Date: Fri, 9 Feb 2007 16:41:50 -0500
Reply-To: Richard Ristow <email@example.com>
Sender: "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From: Richard Ristow <firstname.lastname@example.org>
Subject: Re: Mass Reverse scoring
Content-Type: text/plain; charset="us-ascii"; format=flowed
At 12:36 AM 2/9/2007, Mark Webb wrote:
By the way, if you have a 5-point scale named MY_ITEM, the reverse of
COMPUTE REV_ITEM = 6 - MY_ITEM.
But if you have user-missing values, you don't want to change them, so
you should use RECODE instead.
>Try recode in same variables or recode into new variables depending on
I was going to say, never do the former, because it's totally
confusing: there's no way to tell that the variables' coding has been
reversed, you'll forget whether you've done it...
I recently worked on a small study, following that advice. For
questions that were reverse-coded, we created scratch variables with
the coding reversed accordingly, and used the reversed versions in
But on a bigger study finished about a year ago, I did recode
reverse-coded questions 'in place' - recoded the original variables.
The survey had 60 forward-coded and 38 reverse-coded questions, and a
great many scales. (Likely too many, but that's another question.)
Remembering when to enter to original variables and when to enter
reverse-coded forms seemed even more error-prone than recoding the
But it took a lot of precautions to feel comfortable doing it:
. There was a very clear data path from one stage of processing the
data to the next. It was very clear, and clearly documented, at which
stage the reverse-coded variables had been recoded.
. Variable labels identified forward- and reverse-coded questions:
s02q01 '-- Everybody knows who's on the team'
s02q02 'RV Great uncertainty about team goal'
. Value labels were changed to match the recoding for reverse-coded
questions, so at least a FREQUENCIES would be clear:
5 ' 1.Very Inaccurate'
4 ' 2.Smwt Inaccurate'
3 ' 3.Nthr.Acc/Inacc '
2 ' 4.Smwt Accurate '
1 ' 5.Very Accurate '
8 ' 8.NA '
9 ' 9.BLANK '
(Note that the label gives the number originally selected on the
questionnaire; the value is the value into which it's been recoded.)