New data from the Quarterly Census of Employment and Wages was released since I last wrote about comparing the more timely survey data to the “true” but lagged census employment data.
With the addition of data for January, February, and March of 2014 we can see the pattern of undercounting jobs continues in the survey, although the margin is somewhat smaller starting in 2014. Has the survey become more accurate? Well, each year in February the survey estimates are refined through a benchmark process. Past years’ estimates are revised using the latest employment census data, and the model used to create the estimates going forward is recalibrated using the new information. As a result, we tend to see a pattern in the survey estimates where early months are fairly accurate but as time passes the accuracy wanes. In March for example, the difference between the census employment count and the survey estimate is only 300 which is essentially decimal dust when you are tracking 1.3 million jobs. But March is estimated only a few months after the model is benchmarked.
The June jobs estimate is 1,355,900, 3.5 percent growth above June of 2013. Now that we are out six months from model benchmarking we have to wonder: how accurate is that estimate? Forecasts based not on surveys but generated from models of historical employment patterns estimate June employment growth to be between 3.4 and 3.5 percent. This leads us to believe that even with half the year under our belts the survey estimates could still be on track. On the other hand, digging into the detail we see some signs of potential overestimating. Take for example construction employment. The most recent census data (March) shows construction employment at 73,325 which constitutes 6.7 percent growth over March of 2013. The June survey estimate has construction employment at 82,300, growth of 9.2 percent year over. As the weather warms up from March to June it is commonplace for construction employment to ramp up. But has it grown by almost 10,000 positions in three months? It seems more likely that the June estimate is slightly high.
Ideally, the monthly survey data would accurately estimate employment growth across the state and for all industries. But no statistical model or survey is perfect, so assessing the model’s accuracy whenever possible is essential.