Skip to main content
. 2019 Aug 16;134(4):1747–1791. doi: 10.1093/qje/qjz023

Table IV.

First-Year Treatment Effects: Experimental versus Observational Estimates

Experimental (IV) Observational (OLS)
Outcome variable No controls Post-Lasso No controls Post-Lasso
(1) (2) (3) (4)
Panel A: Medical spending
 Total spending (dollars/month) [admin] 17.7 52.3 Inline graphic137.3** Inline graphic103.8*
(79.0) (59.4) (68.6) (61.9)
N = 3,239 N = 3,152 N = 2,208 N = 2,140
  Drug spending [admin] Inline graphic13.8 Inline graphic12.8 Inline graphic26.3 Inline graphic7.3
(43.2) (20.4) (27.2) (12.0)
N = 3,239 N = 3,152 N = 2,208 N = 2,140
  Office spending [admin] Inline graphic9.9 Inline graphic3.1 12.2 8.7*
(16.2) (6.8) (7.5) (5.1)
N = 3,239 N = 3,152 N = 2,208 N = 2,140
  Hospital spending [admin] 36.1 45.2 Inline graphic118.0** Inline graphic83.4
(50.4) (45.6) (55.7) (51.8)
N = 3,239 N = 3,152 N = 2,208 N = 2,140
 Nonzero medical spending [admin] Inline graphic0.013 0.004 0.061*** 0.036***
(0.018) (0.016) (0.014) (0.012)
N = 3,239 N = 3,152 N = 2,208 N = 2,140
Panel B: Employment and productivity
 Job promotion [admin] Inline graphic0.006 Inline graphic0.009 0.019 0.009
(0.022) (0.021) (0.014) (0.015)
N = 4,146 N = 4,130 N = 2,840 N = 2,828
 Job terminated [admin] Inline graphic0.022 Inline graphic0.023 Inline graphic0.080*** Inline graphic0.063***
(0.018) (0.017) (0.011) (0.011)
N = 4,834 N = 4,753 N = 3,300 N = 3,244
 Sick leave (days/year) [admin] 0.322 0.224 0.275 Inline graphic0.068
(0.398) (0.344) (0.272) (0.251)
N = 4,782 N = 4,712 N = 3,264 N = 3,216
 Management priority on health/safety 0.087*** 0.077*** Inline graphic0.004 Inline graphic0.007
   [survey] (0.023) (0.021) (0.017) (0.016)
N = 3,566 N = 3,514 N = 2,410 N = 2,376
 Productivity index [survey/admin] Inline graphic0.070 Inline graphic0.096 0.069 0.083
(0.092) (0.085) (0.073) (0.067)
N = 3,309 N = 3,300 N = 2,245 N = 2,240
Panel C: Health status and behaviors
 IL Marathon/10K/5K 2017 [admin] 0.003 Inline graphic0.011 0.059*** 0.024***
(0.014) (0.011) (0.008) (0.006)
N = 4,834 N = 4,817 N = 3,300 N = 3,287
 Campus gym visits (days/year) [admin] Inline graphic0.110 0.757 3.527*** 2.160***
(1.309) (0.656) (0.813) (0.425)
N = 4,834 N = 4,817 N = 3,300 N = 3,287
 Ever screened [survey] 0.060*** 0.056*** 0.073*** 0.061***
(0.014) (0.012) (0.011) (0.009)
N = 3,567 N = 3,557 N = 2,410 N = 2,404

Notes. Each row and column reports estimates from a separate regression. The outcome in each regression is specified by the table row, and the (endogenous) focal independent variable is an indicator for completing the screening and health risk assessments (HRAs). For the IV specifications (columns (1)–(2)), the instrument is an indicator for inclusion in the treatment group, and observations include individuals in the control or treatment groups. For the OLS specifications (columns (3)–(4)), there is no instrument and observations are restricted to individuals in the treatment group. The control strategy is specified by the column. Post-Lasso controls include covariates selected by Lasso to predict either the dependent variable or the focal independent variable. The set of potential predictors include baseline values of all available variables in the same family of outcomes, strata variables, and the baseline (2016) survey variables reported in Table I, as well as all two-way interactions between these predictors. Robust standard errors are reported in parentheses. */**/*** indicates significance at the 10%/5%/1% level using conventional inference.