Matching is a class of observational study methods where each treatment unit in a study is linked to one or more control units. In the context of a "quasi" or "natural" experiment (i.e. in the absence of randomization), matching helps to reduce the influence of covariate bias by creating an artificial control group that is comparable in measured features to the original treatment group.
Researchers face two common issues when working with a matched sample: first, distributions of key covariates might still be imbalanced after matching, and second, matching discards large amounts of data in an effort to obtain similar groups.
Weighting by the propensity score is one workaround. Using weighting, researchers can obtain better balance on key covariates and retain all observations from the original sample. Such weighting schemes are intimately tied to the survey sampling literature, both in their technical foundations and general limitations.
Using the population of third grade public school classrooms in New York City, we identify the treatment effect of inclusion on total attendance with a propensity score model:
# propensity score model
m_ps<- glm(treatment ~ PercentBlack + PercentSWD + PercentPoverty + TotalEnrollment +
ENI, family = binomial(), data = working_data)
summary(m_ps)Balance in the propensity score is diagnosed using the cobalt package
#Checking balance before and after matching:
## Call
## matchit(formula = treatment ~ ENI + PercentBlack + PercentSWD +
## TotalEnrollment + PercentPoverty, data = working_data, method = "nearest",
## caliper = 0.25, ratio = 4, family = "binomial")
##
## Balance Measures
## Type Diff.Adj M.Threshold
## distance Distance 0.0595 Balanced, <0.1
## ENI Contin. -0.0445 Balanced, <0.1
## PercentBlack Contin. 0.0058 Balanced, <0.1
## PercentSWD Contin. 0.0148 Balanced, <0.1
## TotalEnrollment Contin. -0.0661 Balanced, <0.1
## PercentPoverty Contin. -0.0328 Balanced, <0.1
##
## Balance tally for mean differences
## count
## Balanced, <0.1 6
## Not Balanced, >0.1 0
##
## Variable with the greatest mean difference
## Variable Diff.Adj M.Threshold
## TotalEnrollment -0.0661 Balanced, <0.1
##
## Sample sizes
## Control Treated
## All 623. 147
## Matched (ESS) 177.34 102
## Matched (Unweighted) 262. 102
## Unmatched 361. 45IPTW and SW weights can be hard-coded, or assigned using the PSweight package:
working_data$treatment_identifier <- ifelse(working_data$treatment == 1, "inclusion", "non-inclusion")
working_data$iptw <- ifelse(working_data$treatment_identifier == 'inclusion', 1/(working_data$propensity_score),
1/(1-working_data$propensity_score))
#stabilized weights
working_data$stable.iptw <- ifelse(working_data$treatment_identifier == 'inclusion',
(mean(working_data$propensity_score))/working_data$propensity_score,
mean(1-working_data$propensity_score)/(1-working_data$propensity_score))
## Weighted Data, using the survey design package
working_data_weighted <- svydesign(ids = ~1, data= working_data, weights= working_data$iptw)
## balance of each in the unweighted sample
data<- working_data_weighted$variables
mod4 <- lm(AllStudents_PA ~ treatment, weights = data$iptw, data = data)Ho, D. E., Imai, K., King, G., & Stuart, E. A. (2011). MatchIt: Nonparametric Preprocessing for Parametric Causal Inference. Journal of Statistical Software, 42(8). doi:10.18637/jss.v042.i08
Li, Fan, Laine E Thomas, and Fan Li. 2019. “Addressing Extreme Propensity Scores via the Overlap Weights.” American Journal of Epidemiology 188 (1): 250–57. https://doi.org/10.1093/aje/kwy201.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent