Dear all
I have 7 formative constructs with 63 items and every constucts has 9 indicators.I run my data for PLS.
I need to taste heterogeneity for the pre-test.
Is any specific test for that? I attached an article to this email. If you look at page 310, line 7, it is mentioned because measurement properties may differ across sub - populations. For example, scale properties may differ across demographic groups, by experience, by education level, etc. We need to check this issue. I know that there are tests such as Anova and T-test for comparing groups, but I look for a simpler way. I think, "Fimix?" in PLS also do this. Do you know any simple way for this issue?
In your idea , is it necessary to check for this issue or not?
Thanks in advnace
Dear Marjan,
Actually it is better to verify homogeneity of variance by SPSS if you are going to compare same variable(s) from different groups with each other, then transfer your data to PLS (PLS is just for the final SEM analysis, not pre-exploratory analyses). If the number of categories over which you wanna make a comparison are two (e.g. gender), follow the paths below in SPSS:
Analyze ˃ compare means ˃ independent samples t test
You can conclude that the categories's variances are homogeneous, if the significant value of Leven test is higher than .05 and not, if otherwise.
And if the number categories are higher than two:
Analyze ˃ compare means ˃ One-way ANOVA ˃
With the same conclusion gained for Leven as in two categories.
No other simpler way is, i think.
The point is that you are not required to consider homogeneity of variance in the pre-test phase of your investigation as there are few number of data. Also, in SEM you are not going to compare the variables for different groups, unless the role of a moderator variables is sought. So, if ur investigation doesn't consider a moderator, skip this test (Leven).
Kind regards compatriot,
Hi,
Refer the following link:
https://books.google.com.sa/books?id=hwkshLwOFM0C&pg=PA73&lpg=PA73&dq=checking+heterogeneity+is+necessary+for+pre-test&source=bl&ots=1hdSPLRP3m&sig=uYMO39TFUikDDPkw2PKPrXp_LWE&hl=en&sa=X&ei=OoLgVM3ZNIfYavm5gIAL&redir_esc=y#v=onepage&q=checking%20heterogeneity%20is%20necessary%20for%20pre-test&f=false
And after that, you can apply ind. t-test and ANOVA test.
All the best for you.
Hi, you can check these websites for this problem:
http://www.univ-orleans.fr/deg/masters/ESA/CH/Geneve_Chapitre1.pdf
http://handbook.cochrane.org/chapter_9/9_5_2_identifying_and_measuring_heterogeneity.htm
In addition to previous comments be sure you check what type of heterogeneity you are interested in.
T-tests or ANOVA will check if the average scores on each of your constructs are the same. I suspect that is not what you want. In fact, you should expect that different groups will have different averages - that would be one of the things that makes them different groups.
The more important issue for your project would be whether the coefficients are consistent across all groups. That is, each of the nine items should have about the same weights across each of your groups.
The old-fashioned method for testing if coefficients are the same is a Chow test. A more recent method is confirmatory factor analysis with separate groups with between-groups comparisons.
Other contributors may have more up-to-date suggestions.
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis