[Marian Dudeck]

Hey Carlo,
thanks for the answer. I will discuss this topic with my mentor.
But just for the semantics: In the dataset are companies which have a constant board size over the whole time or just put one woman more on Board in the observation period. The Variance sometimes equals zero, while revenue grows each year. If i implement this in fixed effect, it is clear that the revenue has more significant effect than the board size or the women's quota. But if I consider all companies in a industry, board size changes more frequently and since it has more variance, you can see more significance of such board variables.
It is also commonly used in literature (p.37: https://ecgi.global/sites/default/fi...skopffinal.pdf)

But I think it is also valid to say, that in respect to revenue, return-on-assets, investments, etc. board compositions itself is not the way to go if you aim for high esg-score.

[Marian Dudeck]

Hey Carlo,
thanks for the answer. I will discuss this topic with my mentor.
But just for the semantics: In the dataset are companies which have a constant board size over the whole time or just put one woman more on Board in the observation period. The Variance sometimes equals zero, while revenue grows each year. If i implement this in fixed effect, it is clear that the revenue has more significant effect than the board size or the women's quota. But if I consider all companies in a industry, board size changes more frequently and since it has more variance, you can see more significance of such board variables.
It is also commonly used in literature (p.37: https://ecgi.global/sites/default/fi...skopffinal.pdf)

But I think it is also valid to say, that in respect to revenue, return-on-assets, investments, etc. board compositions itself is not the way to go if you aim for high esg-score.

[Carlo Lazzaro]

Marian:
if the change in your regression specification is led by the literature in your research field, it makes sense.
Conversely, if the idea underlying this change is due to find statistical significance coefficients, I do not think it is the way to go.
Obviously your mentor will advise you for the best.

[Jared Greathouse]

In the Bible, the Book of Kennedy gives the Ten Commandments. One of them says "Thou shalt not confuse statistical significance with substance".

Your goal shouldn't be to find statistical significance. Your goal should be to estimate the population parameter as best you can for your variable of interest. The statistical tool tu use to estimate this relationship should come secondary to the design you use.

I know Stata is a statistics package, and it's one that I use very well (in my opinion), but even if I knew all about Stata and Mata, that wouldn't make me or anyone else a good scientist or investigator, and this is true if you desire causality or an associative measure. That is, the design, the way you're testing your ideas has to make sense first.

Whether you use fixed effects or some other method, the goal should be to use whatever best informs you about the relationship between your outcome and focal predictors, and sometimes the statistically significant answer isn't always the best.