there are two independent (A & B) variable and two dependent (Y & Z). I want to see if the impact of independent variables vary across dependent variables. I've used SEM and that shows the value of the relationship, where A significantly impacts Y and Z, B significantly impacts Y but not Z. At the same time the regression estimate of Impact of A on Z is less than the regression estimate of B on Z. The reviewer asks for a test for significant change, which I don't understand.
If the reviewer is asking if there is significant difference in the impact, how to do that? Kindly help, thanks
You could formally test the equality of the regression slopes by specifying a model in which the slopes A --> Z and B --> Z are set equal. The fit of the model with equal slopes can then be compared to the fit of the original model with unconstrained (freely estimated) slopes. For example, you could conduct a chi-square difference test to compare the two models statistically. If the chi-square difference is significant, this means that the slopes differ significantly. However, the interpretation of this would most likely only make sense if A and B are in the same metric (same units of measurement).
This difference should be represented by a parameter (coefficient) in your model. If the confidence interval for its estimate does not extend into any significant range, then you have shown that your data is statistically incompatible with any significant difference, what justifies the conclusion that the estimated effect is not significant.
How large an effect would be considered significant is an expert judgement.
Note: being "(not) significant" is not the same as being "not statistically significant". The first refers to the (estimated) size of a parameter, the second to the likelihood of the observed data under a specific model.
Thank you for your reply, I would like to add that I have used SEM (AMOS) for analysis and want to know what further should be done to cater to the reviewer. Kindly answer the solution in a simpler way as I am new to this and I seek support. Thanks and regards.
Just do student T test to compare the means between two groups and ANOVA if you need to compare the means of three or more groups.
- Badges
- Science topic
- Similar topics
- Counseling Psychology
- Confusion