Wednesday, July 22, 2009

Selecting and Confirming - A Fool's Errand

Read this story here.

Did you see the same pattern the author did? George P. Burdell, a man known for making significant patterns, didn't either. You are wise, too.

What you are witnessing in that article is a nationally distributed example of selection and confirmation bias, a fool's errand - looking for something that really is not there.

If you remove the outlier in each graph, the graphs pretty much no longer tell the story the columnist thought he was telling. This type of force-fit linear analysis does little for advancing critical insight. The first graph shows that there is no real favoritism being demonstrated. The columnist is just wrong. But so is Obama. No one (rich or poor) is really being helped significantly more than anyone else.

However, if you look at the last three graphs, the real story is that the stimulus is having little effect. There is virtually no response in unemployment and bankruptcy for stimulus dollars spent (at least up to the time this data is purported to represent); there is only a slight response to foreclosure rate.

So what happened? George and I can't judge the author's intention, but if I had to let my suspicions be made known, I would say the author went looking for a story, simplistically used linear analysis tools (which are highly sensitive to outliers), and found the story he wanted to tell. Unfortunately, he missed the real story that potentially demonstrates the greater policy implications.

We face a number of really important debates right now in this country. Obscuring clear thinking and true exploratory dialog will not be helped with this Fox correspondent's lack of rigor.

More Linear Thinking

I conducted a simple analysis about a week ago based on numbers published at Recovery.Gov. The site actually tells you how the Obama administration developed their forecast. (Editor's note: the links are apparently no longer available.) Essentially, they describe what amounts to a linear distribution based on the population of the states. I wanted to see if they were telling the truth and if there were any significant departures from it.

I scraped the jobs saved forecasts from the recovery website and divided them by state populations (rounded to nearest thousands) to develop the per capita forecast of jobs savings. Then I ordered the resultant per capita saving rate from lowest to highest.

The results show that, indeed, the stimulus money is being distributed (or at least, purported to be) proportionately to state population. The glaring exception is Washington DC, which is receiving funds at a rate 71% higher than the average and 76% higher than the median.


The thought behind the distribution of the stimulus funds is really no more sophisticated than a simple population pro rata.

The other interesting pattern that occurred is that the richest states are clustered among those receiving the higher rates of funds distribution, and the poorer states are clustered among those receiving the lowest (wealth being measured as reported at CNN). However, the clustering is not strong enough to warrant the claims of another recent article by Fox News that indeed richer states are being paid at a disproportionately higher rate.


Blue states are the poorest states, and red states are the richest.


This, combined with the USA Today article lead me to conclude the following. Except for DC, the Obama administration is following the pattern they described. (We might want to ask what kind of jobs are being saved or even created in DC. The answer is, of course, more bureaucrats.) Richer states are more industrialized states. Industrialized states are more urban. Urban populations disproportionately tend to vote for more interventionist/socialist policy makers. So what we’re seeing is not so much a direct nefarious attempt at political repayment as much as we’re see the self-selecting and compounding effects of interventionist policies securing interventionist representation. It represents the mechanism by which votes are bought on a large scale over time.

It also demonstrates that the poorer among us are not being lifted up at a higher rate by these direct interventionist activities. It has long been noted that interventionist/socialist programs do not benefit the people they claim to benefit. I wouldn’t use this analysis as proof to support that claim, but it does fit into the postulated pattern.

Finally, I have to say that I was really shocked at just how linear the distribution of funds is. I suspected that the description at Recovery for the rationale for disbursements was rough. I expected to see less correlation. Now, if the Obama administration had a econometric model that identified population as the key driving variable for stimulative effects related to employment, I could understand that. But they did not have any such empirically based model. Regardless, they forcefully asserted that their efforts would have the direct causal effects they sought. If we do believe that stimulus has causal effects, the effects were demonstrated to be embarrassingly strongly inverse, at least up this point; that is, if you believe the correlation is due to some underlying causal mechanism, even if it is the opposite of what the Obama administration postulated and forcefully advocated.

I don’t believe there is a causal relationship, direct or inverse, to the current stimulus efforts to recovery. I think the stimulus money has been absorbed more or less by state governments and dissipated in the oh so efficient way that governments work, and in parallel, the economic forces at work shed labor at a much higher rate than anyone anticipated would happen. In fact, the evidence is that the time rate at which the economy has produced unemployment is the fastest in this country’s recorded history.

Ultimately, the Recovery.Gov figures are less about jobs saved and more about the way money is redistributed (i.e., embarrassingly simplistic), nothing more.