Advanced Models
The Advanced Models feature enables you to perform nearest neighbors propensity score matching within a specified caliper distance. When the method is designated to be Nearest Neighbors, more features are shown.
Propensity score matching is a causal inference method that pairs each treated observation with one or more untreated counterparts based on estimated propensity scores. In each matched pair, the untreated observation is considered a “control” for the treated observation because it was approximately just as likely to receive treatment as the treated observation. The sample of treated and control observations is called the matched sample. A linear regression is fit on the matched sample, with treatment as the only predictor, to estimate a causal effect as the coefficient of the slope.
When the matching technique is nearest neighbors matching, each treated observation is matched to the untreated observation with the closest propensity score. The distance metric is absolute difference. The caliper is the farthest propensity score distance two matched observations can be from one another. If no untreated observations have a propensity score that is within the absolute caliper distance, no matches are chosen for the treated observation and it is excluded from the analysis.
The Advanced Models option enables you to specify the following parameters:
Number of Matches per Observation
Specifies how many controls should be matched to each treated observation.
Propensity Score Minimum
Specifies a minimum propensity score. Any observation with a propensity score lower than the minimum is excluded from the matched sample.
Propensity Score Maximum
Specifies a maximum propensity score. Any observation with a propensity score higher than the maximum is excluded from the matched sample.
Caliper
Specifies the farthest absolute propensity score difference between matched treated and untreated observations. Any treated observation without a control within the specified caliper distance is excluded from the matched sample.
Matching Statistic
Specifies the matching statistic that is used. You can specify either the propensity score or the logit-transformed propensity score.
Match with Replacement
Specifies whether a single untreated observation can be matched to multiple treated observations.