Dispersion as pooled across a treatment and a control group. By default,
the measure of dispersion calculated within each group is not the
ordinary standard deviation as in stats::sd but rather the robust alternative
encoded in stats::mad. The dispersion measurements are combined
by squaring, averaging with weights proportional to one minus the sizes of
the groups and then taking square roots. Used in match_on.glm.
Details
A non-NULL svydesign_ parameter indicates that the dispersion
calculations are to be made respecting the weighting scheme implicit in
that survey.design2 object. If standardizer is NULL,
one gets a calculation in the style of stats::mad but with weights,
performed by optmatch:::svy_sd; for a pooling of weighted standard
deviations, one would pass a non-NULL svydesign_ parameter along
with standardizer=optmatch:::svy_sd.
(More generally, the provided standardizer
function should accept as a sole argument a survey.design2 object,
with nrows(svydesign_$variables) equal to the lengths of x and
trtgrp. This object is expected to carry a numeric variable ‘x’,
and the standardizer function is to return the dispersion of this variable.)