Path analysis with continuous and categorical variables

[Joe Glass]

Dear list,

I am enjoying learning lavaan, and I have been trying to specify a model that contains a combination of continuous and categorical variables.  I seem to run into a problem when I have two dependent variables, one that is latent and the other is a continuous variable.  The latent has ordinal indicators.  I receive the error:

Error in muthen1984(Data = X[[g]], ov.names = ov.names[[g]], ov.types = ov.types,  : 

  unknown ov.types:integer

Below is my syntax:

structure <-   '  withdrawal =~ q21 + q22 + q23 + q24 + q26

                  pas =~ q50 + q51 + q52 + q53 + q55 + q56 + q58

                  continuousdv ~ withdrawal + pas

                  withdrawal ~ pas + audit10 + female + k6raw'

fit <- sem(structure, data=stig1, ordered=c(withdrawal.ordered, pas.ordered))

This error only occurs when I include the continuous dependent variable.  If I replace that variable with a latent, everything works fine.  Would this be a problem with my code?

Thanks,

Joe Glass

Joseph E. Glass, Ph.D.

Assistant Professor

School of Social Work

University of Wisconsin-Madison

1350 University Ave.

Madison, WI 53706

[yrosseel]

On 11/21/2012 08:20 PM, Joe Glass wrote:
> Dear list,
>
> I am enjoying learning lavaan, and I have been trying to specify a model
> that contains a combination of continuous and categorical variables. I
> seem to run into a problem when I have two dependent variables, one that
> is latent and the other is a continuous variable. The latent has
> ordinal indicators. I receive the error:
>
> Error in muthen1984(Data = X[[g]], ov.names = ov.names[[g]], ov.types =
> ov.types, :
> unknown ov.types:integer
>
> Below is my syntax:
> structure <- ' withdrawal =~ q21 + q22 + q23 + q24 + q26
> pas =~ q50 + q51 + q52 + q53 + q55 + q56 + q58
> continuousdv ~ withdrawal + pas
> withdrawal ~ pas + audit10 + female + k6raw'
> fit <- sem(structure, data=stig1, ordered=c(withdrawal.ordered,
> pas.ordered))

I'm not sure, but could you remove the 'ordered' argument in your sem()
function call? From you explanation, I get that only the q21/q22/...
indicators are ordered. But not 'withdrawal.ordered' and 'pas.ordered'
(which do not appear under these names in your model syntax)?

Would you be able to send me your full R script and a snippet of the
data (so I can reproduce the error)?

Yves.

[Aaron Franzen]

Hello All, 

I seem to get a similar error message and am trying to figure out how to fix it/figure out what it is telling me. I have specified my dichotomous variables in the data frame already. My syntax and subsequent error code is below (with dummy variables in bold... since my 'ordered' specification was not in the model):

bread.model3 <-'eg=~q20g+q20h+q20c+q20d+q21c+q21d

j1=~q20e+q20f+q21b

j2=~q21m+q21n+q21o

improve~eg+j1+j2+educ+south+married+white+Male+q68+q79+q4+q42+evang

j1~1*j2

eg~j1

eg~j2

'

bread.fit3<-sem(bread.model3, data=data2)

Error in muthen1984(Data = X[[g]], ov.names = ov.names[[g]], ov.types = ov.types,  : 

  unknown ov.types:integer

The same model did run before I specified which variables were dummy variables.  

R: 2.15

lavaan: 0.5-10

Aaron 

[yrosseel]

On 12/11/2012 02:57 AM, Aaron Franzen wrote:
> Hello All,
> I seem to get a similar error message and am trying to figure out how to
> fix it/figure out what it is telling me. I have specified my dichotomous
> variables in the data frame already. My syntax and subsequent error code
> is below (with dummy variables in bold... since my 'ordered'
> specification was not in the model):
>
> bread.model3 <-'eg=~q20g+q20h+q20c+q20d+q21c+q21d
> j1=~q20e+q20f+q21b
> j2=~q21m+q21n+q21o
>

> *improve*~eg+j1+j2+educ+*south*+*married*+*white*+*Male*+q68+q79+q4+q42+*evang*

>
> j1~1*j2
> eg~j1
> eg~j2
> '
> bread.fit3<-sem(bread.model3, data=data2)
>
> Error in muthen1984(Data = X[[g]], ov.names = ov.names[[g]], ov.types =
> ov.types, :
> unknown ov.types:integer

Could you send me a snippet of the data (say, the first 40 cases) so I
can replicate this?

In any case, it seems to me that only 'improve' should be declared as
ordered (only dependent variables!). Everything else should be numeric.

Yves.

[TW]

Dear all,

unfortunately, I receive a similar response when trying to predict a continuous variable by a model consisting of categorial variables (5-point Likert scale). I also used the 'ordered' argument, hoping that it's working now.

That's what I got:

Error in muthen1984(Data = X[[g]], ov.names = ov.names[[g]], ov.types = ov.types,  : 
  unknown ov.types:factor
lavaan WARNING: unordered factor(s) with more than 2 levels detected in data

How can I fix / manage this problem?

[Terrence Jorgensen]

Error in muthen1984(Data = X[[g]], ov.names = ov.names[[g]], ov.types = ov.types,  : 
  unknown ov.types:factor
lavaan WARNING: unordered factor(s) with more than 2 levels detected in data

You didn't post your R script, so I can't be sure, but it looks like your data contains variables of class "factor" instead of class c("ordered","factor").  That's something you can see how to change:  ?ordered

Also, you say that you are predicting a continuous outcome?  Does that mean that your categorical variables are exogenous predictors?  If so, you need to code them as dummy codes rather than using a factor variable as a predictor:

http://lavaan.ugent.be/tutorial/cat.html

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

[May Yang]

I have the same issue here.