Discussion:
should I change negative sign in factor loading into positive before running analysis?
(too old to reply)
zencaroline
2007-02-03 18:30:18 UTC
Permalink
Hello,

In my survey questionnaire, I have negative worded question in
order to prevent respondents from answering the questions in a similar
pattern. For the negative worded question item, after running factor
analysis, I will get negative factor loading in the print out of
factor analysis. I will group the items with high factor loadings for
one factor, regardless of negative sign of the item question, due to
negative worded sentence question.

Later I will use the result of factor analysis to run statistic
analysis, here is my question ---------- How should I deal with the
negative worded item question with negative factor loadings? Should
I reverse the negative sign into the positive sign first and then
start to run satistic analysis? Or I just run statistic analysis
directly right after running factor analysis without reversing the
negative sign of the negative worded questions into positive sign? Or
I should do something else?

Thank you very much.

Please take care

Caroline at USC
E***@Helsinki.Fi.INVALID
2007-02-03 19:35:19 UTC
Permalink
zencaroline <***@gmail.com> wrote:

: Later I will use the result of factor analysis to run statistic
:analysis, here is my question ---------- How should I deal with the
:negative worded item question with negative factor loadings? Should
:I reverse the negative sign into the positive sign first and then
:start to run satistic analysis? Or I just run statistic analysis
:directly right after running factor analysis without reversing the
:negative sign of the negative worded questions into positive sign? Or
:I should do something else?

If you use factor scores (estimated by regression method) the resulting
variable is correct without doing anything.

I you use a composite (by summing the most important items into a total
score) you have to reflect the (negative item) scores into same direction.
Otherwise the summing is meaningless. This procedure should be a core
subject in any psychometric course. The reflected variables correlate
with the original once -1. The procedure could be checked by the first
principal component in FA or through reliability module.

Good luck!

Erkki

040-5024491 <http://www.helsinki.fi/people/Erkki.Komulainen/>
Ray Koopman
2007-02-04 06:30:22 UTC
Permalink
Post by zencaroline
In my survey questionnaire, I have negative worded question
in order to prevent respondents from answering the questions in
a similar pattern. For the negative worded question item, after
running factor analysis, I will get negative factor loading in the
print out of factor analysis. I will group the items with high
factor loadings for one factor, regardless of negative sign of the
item question, due to negative worded sentence question.
The easiest way to avoid all those problems is to reverse-score the
negatively worded items. For instance, if your items have 5 response
options and the positively worded items are scored 1,2,3,4,5 then
score the negatively worded items 5,4,3,2,1. Reverse-scoring
simplifies the interpretation for both the writer (you) and the
readers. It is a well-known technique that no one will question. You
should, of course, indicate which items you have reverse-scored.

However, you should also know that a negative loading means that the
relation between the the item (however it was scored) and the factor
is backwards: people with higher scores on the factor generally have
lower scores on the item. The strength of the relation is just the
same as if the loading were positive; only the direction of the
relation differs. Also, note that, regardless of the sign of the
loading, the direction of the implication is from the factor score
to the item score, not the other way.
Post by zencaroline
Later I will use the result of factor analysis to run statistic
analysis, here is my question ---------- How should I deal with
the negative worded item question with negative factor loadings?
Should I reverse the negative sign into the positive sign first
and then start to run satistic analysis? Or I just run statistic
analysis directly right after running factor analysis without
reversing the negative sign of the negative worded questions into
positive sign? Or I should do something else?
You don't say what kind of statistical analysis you're contemplating,
but my guess is that you should probably leave the factor loadings as
they come out. It would really depend on just what the analysis is.
zencaroline
2007-02-04 21:42:20 UTC
Permalink
Dear Ray,

Thank you so much for your kind reply.

You mentioned, "You don't say what kind of statistical analysis you're
contemplating,
but my guess is that you should probably leave the factor loadings as
they come out. It would really depend on just what the analysis is.

Caroline asks, "if later I will run structural equation model, do I
have to do anything else? Or I can just follow what you just kindly
recommended?"

Thank you very much and look forward to your kind reply.

Please take care

Caroline
Ray Koopman
2007-02-04 22:41:32 UTC
Permalink
Post by zencaroline
Caroline asks, "if later I will run structural equation model, do I
have to do anything else? Or I can just follow what you just kindly
recommended?"
Nothing else is necessary.
h***@gmail.com
2020-05-14 22:05:05 UTC
Permalink
Post by zencaroline
Hello,
In my survey questionnaire, I have negative worded questions.lest say there are 08 questions of customer satisfation(dependent variable) on lickert scale in which few of them are negative worded. i did reliability, correletaion and regression analysis. crown bach alpha was above 85% and all varibales were significant but when i recoded those negative questions my reliability decreased to 55% and all variables became insignificant
my question is what should i do now? is it necessary to recode those questions ? if not then why not ? and is there any other way to increase crown bach alpha?
Rich Ulrich
2020-05-15 03:32:21 UTC
Permalink
Post by zencaroline
Post by zencaroline
Hello,
In my survey questionnaire, I have negative worded questions.
lest say there are 08 questions of customer satisfation(dependent
variable) on lickert scale in which few of them are negative worded.
i did reliability, correletaion and regression analysis. crown bach
< cronbach >
Post by zencaroline
alpha was above 85% and all varibales were significant but when
i recoded those negative questions my reliability decreased to 55%
and all variables became insignificant
Basically: Not possible that you did correctly what you said.

If there were "08" items with a few in the wrong direction, you
did not get an alpha of 0.85. For 80 items, a few could be
swamped. But correcting the direction of good scores only
increases the alpha, as you surely expect.

If alpha goes down when you "correct" the directions, then
I assume that your reversal screwed up. The uncorrected
correlation matrix should have a bunch of negative r's;
the corrected will have (probably) none, if it is right.

For an item scored 1-5, you can reverse it by subracting
from 6. One way that this can screw up is if Missing is not
yet defined and there are Missing scores of 9 or -9, which
then get rescored to odd numbers. There is a bigger effect
on the computed alpha if the unrecognized missing is big,
like "99", because alpha is a ratio of variances
Post by zencaroline
my question is what should i do now? is it necessary to recode
those questions ? if not then why not ? and is there any other
way to increase crown bach alpha?
You quoted from Caroline only the Hello.

Ray Koopman gave the good answer, found at
https://groups.google.com/forum/#!topic/comp.soft-sys.stat.spss/-c04zC3_dNM

In short: Start by reverse-scoring the items with reversed
meaning relative to the latent factor.

Okay, you tried that. Now, check your item means and
frequencies, to make sure that you don't screw up (and
that Missing is properly accounted for). Look at the
correlation matrix.

Everything should work out from there.
--
RIch Ulrich
Continue reading on narkive:
Loading...