Hello Mohammad Razib Hossain

My experience is that, the result may not be exactly the same but the conclusion certainly is

Naturally, it can be a bad indicator and can make harder to publish the study. Nevertheless, it depends on what robustness check are you considering. For instance, if there are contradiction between FE estimates and RE estimates, you should favour the former.

Unfortunately, the common practise of most authors is to report just those robustness checks supporting their hypothesis.

What do you mean by robustness tests? Do you mean specification or misspecification tests? You might also explain what you mean by parallel outcomes. I have never encountered your terminology.in this context.

Dear Godwin Lebari Tuaneh thank you so much for your reply. What will happen if the signs of the coefficients changes after the robustness analysis? Will the outcomes still be acceptable?

Dear José-Ignacio Antón thanks for your reply. If the contradictions after the robustness tests are minor should I proceed with the outcomes?

Dear John C Frain , thanks for your question. Actually by robustness I mean how solid are my outcomes when I compare the outcomes with other dissimilar regression models. By parallel I mean whether the outcomes of the robustness tests mirror the outcomes of the main econometric model.

Dear Mohammad.

Anyways, there could be flaws in your robustness checks.

But since you will be using 2 or more robustness test, at least 2 should favour your original results.

This will verify that your result is okay.

Regards

Dear Kehinde Mary Bello thank so much for your reply.

Mohammad Razib Hossain I think that most of us did not understand exactly what you mean by robustness tests. An example of what I now think you mean is the following.

Suppose you are interested in the change in tax receipts on a consumption item following an increase in the tax on that item. You need a measure of the price elasticity of that item. You have a possible variety of models involving quantity, price, and other variables. You need to assess the sensitivity of the price elasticity when different other variables are included in the estimated model. In an informal sense, I would refer to this as sensitivity rather than robustness but I think that we mean the same. We still need exogeneity as if you exclude any exogenous variables that are correlated with the price variable you will get biased estimates of the elasticity and your results will not be "robust". In this case I would only be interested in the robustness of the price variable. The robustness of the estimates of the other coefficients is irrelevant to the question in hand.

For robustness tests consider https://www.economics.uci.edu/files/docs/micro/s11/white.pdf. or the version of this paper in the journal of econometrics.

I think the similarity of your robustness check to the main method depends on the properties of both estimators. For instance, using augmented ardl as a robust test for bootstrap Ardl could yield similar result. Same might go for LSDV and GMM because they are both dynamic model method and can also correct for endogeneity problem. So the similarity depends on the similarity of the estimators. The coefficient doesn't really matter though but the similarity of the signs of the coefficient is very important..

Regards

Dear Hamid Muili thank you so much for your insightful reply. I do agree with you that the signs are more crucial than the value of the coefficients. However, there are times when the signs also vary. What can be done then?

This is very interesting question, indeed, the robustness selected criteria are very crucial in determining the the outcomes that may differ from one system to another. Therefore, the quantitative robustness can be very questionable when comparing two systems for different outcomes, it is more appropriate to investigate the robustness improvement for the same system and with different parameters in order to ensure conclusive remarks.