Hypothesis Testing PreviousNext
Mplus Discussion > Multilevel Data/Complex Sample >
 Mahima Hada posted on Monday, August 23, 2010 - 5:46 am
I am estimating a two-level model. I want to test hypotheses that the total effect of one variable is greater than the other (testing for trade-offs). I am using MODEL TEST to run Wald Tests, but the test results don't match with my graphed results. I am wondering if I am using the wald test appropriately. Would appreciate your help (Mplus ver 6). Code:

DEFINE: rep_sf = sf_fam*Rep;
hom_sf = sf_fam*Homoph;
nsm_sf = sf_fam*Needs_sim;
RintE_sf = sf_fam*R_intenE;
RintB_sf = sf_fam*R_intenB;


Lcons_av on Rep(rp1)
R_intenB (RgB)
R_intenE (RgE)
Needs_sim (ns1)
Homoph (hm1)
sf_fam (sf1)
hom_sf (hm1_sf)
nsm_sf (ns1_sf)
rep_sf (rp1_sf)
RintB_sf (RgB_sf)
RintE_sf (RgE_sf)
RI_good RI_ex RI_bal;
Lcons_av on Block(rb)

model test:
rp1 + rp1_sf = hm1 + hm1_sf;
!(tests if effect of Rep > Hm when sf=1)

 Linda K. Muthen posted on Monday, August 23, 2010 - 8:00 am
In MODEL TEST, I don't see that the specification shows a greater then relationship. I don't see any indication that sf=1. And I don't think the total effect is the sum of two regression coefficients.
 Mahima Hada posted on Monday, August 23, 2010 - 8:15 am
Thanks, Linda

1) I tried a "greater than" relationship test and got the error message that I cannot use <or> in model test. Hence, the equality test as a first step. And then I figured I can calculate the correct p-value for a one-sided test.

2) SF is manipulated as 0 or 1. 0 is the base, so including hm1_sf includes the effect of Hm on the DV when sf=1.

3)Thanks for the third point - this is where I am struggling with calculating the effect using wald tests. I want to test that when sf=1, what is more important Rep or Hm for my sample. Say,
Y= b0 + b1*Rep + b2*Rep*sf + b3*Hm + b4*Hm*sf

then for calculating relative effects for the wald tets:
dY/dRep = b1 + b2
dY/dHm = b3 + b4

hence my equations.

Appreciate your advice. thanks.
 Mahima posted on Monday, August 23, 2010 - 10:48 am
I think I figured out the problem. The below specification of the test gives results that make sense substantively. Add adding random effects, intercepts etc. does not change the result:

0=(rp1 + rp1_sf) - (hm1 + hm1_sf);
Back to top
Add Your Message Here
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Options: Enable HTML code in message
Automatically activate URLs in message