I have seen this issue twice in MPLUS forum. Once in 2003 for categorical outcomes and in some other time, related to LGC, the estimates of p values.
Linda Muthen explains r/SE is similar to running a z test.
If you ask for TYPE=BASIC, you will get the correlations and also the standard deviations for each correlation. If you divide the correlation by its standard error, this is like a z-test.
This comment, can also be found in other references. the usual output of MPLUS follows the sequence:
Two-Tailed
Estimatef S.E.g Est./S.E.h P-Valuei
And, according to Linda, Est./S.E. should be similar to Z. As I couldn’t find the formula to get the p value from a z score, I used this procedure: I got the same issue as Angela, and i resolve it by using excel.
taking this formula:
t = r * SQRT((n-2) / (1 - r*r))
source: http://faculty.vassar.edu/lowry/tabs.html#r
one can transform the r estimate into a t value. Then, to do test this excel formula is used:
p-value =TDIST(ABS(t),df,2) [where df are N-2, in a correlation]
To see if this was working, i compare the p value obtained by spss over two continuous variables vs, mplus estimates, and the results were fairly similar:
r N t P
0,022 956 0,679 0,496 from MPLUS
0,022 956 0,491 from SPSS
Let me know if there is anything wrong or if you have any comments.
