The impacts of the economic crisis continue to be felt unevenly across the country. I’ve previously looked at the factors associated with higher rates of regional unemployment. But which places have seen the biggest jumps in unemployment since the crisis hit?
To get at this, my colleague Charlotta Mellander conducted a straightforward statistical exercise called a “residual analysis.” It’s a simple way to track how a location performs relative to the performance of all other locations. Basically, the analysis examines to what extent the initial unemployment rate in May 2008 seems to have had an impact on the change in unemployment over the last year. Technically speaking, Mellander ran a regression analysis predicting change in unemployment over this last year (May 2008 to May 2009) as a function of the initial level of unemployment at the beginning of the period (May 2008). She then compared the predicted values to the actual values.
The first graph shows the pattern for U.S. states.
The hardest hit states are ones that were doing badly even before the crisis hit. The fitted line is steep; the correlation between the two is 0.59 and significant; and the R2, 0.345. States below the line experienced a smaller than predicted increase in unemployment levels, while those above the line saw a larger than predicted increase.
Michigan has the highest unemployment rate, but Oregon (+3.0) has taken the biggest relative hit. Alabama (+1.8), Indiana (+1.6), South Carolina (+1.6), and Wisconsin (+1.4) have also taken bigger than expected hits. North Dakota has the lowest rate of unemployment but Alaska (-2.8), Mississippi (-2.1), Arkansas (-1.2), Connecticut (-1.2), Iowa (-1.1), and Nebraska (-1.2) have done better than expected.
The second graph repeats the analysis for U.S. metropolitan regions. It excludes two extreme outliers in California – Yuma and El Centro – which started the period with 20 percent plus rates of unemployment.
The hardest hit metros are also those that were doing badly before the crisis. The fitted line is again steep; the correlation coefficient is high, 0.59; and the R2, 0.351.
The crisis has hit hardest at smaller Rustbelt metros, especially those in Indiana: Elkhart-Goshen, IN; (+7.3); Kokomo, IN (+7.2); Decatur, GA (+3.2); Sheboygan, WI (+2.7); Fort Wayne, IN (+2.3); and Youngstown, OH (+2.2).
While Detroit has faced staggering unemployment, the difference between its actual and predicted unemployment is +1.6. Among large metros, Portland (+3.1), Charlotte (+2.2), and, San Jose (+1.9) experienced even bigger than expected increases in unemployment. Las Vegas (1.5), Boise (1.29), and Orlando (+1.29) have also been hard hit. San Francisco (+.93), Miami (+.49), L.A., Chicago (+.31), Atlanta, and San Diego (+.21) also performed worse than their May 2008 unemployment levels predicted.
Several Oregon metros took worse than expected hits: Bend-(+4.6), Eugene-Springfield (+3.8), Portland (+3.0), Salem (+2.5), Medford (+2.4), Corvallis(+1.9). Metros that border Oregon like Spokane, Washington (+0.8) and Boise, Idaho (+1.3) also have high differentials.
Three Texas cities – Dallas (-1.0), Houston (-0.9), and Austin (-1.0) – performed considerably better than expected. Minneapolis-St. Paul (-0.4) did too. Cities along the Bos-Wash mega-region – Boston (-0.4), D.C. (-0.3), New York (-0.1), and even Philadelphia (-0.3) – also did better than predicted. Surprisingly, Phoenix also outperformed expectations (-.2), albeit modestly.
College towns number among the best performers, doing much better than predicted: Champaign-Urbana, Illinois, home to University of Illinois (-2.2); Iowa City, University of Iowa (-1.81); Manhattan Kansas, Kansas State University (-1.82); College Station, Texas, Texas A&M (-1.74); New Haven, Connecticut, Yale University (-1.54); State College, Pennsylvania, Penn State University (-1.47); Boulder, Colorado, University of Colorado (-.93); Austin, Texas, University of Texas (-1.0); Ann Arbor, Michigan, University of Michigan (-.94); and Ithaca, New York, Cornell University (-.97), among others.
July 21st, 2009 at 11:05 am
What Portland and Oregon have in common with Detroit and Michigan is that their economies are heavy on manufacturing. Oregon is also a big lumber producer (officially considered manufacturing) and took a hit with housing and construction cutbacks.
My guess is that when the recovery comes (in whatever form), Portland will come back much faster than Detroit because we manufacture high tech and they manufacture cars. Microchips don’t need a radical overhaul while automobiles do.
The other thing affecting Portland is that people keep moving here without jobs. I don’t know the exact effect on the numbers, but it’s much different than the 1980’s recession when people were moving away, which reduced the number of jobseekers.
July 21st, 2009 at 1:21 pm
Interesting data. I’m surprised the Oregon college towns (Eugene and Corvallis) aren’t better insulated. Why aren’t those metros doing significantly better than Boise, Medford, and Spokane?
July 21st, 2009 at 1:33 pm
I wonder how useful the predicted values really are… being that they are based on a model that explains only around 1/3 of the total variation, suffers from heteroscedastity (note the spread of the data around the regression line as you move right on the x-axis), and very likely spatial dependence.
July 21st, 2009 at 1:59 pm
I am impressed with Philadelphia having a low change in unemployment. However, I would like to see how the city of Philadelphia compared with the surrounding counties. My studies on Philadelphia tend to show the city to perform much more negativly than the suburbs. What concerns me, is that only 18% of Philadelphia residents hold a college degree (22% is the PA average), and has an unemployment rate of 6%, higher than the PA average.
That being said, I do think the city is poised to be the next big city. It has the ingredients to whether the recession and come out stronger. Infact I think it will be stronger than some of its suburbs.
July 21st, 2009 at 4:34 pm
Would be interesting to see the Canadian equivalent. My gut says the line might go the other way: places that had the lowest unemployment (such as Alberta) are seeing the biggest gains to unemployment.
July 22nd, 2009 at 10:13 am
RS: Have you worked with models that include percentage change before?
July 23rd, 2009 at 8:30 am
Yes, but not in this particular context. However, I do still wonder how useful it is to draw conclusions about which cities are doing better than expected when the expection is based only on the intial unemployment level, no controls for industrial structures, geographic location, ect. Futher, the spread of the data hints at bias in the coefficient estimate, which could drastically change the position and slope of that line. Thats all I was suggesting.
On an unrelated point, I am curious to know which city had the intial unemployment rate of 0 and a change of 0?
July 24th, 2009 at 6:44 pm
[...] Kleen has a good post up today about the market stability of college towns. He references an article from The Creative Class which looks at unemployment rates in various cities. I’m going to stop short of saying that [...]
July 31st, 2009 at 10:02 am
[...] Where unemployment is worse than expected, by Richard Florida [...]