Date: Tue, 5 Jan 1999 13:28:01 -0600
Reply-To: Jolie Kanter <neon@ASAN.COM>
Sender: "SPSSX(r) Discussion" <SPSSX-L@UGA.CC.UGA.EDU>
From: Jolie Kanter <neon@ASAN.COM>
Organization: Newscene Public Access Usenet News Service
Subject: Re: Median-split (an easy one!)
Content-Type: text/plain; charset=us-ascii
Reading your reply made my heart race, as your reaction to my question
seemed quite strong. Why are "dietary self-efficacy" and "Likert" so
Perhaps I should clarify. In the context of my study, dietary
self-efficacy is a measure of one's perceived expectations for success
in modifying food-related behaviors, such as reducing intake of dietary
fat by substituting whole milk with nonfat milk. In other words, how
strongly does a person agree/disagree that he/she would be successful
in adapting and maintaining a healthy diet.
My aim is to test the hypothesis that high dietary self-efficacy
(measured by an 18 item instrument with a 5 point likert scale with
responses ranging from strongly disagree to strongly agree) is a
predictor of dietary fat reduction. Dietary fat intake will be
measured at baseline and again at three years post-treatment. The
treatment is a behaviorally oriented program leading to dietary
change and weight loss. The dependent variables are dietary fat
intake and body weight at post-tx followup. It is hypothesized that
high dietary self-efficacy scores at baseline will be associated with
greater reduction in dietary fat and greater weight loss at followup.
How would you design the data analysis, and what tests would you
employ? (Please don't keep me waiting in suspense. I can hardly
wait to read your reply.)
> What is "dietary self-efficacy"? Does that make as little
> sense as "5-point Likert scale"?
> A Likert Scale is an additive scale made up of many items
> which form a reasonable single dimension. Prototypically,
> the items assessed "agree-disagree". My own brief search of
> literature did not uncover examples of
> Likert-type scaling of items has a neutral center for each item.
> If an item is to form a part of a legitimate Likert-type scale
> (legitimized by the assumption that one can design equal intervals),
> then (I am pretty sure) that the item mode will not be at either
> extreme, for any item. I mention this clarification, since the
> use of "Likert" in the question above was totally mystifying.
> Why do you want to take a scale total and waste information by
> forming a dichotomy, instead of analyzing the continuous score?
> Is that what you want to do? If so, why do you have any problem
> in just doing it? - you know, assign the median-total to the
> side that leaves the Ns most even.
> Rich Ulrich, biostatistician email@example.com
> http://www.pitt.edu/~wpilib/index.html Univ. of Pittsburgh