Login

News

Very small estimate with a very large standard error in univariate modelling.

Posted on
No user picture. Zhiyang Joined: 12/11/2024
Forums
Behavioral Genetics Models

Dear all, 

This is the first time to have a post in this forum, and thanks for the platform.

I am a researcher working on the Finnish Twin Cohorts, and I am doing a univariate twin modeling on depression. I have 113 DZ pairs and 69 MZ pairs. The outcome of depression is basically normal-distributed.

Then in the ADE model, I get the data summary and results like this:

data:
$MZ
      dep1            dep2            age1            age2      
 Min.   :1.375   Min.   :1.375   Min.   :34.43   Min.   :34.62  
 1st Qu.:1.750   1st Qu.:1.750   1st Qu.:36.02   1st Qu.:35.92  
 Median :1.875   Median :1.875   Median :37.14   Median :37.25  
 Mean   :1.893   Mean   :1.891   Mean   :37.30   Mean   :37.31  
 3rd Qu.:2.000   3rd Qu.:2.000   3rd Qu.:38.57   3rd Qu.:38.69  
 Max.   :3.250   Max.   :2.750   Max.   :39.61   Max.   :39.74  

$DZ
      dep1            dep2            age1            age2      
 Min.   :1.250   Min.   :1.250   Min.   :34.55   Min.   :34.49  
 1st Qu.:1.625   1st Qu.:1.750   1st Qu.:35.90   1st Qu.:35.92  
 Median :1.875   Median :1.875   Median :37.16   Median :37.12  
 Mean   :1.889   Mean   :1.869   Mean   :37.19   Mean   :37.18  
 3rd Qu.:2.000   3rd Qu.:2.000   3rd Qu.:38.54   3rd Qu.:38.47  
 Max.   :3.000   Max.   :2.750   Max.   :39.75   Max.   :39.82  
free parameters:
   name   matrix row col   Estimate Std.Error A lbound ubound
1  mean MZ.meanG   1   1 1.88344659  115.8524                
2 beta1     MZ.b   1   1 0.01000000        NA !              
3  VA11    MZ.VA   1   1 0.00194220  303.5822 ! 1e-04!       
4  VD11    MZ.VD   1   1 0.03293261  293.7319 ! 1e-04!       
5  VE11    MZ.VE   1   1 0.05600604  328.6864    1e-04

The estimates of variance of A, D, or E are quite small, but their SE is large. I am wondering why this thing happened? Is it because of small sample size? 

Besides, there is another warning:

The Hessian at the solution does not appear to be convex. See ?mxCheckIdentification for possible diagnosis (Mx status RED).

Thank you very much for your insightful suggestions!

Best

Zhiyang

Replied on Tue, 02/04/2025 - 16:17
No user picture. lf-araujo Joined: 11/25/2020

Without looking at the script it is harder to give suggestions. But this type of situation usually comes from poor starting values or lack of power. 

You can try to improve the fit by running

model <- mxAutoStart(model)

before

model <- mxTryHard(model)

and see if this resolves the issue with SEs.

 

Log in or register to post comments
Replied on Fri, 02/21/2025 - 10:13
Picture of user. AdminNeale Joined: 03/01/2013

It's near impossible to say the reason for the results without seeing the R script.  Some things I see:

3  VA11    MZ.VA   1   1 0.00194220  303.5822 ! 1e-04!       
4  VD11    MZ.VD   1   1 0.03293261  293.7319 ! 1e-04!       
5  VE11    MZ.VE   1   1 0.05600604  328.6864    1e-04 
These estimates are all very small, suggesting that the variance of your depression variable is very small.  Try multiplying the data values by 10, which should make the variance greater than 1.0. Small variances sail close to regions where the covariance matrix is not positive definite or difficult to invert accurately.  These issues in turn can make optimization perform poorly.
That the b estimate seems to be stuck at its starting value of .01 and is not identified, which suggests a scripting error - but one that can't be solved without seeing the script being used.
Log in or register to post comments
  Login

News