Self-employment and trade shock mitigation
Online DataAppendix
A1. Industry structural control variables in the lagged year 1990
As discussed in section 3.3, we need to control for lagged regional industrial structure of counties. We calculate the employment shares of the ten SIC divisions in each county for the year 1990 from the County Business Pattern dataset. The division is the highest level of the SIC classification, and includes these ten industry groups:
Table A1. SIC divisions
Division / SIC Code Range / Industry TitleA / 01-09 / Agriculture, Forestry & Fishing
B / 10-14 / Mining
C / 15-17 / Construction
D / 20-39 / Manufacturing
E / 40-49 / Transportation, Communications, Electric, Gas & Sanitary Services
F / 50-51 / Wholesale Trade
G / 52-59 / Retail Trade
H / 60-67 / Finance, Insurance & Real Estate
I / 70-89 / Services
J / 91-99 / Public Administration
Source: United States Department of Labor.
In our model, the measure of trade shock ∆IPWis derived based on the local industry specialization of traded sectors, as shown in equation (1). This means the information on the structure of traded industries has already been incorporated into the trade shock measure. And the trade shock ∆IPWcan be expressed as a function of the employment shares of traded sectors. Thus, in order to avoid strong multi-colinearity, we aggregate the three traded sectors (A, B, and D) of Table A1 into a single sector of "Traded sector". Table A2 shows the adjusted eight SIC divisions that we include in our model as industrial structure control variables as well as their descriptives for metro and non-metro counties.
Table A2. Industrial structure control variables and descriptives (1990)
(% of county employment)
Industry Title / Metro / Non-metroM / SD / M / SD
Traded sector / 26.13 / 13.90 / 28.91 / 17.08
Construction / 6.55 / 4.26 / 4.85 / 4.13
Transportation, Communications, Electric, Gas & Sanitary Services / 5.65 / 4.64 / 5.80 / 5.26
Wholesale Trade / 5.69 / 3.28 / 6.67 / 5.82
Retail Trade / 23.97 / 6.40 / 24.68 / 8.38
Finance, Insurance & Real Estate / 5.47 / 2.95 / 5.12 / 3.48
Services / 26.03 / 8.88 / 23.07 / 10.00
Public Administration / 0.51 / 0.53 / 0.89 / 1.70
Data source: County Business Pattern 1990
A2. Wu-Hausman test for the endogeneity of import penetration at the county level[1]
The OLS estimation for the Wu-Hausman test is based on the original model of equation (3), which includes the full set of control variables of Table 1 as well as the lagged industrial structure and the state dummies. Then for the 2SLS method, we follow Autor et al. (2013) and instrument the using contemporaneous changes of Chinese imports to other high-income countries, , which is calculated as:
(A1)
Equation (A1) differs from the expression of in two ways. First, import changes are for other developed countries. In our model we include: Japan, Australia, France, Germany, and Finland. These five high-income countries together have an economic scale comparable to the US, and all had relatively stable macro economies during 2000-2007. And they are all non-North American countries so that they are suitable instruments for our analysis. The second difference is that, in equation (A1) the three labor-related variables Li, Li,j, and Lus,j are all taken as one decade lag values (1990), as the subscript t-1indicates. Thus the 2SLS estimation for the Wu-Hausman test is based on the original model of equation (3) as described above, with the trade shock instrumented by .
Based on these settings, the Wu-Hausman tests yield (p=0.1863) for metro counties and (p=0.9243) for non-metro counties, and we cannot reject the H0 that there is no endogeneity.
1
[1] Again, we especially thank our reviewer for important suggestions on the endogeneity issue.