You've asked this twice. I'll respond to the earlier one, which was
somewhat more descriptive.
If you have correctly described what was done, the procedure is not
logical nor correct. What the intended result should have been was
to convert the range of values {-3,...,+3} to the range of values
{1,...,7} for positively worded statements (large numbers denoting more
positive responses and small numbers denoting more negative responses),
usually for the reason that one is much less likely to confuse a "2"
with a "6" than to confuse a "-2" with a "+2". For negatively worded
statements, the intent should be to convert the range {-3,...,+3} to the
range {7,6,...,2,1} for negatively worded statements (large numbers
still denoting more positive responses -- that is, greater disagreement
with a negative statement -- and small numbers denoting more negative
responses -- that is, agreement with a negative statement): this would
be done not by adding -4, as you say (that yields a range of
{-7,...,-1}, with the larger (= less negative) numbers denoting more
negative responses), but by subtracting the original score from +4.
(Or, you could get the same effect by adding -4 and then multiplying by
-1. Are you sure the investigators you're reading did not take this
last step, that is, multiplying by -1? Or, perhaps, multiplying by -1
in the first place and then adding -4?) -- DFB.
If there happen to be equal numbers of positive and negative statements
in the instrument, I can imagine wanting to convert the scores to give
{1,...,7} for the positive statements and {-1,...,-7} for the negative
statements, with larger numbers (+7 or -7, e.g.) representing greater
agreement with the statement and the sign representing the direction of
the statement. A score summing all the statements in the instrument
would then have an expected value of 0 under one imaginable null
hypothesis. To attain this result would entail (1) adding the original
scores to +4 for positive statements and (2) subtracting the original
score from -4 for negative statements. Or equivalent arithmetic.
(I don't think I'd bother with so elaborate a scheme, but others might
find it alluring.)
On Tue, 17 Jun 2003, Dianne Worth wrote:
> I have seen responses from surveys that are converted in a number of
> ways, but do not understand the logic. Example: A 7-pt. Likert-type
> scale that asked how much one disagrees/agrees with the statement.
> It was originally scored at -3/+3, with 0 being neutral. It was
> converted by adding +4 if the statement was written in a "positive"
> sense and -4 if "negative."
> The model was multiple regression.
Well, no, I don't think so. The method of analysis may have been
multiple regression, but the model would have been something like
Y = b0 + b1*X1 + b2*X2 + ... + bp*Xp + e
for some response variable Y and Likert items X1,...,Xp.
Or, possibly, something of much the same form where X1 = the total of
all the Likert items, X2 = some other variable of interest, etc. Or one
may have used (e.g.) X1 for the sum of all the positive Likert items and
X2 for the sum of all the negatively worded statements.
On reflection, it is also possible that the regression model was of the
first kind above (each Likert item being one of the variables), and part
of the question of interest lay in seeing whether the coefficients bj
associated with the negative items were in fact negative (and those
associated with positive items, positive); the sign of the _values_ of
the variable in question would indicate whether one had thought
beforehand that the item should behave positively or negatively. Such
an approach would lead to the conversion you describe.
> My question: why is it necessary to do this? What is the benefit?
> Thanks. DW
This would have been a LOT easier to answer if you'd supplied more
details of what the investigators actually did with their analyses, not
just with their data.
[Hmm. The more I think about this, the more it feels backwards to me.
Were I conducting such a survey, I'd have coded the original responses
{1,...,7}, so that keyboard transcribers would have only one character
to record for each item (the responses {-3,...,+3} require 2 characters
per item). I might then have transformed responses X to (X-4) to get
values from -3 to +3 for positive statements, and transformed X to (4-X)
to get values from +3 to -3 for negative statements.
But chacun a son gout... ]
Have I confused you utterly? -- DFB.
-----------------------------------------------------------------------
Donald F. Burrill [EMAIL PROTECTED]
56 Sebbins Pond Drive, Bedford, NH 03110 (603) 626-0816
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
. http://jse.stat.ncsu.edu/ .
=================================================================
