When I run quantile regression instead of r2 Stata gives pseudo r2. Is pseudo r2 something formal(not ad hoc) and can I report it in my PhD thesis?
Yes, psuedo r-squared is a legitimate thing that can be reported. However, you want to find out from the software documentation which pseudo r-square it's reporting. (Some common ones are listed here: https://stats.idre.ucla.edu/other/mult-pkg/faq/general/faq-what-are-pseudo-r-squareds/ ).
Yes, psuedo r-squared is a legitimate thing that can be reported. However, you want to find out from the software documentation which pseudo r-square it's reporting. (Some common ones are listed here: https://stats.idre.ucla.edu/other/mult-pkg/faq/general/faq-what-are-pseudo-r-squareds/ ).
You have to be careful of when you report R2 – pseudo or actual.
The idea that some model could "explain" 100% of variance is flawed. The existence of measurement error means that the maximum that a perfect model can "explain" is not 100% (and, indeed, you can factor in prior knowledge about the error structure of your variables to your model, though I've never seen it done in real life).
Variables that are measured with little error (age, for example) are therefore at an advantage compared with variables that are measured with a lot of error (depression, say). People deride the low R2 values for variables like attitudes, forgetting that the R2 isn't for attitude, but for the measure of attitude that we used.
So what's a big R2? Or what's a big improvement in R2?
These questions are so fraught with imponderables that I, for one, simply don't report R2 at all on the grounds that it's a figure I cannot interpret if asked (though, strangely, no-one ever asks…).
I agree with Prof. Conroy especially on his explanation of the Coefficient of Determination R^2.
I agree with the above colleagues. In addition, there are arguments that, for any models, pseudo R2 has no real meaning; see the following publications:
1- https://www.stata.com/statalist/archive/2006-08/msg00792.html
2 - https://www.stata.com/support/faqs/statistics/pseudo-r2/
3-http://eprints.staffs.ac.uk/1962/1/ABDIXHIKU%20Lumir%20%2C%20Theses%20Final%20.pdf
The third item related to the tobit model.
@LinekerPassos, honestly, I don't see anything in those links that justify saying that pseudo r-square has no meaning, other than someone claimed it in a question on the internet.
In the thesis document, the Wooldridge quote that is used for justification doesn't support the idea that pseudo r-square is meaningless, only that it isn't the criterion used to determine estimates in tobit models. I don't see that as much of an argument.
I don't have any experience using pseudo r-square with tobit models, so maybe there is something specific to these models that justifies not using pseudo r-square.
Salvatore S. Mangiafico, I did not mean that Pseudo R2 has no meaning, so much so that I used the expression "there are arguments" (not necessarily in papers); just wanted to add elements to the discussion. Several sources explain the usefulness of Pseudo R2 in the models, including Wooldridge. However, they assert that in some models (probit, logit, tobit etc.) the quality of model adjustment (measured by Pseudo R2) is less important than obtaining coefficient estimates. I also point out that I do not have any Pseudo R2 experience with quantile regression.
Pseudo R2 is a measure of how well variables of the model explain some phenomenon. If we catch with our variables more than 0,5 we can form our expectation for the model, but there are other unexplained issues and then try to find other factors that can explain and test our thesis.
- Badges
- Science topic
- Similar topics
- Biological Science
- Kinesiology
- Running