Technical Documentation: Industry Data and Associated Visualizations

View Dataset

Data Sources and Overall Methodology:

Our core data are the 2016 EEO-1 surveys of medium and large firms. In the EEO-1, firms with 100 or more employees, report to the U.S. Equal Employment Opportunity Commission (EEOC) a yearly snapshot of the employment diversity at the firm level and also on all of their workplaces with 50 or more employees.

NOTE:  Each of the following variables is first calculated at the workplace level, before averaging/aggregating up to the level of national 3 digit NAICS industry over all city (or rural state for non-city EEO-1 firms)-industry cells, while weighting by workplace total employment.

A.  Industry Employment Proportions within EEO-1 firms

We first calculate the demographic representation in EEO-1 job categories in EEO-1 reporting workplaces as an employment proportion by dividing the totals of demographic level employment within each workplace over the total employment within each workplace. We calculate these proportions over all workers for major EEO-1 job categories (i.e. roughly ten broad occupations), and by race and gender for each broad occupation group.

B.  Industry Employment Representation Relative to Local Labor Force Composition:  comparing EEO-1 firms with ACS local labor supply estimates

We create comparison baselines of the population available to be hired from the American Community Survey (ACS) for each city (CBSA) for workplaces within CBSAs or rural areas in each state for all non-city EEO-1 firms. Because we focus on 2016 employment data from the EEOC, we use the previous five years, 2011-2015, ACS to estimate local labor forces potentially available to have been hired by EEO-1 reporting firms.

We then merged the EEO-1 data on employment proportion (see A. above) by gender, race, and with ACS estimates of local labor market demographic composition.  We used this merged ACS and EEO-1 data to create industry level estimates of industry employment representation relative to local labor market composition for specific occupations. We refer to this as relative representation for short.

We calculate relative representation by dividing the demographic representation in EEO-1 total and in each occupational group in EEO-1 reporting workplaces over the proportion of each demographic group in the city (or rural-state, non-city) labor force, subtracting one from this value and multiplying the resulting statistic by 100. This creates a percentage overpresented (positive numbers) or underrepresented relative to the local labor force.  We refer to these estimates as national industry relative representation. Relative representation answers the question for any particular occupation or industry “given the local labor supply is this demographic group under or over represented in the target occupation or industry?”

C.  Industry Internal Firm Segregation:  Estimated using EEO-1 firms

i.          Segregation in the Pipeline for Top Executive Jobs

For each race and gender demographic group, we calculate the average share of top executive employment relative to the group’s presence in mid-level manager and professional employment within EEO-1 reporting workplaces.  We do this by dividing the demographic proportion in executive jobs in EEO-1 reporting workplaces over the proportion of these same demographic groups in mid-level manager and professional jobs in the same EEO-1 reporting workplaces, subtracting one from this value and multiplying the resulting statistic by 100 to create a percentage.  We refer to these estimates as national industry top leadership pipeline segregation. This measure tells for any particular industry “given the share of this group in professional and manager jobs is this demographic group under or over represented in top executive jobs in this industry?”

ii.         Segregation in the Managerial Jobs Pipelines

For each race and gender demographic group, we estimate average share of top executive and mid-level manager employment relative to the group’s presence in all other jobs within EEO-1 reporting workplaces.  We do this by dividing the demographic proportion in all executive and mid-level manager (i.e. firm leadership positions) in EEO-1 reporting workplaces over the proportion of these same demographic groups all other jobs (i.e. outside of firm leadership positions) in the same EEO-1 reporting workplaces, subtracting one from this value and multiplying the resulting statistic by 100 to create a percentage.  We refer to these estimates as national industry managerial leadership pipeline segregation. This measure tells for any particular industry “given the share of this group in jobs outside the top leadership of firms is this demographic group under or over represented in top executive-managerial leadership jobs in this industry?”

Sampling Restrictions:

We drop all people who claim “Two or more Races”.  In the EEO-1 data, we repress any cells in which there are less than 10 workplaces or any one workplace that comprises more than 50% of employment in that cell.

In the ACS data set, we drop CBSAs with less than 50,000 labor forces in the ACS baseline.  We also drop all race by gender by industry cells if their sample size is below 1 percent of the overall national industry sample size.

Industry Concordances between EEO1 files and ACS 2011-2015:

Between both the EEO1 files and ACS 2011-2015 data, we have 99 three-digit NAICS 2012 industries.

Within the EEO1 files, there are a maximum of 97 industries, as the two industry 924 “Administration of Environmental Quality Programs” and 928 “National Security” are missing in the dataset itself.  This could reflect that the EEO1 only covers private-sector workplaces, and there are no private-sector establishments within these particular industries.  Within the ACS 2011-2015 data, one industry 521 “Monetary Authorities/Central Bank” is missing.

Furthermore, while the EEO1 files are quite detailed at the level of three-digit industry, we were forced to make a few specific matching decisions to expand the industrial coverage of the ACS 2011-2015 data to match all but one industry 521 “Monetary Authorities/Central Bank” (i.e. no adjustments could be made as information on that industry was simply completely absent from the ACS 2011-2015 data).

Specifically, the ACS 2011-2015 lacked detailed responses for the specific construction industries of 236 “Construction of Building”, 237 “Heavy and Civil Engineering Construction” and 238 “Specialty Trade Contractors”.  However, it did include information for individuals within the overall two-digit NAICS 2012 industry of 23 “Construction”.  Thus, I calculated the two-digit NAICS 2012 “Construction Industry” equivalent of each ACS based wage and hour variable and include these results within each of the three-digit NAICS 2012 construction industries listed above.  This substitutes lack of variation for no information for these three-digit industries. 

For the specific three-digit NAICS 2012 industry of 551 “Management of Companies and Enterprises”, the ACS 2011-2015 data does not have any information.  However, it does have information on the highly similar two-digit NAICS 2012 industry 55 “Management of Companies and Enterprises”, which appears virtually identical. We treat these as equivalent codes and merge the information from both sources. This is visible in the data file as NAICS 551.

While the EEO1 files are missing industry 924 “Administration of Environmental Quality Programs” but does have information on 925 “Administration of Housing Programs, Urban Planning, and Community Development”, the ACS 2011-2015 only has information on a combined “Administration of Environmental Quality, Housing Programs, Urban Planning, and Community Development”.  Thus, I include the same combined value within the ACS based wage and hour variables in both the rows for 924 and 925.  Similar to the case of construction, we substitute reduced variation for absence of information. 

While the EEO1 files include both 926 “Administration of Economic Programs” and 927 “Space Research and Technology”, the ACS 2011-2015 only has information on a combined “Administration of Economic Programs and Space Research and Technology”.  Thus, I include the same combined value within the ACS based wage and hour variables in both the rows for 926 and 927.  Again, I substitute reduced variation for absence of information. 

Variable Definitions:

NOTE:  All of the EEO-1 proportion, representation, and internal segregation variables below are calculated across all race/sex groups.  Furthermore, the proportion and representation variables are also calculated across all ten broad EEOC occupation categories.  We use Men (and Male executives in occupation based variables) as just one example of the calculations underlying these variables.