[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: RE: dirifit

From   "Nick Cox" <>
To   <>
Subject   st: RE: dirifit
Date   Thu, 6 Mar 2008 15:10:29 -0000

In addition to Maarten's comment, I'd like to flag two points 
from this: 

1. -dirifit- is working as advertised with your data. You're
understanding the restriction
that each proportion must be from (0,1) correctly. 

2. In a few research problems, there might be a case for shrinking
observed proportions
inward if you can't take observed 0s and 1s as exact, loosely as some
people smooth contingency tables
with sampling zeros. But that doesn't sound the case here,

So, -dirifit- does not sound much use for your problem. You want a
but -dirifit- is a chisel. 


Viktor Slavtchev

Dear Nick,
first of all thank you very much for your kind response.

Indeed, I do have a lot of cases where individuals spend some time on a 
particular activity, while do not consider other possibilities at all.
Hence, there are a lot of cases where at least one of the y's is 0 as 
well as cases where only one of the y's is 1. If I've got you correctly,

-dirifit- ignores such cases.
This issue is actually stated in the help for -dirifit-:
"Note that cases will be ignored if the one or more of the dependent 
variables has a value less than or equal to zero or more than or equal 
to one or if the dependent variables don't add up to one."
So I really apologise for my ignorance.
Indeed, in only 60 cases (of 391) none of the y's is 0 or 1:

. count if y1!=0 & y2!=0 & y3!=0 & y4!=0 & y5!=0

Exactly these have been used for estimation:

. dirifit y1 - y5
<output skipped>

. gen dum1=1 if e(sample)
(331 missing values generated)

. gen dum2=1 if y1!=0 & y2!=0 & y3!=0 & y4!=0 & y5!=0
(331 missing values generated)

. gen diff=dum1-dum2
(331 missing values generated)

. tab diff

       diff |      Freq.     Percent        Cum.
          0 |         60      100.00      100.00
      Total |         60      100.00

*So the question would be, how to proceed when at least one of the 
responses is allowed to be 0, but the model is not really a model of 
competing alternatives (-mlogit-)?

*   For searches and help try:

© Copyright 1996–2019 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index