Cobb Douglas with percentage shares

[ARDE]

Dear all

I am trying to model an expenditure (E) variable by an income variabel (inc) and a list of percentage (per) variables (such as physician density, share of population older than 60 and so on). Some authors recommend to include all percentage variables only in a natural exponential form, i.e.
E = inc^a * exp(b per) * exp(u)
or equivalently
log E = a log inc + b per + u

See e.g. http://www.york.ac.uk/media/economics/d ... 9/9920.pdf

However, my model appears to perfom much better when I allow the logarithms also for the percentage variables, i.e.
E = inc^a * per^b * ...
I know that this means that the model breaks down as soon as per=0, but a share of zero is not within the relevant range of our data.
My question is whether you know of any relevant publications that also use logarithms for variables in percentage terms (in higher ranking journals).
By the way: I do not really see the point why it must be wrong to include percentages as a multiplicative factor in a Cobb-Douglas Framework (unless it is zero). Are there any other arguments (apart from the paper that I cited above)?

Thanks a lot for your help!

Best

ARDE