QREGPD goodness of fit

[Lorina Sertio]

Hi!

I am having issues finding a measure of the goodness of fit of my results.
I used the qregpd command and would like to estimate either the pseudo R2, the F statistic, Akaike Inforation Criterion, Bayesian Information Criterion, or any other indicator.
Is there a command I can run?

Any help would be greatly appreciated!

Thank you in advance.
Lorina

[Joao Santos Silva]

Dear Lorina Sertio,

Do you really need to measure goodness-of-fit? The statistics you mention often have little meaning and they have even less meaning in this context.

Best wishes,

Joao

[Lorina Sertio]

Dear Joao,

I am working on a project that requires I present some form of goodness of fit measure, even it doesn't necessarily say much.
Is there something I can calculate? I will appreciate any help.

If it is possible to calculate some measure, why wouldn’t it say much?
I believe if that is the case, I should provide some explanation as to why the measure provides little meaning in this context.

Thank you in advance.

Best Regards,
Lorina

[Joao Santos Silva]

Dear Lorina Sertio,

I would need to know more about your project to be able to advise; can you give more information? More generally, having a good fit is not an assumption in any estimation method I am aware of, so a good fit or a lack of it say little about the validity of the results. For example, it is well known that the so-called spurious regressions tend to have very high R² and there results are still meaningless.

Best wishes,

Joao

[Lorina Sertio]

Dear Joao,

I am currently writing my thesis on the effects of the mobile phone on consumption for different types of Ehiopian farm households.
Using the qregpd command, I have estimated the effects for the 15th, 50th and 85th quantile.

How can I tell that for the chosen quantiles, the estimated model is a good fit?
I understand that, in this context, the R2 may be meaningless but then I want to discuss those limitations in my paper.

I would really appreciate your help.

Thank you.

Best regards,
Lorina Sertio

[Joao Santos Silva]

Dear Lorina,

Thanks for this. May I ask what is the level of the thesis (undergrad, masters, PhD) and why are you using qregpd?

The R² is popular because in some cases it can be interpreted as the proportion of the variance of y explained by the model. When doing quantile regression that interpretation is not valid, so it is not clear what R² could mean. One possibility is to compute the R² as the square of the correlation between y and its fitted values (which is another interpretation of the R²) but there is no reason why that should be high.

Best wishes,

Joao

[Gopal Trital]

My apologies for posting on an old topic here, but since I have landed up in a similar conundrum, I was wondering how best would I be able to estimate the model fit (in choosing between the un/constained models) in STATA. Would be grateful for the answer. Thank you.

Just an additional info: I'm writing Master's Thesis on: Explaining Cross-Province Differentials in Child Nutritional Outcome in Nepal: An Application of Quantile Regression [Nature of analysis: Cross-sectional] [Will be running bsqreg]