metaSEM
http://openmx.psyc.virginia.edu/taxonomy/term/44/0
enThis forum is about the metaSEM package for meta-analysis
http://openmx.psyc.virginia.edu/thread/2946
<p><a href="http://courses.nus.edu.sg/course/psycwlm/internet">Mike Cheung's</a> metaSEM package is introduced <a href="http://courses.nus.edu.sg/course/psycwlm/Internet/metaSEM">here</a></p>
<p>Post questions to this forum</p>
http://openmx.psyc.virginia.edu/thread/2946#commentsmetaSEMSat, 26 Apr 2014 21:26:16 +0000tbates2946 at http://openmx.psyc.virginia.eduLooking for a global factor for the latent factors
http://openmx.psyc.virginia.edu/thread/4185
<p>Dear Mike,</p>
<p>thank you very much for your metaSEM-package, I really enjoy working with it!</p>
<p>I tested a model with 8 variables and 3 latent factors and now I would like to check, if there is a common factor for these three latent factors at the superordinate level - is it possible with metaSEM? I was looking for specifications like Gamma matrix in LISREL, but I unfortunately couldn't find anything.</p>
<p>Thank you!</p>
<p>Best wishes,<br />
Nastja</p>
<table id="attachments" class="sticky-enabled">
<thead><tr><th>Attachment</th><th>Size</th> </tr></thead>
<tbody>
<tr class="odd"><td><a href="http://openmx.psyc.virginia.edu/sites/default/files/Rplot_X.png">Rplot_X.png</a></td><td>8.94 KB</td> </tr>
<tr class="even"><td><a href="http://openmx.psyc.virginia.edu/sites/default/files/Script_5.R">Script.R</a></td><td>802 bytes</td> </tr>
</tbody>
</table>
http://openmx.psyc.virginia.edu/thread/4185#commentsmetaSEMFri, 16 Sep 2016 16:38:59 +0000nastjuscha4185 at http://openmx.psyc.virginia.eduPossible bug and question: Variable name interference with dummy variable name & Average effect size
http://openmx.psyc.virginia.edu/thread/4182
<p>Hi Mike,</p>
<p>I have two questions, both are related to finding the average effect size of categories in covariates with metaSEM. I've been trying to find the correct way to find the average effect sizes and have found a variety of results for different approaches. One of the reason for these differences seem to be related to my first question/issue, and may be a bug. The second question is simply about the correct way to find average effect sizes.</p>
<p>1. When running one of my 3-level mixed-effects models I use the same name for a dummy variable to indicate categories as for a non-related variable in the data set ("Parent"). This seems to interfere with the results of the 3-level mixed-effects model.</p>
<p>When using the exact same name for the dummy variable ("Parent") as for the non-related variable in the data set, the intercept effect size is -.55. When giving the dummy variable name low caps ("parent") the intercept effect size of the same analysis is -.46 (same when giving a completely different name). The same difference in effect size appears when changing the name of the non-related variable in the data set. I attached an Rscript and two csv files for you to test this. Both csv data sets are identical, apart from the non-related variable name for which one uses a capital ("Parent") and the other does not ("parent").</p>
<p>2. I want to find the average effect sizes with metaSEM for each category for every covariate in my meta-analysis. I understand the following 3 approaches should give the same average effect sizes (as example I'll use the covariate "Parent", consisting of groups "Mother" and "Father"; note, I use a 3-level mixed-effects model in my analysis):</p>
<p>-Run a 3-level mixed-effects model, and use Mother as a reference group, the intercept will give me the average effect size for Mother. I could get the effect size of Father by adding slope1 estimate of those results to the effect size of Mother, or simply run the same model with Father as reference group.<br />
-Run the 3-level mixed effects model while constraining the intercept to 0 and get the effect sizes of Mother and Father both at once.<br />
-Simply run a 3-level mixed-effects model with a data set that ONLY has effect sizes related to Mother, and do the same for Father separately.</p>
<p>The first and second approach mostly give the same or similar effect sizes. The third approach gives different, but more plausible average effect sizes. Which of these methods am I supposed to use to find the average effect sizes of each group with metaSEM?</p>
<p>Thank you kindly in advance Mike.</p>
<p>Best,<br />
Jasper</p>
<table id="attachments" class="sticky-enabled">
<thead><tr><th>Attachment</th><th>Size</th> </tr></thead>
<tbody>
<tr class="odd"><td><a href="http://openmx.psyc.virginia.edu/sites/default/files/metaSEMCAP.csv">metaSEMCAP.csv</a></td><td>2.4 KB</td> </tr>
<tr class="even"><td><a href="http://openmx.psyc.virginia.edu/sites/default/files/metaSEMnocap.csv">metaSEMnocap.csv</a></td><td>2.4 KB</td> </tr>
<tr class="odd"><td><a href="http://openmx.psyc.virginia.edu/sites/default/files/metaSEMnames.R">metaSEMnames.R</a></td><td>1.3 KB</td> </tr>
</tbody>
</table>
http://openmx.psyc.virginia.edu/thread/4182#commentsmetaSEMMon, 12 Sep 2016 12:07:49 +0000jasperd4182 at http://openmx.psyc.virginia.edumissing data
http://openmx.psyc.virginia.edu/thread/4178
<p>How to deal with the missing data using MetaSEM package? (Almost all the variables have missing data and these data are missing at completely random)</p>
http://openmx.psyc.virginia.edu/thread/4178#commentsmetaSEMWed, 24 Aug 2016 14:54:06 +0000crystalzsq4178 at http://openmx.psyc.virginia.eduOverlap of input matrices (missing variables)
http://openmx.psyc.virginia.edu/thread/4177
<p>Hello, </p>
<p>I am Valerie, 25 years old and absolutely fascinated by applying Metasem respectively TSSEM by using metaSEM. When I was doing my first analysis, I came across a problem concerning the input data respectively the input correlation matrices and did not find an solution so far: </p>
<p>Let's say, I have two studies, which are dedicated to the same topic, but slightly differ in their observed variables, what is also reflected in their correlation matrices.<br />
1) Study1 shows a correlation matrix with the variables x1,x2,x3<br />
2) Study2 shows a corrleation matrix with the variables x1,x2,x4<br />
This data is different when compared to the datasets such as Hunter83 provided in the book (Cheung 2015), because in this case, there is no correlation matrix, which is "complete", as variables (and as a result also the correlations) are missing in the studies.<br />
Then input matrix would look as follows (taking random values for correlation coefficients):<br />
1<br />
0.2 1<br />
0.3 0.5 1<br />
NA NA NA NA<br />
1<br />
0.7 1<br />
NA NA NA<br />
0.4 0.2 NA 1 </p>
<p>When I imported the data using the readLowMat() function, it worked out. However, when applying tssem1(), there is an error message in case of method="FEM" (Error in if (!all(isPD)) warning(paste("Group ", (1:no.groups)[!isPD], :<br />
missing value where TRUE/FALSE needed) and in case of method="REM" (Error in if (all.equal(covMatrix, t(covMatrix))) { :<br />
argument is not interpretable as logical). My guess was that the kind of input data is not suitable for TSSEM as there is not "complete matrix" available, when it comes to the equality constraints. But I do not the correct procedure to apply now. I have read about pairwise or listwise deletion, but isn't this an issue, TSSEM is able to avoid? Or do I mix up the cases of missing variables and missign correlations?</p>
<p>I really would appreciate any help and good advice. I send best wishes from Germany! </p>
http://openmx.psyc.virginia.edu/thread/4177#commentsmetaSEMMon, 22 Aug 2016 16:33:22 +0000grafvale4177 at http://openmx.psyc.virginia.eduOutput interpretation
http://openmx.psyc.virginia.edu/thread/4176
<p>Dear Mike,</p>
<p>I fitted a saturated path model over 3 exogenous variables (MO, EO, LO), 1 mediator (INN) and 1 response variable (PERF), but the results are contradictory. </p>
<p>I want to determine the nature of the mediation role of INN (full mediation or partial mediation). </p>
<p>The analysis through direct effects are no consistent with the indirect effects. </p>
<p>I attach the dataset and the commands in R. </p>
<p>Is the model well specified? How can I interpret this output? </p>
<p>Thanks in advance for your attention, I really appreciate your opinion. </p>
<table id="attachments" class="sticky-enabled">
<thead><tr><th>Attachment</th><th>Size</th> </tr></thead>
<tbody>
<tr class="odd"><td><a href="http://openmx.psyc.virginia.edu/sites/default/files/saturatedmodel.jpg">saturatedmodel.jpg</a></td><td>153.75 KB</td> </tr>
<tr class="even"><td><a href="http://openmx.psyc.virginia.edu/sites/default/files/MOEOLO-INN-PERF(SAT).R">MOEOLO-INN-PERF(SAT).R</a></td><td>2.31 KB</td> </tr>
<tr class="odd"><td><a href="http://openmx.psyc.virginia.edu/sites/default/files/dataset155.txt">dataset155.txt</a></td><td>7.19 KB</td> </tr>
</tbody>
</table>
http://openmx.psyc.virginia.edu/thread/4176#commentsmetaSEMSat, 20 Aug 2016 17:59:39 +0000Juanda4176 at http://openmx.psyc.virginia.eduParameter Based MetaSEM
http://openmx.psyc.virginia.edu/thread/4171
<p>Hello all, </p>
<p>I am trying to conduct a parameter based MetaSEM. My model includes several mediators which are in parallel, and one mediator in series with the others (Attached is the figure summarising the model). </p>
<p>I have tried to run the code, however, R is giving me the following error message:<br />
Error in (function (x, n, standardized = TRUE, direct.effect = TRUE, run = TRUE) : length of 'dimnames' [1] not equal to array extent<br />
Error: object 'indirect1' not found</p>
<p>I have attached the code I am currently using.<br />
Based on the figure, have I respected the order of the variables in my code? Also, if I want to include a variable (covariate) predicting all the other variables, but I am not interested in its indirect effects, how can I do so?<br />
I would highly appreciate the help. </p>
<p>Thank you in advance.<br />
Regards,<br />
Arin </p>
<table id="attachments" class="sticky-enabled">
<thead><tr><th>Attachment</th><th>Size</th> </tr></thead>
<tbody>
<tr class="odd"><td><a href="http://openmx.psyc.virginia.edu/sites/default/files/Output.doc">Output.doc</a></td><td>30.5 KB</td> </tr>
<tr class="even"><td><a href="http://openmx.psyc.virginia.edu/sites/default/files/ParameterBased_Aug7.doc">ParameterBased_Aug7.doc</a></td><td>30 KB</td> </tr>
<tr class="odd"><td><a href="http://openmx.psyc.virginia.edu/sites/default/files/Figure.doc">Figure.doc</a></td><td>89.5 KB</td> </tr>
</tbody>
</table>
http://openmx.psyc.virginia.edu/thread/4171#commentsmetaSEMMon, 08 Aug 2016 17:17:31 +0000Arin A4171 at http://openmx.psyc.virginia.eduMissing values error without missings in the data?
http://openmx.psyc.virginia.edu/thread/4168
<p>Dear Mike,</p>
<p>I'm afraid I have to ask you for help again.<br />
I receive a following new error warning every time I try to run an FEM regardless of the constellation of variables from my dataset:<br />
> summary(fixed1)<br />
Error in if (pchisq(chi.squared, df = df, ncp = 0) >= upper) { :<br />
missing value where TRUE/FALSE needed<br />
Even if I try to run it only with first three variables (there are 146 full matrices without any missings) I have the same problem.<br />
I appreciate any idea you can provide.</p>
<p>Regards,<br />
Nastja</p>
<table id="attachments" class="sticky-enabled">
<thead><tr><th>Attachment</th><th>Size</th> </tr></thead>
<tbody>
<tr class="odd"><td><a href="http://openmx.psyc.virginia.edu/sites/default/files/DT_FEM.R">DT_FEM.R</a></td><td>712 bytes</td> </tr>
<tr class="even"><td><a href="http://openmx.psyc.virginia.edu/sites/default/files/dt.dat">dt.dat</a></td><td>8.27 KB</td> </tr>
</tbody>
</table>
http://openmx.psyc.virginia.edu/thread/4168#commentsmetaSEMWed, 03 Aug 2016 15:04:21 +0000nastjuscha4168 at http://openmx.psyc.virginia.eduCI of indirect effects
http://openmx.psyc.virginia.edu/thread/4161
<p>Dear all, </p>
<p>Being very new to MetaSEM as well as R, please accept my apology for such simple question. </p>
<p>Mike kindly corrected my code to conduct a correlation based MetaSEM. </p>
<p>I just wanted to be sure that the coefficients are the standardised coefficients. Also, I was wondering whether it is possible to get the CI of the indirect effects, to know whether these effects are significant or not.<br />
I attached the code and the results. </p>
<p>Thank you in advance,<br />
Regards,<br />
Arin </p>
<table id="attachments" class="sticky-enabled">
<thead><tr><th>Attachment</th><th>Size</th> </tr></thead>
<tbody>
<tr class="odd"><td><a href="http://openmx.psyc.virginia.edu/sites/default/files/Arin4July2016_0.pdf">Arin4July2016.pdf</a></td><td>175.35 KB</td> </tr>
</tbody>
</table>
http://openmx.psyc.virginia.edu/thread/4161#commentsmetaSEMSun, 10 Jul 2016 14:32:08 +0000Arin A4161 at http://openmx.psyc.virginia.edux is not positive definite! Help
http://openmx.psyc.virginia.edu/thread/4156
<p>Dear Mike,</p>
<p>I'm a beginner in R and metaSEM. I'm performing a meta-analytic path model with 3 predictor variables, 1 mediator and 1 dependent variable (5 variables in total and 158 primary studies) as you can see in the figure.</p>
<p>I run the fix effects in the full dataset (158) but it is not possible because the missing values, but neither is possible to run the random effects (1stage).</p>
<p>I've notice the missing values problem due to the function is.pd, so I reduced the full dataset to 38 studies that cotains at least 1 value in the variable mediator, but it doesn't work. Curiously with a random set of 43 studies it was not a problem to run the random effects (1stage) (unfortunely I lost the dataset to probe it). </p>
<p>The problem is "x is not positive definite!". I don't understand why it happened in one case but in the other not. </p>
<p>1) Is there a way to solve this problem from the full dataset (158) without removing correlation matrices? </p>
<p>2) Taking in account the reduced dataset (38) is it possible to fix the "x is not positive definite!"?</p>
<p>3) By the way, following the rest of the code, is right specified? </p>
<p>I thank you in advance for your kind attention to this message and all the help you can provide me.</p>
<table id="attachments" class="sticky-enabled">
<thead><tr><th>Attachment</th><th>Size</th> </tr></thead>
<tbody>
<tr class="odd"><td><a href="http://openmx.psyc.virginia.edu/sites/default/files/model.png">model.png</a></td><td>29.76 KB</td> </tr>
<tr class="even"><td><a href="http://openmx.psyc.virginia.edu/sites/default/files/dataset158.txt">dataset158.txt</a></td><td>7.33 KB</td> </tr>
<tr class="odd"><td><a href="http://openmx.psyc.virginia.edu/sites/default/files/scriptdataset158.R">scriptdataset158.R</a></td><td>2.81 KB</td> </tr>
<tr class="even"><td><a href="http://openmx.psyc.virginia.edu/sites/default/files/dataset38.txt">dataset38.txt</a></td><td>1.63 KB</td> </tr>
<tr class="odd"><td><a href="http://openmx.psyc.virginia.edu/sites/default/files/scriptdataset38.R">scriptdataset38.R</a></td><td>2.21 KB</td> </tr>
</tbody>
</table>
http://openmx.psyc.virginia.edu/thread/4156#commentsmetaSEMThu, 07 Jul 2016 11:58:40 +0000Juanda4156 at http://openmx.psyc.virginia.eduUsing standard error in meta() and meta3()
http://openmx.psyc.virginia.edu/thread/4152
<p>Hi Mike and others,</p>
<p>I've been running some three-level meta-analyses using meta3() with standard error as input instead of variance. However, the description of meta() and meta3() only mention the use of variance as input. Although I am aware it is easy to convert standard error to variance I was wondering whether it is ok to use standard error with meta() and meta3().</p>
<p>Thanks for the help all.</p>
<p>Best,<br />
Jasper</p>
http://openmx.psyc.virginia.edu/thread/4152#commentsmetaSEMSun, 03 Jul 2016 23:23:14 +0000jasperd4152 at http://openmx.psyc.virginia.eduCombining three-level and TSSEM
http://openmx.psyc.virginia.edu/thread/4123
<p>Hi Mike, and others,</p>
<p>I have a question on three-level meta-analysis and TSSEM. I'm working on a meta-analysis, in which I want to test the effects of three predictors on three outcomes, preferably using TSSEM. However, several studies provide multiple effect sizes for a single relation. To overcome this dependency, I would like to use three-level meta-analysis. However, to me as a beginner it looks like the 2 methods are not compatible. Therefore, to me, I have two options if I want to use MASEM: 1) Conduct analyses on bivariate relationships by fitting 3-level RE models, and use the correlation matrix as input, or 2) Aggregate effect sizes and conduct TSSEM. (Total effect sizes before aggregation = 260, after aggregation = 160)</p>
<p>What would you recommend? </p>
<p>Thanks in advance!</p>
http://openmx.psyc.virginia.edu/thread/4123#commentsmetaSEMFri, 01 Apr 2016 12:08:32 +0000Rutger4123 at http://openmx.psyc.virginia.eduNested models
http://openmx.psyc.virginia.edu/thread/4121
<p>A fixed effect model is a more restricted version of a random effects model, when meta-analyzing correlations using tssem1(), correct? Is there a way to do a nested model test? I know the degrees of freedom for the difference test would be the number of random effects estimated. I'm tripped up because the fixed effect model gives many fit stats, whereas the random effects model only gives a -2 log likelihood. </p>
<p>Thank you!</p>
http://openmx.psyc.virginia.edu/thread/4121#commentsmetaSEMSat, 26 Mar 2016 00:28:33 +0000k.corker4121 at http://openmx.psyc.virginia.eduasycov missing values
http://openmx.psyc.virginia.edu/thread/4119
<p>Hi Mike and other users,</p>
<p>I have a question concerning the the function asycov in the preparatory process for a Meta-Analysis using SEM and the WLS-Function. I have a manually created pooled correlation matrix (6x6) as there are some correlations I didnâ€™t compute by myself. The problem is, that there are 4 missing values which resulted in the following error term when trying to calculate the covariance matrix with the asycov function:<br />
if (!is.pd(x.new)) stop("x is not positive definite!\n"). I tried another smaller correlation matrix with missing values and it worked. Do you know why it does not work in the other case and how I could solve this problem? </p>
<p>I really appreciate your help on that.<br />
Greta</p>
http://openmx.psyc.virginia.edu/thread/4119#commentsmetaSEMWed, 23 Mar 2016 08:57:25 +0000Forscher_CG4119 at http://openmx.psyc.virginia.eduDifferences between meta-analytic indirect effects estimated using meta() and tssem2()
http://openmx.psyc.virginia.edu/thread/4106
<p>Hi Mike and metaSEM users,</p>
<p>First of all, thank you, Mike, for maintaining such an active presence on this forum! It makes a HUGE difference as a user to be able to get questions answered from the package maintainer within a pretty reasonable timeframe. :)</p>
<p>I had another question about the meta-analytic indirect effects that are estimated by indirectEffect() (and then meta-analytically combined using meta() ).</p>
<p>As most people already know, in a single sample, a mediation model can be represented as a path model in the attached image. This path model can be represented using the following two regression models:</p>
<p>(1) M ~ aX + e_1<br />
(2) Y ~ bM + cX + e_2</p>
<p>The direct effect of X on Y, not through M, can be estimated using c and the indirect effect of X on Y through M can be estimated using the product ab.</p>
<p>For a meta-analysis I'm doing, I wanted to present coefficients a, b, and c so that readers could understand the relative contributions of a and b to the indirect effect. I had thought that a good way to do this would be to estimate the meta-analytic pooled correlation matrix using tssem1(), which I could use to find meta-analytic coefficient a, and fit the regression model (2) with this pooled correlation matrix using tssem2(), yielding coefficients b and c. You can find my code using 10 arbitrary correlation matrices attached.</p>
<p>My problem is that the product ab calculated using the method I described above departs quite substantially from the meta-analytic indirect effect estimated by first calculating the indirect effect within each correlation matrix, then pooling these indirect effects. The coefficient c obtained from tssem2() also departs from the direct effect obtained from indirectEffect() and meta(). I haven't done a formal simulation study, but I've played with different correlation matrices and found that the indirect effects estimated in these different ways depart from each other pretty consistently.</p>
<p>What's going on here? Is there a flaw in my thinking? Is there a better way of obtaining meta-analtyic estimates of the quantities a, b, and c?</p>
<table id="attachments" class="sticky-enabled">
<thead><tr><th>Attachment</th><th>Size</th> </tr></thead>
<tbody>
<tr class="odd"><td><a href="http://openmx.psyc.virginia.edu/sites/default/files/mediation.png">mediation.png</a></td><td>7.33 KB</td> </tr>
<tr class="even"><td><a href="http://openmx.psyc.virginia.edu/sites/default/files/metaSEM script.R">metaSEM script.R</a></td><td>1.97 KB</td> </tr>
</tbody>
</table>
http://openmx.psyc.virginia.edu/thread/4106#commentsmetaSEMTue, 23 Feb 2016 20:23:53 +0000forscher4106 at http://openmx.psyc.virginia.edu