regression after this...
Zane mentioned that I should try to transform the DV by normal means by
logging it to see if that works--the reason I've been eyeing those other
wacky models, though, is that the transformation doesn't work. Too many
0's. So I started to wonder if poisson is appropriate--but it doesn't
meet the assumptions of the model, anyways. I'll play with it more when
I get back to the states and have Ying's notes on hand, but I'm really
not sure what to do with it.
It becomes normal if I log the un-recoded variables--but the problem is
that all the 0s disappear and become missing data. And if I recode the
data so there are no 0s (just add 1 to the variable) then the
transformation doesn't work. I get this normal distribution with a huge
spike way off to the left side. Wish I could post a picture with the
e-mail...
I've been thinking that this (poisson model) might make sense, because
at least one of the outcome variables--the number of municipal employees
in forestry--is count data, and the distribution resembles a poisson
distribution.
But basically, I'm trying to deal with this badly skewed distribution
that I can't log into Gaussian normality.
On another note, the size variable that I was hoping would work out
(that was significant before I did some data cleaning, and now is
insignificant again) has this very clear significant and positive
relationship with conflicts, with two huge outliers that make the
relationship insignificant again. The two cases have something in
common (they're different from the other munis because of size, but also
because of, among other things (a) climactic zone, (b)
narco-trafficking, (c) shady border-town operators, (d) number of
Zapatistas, and (e) frequency of Ninja attacks).
So, I've been thinking that maybe I should pay a visit to Melchor de
Mencos or Santa Ana, Petén.
No comments:
Post a Comment