If you have any feedback for the author, E-mail me
REQUIRED TEXT: Experimental Methodology eighth edition
by Larry B. Christensen, Allyn and Bacon, 2001
ISBN #0-205-30832-5
------------------------------------------------------------
Chapter 10
Quasi (almost)-Experimental Designs
Nonequivalent Control Groups Design:
No random assignment of participants to experimental & control
groups. Pre-Post comparisons with treatment administered only to the
experimental group but membership in groups is not controlled by
experimenter, e.g. consumers select payment plan themselves and measure
plans influence on increased usage of product, or reading methods across
school rooms, patients' health on high blood pressure medication,
'improved' teaching techniques as semester continues.
Must Eliminate Rival Hypotheses:
I Exp gr begins higher and increases pre to post while con gr
remains the same.
a) Selection-maturation effect: Exp gr were progressing faster
before pre-test, i.e. brighter. If match grs, may choose lowest of one
gr and highest of the other. :. Statistical regression error
b) Local History effect: an extraneous variable that affects one
gr and not the other.
II Rates of increase in DV from pre to post-test occurs in both grs
but faster in exp gr.
a) If one gr scores higher then other at pre-test, Selection-
Maturation Interaction
III The two grs differ substantially of measures of DV at pre-test and
one gr increases / decreases dramatically over the other. May not be a
treatment effect if a) local history and/or b)statistical regression
effect cannot be eliminated. Try repeated pre-testing
IV Crossover effect: Interaction - Cannot be a) Statistical Regression
since it is unlikely for the gr with lower pre-test scores to improve
that much. b) Selection-maturation effect since the higher scoring pre-
test gr should advance even faster then lower scoring gr. = opposite of
effect
Interrupted Time-Series Design: Repeated measures of both grs both pre-
and post-treatment. Compare stable performance and its change. The
fewer the measurements, the probability of concluding that a treatment
is effective when it really is not (type I error) is increased. See
graphs on p 275
Weakness = lack of control of History effects
------------------------------------------------------------
My office is in room 116C of HGH (826-4571) and I invite you
to consult with me on any matter during my office hours or
any time that is agreeable to you and me.
Go back to the beginning
Copyright © 2002, Dr. John M. Morgan, All rights
reserved -
This page last edited Nov 25, 2002
If you have any feedback for the author, E-mail
me