Cj. Kowalski et al., ASSESSING THE EFFECT OF A TREATMENT WHEN SUBJECTS ARE GROWING AT DIFFERENT RATES, International journal of bio-medical computing, 37(2), 1994, pp. 151-159
The analysis of covariance is often used in the context of premeasure/
postmeasure designs to compare treatment and control groups in both ra
ndomized [1] and nonrandomized [2] studies. The intent is to adjust th
e difference between the changes in the 2 groups for any difference wh
ich might exist at baseline, i.e., for any difference between the prem
easures in the 2 groups. An important assumption underlying the use of
the analysis of covariance is that the slopes of the lines for the re
gression of the postmeasure on the premeasure in the 2 groups are equa
l. In this paper we describe a program which can be used to test the h
ypothesis of equal slopes; and performs an alternative analysis which
does not depend on this assumption. This is done in the context of com
paring treatment and control groups with respect to a measurement subj
ect to natural maturation as in [3]. Equal slopes in this context mean
s equal growth rates; unequal slopes implies that the 2 groups are gro
wing at different rates. The method, known as the Johnson-Neyman proce
dure [4] is, however, more general than this, and can be used in any t
wo-sample comparison where an alternative to the usual analysis of cov
ariance is deemed appropriate. The procedure identifies a 'region of s
ignificance' which is especially useful in practice. This region consi
sts of a set of values of the premeasure for which the treatment and t
he control groups are significantly different with respect to the post
measure.