[Carlo Lazzaro]

Ial:
your post is actually too long to follow in detail.
That said, if you're using an -fe- specification, what's the gain in investigating -industry- as usually firms are assumed to remain in the same industry all their life long?

[lal mohan kumar]

Dear Professor Carlo Lazzaro
Thanks for your reply. I thought of posting the entire output that Stata gave me so that community can know if there are any errors in my model specification which I couldn't elaborate on.
,

if you're using an -fe- specification, what's the gain in investigating -industry- as usually firms are assumed to remain in the same industry all their life long?

You are absolutely correct. Though there are many cases where the firms change their industry affiliation, a lot more firms in my sample remain in the industry during the sample period. Thus, firm fixed effects that capture industry fixed effects will be fine, hence as you hinted, fixed-effects at the firm level (id) will be fine, Right?

However, just out of curiosity I would like to know why the results are different if there are any empirical reasons for it?
For instance, while we use firm fixed effects, I guess that intercepts are created for no: of firms while if we use industry effects, a lot more intercepts get reduced (since industries engulf firms). Will this matter the sign and size of the estimates?

Also, some research articles instead of using localized fixed effects (in my case firm is the lowest unit) use fixed effects at a higher level (industry in my case) without explaining why they choose Industry (or that matter say States in development economics) over the firm (counties in the case of development economies). Also, I read somewhere that sometimes "local" fixed effects can wipe off useful variation hence at times it better to include fixed effects at a higher lever where there are variations.

My doubt is that are the fixed effects at say local level always a norm or are there instances where we can have fixed effects at some higher levels? Are there any ways to choose which fixed effects should we use say from the structure of data.

Professor Andrew Musau similar to your view says

Since firms are within industries, the industry effect is captured by the firm dummies: industry dummies are therefore not necessary in the presence of firm dummies. https://www.statalist.org/forums/for...01#post1294001

But my point is not including both effects simultaneously but which one is ideal based on situations
Any thoughts in this regard will be quite useful even if there are strict answers

[Carlo Lazzaro]

Ial:
your (more tha legal) curiosity cab be satisfied with https://stats.idre.ucla.edu/stata/fa...d-by-xtreg-fe/
That said, going -fe-. I think you should consider -firm- as your -panelid- and forget -industry-.

[lal mohan kumar]

Dear Professor Carlo Lazzaro
Roger that. I will go with firm fixed effects
Have a good day

[Carlo Lazzaro]

Ial:
you should also compare -fe- with -re- specification and investigate whether your regression model is misspecified or not.
More substantively, please call me Carlo, like all on (and many more off) this list do. Thanks.

[lal mohan kumar]

Hi Carlo Lazzaro
My sincere apologies for the wrong salutation. With regard to the model misspecification, I can only use -

xtoverid-

, right?,as I have invoke vce(robust) option

[Carlo Lazzaro]

Ial:
nothing to apologize, obviously: you were respectful indeed.
You're right with going -xtoverid- to check which specification is better for your dataset.
Misspecification is a bit different, as it implies testing if the regressand functional form is correctly specified.