 Testing tnstruments    Message/Author  Anna Manzoni posted on Tuesday, August 25, 2015 - 9:13 am
Two questions:
1. Consider the model
Y on lagY X lagX g1 g2 z1
X on lagX M lagM g1 g2 z2
M on lagM g1 g2 z3
I am interested in Y and X, and mostly in X&#61664;Y. I wonder whether I need to specify an equation for M too or not (note: I am not interested in interpreting parameters related to what influences M, I just wonder what�s correct). I tried including and excluding the M equation and I noticed that results on parameters and SE are identical and the only difference is in model fit measures.

2. It is my understanding (but I am not sure I got this right) that in Mplus a Wald test for the coefficients of the instruments being equal 0 is asymptotically equivalent to an F test (and if F>10 I should be fine in terms of instrument strength). So consider the model
Y on lagY X lagX z1(a1);
X on lagX Y lagY g1 g2 z2(a2);
Model test:
a1=0;
a2=0;
This would give as output
Wald Test 47.273, df 2, P-Value 0.000
So I would conclude instruments are valid.
However I am unable to do this when I use more than one instrument in each equation, as I get an error message. So if I have for example
Y on lagY X lagX z1(a1) z3(a3);
X on lagX Y lagY z2(a2) z4(a4);
Model test:
a1=0;
a2=0;
a3=0;
a4=0;
Mplus gives me an error. Any insight?
Thank you.  Bengt O. Muthen posted on Tuesday, August 25, 2015 - 10:50 am
1. No need to include the M equation.

2. You need to have only one label on each line, like:

Y on lagY X lagX
z1(a1)
z3(a3);

The Wald test is a chi-square test and if the p-value is small you reject the hypothesis.    Topics | Tree View | Search | Help/Instructions | Program Credits Administration