Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.

RE: st: Is this a situation for mlogit? xt-flavor of mlogit?

 From "Ploutz-Snyder, Robert (JSC-SK)[USRA]" To "statalist@hsphsun2.harvard.edu" Subject RE: st: Is this a situation for mlogit? xt-flavor of mlogit? Date Mon, 2 Aug 2010 16:26:50 -0500

```Thank you David,

Indeed, your summary is correct.  My problem is that my data are not 1,0, but rather the percent in each of 4 different categories, summing to 1.

I can classify the 4 cell types by using 2 additional indicator variables (1,0 on each) so that the 4 cell types are defined by:  Has Characteristic A (1/0) x Has characteristic B (1,0).  Thus the 4 cell-types are defined as (1,1) (1,0) (0,0) and (0,1) on the two different characteristics of the blood.

So with vars called A and B, each taking  on 1,0 from these two characteristics...how would I model that with xtmixed?

??  ->  xtmixed y group##time##A##B || child: || time: ||  ???

Also, can you help me understand the r. notation?  How does  your hypothetical model (assuming for the moment that I didn't have the sume-to-100% problem) differ from

xtmixed y group##time##celltype || child: || time: ||celltype:

Rob

-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Airey, David C
Sent: Monday, August 02, 2010 4:07 PM
To: statalist@hsphsun2.harvard.edu
Subject: re: st: Is this a situation for mlogit? xt-flavor of mlogit?

.

You have two groups with child nested in group.

Then you have within each child time and cell type crossed with child.

So you have one between subject factor (group) and two within subject factors (time by cell type).

If you had a continuous measure that was not a proportion, and did not sum to 1, you could use xtmixed and issue the command similar to

xtmixed y group##time##celltype || child: || child:R.time || child:R.celltype

But your actual measures are proportions, and you say they sum to 1, so that might cause a hiccup in regression type models.

I had a similar problem that was amenable to xtlogit, but the data were scored as 1 and 0s...

> Stata Colleagues;
>
> I'm at a loss for how to analyze the following data and am looking for suggestions.  Everything that I think of seems to have flaws... maybe I'm over thinking things a bit?  Maybe  I just need more coffee?
>
>
> Experimental Design: 3(times)x2(groups)x4(celltypes) mixed factorial. (random ID: repeated on time & celltypes)
>
> More specifically
>    ->Blood data are collected from children at roughly three times per child (varname = age).
>           (Experimental protocol was to collect data at approx 12, 18 & 24 months of age,
>            but we have true age recorded in the dataset. There are some missing data.)
>
>    ->Two groups of kids (independent measures factor. Varname = group)
>
>    ->At each of the three repeated times, my outcomes data are the proportion of cells
>      belonging to each of 4 categories.  Thus the 4 "categories" of blood cells total to 1 (within
>      rounding error) per child, per time point.  (varname=y)
>
>
> We want to examine/compare the effects of group, age  and group#age on the distribution among the 4 cell type.
>
>
> I don't typically work with this kind of data, but I think it is a situation in which mlogit would be appropriate, Yes?
>
> If so, is there an xt flavor of mlogit that would be better than mlogit with robust errors?
>
> Rob
>

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/
```