Labor market impacts of destructive California wildfires

California fire incident data from the California Department of Forestry and Fire Protection (CalFire) between January 2003 and March 2021 are aggregated at the county level and merged with monthly employment data from the U.S. Bureau of Labor Statistics (BLS) Quarterly Census of Establishment and Wages (QCEW). The data are merged at the county-by-month level.

CalFire data

This article uses a dataset of about 600 individual records of fire events between January 2003 and March 2021 that exceeded 1,000 acres. The dataset was extracted from CalFire and other administrative unit agencies for non-CalFire incidents. The assembled dataset includes the name of fires, counties, start and end dates, acres of event, and number of destroyed structures. This list is then aggregated to the county-by-month level.

QCEW data

California statewide and county-level monthly data between January 2003 and March 2021 are used. The dataset includes only private establishments (it excludes local, state, or federal government). Because of the heterogeneous seasonality across counties and industries, these data are first seasonally adjusted with the use of X-12-ARIMA that stabilizes the variance at the statewide, statewide by industry, county, and county by industry levels. Other common standard practices in the literature include adding seasonal controls directly into the model for some longitudinal data. However, for this dataset, because of the heterogeneous seasonality across counties and industries, broad seasonal controls do not sufficiently capture these microseasonality differences. Counties with a population under 50,000 are excluded from the analysis because of data limitations.

To measure the extent of damage from a fire event at the county level, I weigh the number of destroyed structures by the county population, that is, the number of destroyed structures per 1,000 county population, or the local destruction index (LDI). Destructive fires that affect large acreage but few structures or fire events that destroy many structures but in a populous county may not have a large LDI. Furthermore, the analysis in this article defines the size of a wildfire by a combination of LDI and minimum number of structures destroyed:

· “Large” are those fires with an LDI greater than or equal to 1.0 and that destroy 300 or more structures.

· “Medium” are those fires with an LDI greater than or equal to 0.1 and that destroy 100 or more structures and are not large.

· “Small” are those fires with an LDI greater than or equal to 0.01 and that destroy 10 or more structures or equal to 10 and are not large or medium.

Fires that have less than 10 destroyed structures or have an LDI less than 0.01 are measurably indistinguishable from no fires and together form the baseline group.

To form a more comparable set of counties and months, the analysis further restricts the data to only counties that are most prone to wildfires. Counties that have above-average rates of significant fires (fires that burned over 1,000 acres) over the 18-year analysis period are used. Furthermore, the data are further restricted to only the core wildfire season months, June through October. Analysis for the full dataset was also run as part of robustness checks (see the section that follows), and results were mostly unchanged.