7 May
2003
7 May
'03
7:30 p.m.
When i asked gary in lecture, he said it was a rare event because its a 0/1
variable and out of like 900 obs only 50 are 1. So he said use relogit
because regular logit would be biased.
----- Original Message -----
From: "Kosuke Imai" <kimai(a)fas.harvard.edu>
To: "Traci Burch" <tburch(a)fas.harvard.edu>
Sent: Wednesday, May 07, 2003 5:13 PM
Subject: Re: [gov2001-l] Rare Events, Relogit, and Predicted Values
I'm not sure what your question is... If
it's a random sample, why do you
want to use relogit?
Kosuke
On Wed, 7 May 2003, Traci Burch wrote:
How would one go about producing predicted values
for a model run using
Relogit or its equivalent in R? I have survey data that appears to be
randomly selected (no case control or anything like that) and time is
not a factor. Is it possible to do so (its not listed as a potential
quantity of interest in the lecture notes)? if not, what other quantity
of interest can I use to evaluate model fit?
Traci
7 May
7 May
8:07 p.m.
Traci is right w.r.t. small sample bias correction.
Kosuke is right w.r.t. case-control bias correction.
Gary
On Wed, 7 May 2003, Traci Burch wrote:
When i asked gary in lecture, he said it was a rare
event because its a 0/1
variable and out of like 900 obs only 50 are 1. So he said use relogit
because regular logit would be biased.
----- Original Message -----
From: "Kosuke Imai" <kimai(a)fas.harvard.edu>
To: "Traci Burch" <tburch(a)fas.harvard.edu>
Sent: Wednesday, May 07, 2003 5:13 PM
Subject: Re: [gov2001-l] Rare Events, Relogit, and Predicted Values
I'm not sure what your question is... If
it's a random sample, why do you
want to use relogit?
Kosuke
On Wed, 7 May 2003, Traci Burch wrote:
_______________________________________________
gov2001-l mailing list
gov2001-l(a)fas.harvard.edu
http://www.fas.harvard.edu/mailman/listinfo/gov2001-l
How would one go about producing predicted values
for a model run using
Relogit or its equivalent in R? I have survey data that appears to be
randomly selected (no case control or anything like that) and time is
not a factor. Is it possible to do so (its not listed as a potential
quantity of interest in the lecture notes)? if not, what other quantity
of interest can I use to evaluate model fit?
Traci
9:17 p.m.
Yes, after you correct small sample bias, you can simulate predicted
probabilites just like usual logit.
Kosuke
On Wed, 7 May 2003, Gary King wrote:
Traci is right w.r.t. small sample bias correction.
Kosuke is right w.r.t. case-control bias correction.
Gary
On Wed, 7 May 2003, Traci Burch wrote:
When i asked gary in lecture, he said it was a
rare event because its a 0/1
variable and out of like 900 obs only 50 are 1. So he said use relogit
because regular logit would be biased.
----- Original Message -----
From: "Kosuke Imai" <kimai(a)fas.harvard.edu>
To: "Traci Burch" <tburch(a)fas.harvard.edu>
Sent: Wednesday, May 07, 2003 5:13 PM
Subject: Re: [gov2001-l] Rare Events, Relogit, and Predicted Values
_______________________________________________
gov2001-l mailing list
gov2001-l(a)fas.harvard.edu
http://www.fas.harvard.edu/mailman/listinfo/gov2001-l
I'm not sure what your question is... If
it's a random sample, why do you
want to use relogit?
Kosuke
On Wed, 7 May 2003, Traci Burch wrote:
_______________________________________________
gov2001-l mailing list
gov2001-l(a)fas.harvard.edu
http://www.fas.harvard.edu/mailman/listinfo/gov2001-l
How would one go about producing predicted values
for a model run using
Relogit or its equivalent in R? I have survey data that appears to be
randomly selected (no case control or anything like that) and time is
not a factor. Is it possible to do so (its not listed as a potential
quantity of interest in the lecture notes)? if not, what other quantity
of interest can I use to evaluate model fit?
Traci
7682
days inactive
7683
days old
gov2001@lists.gking.harvard.edu
2 comments
3 participants
participants (3)
-
Gary King -
kimai＠fas.harvard.edu -
tburch＠fas.harvard.edu