@Jae Eun To answer your query, you can refer to below threads.

https://www.researchgate.net/post/How-to-control-variables-in-PLS-SEM-model

https://forum.smartpls.com/viewtopic.php?t=3745

I assume your age is an ordinal type variable (1 = '5-12'; 2 = '13-18', 3 = '18-35', etc.); in that case, you can treat it as a numeric variable as long as your coding scheme reflects the intrinsic ordering to the categories.

If you have a nominal variable with no intrinsic ordering to the categories, then you need to dummy code each level of the values listed.

Regardless, translating any observed variable to 5+ dummy codes makes no sense. If you have a situation like this, I encourage you to create a separate latent variable summarizing those variables, and use that latent variable into your SEM.

There are many mixture modeling you can consider to summarize categorical variables (e.g. latent class analysis). Good luck!

Hello Jae Eun,

The answer depends on how you wish to treat the categorical variable within your analysis. That is, what proposed model you are trying to evaluate.

First, a dichotomous categorical variable (e.g., sex at birth) can be treated just like an interval strength variable, with values such as 0/1, or 1/2. No additional modification is required. Obviously, the scores don't represent a continuum, however.

Second, if you have a categorical variable with more than two levels, then ordinarily one would recast that variable into c - 1 dummy variates (each most likely coded 0/1), where c is the number of levels of the categorical variable. You'd have to subsequently include all of these dummy variates to capture the variance accounted for by the categorical variable.

Here's an example: Marital status: (a) Single; (b) Married; (c) Divorced; (d) Widowed; (e) Living with a partner, but not married.

Four dummy variates would be required to capture all the levels (e.g., d1, d2, d3, d4).

For a single person, code d1 as 1, d2 as 0, d3 as 0, and d4 as 0

For a married person: 0, 1, 0, 0

For a divorced person: 0, 0, 1, 0

For a widowed person: 0, 0, 0, 1

For the last category: 0, 0, 0, 0

Thus, each category has a unique pattern across the dummy variates.

In the future, it makes more sense to avoid taking naturally numeric variables (such as age) and collapse them into categories...you not only lose information, you have to recode the values (which can make subsequent interpretation of results a bit more challenging).

Good luck with your work.

To study in the demographics in SmartPLS you have to first clarify whether you afe testing moderating effect or multigroup analysis. In simple are you trying to find the impact of demo variables on the path of IV-DV or you just want to identify the the statistical differece using MGA.

For Moderating there is an option to simply create a moderating variable in the Model itself. But for MGA you can only identify differences of 2 or 3 not more than that. For groups such as Income and Age you have to Sample your variable into two categories such as Income high and Low, age as Young and Adult and so on.

The new edition of the Primer on PLS-SEM covers this topic: Book A Primer on Partial Least Squares Structural Equation Modeli...