Before the experiment – Recognizing a
pattern
What might cause the pattern? - Hypotheses
Setting up a test – Treatments
Completing the design – Predictions
How could experimentation as a
topic be featured in exams?
In science we seek to explain the phenomena we observe by reference to the action of specific causal factors, that is to say we try to establish a “cause-and-effect” relationship between environmental variables and the patterns of nature. Simply observing that events are correlated is not the same as demonstrating that one causes the other; experimentation is the only way to establish a causal connection. The rules for doing this are very well worked out, based on centuries of practical experience. Deep down, one experiment is much like another, though superficially the techniques employed by different scientific disciplines may look quite varied; this document is intended to lay out the parts of an experiment so you can understand why they are necessary and learn how to apply them yourself.
Experiments
don’t exist in a vacuum, they must be based on a real or at least a theoretical
set of observations. How do we know that an observation is worthy of
experimental testing? Single unique observations cannot easily be subjected to
analysis, so we are usually looking for widely-observed patterns involving
multiple individual organisms, or multiple occasions of observation. An
observation pattern may be simple (“Soft fleshy plants shrivel up in the
desert”) or more complicated (“Fish preferentially form schools with others of
the same size and appearance, unless the water is murky”). Some patterns may be
explained readily by invoking only one or two factors, while others may require
many contributory factors to be incorporated before an explanation can
withstand scrutiny.
When you are asked to design experiments in this course, the patterns will be
constructed to lead you to fairly clear explanatory factors, and indeed you may
be told explicitly which factors to use. When you go beyond Biology 121 to
consider experiments in the real world, you will very quickly realize how
difficult it can be even to frame a possible explanation, much less determine
whether it is the correct one! It is surprisingly difficult even to decide
whether a pattern exists – and you can ask about the causation of patterns only
if you know they are present. [Always remember that nature doesn’t come “pre-cut”
into neat, easily-recognized patterns, we are always obliged to extract
patterns from a noisy and confusing background.] For simplicity, let’s begin by
taking a pattern already mentioned above, “Soft fleshy plants shrivel up in the
desert”, and work on that.
For any pattern, it will be
necessary to draw on knowledge outside it to explain it. In the case of “Soft fleshy plants shrivel up in the desert”,
this involves using knowledge of desert climate, soil conditions, plant
physiology, and perhaps other areas also. Somewhere within those sets of ideas
there will be potential explanatory principles worth testing for. We call these
provisional or possible explanations hypotheses. A good hypothesis will derive
from the specific application of a generally accepted (or at least plausible)
body of knowledge.
So a fleshy
plant shrivels: is this due mainly to the climate?; to the soil?; to its own
physiology or structure?; to some combination of these? What do we mean by a
“fleshy plant”, anyway? Does one part of the plant shrivel first, the rest
following? Does the shrivelling of one part cause the rest to shrivel, or are
all parts made to shrivel independently by outside forces? Are some individual
plants shrivelled by one force but different plants by other forces? Do plants
of different ages/sizes fare equally poorly against shrivelling? Are all plant
species affected by the same, or by different, forces in their shrivelling?
Rather than just rephrasing the original observation, we shall need to narrow
it down and make it much more specific before it can become a hypothesis. We
could never test for all the factors suggested in this paragraph in a single
experiment, anyway.
“Fleshy (soft,
moist-tissued) plants shrivel in the desert because there is insufficient water
available.” This would be a workable hypothesis, and so would this: “Fleshy
(soft, moist-tissued) plants shrivel in the desert because they lack a waxy
cuticle layer on their structures, which would retain moisture.” Several other
possible accounts could be put forward… why not try making some for
practice?
The essence of science is testability: unless a suggested explanation can be “put to the test” and potentially disproven, it is of little value. You have to ensure that any hypothesis you generate (or, in the course, any hypothesis you are given) satisfies this condition. Both of the suggested hypotheses at the end of the previous section can be evaluated with a test, so both are acceptable. Consider another: “Persons who dislike fleshy plants secretly sneak into the desert and blow hot air onto plants using huge fans, causing them to shrivel and leaving no indications of this cause” – this would not be very plausible to account for the fate of plants in deserts generally! Why would anyone go to the trouble of hurting plants in this manner? What purpose would be served? Why would it be secret? Most importantly, if great pains were taken to hide the cause, it would be inherently untestable.
Untestable ideas may sometimes be correct, and even if they aren’t correct they may still be widely accepted. For example, if one claimed: “Fleshy plants were not intended by the Creator to live in deserts, so He makes them shrivel there”, even if it were true, it would not be a scientific hypothesis. No serious scientist would attempt to demonstrate what a Creator might intend, much less to show what He did not intend! The main problem with this “pseudohypothesis” is that even if it were somehow shown to be true, this would not add to our understanding of anything in nature. It would not lead to predictions about other organisms, or other conditions, and would not allow us to say what might happen if conditions changed. Claims of the “Creator intended” type were frequently offered in biology until 150 years ago, and at times in other sciences also, but mainly in cases where so little was understood that it was simply not possible to frame testable hypotheses. We will never know everything, of course, but as we learn more we can be more confident of having good hypotheses, and of using them to improve our understanding further.
So we have a hypothesis which we think is acceptable, say: “Fleshy (soft, moist-tissued) plants shrivel in the desert because there is insufficient water available.” In order to test it, we need to identify the variable(s) whose value may determine the effect, and the variable(s) which must be measured to determine whether the effect is occurring when we change conditions.
The variable(s) whose value may determine the effect, i.e. the possibly-causal variable(s), is(are) the one(s) we must manipulate in the experiment. If you suppose that factor X is causing an effect, you would manipulate to remove factor X; if your supposition is correct, the effect should disappear in the absence of factor X. On the other hand, if you think an effect occurs because factor X is not in operation, the obvious manipulation is to add factor X, and if you are correct the effect should disappear. We call the variable(s) to manipulate the independent variable(s), because in the manipulation we set the value of such variable(s) independently of what it(they) would naturally be. A simple test will have one independent variable, while more complex tests can have several. In this test, since we are interested in one variable – insufficient water supply – the appropriate manipulation would be to add water.
[In general, you will always manipulate the independent variable “against” the effect you suppose it has – at least in any hypothesis which states the direction of an effect, this will be so. If the hypothesis made no directional claim (e.g. : “Water affects the growth of a plant”), then it would be necessary to manipulate water both by increasing and by decreasing.]
The variable which we will measure to determine if the changed value of the independent variable did affect the pattern may be referred to as the result-variable, or the output-variable, but technically the dependent variable, since it adopts a value depending upon the value at which we set the independent variable(s). Here it must be some measure of the degree of “shrivelled-ness”, since that is the symptom of water-shortage named in the hypothesis. (How would you go about measuring “shrivelled-ness”…?)
With a single independent variable, there is only one “dimension” along which we need vary conditions. At one extreme will be the “normal” state, or the unmanipulated state, of the desert: low water availability. We refer to this condition as the unmanipulated or control treatment. Then we manipulate by adding water. Maybe there will just be one more treatment: “Wet”. More likely we would add increments of water, so we could see if just a little extra could help the shrivelling, or if shrivelling improved only when massive amounts were added. These will be manipulated treatments, or more simply manipulations.
We need a control treatment in part because it serves as a baseline against which to compare the effect our manipulations, and in part because it allows us to keep track of extraneous variables outside our experimental design. For instance, if we see surprisingly strong growth of fleshy plants in the control, we might suspect that it was an unusually wet year in the desert; if we didn’t have a control, we could not know this.
In a test with two or more factors being varied, it is necessary to carry out treatments involving all possible combinations of the varied factors, to see if they are of equal or different weight, whether they interact by addition or cancel each other out. This can lead to a large number of combinations, and thus a major logistical challenge for the experimenter, but sometimes it really is necessary.
Having the variables characterized and knowing how they are to be changed is just a starting point. How are we to do it? Students in Biology 121 are not expected to be plant physiologists or experts in desert ecology, but there are some key aspects of carrying out the experimental test – which we can group together as methodology – which must be considered at least briefly.
What plant are we to use? Shall we assume that an easy-to-grow species, one commonly used in experiments, is adequate, for instance lettuce? There is no doubt an imposing literature on growing lettuce, since it is an important agricultural crop, and we can safely assume that we could turn to that literature for guidance on how much water to provide, and so forth. We can also assume that knowing the size of a lettuce plant, we can make test-pots or growth-chambers of such a size to include enough plants for a meaningful test. There is no need for the student to know what values to use, merely to make reasonable assumptions about them. Every experiment will require assumptions.
It is also vital to recognize the necessity of doing the experiment on multiple samples. Imagine that you have one control pot and one “Wet” pot set up, but while you are away from the lab a technician (not knowing about your experiment) mistakes them for pots that need to be cleaned out. Poof! – no more experiment. If you had set up several pots of each type – what we refer to as replicates – it’s less likely such a fate would befall all of them. We replicate also because there may be unrecognized variation among plants (varying genotypes), or among plant-pots (varying drainage), or among bags of soil (varying composition); if we had only single cases in each treatment-type, we might be misled by the result. Replication is especially important in field experiments, because there is no certainty that all spots in the field are similar, and you therefore need to spread all treatments through all areas to balance out biases. How many replicates should there be for each treatment? For our purposes, all that’s necessary is to say: “I would do replicates of each treatment”.
Then there is the question of how the work should be done: will watering be done at just one time of day, or continuously? On the soil surface, or from beneath? By misting, drip-irrigation, or watering-can, or poured from a bucket? How will “shrivelling” be evaluated? These and other method issues cannot be known to you in detail, and may end up requiring further assumptions.
An experimental design is incomplete without one other feature: you must say what pattern you expect to observe in your results if the hypothesis is true. (Normally, if the hypothesis is not true and the factor you manipulate has no effect, the pattern will be simply “Manipulations look similar to controls.”)
Making these predictions is simply an exercise in extrapolating the claims of your hypothesis to the specific situation you set up in the experiment. “If low water-availability causes shrivelling, there should be measurably less shrivelling in the “water-added” plants than in control plants.” (You would of course state this including whatever measure of shrivelling you decided was important.) In a prediction you do not make claims about what you think the effect of this factor “should be” relative to others! You would never say: “I don’t think adding water will help.” Maybe adding water isn’t the key thing to do, but in this particular experiment the only thing about which you can make a prediction is the effect of adding water, as suggested in the hypothesis.
[If an exam question invited you to express an opinion about whether water-adding, or the provision of shade, or changing soil quality, was more likely to reduce shrivelling, then that would be another matter, and indeed not really an experimental question.]
Of course you might easily expect to be asked to do a complete experimental design, either for a given hypothesis, or for the testing of a hypothesis which you are asked to generate from data provided. But this is definitely not the only way of testing your understanding of experimental principles.
You could be given a complete design, and asked to critique it, stating at each step what you liked and didn’t like, and why.
You could be given a set of results and a conclusion, and asked to provide the hypothesis and experimental approach which would have generated such results and conclusion.
It’s also possible that you could be asked to give several alternative hypotheses to account for a given observed pattern, and then not design an experiment at all, or design a brief test of one of your hypotheses.
Finally, you could be asked to explain an experimental-design concept (e.g.: “Why is replication especially important in field experiments?”)
You can expect that 10-20% of any exam will be based either directly or indirectly on experimentation. Why so much? Because, as suggested at the beginning of this document, science really is experimentation. Whether you are a Science student or not, whether you go on to become a professional scientist or not, the experimental mindset is a very useful thing to cultivate. Careful consideration of data and cautious interpretation of observations are beneficial habits in any walk of life!