What Corruption?

Now there's something fishy going on here. Dreznel is arguing that the size of the contract has nothing to do with campaign contributions, because (for example) General Electric isn't getting more money than Halliburton, even though GE donated more money to the Bush campaign.

But this is attacking a strawman. Of course the Bush Administration can't literally give out money solely on the basis of contributions; there must be some surface plausibility to the awards. Thus Halliburton, in the capital-intensive oil and refining business, can more plausibly be given billions for its work in Iraq rather than, say, the 700 Club. Indeed, no matter how much money or support Pat Robertson may have given to Bush, it would be politically impossible for Robertson to be awarded $1 billion to hand out Bibles in Iraq . It just wouldn't fly. But to acknowledge this truism doesn't thereby prove that corruption is impossible; of course it's possible. It is still very suspicious that Halliburton has gotten $2.3 billion, and this is true despite GE's smaller handout of $5.9 million.

(Two related points on GE: One, is this the only government contract that GE is receiving? Does Dreznel think that GE is actually losing money overall, and is thus contributing to politics out of a sense of civic duty? Two, wouldn't it be a bit embarrassing if GE got over $3.2 billion, and yet the Iraqis still had no power? Again, some things are politically impossible. This doesn't prove the absence of Bushian corruption.)

Dreznel goes on to make one final, apparently 'scientific,' observation:

If the corruption argument is true, then the size of campaign contributions should be strongly and positively correlated with the size of government contracts.

Running the numbers, the good news for the Center for Public Integrity is that there is indeed a positive correlation'The bad news is, the correlation coefficient turns out to be 0.192 and not statistically significant . . . . An old joke among statistically minded social scientists is that 'the world is correlated at 0.3.'

Again, I must object. What we really want to test is whether the individual firms would have received such large contracts had they not ponied up money for Bush's campaign. No analysis of statistics will enlighten us of this counterfactual.

However, if we want to feel like objective scientists, the next best thing would be to isolate each contract, then run a regression on the money that various bidding companies got versus how much they contributed to Bush. This would give a much better indication of the level of corruption. (Dreznel acknowledges as much in a footnote, but thinks that what he did instead 'is still a fair test of whether there is systemic corruption.')

To understand my point, imagine that Halliburton has four competitors who bid on the contracts, and that they each gave Bush nothing. Imagine that GE has two competitors who also bid on the relevant contracts, and that they also gave Bush nothing. Etc. Now if it turned out that the Bush Administration picked the recipients of contracts on the basis of (1) a company that could actually fulfill the contract reasonably well and (2) the company that gave the most money to the campaign, then the results would be perfectly consistent with the numbers in Dreznel's Table 2. Of course I'm not saying that this is how the contracts were awarded, but the fact that such an outcome is consistent with Dreznel's low correlation coefficient just proves how irrelevant his regression is.

Perhaps I can drive the point home with a different example. Suppose that Bush gave every unemployed member of his family between the ages of 20 and 50 a job in the federal government. Let's say that there are 200 such people.

Now Noam Chomsky comes along and charges Bush with nepotism. In response, a professor from the University of Chicago runs a regression, with income from the federal government as the dependent variable, with years of education as an independent variable, and finally with another variable set to 0 if the person is unrelated to Bush, and set to 1 if the person is in the Bush family.

I submit that if he were to run such a regression, this professor would come up with a very small correlation coefficient on this last variable, because it would largely capture the difference (if any) in average education among Bush family members versus the population of government employees. (Indeed, if the average Bush had more education than the average federal worker, the correlation coefficient might even be negative!) The Chicago professor would then announce that a person's relation to Bush has very little explanatory power when it comes to a person's job status with the federal government. Furthermore, the professor could draw up a chart depicting the Top 10 Salary Earners in the Federal Government. Presumably Bush's second cousin (who can't multiply numbers bigger than 9) would not be at the top of the list, and hence (the professor would conclude) Noam Chomsky's clich'd charge of nepotism is unfounded.

In summary, the CPI's report is troubling indeed. The admittedly knee-jerk suspicion of the connection between money and politics seems to have some basis in fact, after all. Daniel Drezner's attempt to prove otherwise is just another example of bad statistical reasoning.

Your rating: None
Bob Murphy's picture
Columns on STR: 62