After conducting EFA and CFA and achieving acceptable model fit indicators, which approach is more accurate for testing hypotheses: SEM or linear regression? given that the two approachs yeild different results regarding mediation hypothesis (questios are on Likert scale).
Hello Tarkek,
As long as your model does not include latent variables and all variables are metric strength (vs. ordinal or less), either SEM (here, path analysis) or linear regression should yield comparable results as regards tests of specific path/regression coefficients being equal to zero.
However, with single indicator variables on a Likert-type scale, you'd be best served by using either ordinal regression or an ordinal SEM model.
Good luck with your work.
Hi Tarek
I think its better to go for SMARTPLS if you have to explain complex relationships
In theory, SEM with latent variables would be expected to yield more accurate results than linear regression with manifest variables because SEM with latent variables corrects for measurement error.
SEM has the advantage of providing estimates for goodness of fit.
Also, if you hypothesize moderation (as against mediation which seems to be what your model proposes), tests of interaction in multiple regression are notably lacking in statistical power. Brian O'Connor at University of British Columbia in Canada provides programs which counteract this by applying tests of significance based on bootstrapping.
- Badges
- Science topic
- Similar topics
- Phytochemistry
- Extracts