One of the earliest applications of regression discontinuity design was Holmes (1998). It compared the employment share of manufacturing for counties near a border that divided a right-to-work state from a non-right-to-work state. The article used data from the 1992 County Business Patterns. This assignment will update several key tables from Holmes (1998) using data from 2011, the year before Indiana and Michigan enacted right-to-work laws.
Procedure and Details
Prepare the Census data by downloading the 2011 Complete County File from the County Business Patterns download site. The data file will give employment and plant counts for every industry. Only the numbers associated with all manufacturing (naics=="31----") and total employment (naics=="------") are needed. Some employment counts are censored to zero to protect confidentiality; these counts will show an employment value of 0 and a positive establish count (est>0). Create a variable that estimates employment by summing the product of the number of plants in each size category and the minimum employment for the size category. (Specifically, let estimatedemployment= 1*n1_4+ 5*n5_9+ 10*n10_19+20*n20_49+ 50*n50_99+ 100*n100_249+ 250*n250_499+500* n500_999+1000* n1000. This will underestimate employment in censored counties and thereby introduce measurement error. For the purposes of this assignment, these estimates are close enough.) Use the actual employment when available and the estimated employment for censored counties. Calculate the manufacturing share of employment by dividing manufacturing employment by total employment. (If no manufacturing plants are present, County Business Patterns will not have an observation line for manufacturing. The manufacturing share in these counties is 0.)
Download county distances to the state borders (cntydist.asc) from Thomas Holmes's online data appendix. The borders that will be analyzed have a border index numbers of 103 8 6 80 70 75 76 36 40 47 37 78 55 57. Merge in these distances to have a data set ready for the analysis. Virginia may be dropped to make merging data easier. (Virginia has independent cities that complicate data merging. One solution is to use sum Census data up to REIS counties. An easier solution is to ignore Virginia. Other data anomalies that complicate merging are the renumbering of Dade County, Florida's FIPS code, different FIPS codes for counties contained in national parks, and a new county created in Colorado. All these may be ignored for the purposes of this assignment. Employment registered outside of counties (with county FIPS code of 999) also may be deleted.)
Find the average manufacturing share of employment within 25 miles of the policy border state by state in 2011.
Find the total average manufacturing share of employment on either side of the right-to-work border in 2011.
Use only counties within 100 miles of the border, regress manufacturing share of employment on an indicator for being in a right-to-work state.
Answer the following questions:
- Why not use data from every county instead of limiting the analysis to counties near the right-to-work border?
- Does this exercise identify the effect of implementing a right-to-work law? What effect is being identified?