 40%.


That leaves 54% who oppose all or most of the law. So you're still at 54%-46% opposed, and this is the best case since it's possible that making the law more liberal might also have turned some of the favorers into opposers. Of course, you can argue that this is still slightly better than 40%-30% opposed, but it's a pretty iffy thing.


This is just wrong. In the calculation of 54% Drum is counting people who "aren't sure" as opposed to the law. That's just not accurate. Also, in the comparison, he only counts the 30% who are opposed to the law not the 60% who are opposed or aren't sure.

The correct calculations are 46%% vs less than 40% (but not calculable given the published results) as opposed to 30% vs 40% or 46% vs 54% vs 30% vs 70%.

Drum correctly notes that the data can't be used to support the claim that a bill which went further would have had more popular support -- some who support the ACA would oppose a bill that went further. This is correct and settles the question of whether this poll can settle the question (it can't). But then he goes on to perform a incorrect calculation from the data which he correctly notes can't be used to answer a question, in order to get an answer to the question which he thinks is reasonable (that answer being "we don't know").

Drum might also note that in polling they never ask questions such that there is combination of answers given by a very small set of people. So if to count as a yes one can answer either the first question one way or the second one way, then the results are biased towards yes. The fact that the correct calculations show as high or higher support for yes or didn't go far enough than for yes is a tautology. This must be so no matter what people think. Drawing conclusions about public opinion from a tautology is not sound social science. Drum is right that some people used a tautology as evidence for their claim about public opinion.

By the way Drum suffered a lapse of reading comprehension. 14% or resondents whom Drum counts among the the 54% who, he says "all or most of the law" answered "I oppose a few of the changes in the law." "a few" does not mean "most" as definitely asserted by Drum.

Here Drum is contesting a conclusion which, he correctly notes, is not proven by the cited data. However, knowing the answer, he is sloppy about the intermediate steps.

His valid argument is very simple. We can't know how people would have reacted to a futher reaching reform, because there wasn't a further reaching reform. That argument is too brief for a typical Drum blog post, so he slipped into adding errors to flesh it out.

No comments: