mikewlcheung
tssem2 error
Hello Mike,
I am performing a meta-analysis (with 13 variables) using the “metaSEM” package.
I don't know if this is because the model is complex or because there are many missing correlations, but when I run “tssem2” I get a bunch of problems with error messages (tssem1 takes a long time (about an hour) but produces results normally).
The error messages are:
“1: sqrt(Diag(vcov.wls(object))) NaN”
“2: print.summary.wls(x): OpenMx status1 is neither 0 or 1. You are advised to ‘rerun’ it again.”
and sometimes,
“Polite note from mxTryHard: Hessian not checked as model contains mxConstraints”
I keep trying to rerun it, but I keep getting NaNs (especially when using likelihood-based confidence intervals), and I get values that are impossible to come up with (e.g. -2.113, or a negative value when it shouldn't be possible to come up with a negative value at all).
Could you please take a look at the attached my data and code and see what the problem is? Thank you.
Could you please include the R code to read the data? It is hard to check without reproducing the errors.
I've attached a text file of the entire process, including the R code. I would really appreciate it if you could take a look and let me know what's causing the problem. Thank you. lastsave.txt
It works better if you use diag.constraints=FALSE in this model. The following code works for me.
stage2 = tssem2(stage1random, Amatrix=A, Smatrix=S,diag.constraints=FALSE,intervals.type="z")
summary(stage2)
> summary(stage2)
Call:
wls(Cov = pooledS, aCov = aCov, n = tssem1.obj$total.n, RAM = RAM,
Amatrix = Amatrix, Smatrix = Smatrix, Fmatrix = Fmatrix,
diag.constraints = diag.constraints, cor.analysis = cor.analysis,
intervals.type = intervals.type, mx.algebras = mx.algebras,
mxModel.Args = mxModel.Args, subset.variables = subset.variables,
model.name = model.name, suppressWarnings = suppressWarnings,
silent = silent, run = run)
95% confidence intervals: z statistic approximation
Coefficients:
Estimate Std.Error lbound ubound z value Pr(>|z|)
b106 0.620805 0.036968 0.548349 0.693260 16.7932 < 2.2e-16 ***
b109 0.279495 0.040409 0.200296 0.358695 6.9167 4.623e-12 ***
b111 0.207178 0.064638 0.080489 0.333866 3.2052 0.0013497 **
b112 0.091933 0.056985 -0.019756 0.203622 1.6133 0.1066846
b113 0.021483 0.060856 -0.097794 0.140759 0.3530 0.7240857
b116 0.232994 0.051435 0.132183 0.333805 4.5299 5.902e-06 ***
b119 0.381586 0.038774 0.305591 0.457581 9.8414 < 2.2e-16 ***
b1210 0.361437 0.051973 0.259571 0.463303 6.9543 3.544e-12 ***
b1211 0.359614 0.046320 0.268828 0.450399 7.7637 8.216e-15 ***
b129 0.216031 0.046035 0.125805 0.306257 4.6928 2.695e-06 ***
b1310 0.236028 0.047946 0.142055 0.330001 4.9228 8.533e-07 ***
b1311 0.269357 0.043248 0.184592 0.354121 6.2282 4.719e-10 ***
b1312 0.431349 0.062691 0.308476 0.554222 6.8805 5.964e-12 ***
b94 0.092657 0.112126 -0.127107 0.312421 0.8264 0.4085992
b95 0.685681 0.186260 0.320617 1.050745 3.6813 0.0002320 ***
b96 -1.201594 0.343349 -1.874547 -0.528642 -3.4996 0.0004659 ***
b97 1.111854 0.251584 0.618758 1.604951 4.4194 9.897e-06 ***
b98 0.138821 0.143336 -0.142112 0.419754 0.9685 0.3327938
p62 0.536688 0.036330 0.465482 0.607894 14.7725 < 2.2e-16 ***
p72 0.609550 0.030796 0.549191 0.669908 19.7934 < 2.2e-16 ***
p42 0.504761 0.022788 0.460098 0.549424 22.1507 < 2.2e-16 ***
p82 0.492029 0.050320 0.393403 0.590654 9.7780 < 2.2e-16 ***
p32 0.559028 0.045737 0.469385 0.648671 12.2226 < 2.2e-16 ***
p52 0.616778 0.036728 0.544792 0.688763 16.7931 < 2.2e-16 ***
p76 0.817995 0.040601 0.738418 0.897572 20.1471 < 2.2e-16 ***
p86 0.600213 0.035588 0.530463 0.669964 16.8657 < 2.2e-16 ***
p87 0.586971 0.049288 0.490368 0.683574 11.9089 < 2.2e-16 ***
p64 0.604648 0.035375 0.535314 0.673981 17.0926 < 2.2e-16 ***
p74 0.608988 0.020168 0.569460 0.648516 30.1961 < 2.2e-16 ***
p84 0.540567 0.048510 0.445488 0.635645 11.1434 < 2.2e-16 ***
p54 0.612588 0.019861 0.573660 0.651516 30.8430 < 2.2e-16 ***
p21 0.561069 0.033812 0.494799 0.627338 16.5940 < 2.2e-16 ***
p61 0.576889 0.035386 0.507534 0.646244 16.3028 < 2.2e-16 ***
p71 0.651274 0.031139 0.590242 0.712306 20.9148 < 2.2e-16 ***
p41 0.522513 0.030269 0.463187 0.581840 17.2622 < 2.2e-16 ***
p81 0.526480 0.033625 0.460577 0.592384 15.6575 < 2.2e-16 ***
p31 0.647821 0.047624 0.554480 0.741161 13.6029 < 2.2e-16 ***
p51 0.592772 0.042540 0.509395 0.676149 13.9344 < 2.2e-16 ***
p63 0.579444 0.033002 0.514761 0.644127 17.5578 < 2.2e-16 ***
p73 0.634354 0.034057 0.567604 0.701104 18.6264 < 2.2e-16 ***
p43 0.532117 0.053182 0.427882 0.636352 10.0055 < 2.2e-16 ***
p83 0.495409 0.037836 0.421251 0.569567 13.0934 < 2.2e-16 ***
p53 0.544873 0.037867 0.470654 0.619091 14.3890 < 2.2e-16 ***
p65 0.730608 0.044916 0.642575 0.818642 16.2662 < 2.2e-16 ***
p75 0.620114 0.040664 0.540414 0.699815 15.2496 < 2.2e-16 ***
p85 0.555708 0.033339 0.490364 0.621052 16.6682 < 2.2e-16 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Goodness-of-fit indices:
Value
Sample size 78047.0000
Chi-square of target model 145.6214
DF of target model 32.0000
p value of target model 0.0000
Number of constraints imposed on "Smatrix" 0.0000
DF manually adjusted 0.0000
Chi-square of independence model 24163.1397
DF of independence model 78.0000
RMSEA 0.0067
RMSEA lower 95% CI 0.0057
RMSEA upper 95% CI 0.0079
SRMR 0.0476
TLI 0.9885
CFI 0.9953
AIC 81.6214
BIC -214.8607
OpenMx status1: 0 ("0" or "1": The optimization is considered fine.
Other values indicate problems.)
Thank you for your reply.
I also tried with “diag.constraints=FALSE” before, but it yielded strange values of 1 or more, as in b96 (-1.202) and b.97 (1.112) in the result you provide. Is this possible? For b96 (-1.202), all correlations are positive, and when simplifying the model, such as reducing the variables, the values become positive.
The problem I ran into was (1) it generates NaNs (2) even if it doesn't generate NaNs, some of the values are unreliable.
A standardized coefficient can be larger than 1. See, for example, the following discussion: https://stats.stackexchange.com/questions/120201/magnitude-of-standardized-coefficients-beta-in-multiple-linear-regression
There are 13 variables and only a few studies on some of these cells exist. Thus, there may not be sufficient information to estimate all of them.
> pattern.na(cormat, show.na = FALSE)
rlb acc scr psn ufd ast cmp rsp peu enj pu att bi
rlb 45 23 15 12 15 8 6 19 31 19 23 9 30
acc 23 59 10 15 11 4 15 17 46 24 43 16 48
scr 15 10 60 11 16 7 8 18 46 24 50 20 41
psn 12 15 11 32 9 7 4 17 20 13 18 8 18
ufd 15 11 16 9 39 7 4 17 27 12 22 8 28
ast 8 4 7 7 7 24 4 8 17 11 14 10 17
cmp 6 15 8 4 4 4 37 8 34 6 33 13 34
rsp 19 17 18 17 17 8 8 47 31 13 23 12 31
peu 31 46 46 20 27 17 34 31 116 48 102 46 94
enj 19 24 24 13 12 11 6 13 48 63 41 21 45
pu 23 43 50 18 22 14 33 23 102 41 113 43 94
att 9 16 20 8 8 10 13 12 46 21 43 51 43
bi 30 48 41 18 28 17 34 31 94 45 94 43 124
Thank you for your response! I'll gather more data and try to analyze it again.