stop.methods: Rules for selecting the propensity scores
Description
A list of stop.method objects built into the twang package
that encode rules for selecting propensity score weightsDetails
The ps function uses a stop.method object for instructions
on how to select the propensity score weights. twang has some
stop.method objects built in but the user may implement their own if
they wish and pass them to ps for it to optimize.
A valid stop.method object is a list that defines the following
- metric
{a function that evaluates the similarity of the distribution of
a variable across the treatment and control groups. Currently, the
twang package has functions es.stat,
ks.stat, and strata.stat. The user may
implement their own.}
- rule.summary
{a function that takes the vector of results from the
metric function and summarizes them into a single number.
twang currently utilizes mean and max
for rule.summary}
- direct
{logical. If TRUE then ps will try to
optimize the weights directly rather than utilizing gbm}
- na.action
{a character string indicating how bal.stat should
handle missing values. Current options are "level",
"exclude", or "lowest"}
- name
{a character string, preferably unique from other
stop.methods for labeling the resulting weights}
In addition, the object must have class(mystopmethod) <- "stop.method"