Generate dataset with selected path coefficients

[gaia...@gmail.com]

Dear users

I want to generate a dataset of 10.000 rows (inclding continuous and binary variables) whose path coefficients take the values we decide. is it possible to do it with lavaan? If the answer is yes, where can I consult a tutorial?

One the dataset is generated, I want to run the model with all dataset (expecting to have the path coefficients I defined) and with subsamples and to obtain the values of path coeffcients that I set in advance. 

Thanks in advance

[Rönkkö, Mikko]

Hi,

It is possible with the simulateData function, but why would you want to do so? Would it not be more straightforward and transparent to generate data using R’s built in random number functions? 

For example, you can generate a common factor model with binary indicators using probit link like this:

N <- 10000

eta <- normal(N)

y1 <- (eta + normal(N))>0

y2 <- (eta + normal(N))>0

y2 <- (eta + normal(N))>0

If you want to generate correlated exogenous variables, you can do that with the mvrnorm function from the MASS package.

Mikko

--
You received this message because you are subscribed to the Google Groups "lavaan" group.
To unsubscribe from this group and stop receiving emails from it, send an email to lavaan+un...@googlegroups.com.
To post to this group, send email to lav...@googlegroups.com.
Visit this group at https://groups.google.com/group/lavaan.
For more options, visit https://groups.google.com/d/optout.

[Mario Garrido]

Dear Dr. Rönkkö,

thanks so much for your fast reply.

I want to 'force' the path coefficients connecting variables and generate a dataset following those conditions. Then, I want to assess whether I can obtain those set values (or how close from those values) using different estimation methods under different conditions. However, I am not sure if is possible to do it and whether it has logic to do it. 

I have found a simple example (http://lavaan.ugent.be/tutorial/mediation.html). I think this is a good starting point but realize it is using the most simple mediation model and only with normal variables. I want something a little more complicated. something like the graph

--

Mario Garrido Escudero, PhD

Dr. Hadas Hawlena Lab

Mitrani Department of Desert Ecology 

Jacob Blaustein Institutes for Desert Research 

Ben-Gurion University of the Negev 

Midreshet Ben-Gurion 84990 ISRAEL 

gaia...@gmail.com; gaa...@post.bgu.ac.il

phone: (+972) 08-659-6854

[Rönkkö, Mikko]

Hi,

In the example model you have just one exogenous variable. Why not just do

N <- 10000
x1 <- rnorm(N)

m1 <- 0.8*x1 + rnorm(N)

m2 <- 0.3*m1 + rnorm(N)

m3 <- 0.1*x1 + 0.1*m1 + 0.0375*m2 + rnorm(N)

y1 <-  0.2*m2 + 0.09*m3 + rnorm(N)

data <- data.frame(x1, m1, m3, m3, y1)

If you want non-linearities and non-normal distributions, you need to specify what distributions and non-linear functions to use.

Mikko

[Mario Garrido]

Thanks. Im going to try

[Mario Garrido]

Dear Dr. Rönkkö,

thanks. It seems it works like this!.

I create the data according to the following code and then run a model and obtained the decided values.

However, how can I define as binary some variables? what I have to put here? GAp <- 0.8* sand   + rnorm(what to put here?)

In any case, Im afraid that the concordancy between the values set in data generation and obtained later in the analyses only works for normal distributed variables due to the computation of path coefficients when variables are not normal, is that right?

The data I generated

N <- 10000

sand <- rnorm(N)

GAp <- 0.8* sand   + rnorm(N)

rodens <- 0.3* GAp   + rnorm(N)

fbrdn <- 0.1* sand   + 0.1* GAp   + 0.0375* rodens   + rnorm(N)

Myc <-  0.2* rodens   + 0.09* fbrdn   + rnorm(N)

data <- data.frame( sand  ,  GAp  ,  rodens  ,  fbrdn  , Myc  )

write.xlsx(data, "C:/Users/gaiarrido/Desktop/datageneratedlavaan.xlsx") # write sheet

The model I ran

 DesiredBestModel<- '  Myc  ~f* rodens  +g*fbrdn

fbrdn  ~b* sand  +d* GAp  +e*rodens

rodens  ~c*GAp

GAp  ~a*sand

'

fit <- sem(DesiredBestModel, data = data)

summary(fit)

[Mario Garrido]

sorry, my question is what to put here for a binomial what to put here?(N)

[Rönkkö, Mikko]

Hi,

It depends how exactly you want to transform the normally distributed variable to a binary variable.

To do a probit model, you could do

GAp <-( 0.8* sand   + rnorm(N))>0

To do a logit model, you could do 

GAp <- inv.logit(0.8* sand) > runif(N)

(You can find the inv.logit function in e.g. the boot package.)

There are also other alternatives. 

Mikko

[Mario Garrido]

thanks, I m going to try right now

[Mario Garrido]

Hi again, 

what if I want to simulate a logit variable with two predictors? 

it would be like that

y2dummy <- inv.logit(0.24*x+0.47*y1)> runif(N)

or like this

#y2dummy <- inv.logit(0.24*x) + inv.logit(0.47*y1)> runif(N) 

Thanks

[Rönkkö, Mikko]

The first option.

Mikko

Sent from my iPhone

Mikko

<image.png>