I want to use regression output (b, se B and cov of several predictors) as input for a new analysis.

One example: I want to compute:

SE^2(b1) + SE^2(b2) + 2 COV (b1,b2).

If I do a basic regression in Mplus (which I used because I have a choice of estimators, that I don't have in spss), and I look at the covariance table, the values reported there are extremely different from the covariances I get from the bcov command in SPSS regression.

They are so different that I am convinced they do not reflect the same value (e.g., .484 in mplus, and -.064 in spss).

Can anyone tell me: - What "estimated sample statistics" -> covariates in Mplus vs. "coefficient correlations" => "covariances" in spss refer to? - which one you think might be the cov(b1), referred to in the output? (my instinct says the SPSS one is the one I need) - How I could obtain this in the other program (i.e., if I need the spss one, how can I get this in MPlus)?

Please note that I did use the same estimator in mplus for my mock data on which I tested this, all other estimates (b, seB etc) are the same.

Model in Mplus:

MODEL:  !Main effects
            leka on S;
            leka on P;
            leka on M;
          !Correlations between predictors
            S with P;
            M with P;
            S with M;                      
OUTPUT: tech1; tech4; samp; stand; mod(4); Sampstat Mod(3.84);

Model in SPSS:


Estimated sample statistics in Mplus is the covariances of the variables. In SPSS you have asked for the covariances of the parameters.

In Mplus you need to request tech3 on the output line, i.e.

OUTPUT: tech1 tech3 tech4 samp stand mod(4);

The estimators are not the same, you're using ML in Mplus and OLS in SPSS, but the results should be extremely similar, unless your sample is small.

(Also, make sure you have no missing data, or tell Mplus to use listwise deletion , as SPSS is.)

(Another also: I don't think you need to explicitly add the correlations between predictors in Mplus, it will add them by default.)

| cite | improve this answer | |
  • 1
    $\begingroup$ Thank you for your response. Some follow-up if I may: (1) Am I correct in assuming I probably need the covariances for the parameter as provided by tech3, rather than the ones I have now? I.e., the spss output (covariances) is what I need in the formula I gave right? {(2) Indeed, I have a large sample so the results are the same, and to rule out the possibility that differences may have been due to missings or anything I ran both on an identical sample with no missings. (3) Regarding your comment on correlations: You are right, but I specify them so I can easily do model test on them later.} $\endgroup$ – GerineL Dec 8 '14 at 21:51

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.