Hello, I'm trying to work out the advantages/disadvantages of using xtreg with a panel dataset, or using 'reg' with dummy variables. My problem is that I have two variables with fixed effects.

I'm doing a difference-in-difference analysis to study the effect of removing financial incentives from quality indicators, in primary care clinics.
I have a balanced dataset - for each of 450 clinics, I have an observation for 26 quality indicators, for three years (two indicators are controls, where incentives weren't removed).

Each indicator, and each practice, is likely to have its own 'fixed effect'. (For example, some indicators are harder to achieve; and some practices have more difficult patient populations).

So far, to adjust for indicator and practice fixed effects, I've used the code:

Code:
reg Performance Time##Treated i.Indicator i.Practice
This works, but has the draw back that there is collinearity for one of the indicator variables (which means I can't use the margins dydx() function afterwards, which I need).

Is it worth using 'xtreg' instead?

I think need to create an 'ID' variable, which combines the practice and indicator (each ID would have an one observation per year). And then do:

'xtreg Performance Time##Treated, fe'

This doesn't however isolate the indicator fixed effect from the practice fixed effect - and the subsequent result is not quite the same as when I use 'reg'.

Can someone suggest a possible alternative? Many thanks.