variables prefix

[Duccio Milani]

Hello everyone,

Is there a difference between creating new interaction variables or using the prefixes of the factor variables i. and c.?

So far, I have always preferred to create variables in the dataset and then insert them into regressions, but I have seen many users using prefixes to create interaction variables.
From what I understand, there should be no difference, but I have tested on my dataset and the results are different. Am I doing something wrong, or is there a difference between the two methods?

xtreg AR GDP_dev ceqi T TxGDP Txceqi, fe

xtreg AR GDP_dev ceqi T i.T##c.GDP_dev i.T##c.ceqi, fe

These two regressions give me different results. T is the treatment dummy, while GDP and ceqi are the two variables it should interact with.

Thanks!

[Andrew Musau]

i.var defines a categorical variable "var". If you multiply two variables, that represents a continuous by continuous interaction, so

xtreg AR GDP_dev ceqi T TxGDP Txceqi, fe

is exactly

xtreg AR c.T##c.GDP_dev c.T##c.ceqi, fe

See

help fvvarlist

for more.

[Duccio Milani]

Thank you for the answer. My confusion stems from the fact that i. is used in front of dummy but categorical, which is another case then my treatment effect.

[Andrew Musau]

An indicator (dummy) has two levels. As you cannot include both levels as they are collinear with the intercept, it matters not whether you prefix it with i. or c. in terms of the estimation.

[Nick Cox]

I would add that some commands presuppose that you used factor variable notation in fitting your model.