cic() gains a discrete argument
(default FALSE). When discrete = TRUE, the
discrete CIC estimator of Theorem 4.1 of Athey and Imbens (2006) is used
instead of the continuous estimator (Theorem 3.1). The discrete variant
correctly handles outcomes with mass points (e.g., integer-valued
durations) by integrating the counterfactual over the corresponding
quantile band rather than applying a point mapping. Analytic standard
errors are suppressed with an informative message when
discrete = TRUE, as Theorem 5.1 assumes a continuous
distribution; bootstrap inference is recommended instead.
print.cic() now displays which estimator was used
(continuous (Theorem 3.1) or
discrete (Theorem 4.1)).
ecdf_eval_left(),
integral_quantile(), and compute_cic_cf() to
support the discrete CIC estimator.