With ESEM the model used to detect DIF is more flexible.
Tony LI posted on Friday, September 10, 2010 - 12:01 pm
Thanks Linda, I have a further question on this.
I am analysing DIF for a 14 item scale and its translated verison. The scale is supposed to measure a general factor and uses a summed total score. I started with conventional unidimensional CFA model, which shows very poor fit. I proceed with ESEM model instead, which suggest three factors.(I can not use a higher order multigroup CFA model becuase the translated version has used country specific items).
Further MIMIC model with country as covariate show that there is no DIF presented in any of the items. The effet of the covariate on latent factors is significant as follows:
F1 on country -.84 F2 on country .94 F3 on country -.83
As the test only use a total score, I am wondering if it make any sense to interpret that on averge, one group scores .93 units higher than the other group on the test, as the mean score difference in some of the latent factors are opposite in diretion.