tssem1 model not converging
whitneyR
Joined: 07/29/2020
| Attachment | Size |
|---|---|
| baek.csv | 1.75 KB |
| caspi.csv | 1.82 KB |
| class.csv | 1.95 KB |
| conway.csv | 1.86 KB |
| dejong.csv | 1.89 KB |
| farmer.csv | 1.8 KB |
| forbes.csv | 2.44 KB |
| friedman1.csv | 1.75 KB |
| friedman2.csv | 1.75 KB |
| roysamb.csv | 2.76 KB |
| wrightsimms.csv | 2.36 KB |
| masem_script.R | 3.94 KB |
Hello,
I am having issues pooling correlation matrices with random effects. I have results from a total of ~30 studies I plan to pool. At this stage, I am trying to pool 11 of these to test things out, but when I run tssem the model never converges.
There are a few possible issues I'm wondering about. These are very large correlation tables (i.e., many variables) and none of the data sets include all correlations (but every cell has data from at least one study), thus there are many NAs, and I wonder whether the software is just having difficulty with estimation. I also wonder whether there are too few studies to estimate the random effects for so many correlations. This is also my first time attempting MASEM, so I may have very well made a simple coding error that I'm missing.
I would appreciate any help with this. The code and data files are attached. Please let me know if there are any problems loading the files or if any clarification is needed.
Thank you,
Whitney
Hi, Whitney.
Hi, Whitney.
There is no data in many of the cells. I am afraid that it is not possible to fit the model. If you want to fit it, you need to simplify it a lot.
Best,
Mike
> pattern.na(data$matrices, show.na=F)
ADHD Alc Anorexia ASPD AvPD Bipolar BPD Bulimia Conduct DeprPD DpndPD Drug
ADHD 3 2 0 2 0 1 0 0 3 0 0 3
Alc 2 10 1 6 2 5 3 2 6 1 3 9
Anorexia 0 1 1 1 1 0 1 0 1 1 1 1
ASPD 2 6 1 6 2 2 3 0 3 1 3 6
AvPD 0 2 1 2 2 0 2 0 1 1 2 2
Bipolar 1 5 0 2 0 6 1 2 3 0 1 5
BPD 0 3 1 3 2 1 3 0 1 1 3 3
Bulimia 0 2 0 0 0 2 0 2 1 0 0 1
Conduct 3 6 1 3 1 3 1 1 6 1 1 7
DeprPD 0 1 1 1 1 0 1 0 1 1 1 1
DpndPD 0 3 1 3 2 1 3 0 1 1 3 3
Drug 3 9 1 6 2 5 3 1 7 1 3 10
Dysthymia 1 4 1 2 2 2 2 1 4 1 2 5
GAD 3 8 1 6 2 4 3 0 6 1 3 9
HPD 0 3 1 3 2 1 3 0 1 1 3 3
MDD 3 9 1 6 2 5 3 1 7 1 3 10
Nicotine 2 4 0 3 0 2 0 0 3 0 0 4
NPD 0 3 1 3 2 1 3 0 1 1 3 3
OCD 0 5 0 3 1 4 2 1 1 0 2 4
OCPD 0 3 1 3 2 1 3 0 1 1 3 3
ODD 1 2 0 0 0 2 0 1 3 0 0 3
Pain 0 2 1 2 1 1 2 0 1 1 2 2
Panic 1 7 1 4 2 5 3 2 4 1 3 7
PPD 0 3 1 3 2 1 3 0 1 1 3 3
Psychosis 0 3 0 2 1 2 2 0 1 0 2 3
PTSD 1 6 1 4 2 4 3 1 4 1 3 7
Separation 0 0 0 0 0 0 0 0 0 0 0 0
Social 1 5 1 3 2 3 2 1 4 1 2 6
Somataform 0 2 0 1 0 2 1 1 0 0 1 1
SpecPhob 1 6 1 3 1 5 2 2 4 1 2 6
StyPD 0 3 1 3 2 1 3 0 1 1 3 3
Dysthymia GAD HPD MDD Nicotine NPD OCD OCPD ODD Pain Panic PPD Psychosis PTSD
ADHD 1 3 0 3 2 0 0 0 1 0 1 0 0 1
Alc 4 8 3 9 4 3 5 3 2 2 7 3 3 6
Anorexia 1 1 1 1 0 1 0 1 0 1 1 1 0 1
ASPD 2 6 3 6 3 3 3 3 0 2 4 3 2 4
AvPD 2 2 2 2 0 2 1 2 0 1 2 2 1 2
Bipolar 2 4 1 5 2 1 4 1 2 1 5 1 2 4
BPD 2 3 3 3 0 3 2 3 0 2 3 3 2 3
Bulimia 1 0 0 1 0 0 1 0 1 0 2 0 0 1
Conduct 4 6 1 7 3 1 1 1 3 1 4 1 1 4
DeprPD 1 1 1 1 0 1 0 1 0 1 1 1 0 1
DpndPD 2 3 3 3 0 3 2 3 0 2 3 3 2 3
Drug 5 9 3 10 4 3 4 3 3 2 7 3 3 7
Dysthymia 5 4 2 5 0 2 1 2 3 1 5 2 1 5
GAD 4 9 3 10 4 3 4 3 3 2 7 3 3 7
HPD 2 3 3 3 0 3 2 3 0 2 3 3 2 3
MDD 5 10 3 10 4 3 4 3 3 2 7 3 3 7
Nicotine 0 4 0 4 4 0 2 0 0 0 1 0 1 1
NPD 2 3 3 3 0 3 2 3 0 2 3 3 2 3
OCD 1 4 2 4 2 2 5 2 0 1 4 2 3 3
OCPD 2 3 3 3 0 3 2 3 0 2 3 3 2 3
ODD 3 3 0 3 0 0 0 0 3 0 3 0 0 3
Pain 1 2 2 2 0 2 1 2 0 2 2 2 1 2
Panic 5 7 3 7 1 3 4 3 3 2 8 3 2 7
PPD 2 3 3 3 0 3 2 3 0 2 3 3 2 3
Psychosis 1 3 2 3 1 2 3 2 0 1 2 2 3 2
PTSD 5 7 3 7 1 3 3 3 3 2 7 3 2 7
Separation 0 0 0 0 0 0 0 0 0 0 0 0 0 0
Social 5 6 2 6 1 2 2 2 3 1 6 2 1 6
Somataform 0 1 1 1 0 1 2 1 0 1 2 1 1 1
SpecPhob 4 6 2 6 1 2 3 2 3 2 7 2 1 6
StyPD 2 3 3 3 0 3 2 3 0 2 3 3 2 3
Separation Social Somataform SpecPhob StyPD SzPD
ADHD 0 1 0 1 0 0
Alc 0 5 2 6 3 3
Anorexia 0 1 0 1 1 1
ASPD 0 3 1 3 3 3
AvPD 0 2 0 1 2 2
Bipolar 0 3 2 5 1 1
BPD 0 2 1 2 3 3
Bulimia 0 1 1 2 0 0
Conduct 0 4 0 4 1 1
DeprPD 0 1 0 1 1 1
DpndPD 0 2 1 2 3 3
Drug 0 6 1 6 3 3
Dysthymia 0 5 0 4 2 2
GAD 0 6 1 6 3 3
HPD 0 2 1 2 3 3
MDD 0 6 1 6 3 3
Nicotine 0 1 0 1 0 0
NPD 0 2 1 2 3 3
OCD 0 2 2 3 2 2
OCPD 0 2 1 2 3 3
ODD 0 3 0 3 0 0
Pain 0 1 1 2 2 2
Panic 0 6 2 7 3 3
PPD 0 2 1 2 3 3
Psychosis 0 1 1 1 2 2
PTSD 0 6 1 6 3 3
Separation 0 1 0 1 0 0
Social 1 6 0 5 2 2
Somataform 0 0 2 2 1 1
SpecPhob 1 5 2 7 2 2
StyPD 0 2 1 2 3 3
[ reached getOption("max.print") -- omitted 1 row ]
Log in or register to post comments
In reply to Hi, Whitney. by Mike Cheung
Hi Mike,
Hi Mike,
Thank you for the quick reply. Hopefully once I include all the data sets there will be fewer cells without data. To clarify, there are no other methods aside from removing variables to handle this type of missingness?
Best,
Whitney
Log in or register to post comments
In reply to Hi Mike, by whitneyR
Hi Whitney,
Hi Whitney,
I am afraid not. It is supposed to be a meta-analysis. We cannot create some data from nothing.
Best,
Mike
Log in or register to post comments
In reply to Hi Whitney, by Mike Cheung
Hi Mike,
Hi Mike,
That makes perfect sense, this should have been obvious to me. One final question since you've been so patient and helpful this far is, in principle, can the correlation matrix be estimated with only one study contributing to some cells? So every cell has an observed correlation, but some cells have only a single study?
Thank you,
Whitney
Log in or register to post comments
In reply to Hi Mike, by whitneyR
Sorry, in case it’s relevant,
Sorry, in case it’s relevant, I plan to use the correlation table to do a factor analysis.
Log in or register to post comments
In reply to Hi Mike, by whitneyR
Hi Whitney,
Hi Whitney,
The program allows you to run the analyses with some cells having only a single study. However, I won't recommend to do it. We cannot call it a "meta-analysis" as there is only one study.
Mike
Log in or register to post comments
In reply to Hi Whitney, by Mike Cheung
Hi Mike,
Hi Mike,
I'm hoping you can help me figure out what my issue is now that I have a matrix with at least 2 studies contributing to each cell. There are 30 studies. Attached is a file showing the distribution of studies per cell. When I try and pool the studies with the random effects model, it still doesn't converge. No error messages, it just doesn't produce output even after an hour. Do you have any suggests for trouble shooting this?
Thank you so much,
Whitney
Log in or register to post comments
Could you please post the
Could you please post the data and code?
Log in or register to post comments
In reply to Could you please post the by Mike Cheung
Certainly. The only reason I
Certainly. The only reason I didn't was because of the number of files I would need to upload. Is there a preferred method for uploading all 30 datasets? Or are individual csv files okay?
Log in or register to post comments
You may create a data file
You may create a data file and save it into *.R with the dump() function.
Log in or register to post comments
In reply to You may create a data file by Mike Cheung
Thank you for telling me
Thank you for telling me about this function, super helpful
Attached is the .R file with data.
Log in or register to post comments
There are 21x20/2=210
There are 21x20/2=210 correlation coefficients. Even if we restrict all the covariances of the random effects at zero, we still have to estimate 210 means (average correlation coefficients) and 210 variances of the random effects. Your model is "huge" compared to the data you have.
If you are comfortable using a fixed-effects model, you may try it. It may work, although it will impose stronger assumptions on your data.
Log in or register to post comments