You are here

OpenMX Status: 5

2 posts / 0 new
Last post
sorin.valcea's picture
Offline
Joined: 05/31/2017 - 16:52
OpenMX Status: 5

Hello all,

I am getting status 5 for some analyses and am not sure why. I am including the code and output below. There are 4 clusters and 9 effects. Any idea what is causing it?

Thanks!

#### CODE

clustvar <-c(3.1,3.1,41.1,52.1,52.1,52.1,52.1,56.1,56.1)
effsize <-c(0.0370169,-0.04603249,0.25541281,0.20065009,0.18405043,0.12667322,0.21526454,0.01300073,-0.05304971)
samplvar <-c(0.01020408,0.01020408,0.01298701,0.01538462,0.01538462,0.01538462,0.01538462,0.0078125,0.0078125)
summary( meta3(y=effsize, v=samplvar,cluster=clustvar) )

### End Code
Output below

Call:
meta3(y = effsize, v = samplvar, cluster = clustvar)

95% confidence intervals: z statistic approximation
Coefficients:
Estimate Std.Error lbound ubound z value Pr(>|z|)
Intercept 8.5684e-02 5.5208e-02 -2.2522e-02 1.9389e-01 1.5520 0.1207
Tau2_2 1.4867e-09 NA NA NA NA NA
Tau2_3 6.3242e-03 6.6531e-03 -6.7157e-03 1.9364e-02 0.9506 0.3418

Q statistic on the homogeneity of effect sizes: 9.921885 Degrees of freedom of the Q statistic: 8 P value of the Q statistic: 0.2705515

Heterogeneity indices (based on the estimated Tau2):
Estimate
I2_2 (Typical v: Q statistic) 0.0000
I2_3 (Typical v: Q statistic) 0.3543

Number of studies (or clusters): 4
Number of observed statistics: 9
Number of estimated parameters: 3
Degrees of freedom: 6
-2 log likelihood: -15.12369
OpenMx status1: 5 ("0" or "1": The optimization is considered fine.
Other values may indicate problems.)

Mike Cheung's picture
Offline
Joined: 10/08/2009 - 22:37
Dear Sorin,

Dear Sorin,

I think that the issue is related to the small numbers of studies and effect sizes:

## Number of effect sizes per cluster 
sapply(split(effsize, clustvar), length)
##  3.1 41.1 52.1 56.1 
##    2    1    4    2

If you rerun it, the results seem fine. Tau2_2 is amost zero indicating that there is not much variation within studies.

fit1 <- meta3(y=effsize, v=samplvar,cluster=clustvar)
fit1 <- rerun(fit1, extraTries=50)
summary(fit1)
## 
## Call:
## meta3(y = effsize, v = samplvar, cluster = clustvar)
## 
## 95% confidence intervals: z statistic approximation
## Coefficients:
##              Estimate   Std.Error      lbound      ubound z value Pr(>|z|)
## Intercept  8.5683e-02  5.5208e-02 -2.2522e-02  1.9389e-01  1.5520   0.1207
## Tau2_2     1.0000e-10          NA          NA          NA      NA       NA
## Tau2_3     6.3242e-03  6.6531e-03 -6.7157e-03  1.9364e-02  0.9506   0.3418
## 
## Q statistic on the homogeneity of effect sizes: 9.921885
## Degrees of freedom of the Q statistic: 8
## P value of the Q statistic: 0.2705515
## 
## Heterogeneity indices (based on the estimated Tau2):
##                               Estimate
## I2_2 (Typical v: Q statistic)   0.0000
## I2_3 (Typical v: Q statistic)   0.3543
## 
## Number of studies (or clusters): 4
## Number of observed statistics: 9
## Number of estimated parameters: 3
## Degrees of freedom: 6
## -2 log likelihood: -15.12369 
## OpenMx status1: 0 ("0" or "1": The optimization is considered fine.
## Other values may indicate problems.)

Since there are only a few studies (and effect sizes), the findings are highly questionable.

Mike