Re: observations in different scales
From: John Uebersax (jsuebersax_at_yahoo.com)
Date: 07/09/04
- Next message: Prasad Babu M: "Re: multiple non-linear regression"
- Previous message: Osher Doctorow: "Re: Quantum Entanglement Explained by Jacobson Radical + PI"
- In reply to: Richard Ulrich: "Re: observations in different scales"
- Next in thread: Sergey Tarima: "Re: observations in different scales"
- Reply: Sergey Tarima: "Re: observations in different scales"
- Messages sorted by: [ date ] [ thread ]
Date: 9 Jul 2004 03:22:08 -0700
This is an interesting question, with implications beyond the present
application.
To solve the immediate problem, I agree with Rich. From a purely
pragmatic standpoint, the easiest thing in my opinion is to replace
the "one or more" responses with an imputed value. The most naive,
"mean value" imputation would replace these values with the mean of
all specific reponses of 1 or more. But that is questionable: one
might easily believe that people who would like to answer, say, "90
to 100 times" would be disinclined to say "at least one, but cannot
say how many", as that would be much less helpful than simply saying
"90", "100" or "95." If so, then mean value imputation would not
work. Better would be, as Rich suggested, to take a subset of these
respondents, press them for details, and use that to produce an
imputed value for all "one or more" responses.
A fully general approach, however, is also interesting to consider
because this kind of situation happens often in data collection. For
statistical methods (including survival models) that are based on
maximum likelihood estimation, and which involve a specific likelihood
function, then it seems like the likelihood function can be adapted to
handle both types of data (just as it can analogously be adapted to
handle missing responses).
-- John Uebersax Richard Ulrich <Rich.Ulrich@comcast.net> wrote in message news:<aftre0dvifjba29b2rdf0pikdgv0864e4r@4ax.com>... > You might want to debrief your interviewers, to learn their > opinions. (I've been impressed more than once.) > Was "one or more" a hesitation to elect between 1 and 2, > or was it coyness, an unwillingness to put 10 or 100?
- Next message: Prasad Babu M: "Re: multiple non-linear regression"
- Previous message: Osher Doctorow: "Re: Quantum Entanglement Explained by Jacobson Radical + PI"
- In reply to: Richard Ulrich: "Re: observations in different scales"
- Next in thread: Sergey Tarima: "Re: observations in different scales"
- Reply: Sergey Tarima: "Re: observations in different scales"
- Messages sorted by: [ date ] [ thread ]
Relevant Pages
|
