Sunday, May 31, 2009

I Am Marla's Observations On Artificial Selection In Chrysler Dealerships

Ok. You fuck me, then snub me. You love me, you hate me. You show me a sensitive side, then you turn into a total asshole. Is this a pretty accurate description of our relationship, Barack? This most recent nonsense is only the latest schizoid break. As you are no doubt aware, conspiracy theories are flying to and fro suggesting that the list of dealers that would be confiscated/dissolved/appropriated/killed by Chrysler the Auto Task Force was politically generated. Most versions of this tale mention a sort of Nixon enemies list approach to dealerships and cite snippets of data on the mostly republican political contributions of closed dealerships. We didn't really want to believe that about you. Quips like "Dealer List Targets GOP" sound like dead cats meowing to us.

Such stories interest us, but all stories involving the Chrysler travesty and the pending GM daterape interest Zero Hedge. Zero Hedge is a bathtub of squirming desire and skepticism. Yes, we think there is a lot that goes on that someone isn't telling us (or you). Still, parsing the dealers through an enemies list to determine closures seemed entirely unlikely to Zero Hedge. If true it should (but probably wouldn't) be the end of any administration. That sort of rank thuggery should be the end of any administration, no matter what its denomination.

Regardless, even if some sort of preference was manifesting itself, Zero Hedge doubted that an "enemies list" would be the mechanism. It does seem that Chrysler was not particularly involved in the process. That's suspicious, but not damning without more. It is least political poison though. It is asking for trouble anytime economic decisions like this are made with tainted or potentially tainted political methods. This from Reuters citing Leonard Bellavia, of Bellavia Gentile & Associates, who represent some of the dealers being terminated:

"It became clear to us that Chrysler does not see the wisdom of terminating 25 percent of its dealers," Bellavia said. "It really wasn't Chrysler's decision. They are under enormous pressure from the President's automotive task force."

Control of the list would present the opportunity for a bit of mischief by the automotive task force. Still, Zero Hedge thought it much more likely a "crony" list would be employed, granting automotive pardons for the democratically connected or perhaps to habitual donors. Still, positive Democratic bias in this respect is probably more realistic than negative Republican bias. Megan McArdle expressed similar sentiments last week and Cody Hatch comments over on Megan's site:

Anecdotally, there are Democratic-owned dealerships which are not being closed, but which are seeing all or most of their competitors closed. Equally anecdotally there are profitable-Republican owned dealerships which are being closed. That's troubling, but as must be forever pointed out, the plural of anecdote is not data. Until - and unless - someone does some statistical analysis of both the dealers which are being closed and the ones which are being left alone, this is just a collection of troubling anecdotes. (Okay, a surprisingly large collection of troubling anecdotes, but still...)

We agree. So, as this is a matter of some gravity, we decided to do exactly this.

To start with, we pulled raw donor data from The Center for Responsive Politics / for the 2008 election cycle and extracted ~865 megabytes of 2008 individual contribution ("IC") cycle table entries. Pause here a second.

We cannot say enough about the The Center for Responsive Politics and its decision to open source not just the 2008 cycle data, but basically all the data they collect. Without this sort of disclosure and a public minded attitude, we might never get to the bottom of these kinds of issues. Zero Hedge would like to add that their telephone librarians are sharp, effective and adroit. They sent me right where I needed to be (my cave) and got me exactly what I needed to have (my animal guide).

CRP/, Zero Hedge's message to you: Slide.

As a consequence, and to the best of our knowledge, this particular output is the widest available dataset on contributions. We matched this data against two Chrysler dealer lists:

First, Docket #797 "Document #3" "Schedule of Designated Domestic Dealer Agreements and Cure Costs Related Thereto" (a list of dealers expected to survive).

Second, the famous "Exhibit A" document of dealers to be closed.

Each "Majority Owner" name was matched against donor names in the IC file. Where text matches were close enough we created new entries with address and zip code information to try and quality check our matches and avoid excluding matches with one or two letter typos from our dataset. Where possible we matched the dealship name and the "employer" field from the IC dataset.

Instead of doing this correctly in the first place with a parser written in C, we were lazy enough to run the matching/cleanup in ksh. Hours later we regretted our decision and switched to direct SQL queries. The IC list has in excess of 3.5 million entries. Crosschecking against 3000+ dealer records was an experience.

Once we felt we had a mostly-clean dataset (though this process continues), we matched dealer data against several variables including:

Dealer status: (Safe or Nuked).

The party affiliation of the recipient of the contribution (Republican, Democrat or Silly Party).

The Center for Responsive Politics' individual candidate identification number (CID) which we filtered for one of: Hillary Clinton, John McCain, Barack Obama or Other.

The donation dollar amount.

Zip code.

We ran binary logistic regressions across the variables. The results are interesting but the most dramatic was saved dealers v. donations by candidate and/or party.

This puzzled us. Why would there be an significant noticeable (we have rightly been called out for using significant here) and highly positive correlation between dealer survival and Clinton donors? Granted, that P-Value (0.125) isn't enough to reject the null hypothesis at 95% confidence intervals (our null hypothesis being that the effect is due to random chance), but a 12.5% chance of a Type I error in rejecting a null hypothesis (false rejection of a true hypothesis) is at least eyebrow raising. Most statistians would not call this a "find" as 95% confidence intervals are the gold standard for this sort of work. Nevertheless, it seems clear that something is going on here. Specifically, the somewhat low probability that the Clinton data showing higher survivability of Clinton donors could result just from pure chance. But why not better significance with any of the other variables? Why this stand out?

Then we got to thinking. Steven Rattner, the Car Czar, is married to Maureen White, one-time national finance chairman of the Democratic National Committee. What does Maureen do now? From her website:

Maureen White is currently Chairman of the Board of Overseers of The International Rescue Committee (IRC), a member of the North American Advisory Board for the London School of Economics, and a National Finance Chair of the Hillary Clinton for President Campaign. (emphasis ours)

That website looks dated, but you get the idea.

Again, we want to point out that our findings are preliminary and subject to change. But whatever the result, the Administration has made themselves very vulnerable by taking charge of the dealership closing decisions.

Plot Of Terminated Dealers (Republican Donors)
we haven't begun our in-depth geographic analysis yet, but the
concentration of these in Detroit (almost all of them) is interesting to say the least.

Plot Of Terminated Dealers (Democratic Donors)
clearly just by the numbers there is a rather short supply of these.

If a "pardon list" of a specific class of donors were being used, it would be consistant with the data we observed (noting, however, the significance issue). We would expcet to see no seriously significant correlation to any of the catagories except the Clinton donors. What we can say about the data more difinitively is that it seems pretty clear that the overall democratic donors list was not used. It is also not self-evident that an "enemies list" of republican donors was used to pick beheddings. If a target quota for closures (either $ or number of dealerships) was in use, one does wonder, however, how the difference would be made up if a "pardon list" was utilized.

We also regressed against amounts donated, and found very little in the way of interesting relationships. Again, this would be consistant with the use of a list of donors, and not a dollar weighted ranking system (which would require significant effort on the part of the tabulator of such a list).

Did Maureen White pass a list of Clinton donors to her husband for preferential treatment in the Chrysler dealer beheddings game? We don't know. "Maybe" is about the best answer we can give with this data. But "Maybe" should never have been an option anyone even had to wonder about.

Again, please remember:


There is still clean-up to do in the original data. Zero Hedge believes there are some doubled entries still, but we also believe that the data is unlikely to change substantially even with these corrections. There are a number of other issues I won't belabor for us to cope with before we will call our results "final." Normally Zero Hedge would not post preliminary results, but given the early results, we felt it important to post them immediately with this disclaimer.

Zero Hedge is currently mapping zip codes to average income statistics to try and reverse engineer any local affluence factors that might have gone into the decision, and to correct for any bias in the data on this basis. We will also be taking a very detailed look at geographies and looking for new ways to display our dataset. Suggestions, of course, are most welcome.

Zero Hedge is contemplating posting the entire dataset and analysis for peer review once the data scrub is done and we rerun the numbers. Stay tuned.

Oh, and by the way, you're not getting this GM bond back. We consider it asshole tax.

UPDATE (05/31/2009 2:54 pm et):

1. Like it or not, using 95% as a confidence interval is totally arbitrary. You may not be concerned by a 12.5% chance of a Type I error. We think it significant enough to make us quite a bit more curious.

2. A word about multiple experiments:

We found what I will call the "Clinton Effect" after running the data in separate regressions just with Clinton, Obama, McCain. The rest of the variables we added in later testing. One could make the argument that Zero Hedge was "data mining" or "fishing" with multiple experiments eventually bound to find something. Readers will have to judge the import of this observation for themselves.

Zero Hedge would like to acknowledge the generous assistance of: "J" of the Quad Processor Monster, "B" from Desert parts unknown, "A" of jet-setting dinosaur fame, and "F" security guru extraordinaire. Sphere: Related Content
Print this post
blog comments powered by Disqus