Skip to first unread message
garc...@unlv.nevada.edu
Jul 23, 2018, 6:06:53 PM7/23/18
to lavaan
Hello,
I am currently working on a project where I am fitting several CFA models to imputed data. I am running into an issue where I get warnings when attempting to fit the models even though, according to the output, they seem to be converging normally.
Here is the syntax I am running:
# Model B- Two factor (noncompliance & negative affect)
# Specify Model B
mod.b2 <- 'NonComp.f =~ odd1sc + odd2sc + odd3sc + odds4sc + odds5sc + odds6sc
NA.f =~ odds1sc + odds2sc + odds3sc
NonComp.f ~~ NA.f'
# Fit Model B
fit.mod.b2.wlsmvs.mi <- runMI(model = mod.b2, data = bag.sem, fun = "cfa",
estimator = "WLSMVS", std.lv = TRUE, verbose = TRUE, control = list(rel.tol = 1e-4),
ordered = c("odd1sc", "odd2sc", "odd3sc", "odds1sc", "odds2sc",
"odds3sc", "odds4sc", "odds5sc", "odds6sc"))
Here is the output:
lavaan.mi object based on 30 imputed data sets.
See class?lavaan.mi help page for available methods.
Convergence information:
The model converged on 30 imputed data sets
Rubin's (1987) rules were used to pool point and SE estimates across 30 imputed data sets, and to calculate degrees of freedom for each parameter's t test and CI.
Parameter Estimates:
Information Expected
Information saturated (h1) model Unstructured
Standard Errors Robust.sem
Latent Variables:
Estimate Std.Err t-value df P(>|t|)
NonComp.f =~
odd1sc 0.841 0.020 41.949 Inf 0.000
odd2sc 0.851 0.018 46.639 Inf 0.000
odd3sc 0.795 0.022 36.032 Inf 0.000
odds4sc 0.495 0.037 13.320 730.651 0.000
odds5sc 0.638 0.031 20.569 1093.553 0.000
odds6sc 0.679 0.029 23.579 1050.360 0.000
NA.f =~
odds1sc 0.815 0.021 39.312 785.371 0.000
odds2sc 0.902 0.015 59.176 789.892 0.000
odds3sc 0.694 0.029 23.941 713.587 0.000
Covariances:
Estimate Std.Err t-value df P(>|t|)
NonComp.f ~~
NA.f 0.950 0.016 60.938 660.708 0.000
Intercepts:
Estimate Std.Err t-value df P(>|t|)
.odd1sc 0.000
.odd2sc 0.000
.odd3sc 0.000
.odds4sc 0.000
.odds5sc 0.000
.odds6sc 0.000
.odds1sc 0.000
.odds2sc 0.000
.odds3sc 0.000
NonComp.f 0.000
NA.f 0.000
Thresholds:
Estimate Std.Err t-value df P(>|t|)
odd1sc|t1 -0.710 0.051 -13.877 Inf 0.000
odd1sc|t2 -0.102 0.047 -2.185 Inf 0.029
odd2sc|t1 -0.722 0.051 -14.070 Inf 0.000
odd2sc|t2 -0.027 0.047 -0.580 Inf 0.562
odd3sc|t1 -0.668 0.051 -13.196 Inf 0.000
odd3sc|t2 -0.011 0.047 -0.229 Inf 0.819
odds4sc|t1 -0.143 0.047 -3.050 1261.238 0.002
odds4sc|t2 0.521 0.049 10.637 2043.255 0.000
odds5sc|t1 -0.541 0.049 -10.985 2409.018 0.000
odds5sc|t2 0.100 0.047 2.144 3309.839 0.032
odds6sc|t1 -0.708 0.051 -13.838 754.554 0.000
odds6sc|t2 -0.013 0.047 -0.281 Inf 0.778
odds1sc|t1 -0.636 0.050 -12.665 1664.714 0.000
odds1sc|t2 0.027 0.047 0.578 Inf 0.563
odds2sc|t1 -0.434 0.048 -8.997 2695.879 0.000
odds2sc|t2 0.217 0.047 4.622 Inf 0.000
odds3sc|t1 -0.362 0.048 -7.582 1167.343 0.000
odds3sc|t2 0.344 0.048 7.212 6106.076 0.000
Variances:
Estimate Std.Err t-value df P(>|t|)
.odd1sc 0.293
.odd2sc 0.276
.odd3sc 0.368
.odds4sc 0.754
.odds5sc 0.593
.odds6sc 0.539
.odds1sc 0.336
.odds2sc 0.186
.odds3sc 0.519
NonComp.f 1.000
NA.f 1.000
Scales y*:
Estimate Std.Err t-value df P(>|t|)
odd1sc 1.000
odd2sc 1.000
odd3sc 1.000
odds4sc 1.000
odds5sc 1.000
odds6sc 1.000
odds1sc 1.000
odds2sc 1.000
odds3sc 1.000
And here are the warnings:
Warning messages:
1: In lavaan::lavaan(model = "NonComp.f =~ odd1sc + odd2sc + odd3sc + odds4sc + odds5sc + odds6sc\n NA.f =~ odds1sc + odds2sc + odds3sc\n NonComp.f ~~ NA.f", ... :
lavaan WARNING: not all elements of the gradient are (near) zero;
the optimizer may not have found a local solution;
use lavInspect(fit, "optim.gradient") to investigate
2: In lavaan(slotOptions = structure(list(model.type = "cfa", ... :
lavaan WARNING: not all elements of the gradient are (near) zero;
the optimizer may not have found a local solution;
use lavInspect(fit, "optim.gradient") to investigate
3: In lavaan(slotOptions = structure(list(model.type = "cfa", ... :
lavaan WARNING: not all elements of the gradient are (near) zero;
the optimizer may not have found a local solution;
use lavInspect(fit, "optim.gradient") to investigate
4: In lavaan(slotOptions = structure(list(model.type = "cfa", ... :
lavaan WARNING: not all elements of the gradient are (near) zero;
the optimizer may not have found a local solution;
use lavInspect(fit, "optim.gradient") to investigate
5: In lavaan(slotOptions = structure(list(model.type = "cfa", ... :
lavaan WARNING: not all elements of the gradient are (near) zero;
the optimizer may not have found a local solution;
use lavInspect(fit, "optim.gradient") to investigate
6: In lavaan(slotOptions = structure(list(model.type = "cfa", ... :
lavaan WARNING: not all elements of the gradient are (near) zero;
the optimizer may not have found a local solution;
use lavInspect(fit, "optim.gradient") to investigate
7: In lavaan(slotOptions = structure(list(model.type = "cfa", ... :
lavaan WARNING: not all elements of the gradient are (near) zero;
the optimizer may not have found a local solution;
use lavInspect(fit, "optim.gradient") to investigate
8: In lavaan(slotOptions = structure(list(model.type = "cfa", ... :
lavaan WARNING: not all elements of the gradient are (near) zero;
the optimizer may not have found a local solution;
use lavInspect(fit, "optim.gradient") to investigate
9: In lavaan(slotOptions = structure(list(model.type = "cfa", ... :
lavaan WARNING: not all elements of the gradient are (near) zero;
the optimizer may not have found a local solution;
use lavInspect(fit, "optim.gradient") to investigate
10: In lavaan(slotOptions = structure(list(model.type = "cfa", ... :
lavaan WARNING: not all elements of the gradient are (near) zero;
the optimizer may not have found a local solution;
use lavInspect(fit, "optim.gradient") to investigate
11: In lavaan(slotOptions = structure(list(model.type = "cfa", ... :
lavaan WARNING: not all elements of the gradient are (near) zero;
the optimizer may not have found a local solution;
use lavInspect(fit, "optim.gradient") to investigate
12: In lavaan(slotOptions = structure(list(model.type = "cfa", ... :
lavaan WARNING: not all elements of the gradient are (near) zero;
the optimizer may not have found a local solution;
use lavInspect(fit, "optim.gradient") to investigate
13: In lavaan(slotOptions = structure(list(model.type = "cfa", ... :
lavaan WARNING: not all elements of the gradient are (near) zero;
the optimizer may not have found a local solution;
use lavInspect(fit, "optim.gradient") to investigate
14: In lavaan(slotOptions = structure(list(model.type = "cfa", ... :
lavaan WARNING: not all elements of the gradient are (near) zero;
the optimizer may not have found a local solution;
use lavInspect(fit, "optim.gradient") to investigate
15: In lavaan(slotOptions = structure(list(model.type = "cfa", ... :
lavaan WARNING: not all elements of the gradient are (near) zero;
the optimizer may not have found a local solution;
use lavInspect(fit, "optim.gradient") to investigate
16: In lavaan(slotOptions = structure(list(model.type = "cfa", ... :
lavaan WARNING: not all elements of the gradient are (near) zero;
the optimizer may not have found a local solution;
use lavInspect(fit, "optim.gradient") to investigate
17: In lavaan(slotOptions = structure(list(model.type = "cfa", ... :
lavaan WARNING: not all elements of the gradient are (near) zero;
the optimizer may not have found a local solution;
use lavInspect(fit, "optim.gradient") to investigate
18: In lavaan(slotOptions = structure(list(model.type = "cfa", ... :
lavaan WARNING: not all elements of the gradient are (near) zero;
the optimizer may not have found a local solution;
use lavInspect(fit, "optim.gradient") to investigate
19: In lavaan(slotOptions = structure(list(model.type = "cfa", ... :
lavaan WARNING: not all elements of the gradient are (near) zero;
the optimizer may not have found a local solution;
use lavInspect(fit, "optim.gradient") to investigate
20: In lavaan(slotOptions = structure(list(model.type = "cfa", ... :
lavaan WARNING: not all elements of the gradient are (near) zero;
the optimizer may not have found a local solution;
use lavInspect(fit, "optim.gradient") to investigate
21: In lavaan(slotOptions = structure(list(model.type = "cfa", ... :
lavaan WARNING: not all elements of the gradient are (near) zero;
the optimizer may not have found a local solution;
use lavInspect(fit, "optim.gradient") to investigate
22: In lavaan(slotOptions = structure(list(model.type = "cfa", ... :
lavaan WARNING: not all elements of the gradient are (near) zero;
the optimizer may not have found a local solution;
use lavInspect(fit, "optim.gradient") to investigate
23: In lavaan(slotOptions = structure(list(model.type = "cfa", ... :
lavaan WARNING: not all elements of the gradient are (near) zero;
the optimizer may not have found a local solution;
use lavInspect(fit, "optim.gradient") to investigate
24: In lavaan(slotOptions = structure(list(model.type = "cfa", ... :
lavaan WARNING: not all elements of the gradient are (near) zero;
the optimizer may not have found a local solution;
use lavInspect(fit, "optim.gradient") to investigate
25: In lavaan(slotOptions = structure(list(model.type = "cfa", ... :
lavaan WARNING: not all elements of the gradient are (near) zero;
the optimizer may not have found a local solution;
use lavInspect(fit, "optim.gradient") to investigate
26: In lavaan(slotOptions = structure(list(model.type = "cfa", ... :
lavaan WARNING: not all elements of the gradient are (near) zero;
the optimizer may not have found a local solution;
use lavInspect(fit, "optim.gradient") to investigate
27: In lavaan(slotOptions = structure(list(model.type = "cfa", ... :
lavaan WARNING: not all elements of the gradient are (near) zero;
the optimizer may not have found a local solution;
use lavInspect(fit, "optim.gradient") to investigate
28: In lavaan(slotOptions = structure(list(model.type = "cfa", ... :
lavaan WARNING: not all elements of the gradient are (near) zero;
the optimizer may not have found a local solution;
use lavInspect(fit, "optim.gradient") to investigate
Given that I am new to R/lavaan and my knowledge base for SEM is moderate at best, I am hoping someone is able to explain these warning messages to me and whether they are something I should be concerned about.
Thank you in advance!
Terrence Jorgensen
Jul 24, 2018, 8:35:42 AM7/24/18
to lavaan
I am hoping someone is able to explain these warning messages to me and whether they are something I should be concerned about.
Warnings indicate there might be a problem, not necessarily that there is one. It is up to the user to figure that out. If you want to look into the information suggested by the warning, you can add a user-defined function to your lavaan.mi object using the FUN= argument. See the ordered-categorical example at the bottom of the ?runMI help page to see how you can extract lavInspect() (or other) output from each imputation and inspect results afterward. Instead of "zero.cell.tables" in the example, you could just request "optim.gradient".
But the SEs are all small relative to the estimates, so I am guessing there isn't a problem. If there were a problem, I would imagine different local minima would be found using different imputations (although maybe the fraction of missing information is low), which would increase the between-imputation variance and inflate the SEs. You could also try fitting the model to the unimputed data with missing = "pairwise" to see if those estimates differ greatly from the pooled results over imputations (although differences could be expected if data are MAR rather than MCAR.
Terrence D. Jorgensen
Postdoctoral Researcher, Methods and Statistics
Research Institute for Child Development and Education, the University of Amsterdam
UvA web page: http://www.uva.nl/profile/t.d.jorgensen
garc...@unlv.nevada.edu
Jul 25, 2018, 6:16:50 PM7/25/18
to lavaan
Hello Dr. Jorgensen,
Thank you for your response!
I did as was suggested by both you and the warnings I received and tried the lavInspect(object, what = "optim.gradient") function to attempt to find out more information about what the warnings are telling me. However this returns an error that says "Error in lavListInspect(object = object, what = what, add.labels = add.labels, : object 'optim.gradient' not found".
Additionally, I tried both lavInspect(object, what = "optim") and lavInspect(object, what = "gradient"). Both returned the similar error messages saying that object'optim' and object'gradient' could not be found.
Here is the code I ran:
lavInspect(fit.mod.b2.wlsmvs.mi, what = "optim.gradient")
The fit.mod object is the object I used to fit my model. Is it something on my end that I am doing wrong? Any advice would be much appreciated.
Thank you,
Breanna Garcia
Terrence Jorgensen
Jul 29, 2018, 6:29:15 PM7/29/18
to lavaan
"Error ... : object 'optim.gradient' not found".
That sounds like you forgot to put "optim.gradient" in quotation marks, like this:
example(cfa)
lavInspect(fit, optim.gradient)Message has been deleted
Yves Rosseel
Jul 31, 2018, 5:14:34 AM7/31/18
to lav...@googlegroups.com
> Error in lavListInspect(object = object, what = what, add.labels =
> add.labels, :
> unknown `what' argument in inspect function: `optim.gradient'
> add.labels, :
Indeed, this will not work (yet, 0.6-2). The lavListInspect() function
only offers a tiny subset of the lavInspect() options. The reason is
that the semList() function does not store 'full' lavaan objects, as
this would not scale (if the number of datasets gets large).
For your setting, I would run one imputed dataset at a time, using
regular sem(), and use the lavInspect(fit, "optim.gradient") to check
the gradient afterwards.
Yves.
Shimon Sarraf
Nov 13, 2018, 1:43:51 PM11/13/18
to lavaan
I encountered a similar warning message while running a growth curve model with squared and cubed slopes. The variances of s2 and s3 (the squared and cubed slopes in my model) are both negative. Is this the reason for the error message? I understand that negative variances generally speaking are problematic but want to confirm in this case. Any references about how to address this issue are much appreciated. By the way, all model fit statistics and parameter standard errors seem satisfactory.
Thank you,
Shimon
PS: here's my output...
lavInspect(RR.fit.b1, "optim.gradient")
(eq.con) i~~i s~~s s2~~s2 s3~~s3 i~~s i~~s2 i~~s3 s~~s2 s~~s3
0.000 0.000 0.000 -0.017 -0.956 0.000 -0.001 -0.004 -0.005 -0.035
s2~~s3 i~1 s~1 s2~1 s3~1
-0.251 0.000 0.000 0.000 0.000
Terrence Jorgensen
Nov 22, 2018, 8:54:55 AM11/22/18
to lavaan
I encountered a similar warning message ... Is this the reason for the error message?
Is it a warning or error? Heywood cases produce warnings because there might be a problem that deserves some attention. If your model has decent fit, I typically recommend looking at the CIs to see if your estimates include plausible values, in which case the Heywood case is consistent with sampling error.
In the context of growth modeling, imagine you were using multilevel modeling. Would you allow the quadratic and cubic slopes to be random, or only the linear one? If it is plausible that they are fixed across subjects, then that would mean the population variances of those growth factors are zero. But unlike MLM software, SEM software will not automatically force variances to be non-negative. That is actually a good thing, because forcing a negative maximum-likelihood estimate to be zero would mean your chi-squared statistic is no longer distributed as expected.
Unconstrained estimation is better, and you can interpret the interval estimate (95% CI) instead of the point estimate.
Terrence D. Jorgensen
Assistant Professor, Methods and Statistics
Research Institute for Child Development and Education, the University of Amsterdam
Reply all
Reply to author
Forward
0 new messages