1- Questions,
Hypotheses and predictions
-
A hypothesis is a clear statement
articulating a plausible candidate explanation for observations. It should be
constructed in such a way as to allow gathering of data that can be used to refute
or support this candidate explanation
-
Generate clear questions from pilot
studies (Exploration of the study system conducted before the main body of data
collection in order to refine research aims and data collection techniques).
-
Try to form hypotheses that might
answer the question
-
Make predictions of relationships
that you would expect to observe if your hypotheses were true.
-
Decide what data you need to collect
to either confirm or refute your predictions
-
Now you can design your experiment
and collect your data.
Example (go to the zoo to do a pilot study (or
observation), watching chimpanzees. You notice that they have strong daily
variation in activity patterns)
1)
Research question
Why
does chimp activity vary during the day?
2)
Hypothesis
Chimp
activity pattern is affected by feeding regime
3)
Prediction
The
fraction of time that a chimp spends moving around will be higher in the hour
around feeding time than at other times of day
·
Two kinds of hypotheses are made.
Null hypothesis Ho and alternative hypothesis Ha
·
The null hypothesis can be thought of
as the hypothesis that nothing is going on “Chimp activity is not affected by
feeding regime.” It says that any apparent relationship between chimp activity
and feeding regime is just due to chance. It leads to the prediction that
“There is no difference in the fraction of time that the chimp spends moving in
the hour around feeding time compared to the rest of the day.”
·
Alternative hypothesis states that
the pattern we see is due to something biologically interesting. The original
hypothesis mentioned above is an example of an alternative hypothesis.
·
Statistical tests work by testing the
null hypothesis. If our data allows us to reject the null hypothesis then this give support for our alternative hypothesis.
·
Philosophically science works
conservatively. We assume nothing interesting is happening unless we have
defined evidence to the contrary.
Example (You notice that the snails at a
shore often seem to occur in groups)
1)
Question
Why
do snails group?
It
might be because they want to seek shelter from the mechanical stresses imposed
by breaking waves
2)
Hypothesis
Snails
group for shelter from wave action
The
reason might be clumping in areas of high food
2)
Hypothesis (2)
Snails
group for feeding
If
the first hypothesis is true, we might predict that
3)
Prediction
Snails
are more likely to be found in groups in areas sheltered from wave action
Whereas
if the second is true our prediction would be
3)
Prediction
Snails
are more likely to be found in groups in areas of higher food density
Now
design a study to test both of these predictions
(Both
hypothesis may be true – other hypothesis also may be
true)
Four
possible combinations
Possibility
1: Neither hypothesis is true
Possibility
2: First true and the second false
Possibility
3: Second is true and the first false
Possibility
4: Both are true
Good
predictions will follow logically from the hypothesis we wish to test
Good
predictions will lead to obvious experiments that allow the predictions to be
tested. Untestable predictions are of little use
Question
® Hypothesis ® Prediction
2-
Producing the strongest evidence with
which to test a hypothesis
Consider
the hypothesis:
“Students
enjoy the course in Mycology more than the course in experimental design”
One
way to test this hypothesis would be to look at exam results
Thus
our prediction is:
“Students
will get higher grades in Mycology exam than in experimental design exam”
*
This is a weak test of the hypothesis (even if the grades where higher, other
reasons might have resulted in this)
*
By the use of exam scores to infer something about student enjoyment, we are
using an indirect measure. Indirect measures generally produce unclear
results.
Should use direct measure such as interview the
students.
*
Before doing an experiment you should always ask yourself for every possible
outcome, how such a set of results could be interpreted in terms of the
hypothesis being tested.
*
Avoid experiments that can produce outcomes that you cannot interpret.
*
Do not do experiments where a useful outcome of the work hangs on getting a
specific result. Negative results where you find no effects for example are
valid and useful.
Example,
you want to test the hypothesis “Drug use has an effect on driving.” If you
find no effect, then it is also interesting result. Positive results are
interesting.
Example,
you want to test the hypothesis “Preference for meat or chicken eating is
linked to driving ability.” If you find relationship then it is interesting,
but if you find no relationship, then this is a less interesting result.
3-
Satisfying skeptics
Make
sure that your experiment is designed so that the conclusions that you draw are
as strong as possible.
Example:
A group of observations could be explained by two contradicting mechanisms (A)
and (B). Don’t be satisfied by an experiment that will let you conclude that
“The results of this experiment strongly suggest that mechanism A is operating.
Although it is true that the results are as equally as compatible with
mechanism B, we feel that A is much more likely.”
The design of the experiment should allow us
to conclude: “These results are entirely consistent with the hypothesis that
mechanism A is in operation. Moreover, they are totally inconsistent with
mechanism B. Therefore, we conclude that mechanism A is the one acting.”
You should thing of the people evaluating your work as a highly intelligent but skeptical people. If there is a weakness in your argument, then they will find it. However, they can be made to believe you, but only when they have no reasonable alternative. They will not give you the benefit of any reasonable doubt.
4-
Experimental manipulation versus natural variation
A manipulative
study is where the investigator actually does something to the study system
and then measures the effects these manipulations have on the things that they
are interested in.
A correlational study makes use of naturally
occurring variation rather than artificially creating variation to look for the
effect of one factor on another.
Example
Hypothesis “Long tail streamers seen in many species of birds have evolved to make males more attractive to females.”
Prediction
“males with long tails should get more matings than
males with short tails”
To test this
prediction: First a correlational study
“Catch a
number of bird males. Measure the length of their tail streamers and then let
them go again. Watch the birds for the rest of the season and see how many
females mate with each male. If after doing the appropriate statistical
analysis we found that the males with longer tails obtained more matings, this would support our hypothesis.”
Second, a
manipulative study “Catch the male birds, but instead of just measuring them
and letting them go, we would manipulate their tail length. For example, we
divide them into three groups, the first with no change in length (control),
the second we shorten their tails and the third we increase their tail length.
If after doing the appropriate statistical analysis we found that the males
with longer tails obtained more matings, this would
support our hypothesis.”
In this
example, both studies seem to be effective.
* Arguments
for doing a correlational study
-Easier to
do.
-If we are
dealing with organisms that are likely to be stresses or damaged by handling,
or samples that can be contaminated, this is the way to do the study.
-We can be
sure that we are dealing with biologically relevant variation, because we
haven’t altered it. For example, if we extend the tail streamers by more than
the natural length range, it is doubtful that our experiment can tell us
anything biologically relevant about the system.
* Arguments
for doing a manipulative experiment
- Correlational studies may suffer from two problems: third
variables (extrinsic, confounding factors) and reverse causation.
Third
variable: when we sometimes mistakenly deduce a link between factor A and
factor B when there is no direct link between them. This can occur if another
factor C independently affects both A and B. C is the third variable.
Reverse
causation: When we mistakenly assume that factor A
influences factor B, when in fact it is change in B that drives changes in A.
5- Work in
the field or in the laboratory
·
Will your study organism be
comfortable in the laboratory? If yes then you do not have a problem.
Laboratory studies are easier to control allowing you to focus on the variable
of interest without lots of variations.
·
Remove between-individual variations
by controlling factors such as feeding that affect the organism in the natural
setting.
·
Observation will generally be much
easier in the laboratory.
·
It will nearly always be easier to
make detailed measurements on most organisms in the laboratory.
The controlled nature of the lab
environment is also its major drawback. Individuals will not experience many of
the stresses and strains of everyday life in the field and that can make it
difficult to extrapolate from lab results. “No effect in the lab is different
from no effect in the field.”
“Sometimes
our research question means that a laboratory study is impractical.”
6-There is no
perfect design to all experiments
This is why
biologists rather than statisticians must design biological experiments
Good
experimental design is all about maximizing the amount of information that we
can get, given the resources that we have available.
If what we can conclude has been limited not by how the natural world is but by our poor design, then we have wasted out time and someone’s money too.
More importantly, if our experiment has involved animals these will have suffered for nothing.
·
You cannot design a good experiment
unless you have a clear scientific hypothesis that you can test
·
Make sure that your experiment allows
you to give the clearest and strongest evidence for or against the hypothesis
·
Make sure that you can interpret all
possible outcomes of your experiment
·
Use indirect measures with care
·
Design experiments that give definite
answers that would convince even those not predisposed to believe that result
·
Correlational studies have
the attraction of simplicity but suffer from problems involving third variable
effects and reverse causation
·
Manipulative studies avoid the
problems of correlational studies but can be more
complex and sometimes impossible or unethical
·
The decision of whether to do animal
experiments in the field or the laboratory is determined by whether the test
organism can reasonable be kept in the lab, whether the measurements that need
to be taken can be recorded in the field and how reasonably laboratory results
can be extrapolated to the natural world. All these vary from experiment to
experiment
·
There is no perfect design for all
experiments, but a little care can produce a good one instead of a bad one.