TABLE 3.

Rates of false detection of DIF among scenarios with no simulated DIF computed at the end of each procedure (%DIF wrongly detected) and rates of significant likelihood-ratio tests (%LRT SIG).

			ROSALI-DIF FORWARD		PCMLasso
n	J	Corr	%LRT SIG	%DIF wrongly detected	%DIF wrongly detected
400	4	No	5%	4%	50%
400	4	Yes	6%	3%	50%
400	7	No	7%	6%	44%
400	7	Yes	6%	3%	54%
800	4	No	6%	4%	46%
800	4	Yes	5%	3%	47%
800	7	No	6%	5%	48%
800	7	Yes	4%	3%	46%

%DIF wrongly detected: proportion of datasets where DIF was wrongly detected on at least one item-covariate pair at the end of the procedure (i.e., rate of false DIF detection). %LRT SIG: Proportion of datasets with a significant likelihood-ratio test. Corr: correlation, indicates whether covariates C₁ and C₂ are correlated (=Yes) or not (=No). The procedures converged on all datasets. No identifiability issues were encountered. Results are given according to the simulation characteristics n (sample size), J (number of items) and the presence or absence of correlation between covariates C₁ and C₂.