 Zero standardized indirect effects    Message/Author  Lang Chen posted on Thursday, February 24, 2011 - 2:35 pm
Hello,

I am running a path analysis and find that the standardized indirect effects of one exogenous variable are all zero although the effect was significant and the non-standardized coefficients are normal. I am wondering whether this is because the standardized value are too small (<.0001) to be displayed or something is going wrong here. Also, I am using an old version of Mplus 3.0.

Thanks.  Bengt O. Muthen posted on Thursday, February 24, 2011 - 5:14 pm
Try multiplyin the raw indirect effect with the SD of X and divide by the SD of Y to compute the standardized value yourself.

And you want to upgrade to version 6.1.  Nicki Keating posted on Thursday, April 07, 2016 - 12:49 pm
Hello
I have a similar question. I have a two-level L2X>L2M1>L2M2>Categorical outcome. X, M1, M2 are continuous. The X on Y via M1 is insignificant and has been removed.

MODEL: %WITHIN%
Y on L1covariates;

%BETWEEN%
y ON M2 (b2);
Y ON X (cdash);
M1 ON X (a1);
M2 ON X (a2);
M1 ON M2 (d1);

MODEL CONSTRAINT:
NEW(a2b2 a1d1b2 direct TOTALIND TOTAL);
a2b2 = a2*b2;
a1d1b2 = a1*d1*b2;
direct = cdash;
TOTALIND = a2*b2 + a1*d1*b2;
TOTAL = a2*b2 + a1*d1*b2 + cdash;

Is the Preacher and Kelley paper to be used for each of these paths to calculate standardized indirect estimates in a multi-level model? Bootstrapping is not possible for two-level - how does one go about calculating the CI for the standardized indirect effect?  Bengt O. Muthen posted on Friday, April 08, 2016 - 10:14 am
You need to express the between-level variance of X and Y in Model Constraint using parameter labels in Model. And then express the standardized effects using those corresponding SDs. That gives you the estimates, the SEs, and the CIs. The CIs will be symmetric, however. To get non-symmetric you would have to use Bayes.  Nicki Keating posted on Saturday, April 09, 2016 - 4:02 am
Thank you so much. I think I have made a naive mistake somewhere obtaining my between level variance for X and Y as the standardized estimates are extremely high (although I have high X variance).

MODEL: %WITHIN%
Y on L1covariates;

%BETWEEN%
y (varY); !between-level variance Y CATEGORICAL);
X (varX);!BETWEEN LEVEL VARIANCE X;

y ON M2 (b2);
Y ON X (cdash);
M1 ON X (a1);
M2 ON X (a2);
M1 ON M2 (d1);

MODEL CONSTRAINT:
NEW( a2b2 a1d1b2 direct TOTALIND TOTAL VARDV VARIV STDA2B2 STDIndirect STDDIR STDTOTALIND STDTOTAL);
a2b2 = a2*b2;
a1d1b2 = a1*d1*b2;
direct = cdash;
TOTALIND = a2*b2 + a1*d1*b2;
TOTAL = a2*b2 + a1*d1*b2 + cdash;

VARDV = VARy;
VARIV = VARx;

STDA2B2 = 0.145*VARIV/VARDV; !insert raw co-efficients;
STDIndirect= a1d1b2 = 0.002* VARIV/VARDV;
STDDIR = 0.589* VARIV/VARDV;
STDTOTALIND = 0.1470* VARIV/VARDV;
STDTOTAL = 0.735*VARIV/VARDV;

Sincerely hope you can help, have a lovely weekend.  Bengt O. Muthen posted on Saturday, April 09, 2016 - 6:19 pm
Note that for DVs the residual variance is the estimated parameter, not the full DV variance. Also, you use SDs not variances to standardize, so use SQRT. And you shouldn't insert values like 0.145 but the label for the parameter estimate.    Topics | Tree View | Search | Help/Instructions | Program Credits Administration