1
$\begingroup$

I have removed midpoints from a 5-point Likert scale instrument on my supervisor's advice. When I did some reliability analyses - SPSS showed that almost half of my cases were excluded. So we substituted the means for the missing cases. Is this right?

The midpoint was removed because it is assigned to answers such as: I don't know, neither agree/disagree. This can inflate the means. This is a conceptual argument that was made to me.

$\endgroup$
4
  • 4
    $\begingroup$ What reason did you have for removing part of the measurement scale after collecting data? $\endgroup$
    – RAND
    Apr 9, 2019 at 0:01
  • $\begingroup$ And you did not explain how exactly you "removed" it? What did you do? $\endgroup$
    – ttnphns
    Dec 9, 2019 at 17:53
  • 3
    $\begingroup$ Are you asking whether or not you should replace midpoint responses with the mean? Can't imagine why that would be a good idea. $\endgroup$
    – logistic
    Dec 9, 2019 at 18:01
  • $\begingroup$ Using the mean is bad for many reasons including reduced variation, Multiple imputations is recommended.There is a huge difference between I don't know (which should not be in the code since this is not logically in the middle of the measure) and neither agree or disagree which arguably might be a middle value. Some argue, its called forced choice that you make even likert scale to avoid the practice of people always choosing the middle value. But that is in the original question - not getting rid of that after you asked. $\endgroup$
    – user54285
    Mar 25, 2020 at 23:32

2 Answers 2

8
$\begingroup$

I'm not sure why your advisor thought it was a good idea to remove a response option from the data after collection, as @RNM noted. As well, mean substitution is seldom a really good option for missing data.

If you have to work with the altered data, one option you can do with SPSS if you have access to the MVA procedure is to use the EM imputation option to create an estimate of the covariance matrix of the items and use MATRIX IN to read that into RELIABILITY. That's likely to produce more defensible results.

$\endgroup$
5
$\begingroup$

First, you created missing data where there was none. Perhaps your supervisor had something particular in mind, but I can't see what it could be. The midpoint is not missing.

Second, you used a very poor method of imputation - mean substitution. This makes the missing data much too regular. If you really insist on creating a problem then, at least, you could solve it appropriately, probably using some form of multiple imputation.

But ask your supervisor why creating missing data was suggested.

$\endgroup$
1
  • $\begingroup$ Multiple imputation would not be solving the problem appropriately unless you can somehow justify the MAR assumption. $\endgroup$
    – logistic
    Dec 11, 2019 at 14:05

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.