Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss
rake weight using more than 5 variables (Dim)
160 views
Skip to first unread message
wollyka
Nov 25, 2009, 2:05:28 PM11/25/09
to
Hi
Like the title says, is it possible to do it? or there is no way to do
so..
my problem is that i have 28 data sets, in each data set, rake module
using 3 variables (religion,gender and age) was used to balance the
sample and it worked fine..
However, each data set represent a region and i want to merge them all
in 1 dataset to process results for the entire regions.. but there is
a problem each region has its own distribution when it comes to
religion so i cannot use " global" religion variable to do the
weighting and the rake module only accept 5 variables max (and i have
here 28!)
can it be done? or is there any other method to do it?
Thanks for the help
Like the title says, is it possible to do it? or there is no way to do
so..
my problem is that i have 28 data sets, in each data set, rake module
using 3 variables (religion,gender and age) was used to balance the
sample and it worked fine..
However, each data set represent a region and i want to merge them all
in 1 dataset to process results for the entire regions.. but there is
a problem each region has its own distribution when it comes to
religion so i cannot use " global" religion variable to do the
weighting and the rake module only accept 5 variables max (and i have
here 28!)
can it be done? or is there any other method to do it?
Thanks for the help
JKPeck
Nov 25, 2009, 11:06:26 PM11/25/09
to
Can you just adjust the totals in each region in the first step of the
process and then merge the files? The limit of 5 is a hard limit in
the procedure because of the underlying limit of 5 in the GENLOG
procedure, which is used by RAKE.
However, raking in a large number of dimensions is usually a bad idea,
because the data can become very thin, and some weights become quite
extreme. You are making an increasingly draconian assumption of
conditional independence.
HTH,
Jon Peck
wollyka
Nov 26, 2009, 5:53:28 AM11/26/09
to
On Nov 26, 6:06 am, JKPeck <jkp...@gmail.com> wrote:
> On Nov 25, 12:05 pm, wollyka <woll...@gmail.com> wrote:
>
> > Hi
> > Like the title says, is it possible to do it? or there is no way to do
> > so..
> > my problem is that i have 28 data sets, in each data set, rake module
> > using 3 variables (religion,gender and age) was used to balance the
> > sample and it worked fine..
> > However, each data set represent a region and i want to merge them all
> > in 1 dataset to process results for the entire regions.. but there is
> > a problem each region has its own distribution when it comes to
> > religion so i cannot use " global" religion variable to do the
> > weighting and the rake module only accept 5 variables max (and i have
> > here 28!)
> > can it be done? or is there any other method to do it?
> > Thanks for the help
Hi> On Nov 25, 12:05 pm, wollyka <woll...@gmail.com> wrote:
>
> > Hi
> > Like the title says, is it possible to do it? or there is no way to do
> > so..
> > my problem is that i have 28 data sets, in each data set, rake module
> > using 3 variables (religion,gender and age) was used to balance the
> > sample and it worked fine..
> > However, each data set represent a region and i want to merge them all
> > in 1 dataset to process results for the entire regions.. but there is
> > a problem each region has its own distribution when it comes to
> > religion so i cannot use " global" religion variable to do the
> > weighting and the rake module only accept 5 variables max (and i have
> > here 28!)
> > can it be done? or is there any other method to do it?
> > Thanks for the help
thx for the reply
how can I adjust the totals in each region?
JKPeck
Nov 28, 2009, 10:06:44 PM11/28/09
to
I was assuming that you have regional data. If so, just rake each
regional subset of the data separately. Perhaps I don't understand
the problem, though.
-Jon Peck
0 new messages