Clustering Industry types

[Sabine Dicks]

I have constructed the following one year lagged regression of ESG on ROA. However, I do not get an F-statistic. Probably because I have inserted too many categories for industry. How can I best solve this? Can I cluster some categories together? If so, what categories?
Kind regards,
Sabine Dicks

[William Lisowski]

I do not expect that the number of categories of industry is the source of your problem.

If you omit the robust option, do you get an F statistic?

If you include the robust option and omit l.RD do you get an F statistic?

I'm concerned that numerical precision difficulties in calculating the robust estimates of standard error may have caused the problem.

Generally, I'm concerned when I see coefficients that are 7 orders of magnitude smaller than most other coefficients. If RD is, for example, a quantity measured in dollars, then I would

replace RD = RD/1000000

and then document that RD is measured in millions of dollars rather than in dollars.

[Carlo Lazzaro]

Sabine:
William is right.
You did not get an F-statistic because of non-default standard errors. Just click on the link to get more details about that issue.

[Sabine Dicks]

Dear William and Carla,

Thank you so much for your responses. I have now excluded the robust function and get the following data. Carlo Lazzaro I can't seem to find the link to more details about the issue.


Kind regards,
Sabine

[William Lisowski]

Wow, this is one of the few cases I've seen where presenting a screen capture of output is an improvement over using CODE delimiters. Revisiting post #1, and looking very closely, I see now what Carlo saw - the "F(14, 500)" is in blue, which is Stata's way of signaling a clickable link (in the default color scheme). So rerun the command you ran in post #1 and click on F(14, 500) and see what Stata has to tell you. And please let us know, because without being able to recreate your regression I can't tell where the link leads to.

[Carlo Lazzaro]

Sabine:
in addition to William's helpful advice, you can simply type -help j_robustsingular- to access the page the link would lead you to.

[Sabine Dicks]

Thank you for your responses again . I have clicked on the link and this is what I see:

[Carlo Lazzaro]

Sabine:
"Are you using a svy estimator... i" s the title of the paragraph you shoyod be interested in.

[Sabine Dicks]

Thank you so much. So I understand that I have to many constraints. Do you know how I can best solve this in my model?
Kind regards,
Sabine

[Carlo Lazzaro]

Sabine:
not quite.
The issue rests on the number of coefficients to be estimated, that are >14 ( see F(14; 580); 14 is the max number of coefficients that you can estimate with non-default standard errors).