Thanks again Maarten, I think the brainstorm is getting productive. I
understand your point, but disagree. Here's why: the top panel is
actually a prevalence ratio (the ratio of the prevalences among
exposed and unexposed) and when I use effect coding I am subtracting
this estimate from the 'mean' prevalence rate across all categories of
the dependent variable (not dividing it), obtaining a prevalence ratio
difference (or, a relative risk difference - although i disagree with
the use of this term in cross sectional designs).
I'll do some more reading over the weekend, but I think this sounds
like a more reasonable interpretation. Why do you disagree with it?
On 2/2/07, Maarten buis <maartenbuis@yahoo.co.uk> wrote:
--- Diego Bassani <diego.bassani@gmail.com> wrote:
> but that would apply if we were using dummy variables.
> When I use effect coding I am expressing the change in the beta
> compared to a grand mean of the betas in the population (all
> categories combined). This is what is making me think that it is very
> unlikely to be an odds, relative odds or odds ratio since the
> information from the numerator is used to calculate the reference
> value in the denominator.
I still think that you can think of these as odds ratios. Lets take a
concrete example: the example below. The parameter in the top panel are
the odds ratios of dropping out of highschool versus going to college.
So I would interpret _Irace_2 (black) as: the odds that someone drops
out of highschool versus going to college is twice as large for a black
women then for a women of average race. (The dataset is about women)
All the different codings possible with -xi3- are there to make
interpretation easier. They all lead to exactly the same model, the
results are just displayed in a different way. So if you find effect
coding difficult in this context, then just go back to dummy coding.
*------------- begin example -------------
set more off
sysuse nlsw88, clear
gen ed = grade
recode ed (0/11 = 1) (12 = 2) (13/18 = 3)
label define ed 1 "dropout" 2 "highschool" 3 "college"
label values ed ed
xi3: mlogit ed e.race south age, rrr
nlcom (dropout: exp(-[dropout]_b[_Irace_2]-[dropout]_b[_Irace_3])) /*
*/ (highschool:
exp(-[highschool]_b[_Irace_2]-[highschool]_b[_Irace_3]))
*--------------- end example ---------------
Hope this helps,
Maarten
-----------------------------------------
Maarten L. Buis
Department of Social Research Methodology
Vrije Universiteit Amsterdam
Boelelaan 1081
1081 HV Amsterdam
The Netherlands
visiting address:
Buitenveldertselaan 3 (Metropolitan), room Z434
+31 20 5986715
http://home.fsw.vu.nl/m.buis/
-----------------------------------------
___________________________________________________________
The all-new Yahoo! Mail goes wherever you go - free your email address from your Internet provider. http://uk.docs.yahoo.com/nowyoucan.html
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
--
Diego Bassani
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/