We surveyed faculty members in degree-granting physics departments and those teaching physics in two-year colleges in the US in 2016. We offer our gratitude to the almost 1,700 who provided data on their employment status, rank, tenure status, gender identity, and academic training. We have used those data to estimate a series of regression models that we now use to provide average salaries of physics faculty members in US two-year colleges, colleges, and universities.
The results are average salaries and are not claimed to be actual salaries. Actual salaries will vary above and below the mean. We provide a range which indicates the variation in salaries. The range includes one standard error above and below the mean. The regression is run on the log of salaries (because salaries are skewed), so the range is not symmetric. We use a series of models because no one model meets every purpose. Each model includes all the variables in the previous model.
These regression models explain 85 – 86% of the variation in the 1,694 salaries included in our analysis. There is still 15% of the variation that is not explained. We recognize that salaries are a function of the variables included in the model and other factors. These salaries are not adjusted for required contributions for health insurance, for example.
The results are adjusted to 2018 levels using the US CPI.
Model I: The Basic Model (National estimate)
We use a series of models because no one model meets every purpose. Our most basic model provides a national estimate and includes:
- Public, four-year institution
- Two-year college
- Highest physics degree offered (does not apply if institution is a two-year college)
- Full-time or part-time
- Period salary covers (9-10 months, 11-12 months, or course-by-course)
- Tenure status
- PhD recipient (or not)
- Highest degree earned in the US
- Completed a postdoc
- Year of highest degree
This basic model explains 85.5% of the variation in salaries.
Model II: Adding Gender (Refined national estimate; includes all variables in Model I)
We add gender to the model. We understand that gender is not necessarily binary. However, we have data to estimate the averages salary only for people identifying as men and women. Adding gender alone is marginally significant (one-tail p-value = 0.075). However, adding gender and the interaction of gender and associate professor is statistically significant. The p-value for the F test for the change in R-square is 0.037. The adjusted R-square is 0.855, so this model, too, explains 85.5% of the variation in salaries.
Model III: Adding Population Density (State-level estimate; includes all variables in Model II)
We next add population density to the model. The p-value for the F test for the change in R-square is < 0.0005. We use population density instead of state because we do not have sufficient data from each state to provide estimates using states. However, population density is positively correlated with per capita income (p-value = 0.069), so we use that as a proxy for each state. The R-square for this model is 0.858 indicating that this model explains 85.8% of the variation in salaries.
Model IV: Adding Census Bureau Region (Refined state-level estimate; includes all variables in Model III)
While population density does explain slightly more of the variation in salaries (from 85.5% without it in the model to 85.8% with it in the model), adding the Census Bureau region to the model helps refine the state-level estimate. This model explains 86.3% of the variation in salaries.
This tool is © 2018 by the Statistical Research Center at the American Institute of Physics. For questions, please contact Susan White at swhite [at] aip.org.