Compare IV-DV regression parameter between two seperate models.

[Conal Monaghan]

Hi Everyone,

      Here we wanted to check the strongest method of comparing a single parameter between SEM models. In this case we have the same DV but different IV's and thus measurement models. E.g., a range of mental health causes. Each of these models have the same DV, Depression. We want to compare the strength of the regression parameter between each of these models - that is, which IV has the strongest relationship with depression. 

The solution that comes to mind is to just use t-tests (or Chi-2). However, I am unsure if lavaan or SemTool has a function for this already. 

# Run Models

 model1 <- sem(model1, data = data1) 

model2 <- sem(model2, data = data2) 

 # Extract the SEs of the parameter estimates

 summary1 <- summary(model1) 

summary2 <- summary(model2)

 # Extract the path coefficient estimates and SEs 

coef1 <- summary1$coefficients[["IV -> DV"]] 

se1 <- summary1$se[["IV -> DV"]]

coef2 <- summary2$coefficients[["IV -> DV"]]

 se2 <- summary2$se[["IV -> DV"]]

Then run a t-test between these values. Ideally, it would be nice to identify the difference between regression parameters required to reach significance. E.g., a difference of .15 is < .01. However, this might not be possible given the differences in error between models. 

Kind Regards,

   Conal

[Christian Arnold]

Probably I do not understand your question correctly - sorry. Are these two different data sets? If no and if you are interested in the relative importance of (correlated) IV on a DV (multiple regression) you could apply the Shapley decomposition: https://wernerantweiler.ca/blog.php?item=2014-10-10 This is easily implementable with latent IV/DV.

---

From: lav...@googlegroups.com <lav...@googlegroups.com> on behalf of Conal Monaghan <conal.m...@gmail.com>
Sent: Monday, January 16, 2023 5:15:18 AM
To: lavaan <lav...@googlegroups.com>
Subject: Compare IV-DV regression parameter between two seperate models.

 

--
You received this message because you are subscribed to the Google Groups "lavaan" group.
To unsubscribe from this group and stop receiving emails from it, send an email to lavaan+un...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/lavaan/84e56fa9-a76c-4b1f-944e-b04d023ba4f3n%40googlegroups.com.

[Conal Monaghan]

Thank you for the suggestion here and the recomendation for Shapley. This suggests that we can simply treat regression paths in SEM the same as in standard / non-latent models. Is this correct? This would also suggest we can us something like the linearHypothesis function from the car package: https://kzee.github.io/CoeffDiff_Demo.html

Kind Regards,

     Conal