Civil Service Adoption in America: The Political Influence of City Employees
Civil Service Adoption in America: The Political Influence of City Employees was written by Sarah F. Anzia and Jessica Trounstine in 2025.
City governments in the U.S. have been transformed from
- patronage systems in the early 20th century
- merit-based civil service
- Social Security Act in 1939 required adoption of merit systems
- politically activated workforce backed by powerful unions (esp. teachers, police, firefighters)
The power dynamic has been substantially reversed.
Most prior work has examined civil service as an effect of party competition and issue evolution. The authors seek to instead model city workforces as agents that cause adoption.
There are few predictors available for this type of analysis. The authors point to firefighter organizations. "By mid-century, property owners and fire insurance companies began to push for municipal fire departments staffed by paid firefighters. By 1900, nearly all large cities (those with populations over 25,000) had established fire departments with paid firefighters (Bureau of the Census 1905)." Furthermore: "As early as the 1880s, firefighters organized mutual benefit societies and social clubs." International Association of Fire Fighters (IAFF), founded in 1918, apparently was exceptionally successful in establishing local chapters.
Authors aggregated data from the municipal yearbooks of the International City/County Management Association.
- Yearbooks of 1940-1944 published historic data of when civil service was adopted. Yearbooks of 1945-1962 simply published an indicator for adoption, but in aggregate the year of adoption can be inferred. Relevant data was dropped from publications thereafter.
- A subset of cities are published as always having civil service but no date can be identified. These cases are excluded from analysis.
- Yearbooks published list of local chapters of employee organizations, except police. Even beyond that, some undercoverage is expected, i.e. local organizations that are not a chapter of a national organization are not counted.
Because of the anticipated undercoverage in employee organization indicators, the authors prefer an indicator for IAFF only. Models are estimated for IAFF only, IAFF or other firefighter organization, and any organization.
Authors focus on 1940 as a pivot. They regress the indicator for civil service adoption on the indicators for employee organizations. Controls are drawn from 1930 Census data: "population [logged] and the shares of the population that were illiterate, foreign-born, and Black". Also if the state had passed a civil service law. Uses clustered standard errors by state. Uses Census region fixed effects.
- Authors also run one model with state-level fixed effects, instead of regions.
- In appendix, several alternative models were considered:
Logistic regression instead of OLS
- Regress on lagged IAFF indicator. Interpretation of model is not substantially different and fit is worse.
- Control for number of employees logged. Not significant.
- Alternative controls for population: binning, quintiling, interactions of the independent variable and population (both logged and binned), etc. Authors find no model is robust to the outliers.
- Bins are: Less than 5,000; 5,000-9,999; 10,000-24,999; 25,000-49,999; 50,000-99,999; 100,000+
- Exclusion of states that passed early civil service laws (NY, MA, and OH)
Authors fit a time series model on the data from 1900 to 1940. Only control used is population logged due to "a great deal of missingness". Uses city and year fixed effects.
- In appendix, several alternative models were considered:
- Regress on lagged IAFF indicator.
Authors argue that correlation of civil service adoption and employee organization is real and that causation makes more sense in the direction of organization to adoption. Most significant quantitative evidence they can produce is the worse fit of models with lagged independent variables.
Reading notes
I am very puzzled by the resistance to using logistic regression.