Message/Author |
|
Emily Zhang posted on Monday, September 28, 2015 - 5:26 pm
|
|
|
I have two questions about alignment. First, is it necessary to create a good-fitting baseline multigroup model with no intercept and loading constraints before alignment? Or can your directly subject your items to alignment? Second, I am attempting alignment with a small sample size (2 groups of about 70 participants). I need to establish MI for 2 scales with 9 and 11 items and, due to my sample size constraints, it’s difficult to establish good fit since adding many residual correlations makes it so there are more parameters than subjects. Is it reasonable to do alignment on factors whose indicators are parcels instead of individual items? Thanks! |
|
|
1. You can use alignment directly however, alignment will not improve the model fit. 2. It sounds reasonable. |
|
|
Dear professors, Hello!I have a question about alignment. The fixed alignment fix the factor mean of the first group at 0 so the first group is the reference group. But for free aignment, the factor mean of the first group is freely estimated, so if there is no reference group, what is the metric or reference value of the factor means in free alignment? Thanks very much! |
|
|
There is no reference value. If you need to compare factor means you would form the parameter difference in model constraints. The extra parameter is identified through the alignment process and has no analogue in standard SEM modeling without alignment. |
|
|
Dear Professors, Hello!One more question about alignment. I got the outputs which report the groups with significantly smaller factor mean. I want to ask how the system calculate the significance,with F test? And if the ranking is 1,2,3,4 and group4 is significantly smller, this means group4 is smaller than group3 or group4 is smaller than the entire 3 previous groups? Thank you for shedding light on my questions! |
|
|
Z test based on the pairwise comparison of the two parameters. It doesn't mean that the group 4 is statistically significant from group 1. Even though group 1 has a higher mean value it could happen that it is a very small group and that statistical significance can not be established. |
|
Back to top |