[R] Problems bootstrapping multigroup SEM
John Fox
jfox at mcmaster.ca
Tue Sep 2 22:03:59 CEST 2014
Dear Chad,
> -----Original Message-----
> From: Chad Danyluck [mailto:c.danyluck at gmail.com]
> Sent: Tuesday, September 02, 2014 3:29 PM
> To: John Fox
> Cc: r-help at r-project.org
> Subject: Re: [R] Problems bootstrapping multigroup SEM
>
> Dear John,
>
> Thank you for your insights. I think you do understand what I've been
> trying to do. Because I am doing a multigroup comparison –
> specifically, examining the moderating role of test type on outcome – I
> needed two covariance matrices to pass through the model. I wasn't sure
> how to do this other than by listing these covariance matrices
> together. This list was called to the SEM and worked when the summary
> stats were called. As you point out, however, the covariance matrix did
> not get passed into bootSem() because, rather than use the cov
> function, I simply added the list. On further reflection, the crux of
> my problem, or confusion, seems to stem from not understanding how to
> deal with the need to pass two covariance matrices into bootSem(). I
> could pass "cov(na.omit(stereotype.MAP.data[,-1],
> na.omit(evaluative.MAP.data[,-1])))", but I thought that the two
> matrices needed to be kept separate because I am comparing the models
> produced by each matrix to one another.
>
> Another problem comes to light as I think through the help
> documentation:
>
> "In the case of an msem (i.e., multi-group) model, a list of data sets
> (again in the appropriate form), one for each group; in this case,
> bootstrapping is done within each group, treating the groups as strata.
> Note that the original observations are required, not just the
> covariance matrix of the observed variables in the model. The default
> is the data set stored in the sem object, which will be present only if
> the model was fit to a data set rather than to a covariance or moment
> matrix, and may not be in a form suitable for Cov."
>
> In my case, the data passed through the sem object was
> "c(nrow(stereotype.MAP.data), nrow(evaluative.MAP.data))". Perhaps this
> is not suitable for Cov?
I'm sorry but I don't follow this: c(nrow(stereotype.MAP.data), nrow(evaluative.MAP.data)) are simply the numbers of observations in the data sets, not the data sets themselves. From what you've said, the data sets are stereotype.MAP.data and evaluative.MAP.data. You can use the data and formula arguments to sem(). You need the explicit formula argument because you're apparently omitting one of the variables in each data set; alternatively, you can directly remove the unused variable from each data set, as you've done above. And you need not filter out the missing data (though it doesn't hurt to do so), since the default na.action is na.omit (as is shown in ?sem).
>From ?sem:
"data: As a generally preferable alternative to specifying S and N, the user may supply a data frame containing the data to which the model is to be fit. In a multigroup model, the data argument may be a list of data frames or a single data frame; in the later event, the factor given as the group argument is used to split the data into groups."
and
"formula: a one-sided formula, to be applied to data to generate the variables for which covariances or raw moments are computed. The default formula is ~., i.e., all of the variables in the data, including an implied intercept; if a covariance matrix is to be computed, the constant is suppressed. In a multigroup model, alternatively a list one one-sided formulas as be given, to be applied individually to the groups."
The multigroup example given in ?sem for the HS.data uses a single data frame, which is then classified by the factor Gender. In your case, you'd specify a list of two data frames; if the same variables are to be used in each, then you need give only one formula rather than a list of two, though the latter would also work.
Best,
John
>
> At this point I am spinning my wheels. Any further suggestions would be
> appreciated.
>
> Kind regards,
>
>
> Chad
>
>
>
>
> On Wed, Aug 27, 2014 at 6:59 PM, John Fox <jfox at mcmaster.ca> wrote:
>
>
> Dear Chad,
>
> It's possible that I don't understand properly what you've done,
> but it appears as if you're passing to bootSem() the covariance
> matrices for the observed data rather than the case-by-variable data
> sets themselves. That's also what you say you're doing, and it's what
> the error message says.
>
> Moreover, if you look at the documentation in ?bootSem, you'll is
> that the Cov argument isn't a covariance matrix, but "a function to
> compute the input covariance or moment matrix; the default is cov. Use
> cor if the model is fit to the correlation matrix. The function hetcor
> in the polycor package will compute product-moment, polychoric, and
> polyserial correlations among mixed continuous and ordinal variables
> (see the first example below for an illustration)."
>
> So what is there to bootstrap if bootSem() doesn't have access to
> the original data sets? I suppose that one could do a parametric
> bootstrap of some sort, but that's not what bootSem() does -- in
> implements a nonoparametric bootstrap, which requires the original
> data.
>
> I hope this helps,
> John
>
> -----------------------------------------------
> John Fox, Professor
> McMaster University
> Hamilton, Ontario, Canada
> http://socserv.socsci.mcmaster.ca/jfox/
>
>
>
>
> > -----Original Message-----
> > From: r-help-bounces at r-project.org [mailto:r-help-bounces at r-
> > project.org] On Behalf Of Chad Danyluck
> > Sent: Wednesday, August 27, 2014 12:22 PM
> > To: r-help at r-project.org
> > Subject: [R] Problems bootstrapping multigroup SEM
> >
> > Hello,
> >
> > I am having difficulty resolving an error I receive trying to
> bootstrap
> > a
> > multigroup SEM. The error (below) indicates that the model
> called to
> > bootSem doesn't contain matrices. This is true, sort of, because
> I
> > created
> > a list of two covariance matrices for the model to call. All of
> this
> > syntax
> > works fine (a summary of "MAP.mg.sem" will produce parameter
> estimates,
> > goodness of fit indices, etc.), however, the bootSem function
> does not
> > run.
> > Any ideas on a workaround?
> >
> > MLM.MAP.Data$IAT.factor <- as.factor(IAT)
> > IAT.factor <- MLM.MAP.Data$IAT.factor
> > evaluative.MAP.data <- subset(data.frame(IAT.factor, exp.race,
> > meditation.experience, years.meditate, repeated.iat,
> repeated.ERN, age,
> > acceptance, awareness, FCz.GNG.150.incor, FCz.GNG.150.cor,
> > FCz.stereo.150.incor, FCz.stereo.150.cor, FCz.eval.150.incor,
> > FCz.eval.150.cor), IAT==2)
> > stereotype.MAP.data <- subset(data.frame(IAT.factor, exp.race,
> > meditation.experience, years.meditate, repeated.iat,
> repeated.ERN, age,
> > acceptance, awareness, FCz.GNG.150.incor, FCz.GNG.150.cor,
> > FCz.stereo.150.incor, FCz.stereo.150.cor, FCz.eval.150.incor,
> > FCz.eval.150.cor), IAT==1)
> >
> > MAP.stereotype.cov <- cov(na.omit(stereotype.MAP.data[,-1]))
> > MAP.evaluative.cov <- cov(na.omit(evaluative.MAP.data[,-1]))
> > MAP.cov.list <- list(stereotype=MAP.stereotype.cov,
> > evaluative=MAP.evaluative.cov)
> >
> > #### Specify your MSEM path model: Years Meditating, ERN,
> IAT####
> > MAP.msem.model <- specifyModel()
> > years.meditate -> repeated.ERN, path1
> > years.meditate -> repeated.iat, path2
> > repeated.ERN -> repeated.iat, path3
> > age -> repeated.iat, path4
> > years.meditate <-> years.meditate, var1
> > repeated.ERN <-> repeated.ERN, var2
> > age <-> age, var3
> > age <-> years.meditate, cov1
> > repeated.iat <-> repeated.iat, d1
> >
> > MAP.mg.mod <- multigroupModel(MAP.msem.model,
> groups=c("stereotype",
> > "evaluative"))
> >
> > MAP.mg.sem <- sem(MAP.mg.mod, MAP.cov.list,
> > c(nrow(stereotype.MAP.data),
> > nrow(evaluative.MAP.data)), group="IAT.factor")
> >
> > system.time(bootSem(MAP.mg.sem, R=100, MAP.cov.list))
> >
> > Error in bootSem.msem(MAP.mg.sem, MAP.cov.list, R = 100) :
> > the model object doesn't contain data matrices
> >
> > --
> > Chad M. Danyluck
> > PhD Candidate, Psychology
> > University of Toronto
> > Lab: http://embodiedsocialcognition.com
> >
> >
> > “There is nothing either good or bad but thinking makes it so.”
> -
> > William
> > Shakespeare
> >
>
> > [[alternative HTML version deleted]]
> >
> > ______________________________________________
> > R-help at r-project.org mailing list
> > https://stat.ethz.ch/mailman/listinfo/r-help
> > PLEASE do read the posting guide http://www.R-
> project.org/posting-
> > guide.html
> > and provide commented, minimal, self-contained, reproducible
> code.
>
>
>
>
>
>
> --
>
> Chad M. Danyluck
> PhD Candidate, Psychology
> University of Toronto
> Lab: http://embodiedsocialcognition.com
> <http://embodiedsocialcognition.com/>
>
>
>
>
> “There is nothing either good or bad but thinking makes it so.” -
> William Shakespeare
More information about the R-help
mailing list