Date:         Wed, 23 Oct 2002 08:51:23 -0700
Reply-To:     "McCrea, Pat A EDUC:EX" <Pat.McCrea@gems7.gov.bc.ca>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         "McCrea, Pat A EDUC:EX" <Pat.McCrea@gems7.gov.bc.ca>
Subject:      Mann-Whitney test:  relationship between parental satisfaction an
              d student achievement
Content-Type: text/plain; charset="iso-8859-1"

Hi list,

A while back, I saw a few notes on the use of the Mann-Whitney test in
comparing distinct categories on a likert scale.  I am wondering if I can
use the same test on 'aggregated likert scale results.'  Here is what I am
thinking about doing.

We have parents' rating of satisfaction with their school on a five-point
likert scale.  In order to simplify the comparison of school-level
satisfaction, the two highest categories were summed and divided by the
total number of parents to calculate a measure called 'percent of parents
satisfied.'  The schools were then grouped into somewhat arbitrary
categories:   0-20% satisfied, 20-40% satisfied, 40-60% satisfied, 60-80%
satisfied, and 80-100% satisfied.  I want to compare the 40-60% satisfied
group to the 80-100% satisfied groups.  My hypothesis is the higher level of
satisfaction should be related to higher student achievement.

All grade 4 student took a reading assessment.  Each student's score was
classified as below, meeting, or exceeding expectation.  The student score
were aggregated to the school-level by creating a measure of student
achievement 'percentage of students meeting or exceeding' (an ordinal
scale).

So my questions are
Can I use a Mann-Whitney test to compare school-level achievement (pct
meeting or exceeding) by school-level groups of parents (those with a
moderate percentage of satisfied parents and those with a high percentage of
satisfied parents)?
Does all the aggregation to school level and manipulation into somewhat
arbitrary groups invalidate the Mann-Whitney test?
Does the variability of school population make all this work invalid?  (Both
in the satisfaction measure and in the student achievement test, the
populations in a school may vary from 1 to 80 respondents and by
aggregating, I'm flattening the internal variability.)

Thanks for your collected wisdom on this question and all the others I read
on the listserv.

------------------------------------------------------------
Pat McCrea
(pat.mccrea@gems7.gov.bc.ca)
Data Analysis and Reporting, Ministry of Education, PO Box 9143
617 Government St, Victoria BC V8W 9H1     (250) 356-2440 fax (250) 356-0407