1
$\begingroup$

I am new to the concept of SEM. Yet, after reading about SEM it seems quite promising to give me a better understanding about how influential/important a set of measured variables (in respect to each other) is, regarding a abstract concept, i.e. latent variable. As I am still not 100% sure if I could use SEM for my purpose, let me give a more precise example. I would like to use a set of variables on different scales to describe a habitat. Let´s say I would use tree cover in %, sunlight input in hours, vegetation hight, ... Could I use SEM to tell me how each variable is important for the abstract concept/latent variable habitat quality? If yes, I guess the way to go would be a confirmatory factor analysis?

In a next step I would like to use the information to weight different spatial data to create an output for the abstract concept habitat quality, which I would then like to visualize and quantify.

Looking forward to answers

$\endgroup$
0
$\begingroup$

A latent variable (as usually conceptualized in SEM) means that there is a latent variable, and this latent variable is the cause of the measured variables.

The original latent variable was intelligence, and we theorize that there is a value of intelligence (which is unknown, and cannot be directly measured). The higher an individual's intelligence, the higher the probability that they will answer any question (that measures intelligence) correctly. Or that there is a construct of depression, and the higher an individual's level of depression, the more likely they are to endorse symptoms that represent depression (e.g. if you are more depressed, you are more likely to agree with statements like "I feel sad" or "I don't enjoy things as much as I used to."

You have a latent variable "habitat quality". Is this a latent variable, that causes tree cover %, sunlight in hours, veg height? I suspect not - hours of sunlight cannot be caused by something else. This is a causal indicator model - and you can't (or shouldn't) use a regular CFA model here. (It's also been argued that depression items are not effect indicators, it's not that depression makes you sad, not enjoy life, etc. It's more that being sad makes you depressed, not enjoying life makes you depressed). The classic paper on this is: https://psycnet.apa.org/record/1992-03966-001

You can fit models with causal indicators (or with a mixture, this is called a MIMIC [Multiple Indicators Multiple Cause] model), but it's trickier to identify. There are constraints you need to add to ensure that the latent variable is identified.

I'm not exactly sure what you mean by the next step about weighting.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.