Experimental Design in Aquatic Ecology - Freshwater Ecology: Concepts and Environmental Applications - page 481

the assumptions used to construct it. For this type of analysis, input para-
meters are varied and outputs are compared under different model scenar-
ios. If much uncertainty exists over the value of a parameter that has a
great influence over the behavior of a model, the predictions of the model
are suspect. The scientist may then design the research program to more
thoroughly characterize the uncertain parameter. Another method for val-
idating a model is called hind-casting . In this method, the model is used to
predict a system response that is already known. This is possible in systems
in which sufficient data are available from the past but may be difficult
where conditions are changing over long periods of time (e.g., under global
change).
MANIPULATIVE EXPERIMENTS
To provide more formal proof of a hypothesis, manipulative experiments
are often necessary. To test a hypothesis experimentally, all other factors must
be held constant, the factor of interest must be varied, and the effect must be
noted. If a change is related to variation in several factors, then the hypoth-
esis is not proved because the change may have occurred without changing
the factor of interest (correlation, not causation). Thus, a control, in which
no factors are varied, is necessary for comparison within the treatment. In the
example given in the previous section, the polluter that released the organic
chemical into the stream could argue that the chemical was not necessarily
responsible for the fish and invertebrate deaths and that they would have died
anyway. If the fish and invertebrates in nearby, similar streams did not die,
such an argument would hold less water.
Deciding which is the control and which is the experimental treatment
can be difficult because this designation depends on how the hypothesis is
stated. For example, if we form the hypothesis that ultraviolet (UV) light
lowers fertility of frog eggs, then no ultraviolet light is the control, and ul-
traviolet light exposure is the experimental treatment. Conversely, if the hy-
pothesis is that lack of UV increases fertility of frog eggs, then UV light can
be the control and removal of UV the experimental treatment. Neither way
is right or wrong, but often it makes ecological sense to frame a hypothesis
in one particular way. In the previous example, ambient UV exposure from
sunlight might be a control, and increased UV equal to that expected to re-
sult from decreased ozone in the upper atmosphere could be the treatment.
The next problem that arises is the certainty of any particular result.
Say that we treated one frog egg with increased UV and it did not hatch,
and the control egg treatment with ambient UV did hatch. Have we proven
the hypothesis that increased UV will lower frog egg fertility? Perhaps the
UV-treated egg was not fertilized properly in the first place. Replicating or
repeating the experiment could increase the certainty of the result. If we
did the experiment five more times and always had the same result, this
would increase the certainty that we had proved the hypothesis. This repli-
cation is another key point in experimental design.
Statistics are involved when an experiment is replicated. Statistics al-
low comparisons between experimental and control treatments and formal
expression of the degree of certainty that the hypothesis is true (Tables A.1
and A.2). Many tests can be applied to different experimental designs.