“You cannot judge results without
judging methods”
The Essentials
Although the methods and technology used in scientific experiments might vary
tremendously, from particle accelerators in physics to chainsaws in ecology, the
basic design of an experiment should, where possible, follow a few simple rules.
Here is a description of 'the essentials':
Drugs Education - does it work? Many millions of pounds are spent on health
education programmes every year. But do these campaigns actually work? The
only way to find out is to adopt an experimental approach. Suppose you were
asked to determine if a new way of teaching about the dangers of drug abuse to
teenagers had an effect. You could set up an experiment as follows:
14
What is wrong with this design?
Consider how you would interpret the results you might get:
Result Interpretation
Percentage using
drugs goes down
1. The education has been effective.
2. Something else (e.g. a highly
publicised death, like that of Leah Betts)
has changed behaviour.
3. There is bias e.g. students are lying.
Percentage using
drugs goes up
1. Education has encouraged drug use.
2. Education has discouraged drug use,
without it, even more students would be
using drugs.
So the experiment cannot tell you anything useful, because you could interpret
the results in many ways. The first thing you need to do to improve the design is
to include a control as well as a treatment group, that is a group which does
not receive the treatment, to act as a comparison.
Write down now four characteristics of a good control group for this
experiment
Now consider your interpretation of the possible outcomes:
Result Interpretation
Drug use in the
treatment group goes
down, drug use in the
control group stays the
same
1. The education has been effective.
2. Some other factor has acted to reduce drug use in the
treatment, but not control group (e.g. perhaps one
member of the treatment group was arrested, thus
scaring his friends off drugs).
Drug use in treatment
and control groups
stays the same
1. The education has been ineffective.
2. Some factor would have caused the treatment group
to increase their consumption of drugs, but they did not
do so because of the education.
Although the design is now better, it still cannot answer your question. This is
partly because the question itself is too broad. It is much harder to answer the
question "Does this intervention have an effect on illegal drug use?" than to
address the more focused question, "Does this intervention reduce illegal drug
use?" In general, it is best to make sure that the hypothesis you are testing (in
this case, that this education reduces drug use in teenagers) is as focused as
possible.
15
The other main flaw is the lack of replication, both of treatments and controls.
How do we know that the groups we have chosen are representative? If the
treatment group happened to have a charismatic teacher, drug abuse might
reduce because of this, and not because of the particular educational technique
of interest. However, if we had ten treatment and control groups, chosen
randomly, it is most unlikely that the treatment groups will all happen to have
particularly charismatic teachers, and the control group particularly poor ones. So
this would be a much better design:
Increasing the number of replicates will make the experiment more convincing,
but will increase the costs and work involved. One way round this might be
simply to subdivide the treatment and control groups. For example, if all the
students in a class (rather than the class itself) were regarded as replicates, a
single class would provide perhaps 20 replicates.
Write down what is wrong with this approach.