The State of Working Alabama 2023 – Appendix: Research design and methodology

State of Working - Appendix cover

How we define the ‘auto industry’

In this study, we confined our definition of the auto industry to the three industries that conduct auto manufacturing (as opposed to other auto-related activities):

  • NAICS 3361 – Motor Vehicle Manufacturing
  • NAICS 3362 – Motor Vehicle Body and Trailer Manufacturing
  • NAICS 3363 – Motor Vehicle Parts Manufacturing

Taken together, these three industries account for all aspects of vehicle and parts manufacturing, including core auto plants and their various suppliers (including everything from engines to seat covers). We do not include other related industries within the broader auto sector, including dealers, wholesalers, and maintenance and repair.

Data sources

We relied on two main sources of data for this analysis. First, our analyses of auto industry demographics, earnings and employment (Sections 4, 5 and 6) used Quarterly Workforce Indicators (QWI) from the Longitudinal Employer-Household Dynamics Survey, produced by the U.S. Census Bureau. We also cross-checked QWI earnings and employment with the Quarterly Census of Employment and Wages to ensure the values were broadly similar. Secondly, for our analysis of earnings and occupational segregation (Section 6), we relied on Economic Policy Institute sampling of five-year estimates of American Community Survey (ACS) microdata (2015-2019) to have sufficient sample size to conduct the analysis.

Study period

The study period for data analysis is 2002-2019. We chose 2002 as our starting point due to data limitations in QWI, which could not provide both employment and earnings data prior to that year. We chose 2019 as our end year because disruptions related to the COVID-19 pandemic disproportionately affected the auto industry. These disruptions artificially lowered employment and earnings in ways we felt were not representative of overall trends in the industry. Note that when we ran sensitivity analyses that included 2020 and 2021, these concerns proved valid. QWI data for 2022 and 2023 were not yet available at the time of the analysis in this report.

Earnings analysis

Using our QWI data, we estimated the average yearly earnings for the auto industry by taking a weighted average of the earnings across 3361, 3362 and 3663, weighting according to each industry’s share of combined auto employment. We then adjusted each year of average earnings for inflation. All earnings values in the report are presented in 2021 constant dollars.

Occupational segregation analysis

We examined five occupations within the auto industry that had sufficient sample size to incorporate into our analysis. These included by Census Occupational Code:

  • 7700 – First-Line Supervisors of Production and Operating Workers
  • 8140 – Welding, Soldering, and Brazing Workers
  • 8740 – Inspectors, Testers, Sorters, Samplers, and Weighers
  • 7750 – Miscellaneous Assemblers and Fabricators
  • 9620 – Laborers and Freight, Stock, and Material Movers

The QWI survey does not provide occupational earnings or employment by industry, race or sex, so we relied on the ACS microdata to provide us with the share each demographic group’s employment comprises of each occupation. We then applied these shares to the employment counts in QWI to determine how many employees of a demographic work within each occupation. To calculate the average earnings in each occupation, we took the percentage of each occupation’s earnings as a total of the average auto industry earnings (as presented in the ACS microdata) and applied that percentage to the QWI industry average earnings. This gave us estimated earnings levels for each occupation (and each demographic) consistent with and comparable to our overall earnings analysis using QWI data.

IMPLAN analysis

We used IMPLAN, a proprietary economic impact analysis software package, to estimate the economic impact of the key pay gaps discussed in Section 5.

IMPLAN is an input-output modeling program that permits researchers to estimate the projected effects of an exogenous change in demand – such as increase in overall labor income due to closing the pay gap between Black and white autoworkers – in a specified geographic region, such as Alabama. For our study, we used IMPLAN to model the implicit counterfactual of how much additional spending would have flowed into Alabama’s economy if auto employers had paid their workers the same average amount in 2019 as they did in 2002, and paid men the same as women, and Black and Hispanic workers the same as white workers.

The software is typically used by economic developers looking to assess the impacts of new plant locations or tax changes. But it also has been used to model the impacts of minimum wage changes in other states and increased labor income resulting from greater female labor force participation in Alabama, and so is well suited to assessing changes in earnings like those we examine.

Specifically, we used IMPLAN to estimate the following four pay gaps:

  • The difference between the inflation-adjusted average annual earnings by autoworkers in 2002 and in 2019 (in 2021 constant dollars).
  • The difference between the average annual earnings of Black autoworkers and white autoworkers in 2019 (in 2021 constant dollars).
  • The difference between the average annual earnings of Hispanic autoworkers and white autoworkers in 2019 (in 2021 constant dollars).
  • The difference between the average annual earnings of male autoworkers and female autoworkers in 2019 (in 2021 constant dollars).

We ran four separate models, one for each pay gap. In each of these models, we used the pay gap amount as the input and then modeled each of them as an Industry Impact Analysis (Detailed – labor income change) event for the auto industry, specifying the year of impact as 2019 (to match our QWI data), using 2021 dollars. To check our work, we also ran all four models as a standard labor income event (with 100% of our earnings change assigned to employee compensation), and as expected generated identical results to the original Industry Impact Analysis.

In terms of results, for labor income changes like these, the software only reports “induced effects,” which capture tail-end economic ripple effects of shifts in household spending resulting from the total change in earnings we’re modeling (e.g., resulting from closing the pay gaps). But we needed to estimate the total impact on Alabama’s economy, which includes the original amount of the new labor income that would flow to Alabama if these pay gaps were closed (e.g., $348,926,144 for the historical pay gap).

Though it is unusual to combine original and induced effects in this way for labor income changes in particular, staff economists at IMPLAN advised us it would be appropriate in this case. This is because the economic change is not associated with a separate Industry event (like a new plant facility opening), and we need to reflect the full range of an economic change that includes both new earnings (closing the pay gaps) and the induced effects (the ripple effects of the input) together.

Our results tables are reported here:




Additional sections

Summary Executive summary


Section 1 Introduction


Section 2 The high stakes and big bet on Alabama’s auto industry


Section 3 The ways the bet on auto benefited Alabama


Section 4 A wheel in the ditch: Autoworkers see falling pay


Section 5 A wheel in the ditch: Pay gaps and occupational segregation


Section 6 A wheel in the ditch: Economic impact of falling wages and the pay gap


Section 7 A wheel in the ditch: Working conditions worsen


Section 8 The auto industry and Alabama’s low-road economic development approach


Section 9 What we should do next