Table 3. Residency Program Outcomes for Unionized and Nonunionized Programsa.
| Outcome | Program, No./total No. (%) | Linear regression, OR or mean difference (95% CI)b,c | IV analysis, difference in probability (95% CI)d,e | |
|---|---|---|---|---|
| Unionized | Nonunionized | |||
| PGY-1 salary, mean (SD), $ | 61 932 (4557) | 57 798 (4652) | Mean difference, 552 (−1115 to 2220) | 7180 (−2182 to 16 542) |
| Vacation length, wk | ||||
| <4 | 2/29 (6.9) | 118/170 (69.4) | OR, 19.18 (3.92 to 93.81)f | 0.77 (0.09 to 1.45) |
| 4 | 27/29 (93.1) | 52/170 (30.6) | ||
| Subsidized childcare | 0/29 | 16/170 (9.4) | NE | −0.07 (−0.50 to 0.37) |
| Housing stipend | 10/26 (38.5) | 9/56 (16.1) | OR, 2.15 (0.58 to 7.95) | 0.62 (0.04 to 1.20) |
| Relocation stipend | 2/29 (6.9) | 20/170 (11.8) | OR, 1.13 (0.16 to 8.18) | 0.24 (−0.26 to 0.74) |
| Technology stipend | 13/29 (44.8) | 76/170 (44.7) | OR, 0.85 (0.34 to 2.12) | −0.60 (−1.46 to 0.26) |
Abbreviations: IV, instrumental variable; NE, not estimable; OR, odds ratio; PGY, postgraduate level.
Each row represents a separate model assessing the association of unionization status with each outcome. Nine programs without available salary data were excluded, resulting in 277 programs available for analysis. For benefits, only programs for which data were available through the American Medical Association FREIDA database or a publicly available union contract were included, resulting in 199 programs available for analysis (29 unionized and 170 nonunionized).
Linear regression model of salary for unionized vs nonunionized programs included program size, program type, census region, urban-rural classification, and county median household income as covariates.
Logistic regression models of unionized vs nonunionized programs included program size, program type, and urban-rural classification as covariates. For the housing stipend model, county-level median rent was also included.
Model estimated as a linear regression model salary for unionized vs nonunionized programs using regional rates of public sector employees covered by labor unions as an instrument for program unionization. The same covariates from the multivariable linear model were included.
Models estimated as linear probability models for unionized vs nonunionized programs using regional rates of public sector employees covered by labor unions as an instrument for program unionization. The same covariates used in logistic models were included. Coefficients represent the difference in probability of the outcome.
Remains significant after Bonferroni adjustment for multiple comparisons.