Table 6

Robustness Check of Estimates of Average Treatment Effect on the Treated for Lake Recreation (Matching within Rural, Small Town, Micropolitan, and Metropolitan Cell)

Participation in 2009 IDifference in Participation IITotal Trip in 2009 IIIDifference in Total Trip IV
Treatment Group 1: Unemployed, Part-time, and Retired
Nearest neighbor with replacement  0.115 (0.073)  0.128 (0.079)  0.78 (1.416)  0.655 (1.732)
Nearest five neighbors with replacement  0.084 (0.053)  0.119** (0.055)  0.438 (1.062)  0.115 (0.908)
Radius matching (caliper = 0.5 ×SD)  0.076 (0.052)  0.122** (0.054)  0.285 (0.975)  0.219 (0.865)
Kernel matching  0.075 (0.053)  0.112** (0.056)  0.099 (1.013)  0.098 (0.876)
Treatment Group 2: Unemployed and Part-time
Nearest neighbor with replacement  0.147* (0.089)  0.131 (0.109)  1.2 (1.425)  1.983 (1.438)
Nearest five neighbors with replacement  0.132* (0.068)  0.159* (0.083)  0.209 (1.126)  0.68 (1.045)
Radius matching (caliper = 0.5 ×SD)  0.129* (0.071)  0.136 (0.085)  0.129 (1.140)  0.195 (1.047)
Kernel matching  0.123* (0.068)  0.150* (0.083)–0.081 (1.098)  0.446 (1.006)
Treatment Group 3: Retired
Nearest neighbor with replacement  0.11 (0.083)  0.066 (0.085)  1.784 (1.739)–0.849 (1.621)
Nearest five neighbors with replacement  0.04 (0.068)  0.056 (0.065)  0.555 (1.452)–0.505 (1.406)
Radius matching (caliper = 0.5 ×SD)  0.041 (0.069)  0.057 (0.066)  0.869 (1.479)–0.239 (1.352)
Kernel matching  0.055 (0.070)  0.073 (0.069)  1.108 (1.514)–0.21 (1.417)
  • Note: Caliper is chosen as one-half of standard deviation of propensity score. Standard errors reported for nearest neighbor estimators are obtained from propensity score weighted regression of outcome variable on treatment indicator based on the matched sample (Marcus 2014). Standard errors reported for radius matching and kernel matching estimators are obtained from bootstrapping with 1,000 replications.

  • *, ** Statistical significance at the 10% and 5% levels, respectively.