Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
| From | Alan Acock <acock@me.com> |
| To | statalist@hsphsun2.harvard.edu |
| Subject | Re: st: SEM becomes unidentified when introducing single item control variables |
| Date | Tue, 15 Jan 2013 11:16:41 -0800 |
Johannes,
You could use
.sem (x1<- X1), reliability(x1 .8)
Then, you could try other estimates of reliability to do a sensitivity analysis. If you assume there is no measurement error, then you would simply use x1 as is and not use a latent variable for it.
Alan Acock
On Jan 15, 2013, at 10:45 AM, Johannes Kotte <johannes.kotte@st.ovgu.de> wrote:
> Hi Billy,
>
> makes complete sense what you say about the covariates - thanks for your help!
>
> What I meant by "I have already seen models with latent single-item variables" is that some authors use single-item latent variables isntead of the observed ones (like I tried to). What I don't understand is how this can work, considering my experience that latent single-item variables cannot be identified.
>
> Best
> Johannes
>
>
> Zitat von William Buchanan <william@williambuchanan.net>:
>
>> Hi Johannes,
>>
>> I'm not sure why you would use several latent variables for observed
>> covariates. If you wanted a measurement model for your covariates it would
>> be something more like:
>>
>> (x16 x17 x18 x19 <- Covariates)
>>
>> But given what you've mentioned about the variables, it doesn't seem like
>> this would be a good idea (e.g., suggesting that some unobservable variable
>> affects someone's gender, age, and what I presume would be other demographic
>> indicators). Why is it not acceptable to include your observed variables as
>> covariates? If you're going to mention how you've seen this done before in
>> other articles/papers it would also be a good idea to reference those papers
>> so others can approach helping you from the same frame of reference. And
>> you should include the output from your command(s) as well as the syntax
>> that you've used to produce them. Sometimes you may have just overlooked a
>> small, but important, piece of information that could explain a lot of the
>> problems you're running into.
>>
>> HTH,
>> Billy
>>
>>
>>
>> -----Original Message-----
>> From: owner-statalist@hsphsun2.harvard.edu
>> [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Johannes Kotte
>> Sent: Tuesday, January 15, 2013 8:50 AM
>> To: statalist@hsphsun2.harvard.edu; JVerkuilen (Gmail)
>> Subject: Re: st: SEM becomes unidentified when introducing single item
>> control variables
>>
>> Thanks for your reply!
>>
>> I looked at the model identification after letting sem iterate for a few
>> times. The df are above 60, so I always thought that identification is no
>> issue.
>>
>> Now this might sound stupid, but I always thought that "(x16 <- CV1) ...
>> (x19 <- CV4)" IS my measurement model for the control variables.
>> However, you are right that CV1-CV4 are unidentified if I run the
>> measurement models alone. As they are single-item variables like gender,
>> age, etc., I (obviously wrongly) presumed that they cannot be unidentified.
>>
>> Nevertheless, they don't have to be latent (I guess), even though I have
>> already seen models with latent single-item variables. So, if I altered
>> model 2 as follows (with x16 x17 x18 x19 being the controls), would that be
>> correct?
>>
>> sem (y1 y2 y3 y4 <- PRAXREL) ///
>> (x1 x2 x3 x4 x5 x6 x7 <- BKA) ///
>> (x8 x9 x10 x11 <- KVSENIOR) ///
>> (x12 x13 x14 x15 <- KVL) ///
>> (BKA PRAXREL <- KVSENIOR KVL x16 x17 x18 x19) ///
>> (PRAXREL <- BKA) ///
>> , standardized method(mlmv)
>>
>> I tried the above sem and it works. However, the estat mindices command
>> results in missing values only, even for the latent constructs
>>
>> Again, thanks a lot!
>> Johannes
>>
>> --------------------------------------- Original e-mail
>> ---------------------------------------
>>
>> Zitat von "JVerkuilen (Gmail)" <jvverkuilen@gmail.com>:
>>
>>> The standard errors being crazy is a sign that the model is not
>>> identified. I'd suspect it's because the latent variables for these
>>> controls aren't identified, and given that it doesn't sound like you
>>> have a measurement model for them I'm not sure how they could be. Why
>>> are they latent anyway?
>>> *
>>> * For searches and help try:
>>> * http://www.stata.com/help.cgi?search
>>> * http://www.stata.com/support/faqs/resources/statalist-faq/
>>> * http://www.ats.ucla.edu/stat/stata/
>>
>> ----------------------------------------------------------------------
>> Datum: Tue, 15 Jan 2013 15:21:36 +0100
>> Von: Johannes Kotte <johannes.kotte@st.ovgu.de>
>> Betreff: SEM becomes unidentified when introducing single item control
>> variables
>> An: statalist@hsphsun2.harvard.edu
>>
>> Dear fellow researchers,
>>
>> I would be grateful for advice with the following problem: I have created a
>> very simple SEM (let's call it 'model 1') that works fine (see below for
>> code). It contains a latent dependent variable called PRAXREL and a latent
>> independent variable called BKA. Moreover, it contains latent control
>> variables called KVSENIOR and KVL. As I said, model 1 works fine
>> (identified, good fit).
>>
>> However, the model becomes problematic when I introduce single-item latent
>> variables (CV1, CV2, CV3, CV4) as control variables ('model2').
>> In this case Stata iterates forever saying “not concave”.
>>
>> WHAT COULD BE THE REASON? I tried many different setups of the model (incl.
>> constraining the path coefficients of the CV to 1 or setting the reliability
>> of the CV to 0.9 or 0.5) but none of them really worked unless I delete at
>> least some of the CVs.
>>
>> The following might be interesting: (i) If I let Stata iterate 15 times and
>> take a look at the output, I find that sometimes the standard errors of CV1,
>> CV2, CV3 and CV4 are extremely high. (ii) Moreover, I found that pairwise
>> correlation of the variables shows that they are mostly correlated - at
>> least at the 10% level, sometimes even 1%. Might there be a collinearity
>> problem?
>>
>> Can anybody give me advice? I would greatly appreciate that!
>>
>> Thanks in advance!
>> Johannes
>>
>> CODE FOR BOTH MODELS:
>>
>> /***MODEL 1***/
>>
>> sem (y1 y2 y3 y4 <- PRAXREL) ///
>> (x1 x2 x3 x4 x5 x6 x7 <- BKA) ///
>> (x8 x9 x10 x11 <- KVSENIOR) ///
>> (x12 x13 x14 x15 <- KVL) ///
>> (BKA PRAXREL <- KVSENIOR KVL) ///
>> (PRAXREL <- BKA) ///
>> , standardized method(mlmv)
>>
>>
>> /***MODEL 2***/
>>
>> sem (y1 y2 y3 y4 <- PRAXREL) ///
>> (x1 x2 x3 x4 x5 x6 x7 <- BKA) ///
>> (x8 x9 x10 x11 <- KVSENIOR) ///
>> (x12 x13 x14 x15 <- KVL) ///
>> (x16 <- CV1) ///
>> (x17 <- CV2) ///
>> (x18 <- CV3) ///
>> (x19 <- CV4) ///
>> (BKA PRAXREL <- KVSENIOR KVL CV1 CV2 CV3 CV4) ///
>> (PRAXREL <- BKA) ///
>> , standardized method(mlmv)
>>
>> --
>> Johannes Kotte
>> Otto-von-Guericke-Universität | Faculty of Business and Economics| Chair of
>> Management and Organization (Prof. Thomas Spengler) | Postfach 4120, 39016
>> Magdeburg | www.ufo.ovgu.de
>>
>> Telefon: +49-173-6371955 | E-Mail: johannes.kotte@st.ovgu.de
>>
>>
>> *
>> * For searches and help try:
>> * http://www.stata.com/help.cgi?search
>> * http://www.stata.com/support/faqs/resources/statalist-faq/
>> * http://www.ats.ucla.edu/stat/stata/
>>
>>
>> *
>> * For searches and help try:
>> * http://www.stata.com/help.cgi?search
>> * http://www.stata.com/support/faqs/resources/statalist-faq/
>> * http://www.ats.ucla.edu/stat/stata/
>>
>
>
>
> --
> Johannes Kotte
> Otto-von-Guericke-Universität | Fakultät Wirtschaftswissenschaften | Lehrstuhl für Unternehmensführung und Organisation (Prof. Dr. Thomas Spengler) | Postfach 4120, 39016 Magdeburg | www.ufo.ovgu.de
>
> Telefon: +49-173-6371955 | E-Mail: johannes.kotte@st.ovgu.de
>
>
> *
> * For searches and help try:
> * http://www.stata.com/help.cgi?search
> * http://www.stata.com/support/faqs/resources/statalist-faq/
> * http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/faqs/resources/statalist-faq/
* http://www.ats.ucla.edu/stat/stata/