Description
Hi,
When running a multiple sensitive item design with ictreg(..., multi.condition = "level"), it is not currently supported by predict.ictreg() to make prediction:
require(list)
## multi.condition = none
results_no_level <- ictreg(y ~ male + college + age + south + age:south, treat = "treat",
J = 3, data = multi, method = "ml")
## prediction succeeds
predict.ictreg(results_no_level, sensitive.item = 1, avg = TRUE)
## multi.condition = level
results_w_level <- ictreg(y ~ male + college + age + south + age:south, treat = "treat",
J = 3, data = multi, multi.condition = "level", method = "ml")
## prediction fails
predict.ictreg(results_w_level, sensitive.item = 1, avg = TRUE)
which gives an error message:
Error in xvar %*% beta : non-conformable arguments
It seems like the reason is that y_i(0) and the extra associated parameter are not taken into account in predict.ictreg().
In general, I manually simulated the expected probabilities of respondents choosing the sensitive items taking y_i(0) and the extra parameter into account (following the replication file PA_Replication.R -- Replication R command file for: Graeme Blair and Kosuke Imai, "Statistical Analysis of List Experiments," Political Analysis, In Press. Created 27 October 2011). It seems like the inclusion of y_i(0) and the associated parameter drastically changes the expected probabilities vis-a-vis the difference-in-means estimate. Will there be any implementation of robust NLS/ML models (i.e.ictreg(...robust = TRUE)) for multiple sensitive item design to ensure their correspondence?
Thank you!