Analyzing in stata with different (splitted) samples

[Stefan Lukic]

Hi all!

Currently, I'm working on my thesis about CSR (archival panel data). I have little experience with stata so I need some help regarding regression analyses
In my study, I'm looking at the effect of misaligning internal CSR actions (what firms actualy do internally about csr. e.g. policies) with external CSR actions (what firms report about csr. e.g. claims and disclosure). My first two hypotheses are:

H1: The sum of internal and external csr actions have a positive effect on market value
H2: The gap of internal and external csr actions (absolute value) have a negative effect on the market value (e.g. due greenwashing , or on the other hand due a lack of transparancy)

However, this looks a lot like a different study. So additionally, I want to look separately (Exceeding internal actions causes lack of transparancy and credibility, where Exceeding external actions may cause to be identified as greenwashing) to the effect on the market value.

So my idea was to develop another 2 hypotheses (H3a Exceeding internal actions is negatively associated with market value & H3b Exceeding external actions is negatively associated with market value). In other words, the original sample will be splitted into firms that exceed in external actions and firms that exceed in internal actions. I've never worked with a split sample and don't know how to apply this in stata (first time that I use stata).

I'd like to hear your thoughts,

Kind regards,

Stefan

PS: if this question already is asked, my apologies, I couldn't find it.

[Phil Bromiley]

You didn't get a quick answer. You;ll increase your chances of a useful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex. Remember, for the most part, we know statistics and Stata but not your substantive area so questions that are really about the substantive area are likely to not get helpful responses.

I would not think of this as a split sample. Instead, think of it as creating two separate variables - one equal the variable of interest when it is positive and zero otherwise, and the other equaling the variable of interest when it is negative and zero otherwise. This can be done manually, or by creating a dummy for positive and using factor variable notation interacting the dummy with the variable of interest to generate the split. Then just run it on the full sample. This is done routinely in aspiration model for example.