Strictly speaking a single Likert item is measured at the ordinal level, so
you ought to use noon-parametric procedures. However, in practice we
usually start by combining several items into a scale, generally by adding
them. Having treated the data inn this way, there seems to be no good
reason for not continuing, and with reasonable sample size most "parametric"
procedures are robust.
I don't remember the details, but the scaling method described by Likert (in
about 1936) used quite an elaborate model to estimate the true item category
boundaries, but it was found empirically that just adding 5y3 bite scores
(having reversed any with a negative meaning) gave essentially the same
results.
----- Original Message -----
From: "Anthony Rodriguez" <[log in to unmask]>
To: <[log in to unmask]>
Sent: Thursday, July 28, 2005 10:00 AM
Subject: Likert-type scales
> Dear Allstaters,
>
> It appears there are differences of opinion on whether parametric or
> non-parametric tests can be used for a survey instrument using a
> Likert-type scale.
>
> Comments?
>
> Tony
> ([log in to unmask])
>
|