Using absolute value in a fixed effects model

[Elizabeth Day]

I'm wanting to find the absolute value for change over time in a fixed-effects model. I'm using Stata MP 16.

My outcome variable is child maltreatment data for states over time (2006-2016). My independent variables are dummy-coded policy variables (e.g., if a state added a centralized intake system in 2009, it gets a 0 for 2006-2008 and a 1 for 2009-2016). I'm hoping to have the xtreg output coefficients reflect the absolute value difference in maltreatment rather than a directional difference. Is there a Stata option for this? Or will we need to manually create a new dependent variable?

Thank you!

[Carlo Lazzaro]

Elizabeth:
welcome to this forum.
If you're interested in variation in the number of chlld maltreatments over years, your regressand should be expressed in counts.

[Elizabeth Day]

Thanks Carlo! We did explore the outcome as counts, but found treating it as continuous (we've calculated rates and log-transformed) seems to be best, based on analyses and even running it with a poisson distribution and Stata automating a message that was incorrect and to interpret with caution!

We are still hoping to find a way to use absolute value of the change over time to make our coefficients easier to interpret.

[Carlo Lazzaro]

Elizabeth:
-I do not think that you can get what you're after via -xtreg,fe-:
-poisson- throws that warning message you reported whenever the regressand takes on non-integer values;
-as -xtpoisson- allows conditional fixed effects (which differ from the fixed effectsw that can be calculated under -xtreg-). Perhaps you can use a -poisson- regression with -i.panlid- among predictors to fit a -fe- specification to your dataset, and the use -predict events, n- to estimate the cases (and their within panel variation) as time goes by.