Is a ordered logit model possible?

[Jang JinHee]

First of all, thank you for reading.

I have a question.
As a dependent variable, we use the average of 8 questions measured on a 5-point scale, and used a ordered logit.
However, using the average, we found that it did not have values of 1,2,3,4,5...it had values of 1, 1.2, 1.5, 2, 2.2, 2.5....
is a ordered logit possible in this case? the marginal effect has too many value.

If the dependent variable has these values, how do I analyze this? OLS(GLS)?
I'd appreciate it if someone could tell me..
thanks

[Maarten Buis]

Linear regression (linear regression is the model, OLS is the algorithm commonly used to estimate it's parameters) is commonly used in this kind of situation. After all, by computing the mean you already assumed interval scale, so the model does not add assumptions...

[Jang JinHee]

even if the dependent variable is not a negative to positive infinity, but has a limited distribution between 1 and 5, is it possible to estimate ols?
Then, estimate the instrumental variable considering the endogenous problems?
I'm confusing. I don't know what to do...

[Maarten Buis]

even if the dependent variable is not a negative to positive infinity, but has a limited distribution between 1 and 5, is it possible to estimate ols?

Then, estimate the instrumental variable considering the endogenous problems?
I'm confusing. I don't know what to do...

I suspect that the source of your confusion is that you were taught very strict rules on "this assumption must be met and that assumption must be met, etc." The thing about analysing real data is that the assumptions are never met. The world itself is messy. On top of that, the process of measuring stuff in that messy world is messy. So real data is a really big mess. It does not fit a neat little model. In a sense that mismatch between the messy data and the neat model is intentional: The purpose of a model is to simplify reality. We humans don't have the capacity to understand the world in its entirety. We must simplify it before we can start to understand parts of it. But simplify is just another word for being wrong in some useful way. So a model does not have to be right, it by definition cannot be right, but it has to be right enough to be useful. So it is not enough to learn a set of assumptions by heart, and than use them as a checklist. Instead you need to understand why those assumptions are there, so you can make an informed decision about whether a deviation from that assumption in your data will hurt your analysis. In your case, a bounded variable will lead to non-linearity. However, it is possible that, within the range of your data, a linear relationship is still an adequate simplification. So that is what I would pay close attention to.