Date:         Thu, 3 May 2007 06:14:36 -0400
Reply-To:     Peter Flom <peterflomconsulting@mindspring.com>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Peter Flom <peterflomconsulting@MINDSPRING.COM>
Subject:      Re: rescaling categorical survey data
Comments: To: David L Cassell <davidlcassell@MSN.COM>
Content-Type: text/plain; charset=UTF-8

David L Cassell <davidlcassell@MSN.COM> wrote
>
>Peter is right.  There is no way to do this.
>
>Even if there is no change in *anything* *whatsoever*, it may be that
>1 maps to 1 and 4 maps to 7, or *not*.  Maybe 1 will have to map to 2
>in the new scale.  This is often a subject-matter area, not a data-analyst
>area.
>
>But there is usually some change other than the scale.  Are the questions
>*exactly* the same?  Are they being asked in *exactly* the same place in
>the survey form?  Are the preceding questions still *exactly* the same?
>Unfortunately, changes in these things can have important effects on
>your instrument.  There's a whole field of study on this type of problem
>in developing surveys.. and there are plenty of other people who study this,
>learning how to make their surveys say whatever they want them to.
>
>So mapping your 1-4 scale to a new 1-7 scale may be really hard.  And
>errors on your part may have impacts on the comparisons you apparently need
>to run...


Writing a good survey is difficult, but most people seem to think it is easy.

Here's an example (I have details wrong, but the idea is right)


Ask people about how much (alcohol) they drink.
Question 1
Are you
a) Not a drinker (less than one drink a week)
b) A light drinker (1 - 3 drinks a week)
c) A moderate drinker (4-7 drinks a week)
d) A heavy drinker (8 or more drinks a week)


Question 2)
Are you
a) Not a drinker (less than one drink a week)
b) A light drinker (1-7 drinks a week)
c) A moderate drinker (8-14 drinks a week)
d) A heavy drinker (15 or more drinks a week)


Ask two randomly drawn samples from the same population.....
Clearly a matches to a
b and c to b
d partly to c and part to d

but nope!  More people answer c) in Question 2 than answer d) in Question 1!


Peter