propensity: Propensity Summary
In dewittpe/pstools: Propensity Score Analysis Tools

Description Usage Arguments Details Value References See Also Examples

Given a logistic regression model, generate the propensity weights for the subjects (rows) in the data set.

1	propensity(fit, weight_method = 1)

`fit`	a regression object such as
`weight_method`	one of three methods for defining the weight for each row of the data set. See Details.

The regression model is expected to estimate the probability of an exposure (Z = 1) given a set of predictors, X, i.e., Pr[Z = 1 | X].

NOTE: for binary predictors coded as 0/1, such as male, the default action of propensity_summary will return a mean (sd), that is, if propensity_summary(geeglm(y ~ male)) will summarize the variable male as if it was continous predictor. To get the summary for the proportion of males, and females will be reported too, use propensity_summary(geeglm(y ~ factor(male))).

weight_method controls the weight given to each row of the data set. Let ps = Pr[Z = 1 | X]. The default setting is weight_method = 1.

weight_method = 1: w = min(c(ps, 1-ps)) / (z*ps + (1-z)*(1-ps))
weight_method = 2: w = 1/(z*ps + (1-z)*(1-ps))
weight_method = 3: w = z + (1-z)*ps/(1-ps))

A pstools_propensity object which is a data.frame with columns summarizing each varaible used as a predictor in the propensity model. The function will determine if each variable is a categorical or continous variable. if the variable is continuous the mean (sd) is returned. If the variable is categorical the the percentage of each level is returned. Standard differences are reported as well. Unadjusted and Matched Weight (see Li and Greene (2012)) values are reported.

Li, Liang, and Tom Greene. "A weighting analogue to pair matching in propensity score analysis." The international journal of biostatistics 9.2 (2013): 215-234.

glm for fitting logistic regression models, or geeglm for fitting GEEs.

## Not run: 
data(pride)
glmfit <- stats::glm(PCR_RSV ~ SEX + RSVINF + REGION + AGE + ELTATOP + EINZ + EXT, 
                     data = pride,
                     family = stats::binomial())
propensity(glmfit)
summary(propensity(glmfit))
plot(propensity(glmfit))

## End(Not run)