Non-normally distributed residuals, advice on how to detrend data

[Ben Cunningham]

Code for fixed effect regression using Driscoll-Kraay standard errors and residual prediction :

xtscc realgvaperh lifesatisfaction anxiety logrealgfcfl2, re

predict newREDK2, resid

qnorm newREDK2

Here is the qnrom :

I think the lack of normal distribution is due to a variable I have that has an upward yearly trend

Not a great graph, can't seem to get the colour coding right, but it's panel data and this is a scatter

scatter logrealgfcfl2 Year

Here is the graph :

Of the data against time, as you can see it sort of trends upwards for every panel variable individually. It's already put in real terms and it's already logged and it's still trending, I've tried taking the data away from a lagged version of itself (difference detrending I think it's called?) and that didn't help much.

My dependant variable's qnorm looks similar, so there's a lack of normality there as well. Could this be causing it?

Dep variable qnorm :

[Nick Cox]

Please post your images here, not somewhere else. If anyone clicks on one of your graphs, they can't easily see the question at the same time. Remember that many people read Statalist on phones, tablets, whatever. FAQ Advice explains how to attach image files.

(Sorry, but I don't do this kind of analysis to address the question.)

[Iuri Gavronski]

Ben,

What happens if you include the time variable in the equation:

 
 xtscc realgvaperh lifesatisfaction anxiety logrealgfcfl2 Year, re

[Ben Cunningham]

@Luri

It actually becomes slightly worse, surprisingly. I would have assumed it'd taper out at least somewhat.

[Iuri Gavronski]

And with year dummies?

i.Year