Fourth District Labor Markets: Cleveland’s Puzzling Data
The Great Recession had very different effects on regional labor markets in the United States, and not surprisingly, the recovery has also proceeded at different speeds across regions. Currently, very few U.S. states have employment levels higher than they were at the start of the recession, and no state has an unemployment rate lower than it was then. States hard hit by the housing bust—Nevada, Arizona, and Florida—have employment levels that are more than 7 percent below their 2007 levels, and only a handful of states have experienced positive employment growth over the period. Many of the labor markets experiencing net employment gains are in states whose output is heavily focused on natural resources, such as North Dakota, Texas, and Alaska.
Like the rest of the country, improvements in the Fourth District’s labor market have been relatively uneven. The Fourth District states of Kentucky, Ohio, Pennsylvania, and West Virginia are spread out across the employment-growth distribution. Ohio has had the lowest employment growth from December 2007 to May 2012 (−5 percent), while West Virginia is close to break-even growth over the period (−0.3 percent). Pennsylvania and Kentucky each have shed roughly 2 percent of their employment over the same period.
Unemployment rates have remained elevated across the country, with almost half of all states having unemployment rates 3 percentage points above their pre-recession levels. In Fourth District states, unemployment rates are between 1.5 and 3 percentage points higher than in December 2007. Relative to the nation, all Fourth District states have unemployment rates either at or below the national rate of 8.2 percent.
Within the Fourth District, large metropolitan areas are spread across the employment-growth distribution, with Cleveland having the lowest employment growth rate (−6.9 percent) and Pittsburgh the highest (+0.8 percent). Columbus and Cincinnati are in between at −1.4 percent and −4.0 percent, respectively. If we look at changes in unemployment rates over the same period, we see a markedly different pattern for the large metro areas of the Fourth District. Of the top 50 metro areas in the country, Cleveland has the smallest rise in the unemployment rate over the December 2007 to May 2012 period, and all four large metro areas in the Fourth District are in the lower tail of the distribution.
For Cleveland, these are strikingly different results. How can we explain such distinct differences in the paths of employment growth and the unemployment rate? One answer is that labor force growth in Cleveland has been relatively weak—declining by an estimated 1.4 percent over the December 2007 to May 2012 period. This weak labor force growth means that the Cleveland labor market did not have to add a large number of new jobs to reduce its unemployment rate.
However, the magnitude of the decline in the labor force is too small to explain the very large difference in the paths of employment and unemployment for Cleveland. A more likely reason for the difference is the fact that the data on unemployment and employment come from two different data sources, and these data sources simply do not agree in the case of Cleveland.
We can see this more clearly by comparing estimates of employment at the metro level from both sources. Unemployment rates are derived from data gathered from a survey of households (the household survey) augmented with administrative data and model estimates, and employment statistics are based on data from a survey of businesses (the payroll survey). While the household survey is used to construct the unemployment rate, the unemployment data also contain an alternative measure of employment. To be sure, employment is defined somewhat differently than in the payroll survey, and coverage and geography are not identical. Importantly, the data from the household survey include the self-employed and measure employment by place of residence, whereas the payroll survey excludes the self-employed, counts the multiple jobs of a single worker, and measures employment by place of work. Nevertheless, the comparison is informative.
At the national level, employment loss from the payroll series exceeds the household series by 0.8 percentage points over the December 2007 to May 2012 period. For Cleveland, on the other hand, the difference between the two series is substantial. Where the payroll statistics show a decline in employment of 7 percent, the household-based statistics show a decline of only 2 percent over the same period. Moreover, the path of employment from the payroll survey shows Cleveland making no progress in expanding employment during the recovery—whereas a solid employment recovery is reflected in the data from the household survey.
This wide gap in the paths of employment from the two survey programs is not present for the other large metropolitan areas in the Fourth District; however, for other metropolitan areas across the country such differences do exist. We can see this in the scatter plot below, which shows employment growth from December 2007 to May 2012 from the two surveys across the 50 largest metropolitan areas in the country. If a metropolitan area is near the line in the plot, it means that the surveys generally agree. If the metropolitan area’s data point is above the line, then employment growth in the household survey exceeds employment growth in the payroll survey. Alternatively, if the metropolitan area’s data point is below the line, then employment growth in the household survey falls short of employment growth observed in the payroll survey.
One can see that there are metropolitan areas above and below the line, but most metropolitan areas lie above it. This agrees with the national data, which show that household employment growth was somewhat above employment growth from the payroll survey. In the scatter plot, Cleveland’s data point is less of an outlier, but it still is on the upper edge of metropolitan areas that lie above the line. Las Vegas is the clear outlier in the chart. In the payroll survey, the decline in Las Vegas’s employment currently stands at a little over 12 percent, whereas in the household survey the decline is only 4 percent. These data paint very different pictures of the magnitudes of employment loss in Las Vegas over the period.
There are some other general patterns in the chart as well. All the major Texas metro areas lie well above the line—showing that estimates of household employment growth exceed payroll estimates in these locales. The opposite is true for metropolitan areas in New York. This suggests that there may be differences in data collection programs at the state level that systematically influence the observed gaps at the metropolitan level. However, in the case of Ohio, no clear pattern emerges. The data series from both surveys were in close agreement for Columbus and Cincinnati, while the Cleveland data series showed wide disagreement.
What lessons do we draw from these data? First, we know that the statistics from both data programs will be revised, so that some of the differences that we observe today may be reduced through the revision process. One should be especially cautious in interpreting movements in the near-term data, as these can be quite volatile in certain states. Second, analysts of labor market data should look at multiple data sources to characterize labor market developments in local areas. The data from these programs rely on statistical samples, and for local areas, sampling variability can be substantial. Data from multiple sources may help to paint a more complete picture of the local labor market.