What's Up with Political Math?

Matthias Shapiro at Political Math made a big splash on August 16 with a post purporting to show that Texas had had the best employment record since the official start of the recession in December 2007.  How big a splash? Well, Ross Douthat endorsed it in his New York Times column and Zach Pandl at Goldman Sachs (no link, but thanks to Pandl for sending more detail on the analysis) used the raw jobs numbers as the dependent variable for explaining why some states had better employment outcomes than others (energy industry, low exposure to housing bubble, and high tech employment, he concluded).

Shapiro was deluged with responses, including 452 comments on his blog post and no telling how many emails, including a couple from me. Shapiro had made a serious error in what he called "My Personal Favorite Chart," which was that he looked at labor force data and acted like it was population data. Since December 2007, "739,000 people fled into Texas," he said. In fact, he had no data on interstate migration at all, and was misinterpreting the change in the size of the labor force to be the same thing as interstate migration. One commenter, "delaustin," pointed him to a link for interstate migration data (http://www.census.gov/hhes/migration/), where we see, for example, that net interstate (including DC and PR) immigration into Texas in 2009 was 130,234. This, of course, was all migrants, not just members of the labor force, and less than half of Texas' total population is employed. As a result, his "Personal Favorite Chart" was GIGO.

An even more serious problem is that the concept of "employment performance" is extremely sensitive to the way it is measured. Shapiro asserted without argument that population growth is, at worst, "a good problem to have." Of course, since what he was actually measuring was labor force growth, I argued earlier that there was no clear reason to think that adding two unemployed people to the labor force for every new employed person was "a good problem to have."

Shapiro did argue that unemployment rates were not the best measure of employment performance because interstate migration meant that states were rewarded for having their labor force fall in size. Thus, his preferred measure was the ratio of employed persons in June 2011 to that of December 2007, i.e. raw job numbers. By this measure, it turns out from Pandl's study of every state that Texas came in fourth, after North Dakota, Alaska, and DC. (Shapiro says Texas did better than North Dakota; I can't account for their different findings without seeing more of Pandl's methodology.)

If we do use unemployment rates, however, we find that Texas' unemployment rate has deteriorated relative to the national average, rising from 88% of the national average in January 2008 to 92.3% of the national average in July 2008 (and it rose again in August). I showed that by an alternate measure of employment performance, ratio of unemployment rate in July 2011 to that of December 2007, Texas was slightly below the median, with the July 2011 unemployment rate 187% of the December 2007 rate. By this measure, North Dakota and Alaska held the top two spots, just as they did with raw employment numbers. DC, however, is right below the median, at 177%.

As I wrote a couple of weeks ago, the employment/population ratio is often used in place of the unemployment rate to measure employment performance. Since population is a component, it would seem that Shapiro would have less to object to with this metric. Felix Salmon showed that by this measure, Texas has performed worse than the U.S. average during the Perry administration.





Source: Felix Salmon

Two caveats are necessary here. First, Salmon uses a non-standard definition of the population denominator, including all persons regardless of age or institutionalized status. Second, even with this measure, we can see that since the recession began, Texas has performed better than the national average (compare the slopes of the two lines after 2007).

To find out much better, I used the standard Bureau of Labor Statistics definition of the employment population ratio, where population is the civilian, non-institutionalized population aged 16 and older, and took the ratio of that ratio for August 2011 (P for preliminary) to December 2007 (R for revised). As we can see, no state has gotten back to its December 2007 employment/population ratio, but North Dakota has gotten the closest.

                           Employment/Population Ratios by State (%)

State		August 2011 P	Dec 2007 R	Ratio of ratios
North Dakota		69.8	71.4	97.8
Vermont		66.1	67.8	97.5
Mississippi		54.2	55.8	97.1
Minnesota		66.8	68.9	97.0
Kentucky		56.1	58.1	96.6
Alaska		64	66.6	96.1
New Hampshire		65.6	68.6	95.6
Virginia		64.1	67.1	95.5
Nebraska		68.1	71.3	95.5
Maine		60.1	63	95.4
Texas		59.8	62.7	95.4
Kansas		64.4	67.6	95.3
South Dakota		67.3	71	94.8
Iowa		66	69.7	94.7
Tennessee		56.8	60.1	94.5
Massachusetts		60.2	63.8	94.4
Oregon		58.7	62.3	94.2
Oklahoma		57.6	61.4	93.8
Connecticut		61	65.1	93.7
New York		56.3	60.1	93.7
Ohio		59.1	63.1	93.7
Hawaii		59.5	63.6	93.6
Pennsylvania		57.6	61.7	93.4
Wisconsin		63	67.6	93.2
New Jersey		59.5	63.9	93.1
Missouri		59	63.7	92.6
Arizona		56.2	60.9	92.3
Arkansas		55.1	59.8	92.1
Montana		59.5	64.6	92.1
Louisiana		54.5	59.2	92.1
Wyoming		64.7	70.3	92.0
South Carolina		54	58.9	91.7
Rhode Island		59.1	64.5	91.6
New Mexico		55.8	61	91.5
Illinois		59.4	65	91.4
Florida		55.2	60.5	91.2
Maryland		61.4	67.3	91.2
Washington		59.1	65	90.9
Alabama		53.1	58.7	90.5
Idaho		58.7	65.1	90.2
West Virginia		48.5	53.9	90.0
Indiana		57.1	63.6	89.8
Michigan		53.5	59.9	89.3
California		55.3	62.1	89.0
North Carolina		55.3	62.4	88.6
Delaware		55.7	63.1	88.3
Georgia		56.5	64.1	88.1
Colorado		61.7	70.2	87.9
Nevada		56.2	64	87.8
DC		58	66.2	87.6
Utah		60.8	69.5	87.5

Source: http://www.bls.gov/lau/ststdsadata.txt

Here we find that Texas comes in 11th place. As noted, North Dakota retains the top spot; Alaska is 6th by this measure and does well on all three measures. DC, however, comes in 50th by this measure, rather than third or 27th. This underscores the difficulty in measuring "employment performance." Clearly, more analysis is needed.

Meanwhile, Shapiro promised on August 23 "to try to do a follow-up in the coming week that addresses more of the data." Since then, he managed to publish a long, multi-chart post on an unrelated topic on August 31, but has not delivered any follow-up. Here we are, more than a month later, and it is unclear if any will be forthcoming. I'm sure I'm not the only person curious as to what points Shapiro thinks still hold up.