Friday, November 06, 2009

Depending on Independents

Some people, including David Brooks, note that independents are more conservative than they used to be and that most of them voted Republican in New Jersey and Virginia. They conclude that the Democrats are in trouble.

This is definitely invalid reasoning. It is based on the assumption that the same people call themselves independents.

It is also a notable fact that the number of self described Republicans is at a record low. Some people who used to call themselves Republican might have switched to calling themselves Democrats but most switched to calling themselves independents.

I won't crunch the numbers (and I don't apologise because Brooks didn't either and I am crticizing him). So consider a hypothetical USA in which the only change is that people who used to call themselves Republicans are so disgusted with the Republican party that they call themselves independents.

If this is the only change, then the opinions on policy of the average independent would move right. Also, unless the disgust is unusually passionate, more independents will vote Republican.

Now I don't claim that this hypothesis is plausible. However, I do claim that people who interpert changes in the distribution of views of independents have not presented any evidence that it is false. More seriously, they haven't considered the effect of people shifting declared orientation from Republican to independent on the data they claim to analyse.

Confession: I read the Brooks op-ed *after* writing my critique. That is I took a blind folded pot shot at him. First, I note I hit him. He did not consider any possible change in the set of people who call themselves independents.

On the other hand, I missed the bull's eye. He wrote something so stupid that I didn't imagine it.

Liberals and conservatives each have their own intellectual food chains. They have their own think tanks to provide arguments, politicians and pundits to amplify them, and news media outlets to deliver streams of prejudice-affirming stories.

Independents, who are the largest group in the electorate, don’t have any of this.


He asserts that liberals and conservatives have something and independents don't. This logically implies that independents are neither liberal nor conservative. This claim is absolutely 100% refuted by the data. There are people who declare themselves to be conservative and independent and there are people who declare themselves to be liberal and independent. Therefore Brooks's claim is demonstrably false. His discussion of polling is as valid as the discussion of geography of someone who claims that the world is flat.

He contrasted independents with conservatives and liberals. This is pure total idiocy. It is possible for people to declare themselves both independent and conservative or independent and liberal.

The categories in the polls abused by Brooks are (Democrat, Republican or independent) and (liberal, conservative or moderate). It is just not at all true that all independents describe themselves as moderates. Brooks's discussion of polling is absolutely inconsistent with the polls.

His implied claim that people are either liberal or conservative or independent is plainly false. There is nothing subtle about this. He has ignored the reported results of the polls he claims to interpret.

update: This, like much of Brooks's work, shows what horrors arise when one writes about numbers using English. It is possible for totally false arguments to sound OK if they include numbers and discuss them in English. This, and not a problem with statistics, justifies the statement that there are "lies, damn lies, and statistics"

I think the problem is with English (and all other naturally evolved languages). It is very easy to base an argument in English (or Italian) on the assumption that all correlations are zero, one or minus 1. Brooks made a career discussing red states and blue states. He managed to argue, in effect, that rich people vote Democratic and not so rich people vote Republican, by noting that more people in richer states vote Democratic. Thus the typical blue state voter is a Democrat and richer than the typical Red state voter. A simple association -- probability of voting Republican increases monotonically in income was reversed by substituting "all" for "most" or "more".

Here again he confutes most and all. Most conservatives call themselves Republican. Therefor no independents are conservative. Most liberals call themselves Democratic. Therefore no independents are liberal. More people call themselves independent the Democratic. Therefore most people reject liberal solutions which most people in polls say they support (e.g. raising taxes on the rich and cutting them on the poor).

When I translate Brooks's arguments into quantitative claims about probabilities, correlations and expected values, it is blindingly obvious that they are total nonsense. Yet it is clear that they are not perceived to be total nonsense.

I think the problem is the mixture. Brooks switches from numbers to English and back in his writings. Sensible things can be written without any reference to numbers (I checked the collected essays letters and journalism of George Orwell).
The quantitative analysis of data can be useful.

Only suckers allow people to go back and forth.

No comments: