Mapping Green Jobs, Fossil Fuel Workers, and Green Energy Production.

How easily will today's fossil fuel workers transition to emerging jobs in green energy production?

The green energy revolution may displace 1.7 million fossil fuel workers in the US. But a Just Transition to emerging green industry jobs offers hope. Here, we use 14 years of data to test if today's fossil fuel workers are appropriately skilled and co-located with expected green job growth. In agreement with political rhetoric, we find that fossil fuel workers are appropriately skilled for green industry jobs. However, co-location appears to be a bigger issue. The visualization below maps projected green jobs according to the US Department of Labor, fossil fuel worker employment in 2022, and the locations of green energy production plants in 2022.

Demo Loading...

Select bars from the bottom plot to see the locations for different types of power plants on the map. Use the Shift key to select multiple types at once.

Data Description

Mapping Current Fossil Fuel Worker Employment

The Bureau of Labor Statistics (BLS) collects employment data for over 750 occupations, 232 of which are present in the fossil fuel industry. Among these, we focus on 11 occupations that fall into the extraction workers category. We focus on these occupations for two reasons. First, they are primarily represented in the fossil fuel sector: about 70% of extraction workers are active in fossil fuel firms (as of 2019). Second, they represent a fundamental core of the fossil fuel business (representing about 27% of workers in the industry) and their livelihood depends the most fundamentally on the fossil fuel sector. The next largest groups -- engineers and material movers -- represent much smaller shares (about 13% and 10% respectively) of fossil fuel workers and have clear outside options. As such, we use a narrow definition of fossil fuel workers in order to focus on the segment that is the most at risk in the clean energy transition. Additionally, we use the distribution of the entire workforce fossil-fuel industry using data from the Business Dynamics Survey (BDS) from Census Bureau. BDS provides the number of workers in metropolitan areas at the industry level, following the North American Industry Classification System (NAICS) two-digit code.

Mapping Projected Green Jobs

To identify green occupations, we rely on the classification established by Dierdorff and colleagues, a measure that has since been used widely, including in the European Union. They identify three types of green occupations. First, green increased demand occupations are those that are found in economic activities that contribute to the reduction of fossil fuel emissions (energy efficiency, renewable energy, etc.) or increase recycling. Second, green enhanced skills occupations are those that exist but require change in tasks, skills, knowledge, and so forth. Third, green new and emerging occupations are new types of jobs that are not defined yet. We focus primarily on the first category since we are interested in modeling the odds of job transition toward established occupations.

We created predictions for green occupation employment in 2029 using historical data from the BLS on occupation employment distribution in MSAs in the US from 2005 to now, combined with Census data (population by age, gender, race, and education) and economic features (state-level GDP per capita by sector). With this historical data, we train the model that predicts the employment projection in 10 years. Since we aim to predict the employment projection in 2029 using the currently available data, we train the model to predict occupational employment using the data from 10 years lagged. We compare the performance of random forest against other classifiers (OLS, Poisson, Lasso). Cross-validation (10-fold) was used to tune the hyper-parameters of the model. We cannot directly evaluate the accuracy of our prediction in 2029. So, we use historical employment data to train and test the accuracy of predictions. To avoid data leakage, we separated the training and test sets using the temporal cutoff. That is, we set aside the 2019 employment distribution data as a test set while training the model with the previous years. We find that the random forest model was the best performer in out-of-sample predictions of historical data (i.e., train the model on pre-2019 data to predict employment in 2019).

Research Team

Junghyun Lim

University of Pennsylvania

Michael Aklin

University of Pittsburgh

Morgan R. Frank

University of Pittsburgh

Contact Us

Pittsburgh, PA, USA

Email: mrfrank@pitt.edu