COMMONWEALTH OF AUSTRALIA

Copyright Regulations 1969

WARNING

This material has been reproduced and communicated to you by or on behalf of

Griffith University pursuant to Part VB of the Copyright Act 1968 (the Act).

The material in this communication may be subject to copyright under the Act. Any further
reproduction or communication of this material by you may be the subject of copyright
protection under the Act.
Do not remove this notice

Course of Study:
(CCJ19) Doing Criminology

Title of work:
The science game; an introduction to research in the behavioural and social
sciences, 7th ed. (2007)

Section:
The Science Game (Chapter 5, pp. 91 to 94) Control-Group Method pp. 91--94

Author/editor of work:
Agnew, Neil McK.; Pyke, Sandra W.

Author of section:
Sandra W. Pyke, Neil McK. Agnew

Name of Publisher:
Oxford University Press
 5_ _ __
Control~Group Method

We have learned that determining cwhat causes "~Nhat" is no simple n1atter

regardless of what method we use. When introducing a paTticulaT "treat
ment" (independent variable) in the before-and-after method; it's difficult
to tell whether a difference between 01 and 0 2 (the dependent variable)
tesults from the treatment or from something else. In the prevlqus chapter
we indicated that the control-group metho'd represents a major break-
through in helping deal with the four rogues: in-the-gap, time-tied,
elastic-ruler, and on-stage. Therefore one of the first questions a consutner
of science should ask is whether a control group or groups were used in
helping select prime suspects. As we will see, control groups don t guaran- -
3

tee that we will come up with the correct suspect, but they do reduce the
risk of error considerably.
Consider what happens if we use two groups of patients instead.of
one when evaluating a new drug treattnent. We sp1it the group of
depressed patients in half. One group gets the new wonder pill, and the
other group gets a placebo, a pill that looks exactly the same as the wonder
pill but contains only sugar. Some people feel better if they take a pill-any
pill. We divide the patients so there is little chance of getting healthier
patients in one group than in the other. Ideally, the gmups should statt out
Identical in as many respects as possible. To protect yourself from bias in
'assigning people to groups, pick the names out of a hat; the first name goes
to Group 1, the second name to Group 2, the third to Group 1, and so on.
Unless you use some such method, you end up with a special collection of
patients in one group (for example, as a result of staff members attempting
to get their patients or relatives into the group that gets the wonder pill).

91
92 Control-Group Method

Having established two comparable groups, we treat thetn exactly the

same way with one exception: One group pets the wonder pill, while the
other group gets the sugar pill or placebo.
The procedure now becomes:

,-,
(~) <f~ ... 'X"~ ... ...... , \Xnl
~~

Group I C-"'
10,1 ~~~
' '"""'
' ,_,
tO,,
~Y ~/

New Drug
r:_ . . ., ~--....... ,r"-\ (]7\ [-"\
Group 2
~~ \Xt; \X21 .. , ,XP1
..,,., -....-- ' ....... . , \Xn1 p,.y
~- '-'
Sugar Pill

Thus the two groups start out supposedly with the same amount of
depression-that is o, = o,a-{)f in other words, one group does not have
more seriously depressed patients than the other does. Following treatment
we see whether 0 2 is less than 0 2 a-whether the level of depression is now
less for Group 1 than for Group 2.

ROGUE SUSPECTS

Historical (IntheGap) Suspects

The main point is to try to run the experiment so that the individuals
in the two groups are treated exactly the same, except for one suspect-the
dl"Ug X". Thus we attempt to make sure that the same X's pour into the gap
between 0 1 and 0 2 as between 0 1" and 0 2 a. To ensure that the nursing staff
does not spend more time with the patients in one group than with the
patients in the other, the patients in the two groups are mixed up or made
indistinguishable as far as anyone who can influence the experiment is
concerned. The nurses and other doctOI"s are not told which patients are
getting the new wonder drug (XJ) and which are getting the sugar pill (Xp)
All patients teceive pills that look identical.
If both gmu ps are to be open to influence by the same historical or
time-tied suspects, then both groups must occupy the same space and time
frame-for example, they must occupy the same hospital ward at the same
time for the smne duration.

lnstrument~Decay (ElasticwRuler) Suspects

If the experiment is run properly, the doctor who meastues the
depression at the beginning and at the end does not know which patients
received Xd and which received Xp, so the physician's biases, or elastic-ruler

1In some studies instead of ghing a sugar pill. the researcher uses the treatment pill in

common use. Thus the researcher can see whether more people improve in the new pill group
than in the group receiving the usual treatment.
 Control-Group Method 93

effects, cannot systematicaLly influence the doctor's assessment of Olle

group over the other, either during the study or when deciding which
patients have improved and which have not. This double-blind procedure,
where neither the patients nor the treatment evaluator knows who got what
pill, helps protect against elastic-ruler and on-stage effects. Only the
researcher knows the code, which is not disclosed until 0 2 and 0 2 a are
completed.

Testing (On-Stage) Suspects

The on-stage effects of having been interviewed would influence
patients in both groups. There will be patients in both groups who want to
impress the doctor that they are well enough to go home, as we11 as some
who merely want to "help the nice young doctor. 11 Thus -we hope that the
resulting influence on 0 2 and 0 2 will be about the same-that is, that both
groups will probably show about the same amount of on-stage improve-
ment apart from any effect of the new drug. .

Maturational (Time~ Tied) Suspects

Furthermore, spontaneous recovery should be about the same for
both groupsJ since the time between 01 and 02 is the same as between ola
and 0 2 Thns time-tied or natural-recovery suspects should affect each
group the same way.

NOW WHAT?

Notice that the control~group method does not eliminate the individuai or
combined influences of the fonr rogues; rather, the control-group design
provides the Togues with equal access to both groups. Thus 0 2 reflects the
influence of the four rogues combined with the druf!i, while 0 2 reflects the
influence of the rogues combined with the sugar pill.
We should not be surprised, therefore, if bod1 groups show some
improvem_ent: 0 2 shows an improvement over 0 1, and 02a shows ~n
improvement over 0 1a. These changes reflect the effects of such suspects as
spontaneous recovery (time~tied), a biased doctor (elastic-ruler), a nice
ward supervisor (in-the-gap), and a desire to go home (on-stage). If, in
addition, Xd has had an effect greater than Xp, we should have 0 2 showing
a greater shift than 0 20 . If the tv-.ro groups were the san~e to begin with, the
difference between 0 2 and 0 2 provides us with a measure of the effect of
Xd over Xp. This is in contrast with the simple before-and-after model
where we have Xd effects all mixed up with the effects of the other suspects
without being able to untangle them. It was this kind of tangle that led
some wit to wisely observe that a good doctor keeps d1e patient occupied
while nature works the cure. It is easier to wait for a natural change when
under the illusion that some potion is bringing it about.
When we divide a group in two to make the two sections as identical
as possible and then give them the same treatment except for one X, we are
using a control-group method-a much more precise sieve than the natural-
94 Control-Group Method

istic observation, after-the-fact, and before-and~after sieves discussed so

far. Representing a remarkable leap forward in helping us produce
packages of durable information, in one stroke the conuol-group method
permits researchers to assess the effects of their ueatment over and above
the effects of the four rogues alone.