A Shift-Share Analysis of the MEGA Program

We commissioned a new statistical analysis of the MEGA program for this study, this time using a different measuring technique. A new approach seemed appropriate given that the results of the previous analysis were so conclusive.

MEGA is meant to create jobs, particularly in manufacturing, so Hicks (mentioned above) employed statistical regression and a longstanding technique known as "shift-share" analysis to evaluate the relationship between a county's manufacturing employment and the dollar value of the MEGA tax credits awarded there.

"Shift-share" analysis is useful because it begins by recognizing that local employment figures may be influenced by trends in the surrounding region. For example, a decline in Oakland County manufacturing jobs may be partly attributable to declines in total statewide employment and partly attributable to statewide trends away from manufacturing jobs and into other business sectors. Hence, only some of the decline in Oakland County manufacturing employment may be due to factors peculiar to that county. In such a case, shift-share analysis would use a basic mathematical formula to calculate what percentage of the decline in Oakland County's manufacturing jobs is reasonably attributable to each of three possible causes: changes in total statewide employment, changes in the statewide mix of manufacturing jobs and changes in manufacturing employment peculiar to the county.[*]

In our new analysis, this last factor — changes in manufacturing employment peculiar to the county — was computed by Hicks for all Michigan counties from 2001 to 2007. He then regressed these county-specific manufacturing job changes against the dollar value of the MEGA manufacturing tax credits that businesses in each county had been awarded from 1995 through 2000. In other words, through shift-share analysis, he isolated and removed state employment trends that might mask the effects of a MEGA tax credit in each county, and he then determined whether there was a statistical relationship between the MEGA manufacturing credits awarded and the county's manufacturing job counts over a six-year period afterward. Ultimately, if MEGA manufacturing credits help stimulate an area's manufacturing job growth (as opposed to growth at just one company), this result should show up in the manufacturing job growth peculiar to the county.

From 1995 through 2000, there were 107 MEGA deals, and the life of the credits ranged from five years to 20 years into the future. For the purpose of this analysis, only deals that resulted in tax credit awards prior to 2001 were included. If a deal was approved by MEGA in 2000, but operations did not begin until 2003, the deal was excluded from the study. To include such deals would have been to demand that MEGA credits have an impact even before a company received them; excluding those deals meant that only the cases where the credits were successfully awarded were being mined for evidence of impact in the years that followed.

In fact, a statistical relationship between MEGA manufacturing tax credits and county manufacturing employment did emerge, but unfortunately, the relationship was negative. Hicks found that from 2001 to 2007, every $1 million in MEGA manufacturing tax credits awarded in a county was associated with the loss of 95 county manufacturing jobs. This result was strongly statistically significant. A subsequent statistical "t-test" also indicated a very high probability that the relationship between MEGA credits and manufacturing employment was in fact negative - not positive or zero. A complete discussion of the statistical results appears in "Appendix C: Technical Appendix for Shift-Share Analysis."

Hicks' findings are especially troubling given that his methodology was designed to avoid accidental negative relationships. By tracking individual MEGA-related projects at the county level, Hicks effectively precluded a chicken-and-egg problem that can occur with MEGA-style development incentives, which sometimes intentionally target payments to distressed areas. Crude statistical analysis might have produced the spurious conclusion that MEGA was ineffective simply because it made payments to firms in areas where job losses were already mounting. Hicks' shift-share-based analysis, however, would not.[†]

[*] More generally, shift-share analysis assigns a change in local employment in a business sector, such as manufacturing, to three possible causes: trends in total regional employment, trends in that business sector's portion of regional employment, and specifically local trends for that business sector. (Technically, this final factor is the rate of change in the local business sector's employment relative to the rate of change in the regional business sector's employment.) The local area might be city, county or state; its "region" might be a county, state or nation. The shift-share equation appears in "Appendix C: Technical Appendix for Shift-Share Analysis."

[†] Note that Hicks and LaFaive's 2005 study also corrected for this concern.