whether the latent class model can identify individually which respondents in each class or not?

14 views
Skip to first unread message

Tony Zeng

unread,
Oct 21, 2021, 7:01:30 AMOct 21
to bio...@googlegroups.com, Tony Zeng
Dear Prof. Bierlaire and forum members,

I am a new user of Biogeme.

I would like to build a latent class model using Biogeme. I studied one of the good examples 07discreteMixture.py regarding estimating the latent class model.

I understand that this model can return all the estimated parameters. However, besides this,  I would like to know whether the latent class model can identify which respondent belongs to which class, such that I could do further descriptive analysis for each class.

Having this question, I came across the conversation related to this example in biogeme googlegroups. The link is here.


Someone asked the same question before "Could we identify individually which respondents in each class?"

The answer to this question is "No. Only the probability. This is why the class is latent."

Meantime,  I found a Youtuber talked about conducting latent class cluster analysis with free software Jamovi. The link is below:


In this video, he uses Jamovi to build a latent class model with 4 classes and shows how the model tells us which respondent belongs to which class.

Here is the screenshot from his video. Currently, Jamovi is only able to do latent analysis for nominal/categorical/qualitative variables rather than continuous variables.

image.png

image.png

Now, I got a bit of confusion.

My question is whether the latent class model is able to tell us which respondent belongs to which class? If yes, could you kindly please advise me on how to implement it in Biogeme since my data includes nominal variables as well as continuous variables.


Many thanks for your great help!


Best Regards,

Tony

Bierlaire Michel

unread,
Oct 21, 2021, 10:23:09 AMOct 21
to tony.z...@gmail.com, Bierlaire Michel, bio...@googlegroups.com

On 21 Oct 2021, at 13:27, Tony Zeng <tony.z...@gmail.com> wrote:

Dear Prof. Bierlaire and forum members,

I am a new user of Biogeme.

I would like to build a latent class model using Biogeme. I studied one of the good examples 07discreteMixture.py regarding estimating the latent class model.

I understand that this model can return all the estimated parameters. However, besides this,  I would like to know whether the latent class model can identify which respondent belongs to which class, such that I could do further descriptive analysis for each class.

Yes. You need to use Biogeme in "simulate” mode to generate the class membership probability for each class and each individual.


Having this question, I came across the conversation related to this example in biogeme googlegroups. The link is here.


Someone asked the same question before "Could we identify individually which respondents in each class?"

The answer to this question is "No. Only the probability. This is why the class is latent."

Exactly.


Meantime,  I found a Youtuber talked about conducting latent class cluster analysis with free software Jamovi. The link is below:


In this video, he uses Jamovi to build a latent class model with 4 classes and shows how the model tells us which respondent belongs to which class.

In the video, it is not said how this “membership” variable is calculated. My guess is that they select each individual with the class with the highest membership probability. If so, this is wrong. It biases the results a lot. Moreover, you lose information. It is more informative to know that an individual belongs to a class with some probability than to pretend that we are absolutely sure that s/he belongs to this class, when we know as a fact that the probability is not one. 




Here is the screenshot from his video. Currently, Jamovi is only able to do latent analysis for nominal/categorical/qualitative variables rather than continuous variables.

<image.png>

<image.png>

Now, I got a bit of confusion.

My question is whether the latent class model is able to tell us which respondent belongs to which class? If yes, could you kindly please advise me on how to implement it in Biogeme since my data includes nominal variables as well as continuous variables.


Many thanks for your great help!


Best Regards,

Tony


--
You received this message because you are subscribed to the Google Groups "Biogeme" group.
To unsubscribe from this group and stop receiving emails from it, send an email to biogeme+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/biogeme/CAOyGxAteJ%2BHZq0-f%2Bx0ppS%2BSzMZcnLknyG67Rj3uSKx8KiKvuw%40mail.gmail.com.

Reply all
Reply to author
Forward
0 new messages