Simulate Dichotomous Data

[Monica Casella]

Hello,

I post the following question also in this group:

I’m working on my degree thesis and I’m using R package “Lavaan” and the SimulateData () function to simulate data from this model: 

Model <- '

f1 =~ V1 + V2 + V3 + V4 + V5 + V6 + V7 + V8 + V9 + 0.2*V10 + 0.2*V11 + 0.2*V12 + 0.2*V13 + 0.2*V14 + 0.2*V15

f2 =~ V16 + V17 + V18 + V19 + V20 + V21 + V22 +V23 + V24 + 0.2*V25 + 0.2*V26 + 0.2*V27 + 0.2*V28 + 0.2*V29 + 0.2*V30

f3 =~ V31 + V32 + V33 + V34 + V35 + V36 + V37 + V38 + V39 + 0.2*V40 + 0.2*V41 + 0.2*V42 + 0.2*V43 + 0.2*V44 + 0.2*V45

 

O1 ~ 1*f1 + 0.2*f2 + 0.2*f3 '

 

The observed variables and the criterion variable O1 should be dichotomous.

I know that the "|" operator can be used to define the thresholds of categorical endogenous variables, but how should I choose the threshold values? Does this way work for the criterion variable? Is it possible to simulate dichotomous data in Lavaan?

Thanks for any help,

Monica

[Nickname]

Monica,
Taking your questions in reverse order:

Yes, simulateData() will simulate dichotomous variables.

Yes, this will work for both exogenous and endogenous variables.  In some sense, all dichotomous variables are endogenous because they are modeled through a latent continuous variable.

Choose the thresholds based on the desired frequencies.  The latent continuous variable is (normally) parameterized as a standard normal variable.  So, you can use the standard normal distribution to calculate threshold values corresponding to a desired proportion of cases.  For example, if you wanted 1/3 of the cases to have the lower value, you could use qnorm(1/3) to obtain the value of -0.4307273.  For an even split, use zero.

Keith
------------------------
Keith A. Markus
John Jay College of Criminal Justice, CUNY
http://jjcweb.jjay.cuny.edu/kmarkus
Frontiers of Test Validity Theory: Measurement, Causation and Meaning.
http://www.routledge.com/books/details/9781841692203/

[Terrence Jorgensen]

 to simulate data from this model: 

Model <- '

f1 =~ V1 + V2 + V3 + V4 + V5 + V6 + V7 + V8 + V9 + 0.2*V10 + 0.2*V11 + 0.2*V12 + 0.2*V13 + 0.2*V14 + 0.2*V15

f2 =~ V16 + V17 + V18 + V19 + V20 + V21 + V22 +V23 + V24 + 0.2*V25 + 0.2*V26 + 0.2*V27 + 0.2*V28 + 0.2*V29 + 0.2*V30

f3 =~ V31 + V32 + V33 + V34 + V35 + V36 + V37 + V38 + V39 + 0.2*V40 + 0.2*V41 + 0.2*V42 + 0.2*V43 + 0.2*V44 + 0.2*V45

 

O1 ~ 1*f1 + 0.2*f2 + 0.2*f3 '

Half of your specified loadings have no population values.

The observed variables and the criterion variable O1 should be dichotomous.

I know that the "|" operator can be used to define the thresholds of categorical endogenous variables, but how should I choose the threshold values?

That's up to you.  If you want the categories to have roughly equal counts, you can set thresholds to zero (V1 | 0*t1) 

Terrence D. Jorgensen

Assistant Professor, Methods and Statistics

Research Institute for Child Development and Education, the University of Amsterdam

http://www.uva.nl/profile/t.d.jorgensen

[James Grace]

I tried to translate advice in this thread to a similar, but simpler problem. Observed variables only, dichotomous Treatment and two continuous responses. Admit that I really don't know what I am doing with the syntax.

Any advice for how to get dichotomous values for treatment (50:50 split) would be greatly appreciated.

## Sim Model

sim.two <- '

Grazers1 ~ -1.5*(Trt | 0*t1) 

Grazers2 ~ -0.8*(Trt | 0*t1)'

## Generate data

set.seed(2)

simdat.two <- simulateData(sim.two, sample.nobs=10000L)

[James Grace]

I found the answer to my own question about syntax.

Using

## Sim Model

sim.two <- '

Grazers1 ~ -1.5*Trt 

Grazers2 ~ -0.8*Trt

Trt | 0*t1'

does the trick!

On Tuesday, March 31, 2020 at 10:05:35 AM UTC-5, Monica Casella wrote: