6.4 Reflections on the Exhibition
Recapitulating our definition, an experiment changes the state of a system to produce an organized response, in purpose to answer a specific question. We added that the response of interest must arise from the treatment imposed by the experimenter. If these criteria are not fulfilled it is an observational study, where measurements are made without attempting to influence the response. One reason why observational studies are sometimes mistaken for experiments is that building and operating measurement devices often require a great deal of practical manipulation, though measurements in themselves are only a way of observation. Looking at the experiments in our gallery we see that they all change the state of a system in order to study the resulting response. For example, Harvey obstructed blood flows in various ways to observe how the vessels were emptied or filled, and Feynman put an O-ring in cold water to see if its resiliency was affected.
These experiments are captivating in that they use rather simple means to interact with things that often cannot be seen or touched: the minute capillaries that transfer blood from the arteries to the veins, the DNA of viruses, subatomic particles like electrons, not to mention female birds’ taste in males. Mendel even studied the effects of something that he did not know existed – genes. It may not always be possible to replace an observational study with an experimental one, but it is equally true that many experiments reveal things that would simply be impossible to find out in observational studies. The only example from the gallery where I can see a possibility for this is the widowbird study in Example 6.7 – but just think of the immense amounts of observational data that would have to be collected to support its conclusion.
We have seen that even experimenters must take precautions against background variables that may muddle their analyses. One common way to do this is to use a control group, as in the widowbird experiment. Andersson realized that a change in mating success could be due to several factors, such as changes in the birds’ behavior, the females’ ability to recognize the males as birds of their own species, and so on. He included control birds that were not manipulated in order to rule out such influences. If a control group is used, the study is called a “controlled experiment”. A similar type of precaution was used in the engine experiment in Example 6.6, where it was known that the UHC emissions were affected by the engine load – a variable that was not directly connected to the research question. For this reason, a reference curve of UHC was acquired as function of engine load. The results of the experimental treatments were then compared to this curve to judge if there was an effect or not.
When looking closer at the examples, we actually find that many of them have no control or reference sets. Does this mean that they are poor experiments? For example, in the Hershey–Chase blender experiment (Example 6.4), radioactively labeled phage were manipulated in various ways to see where the radioactivity went during infection. A control group of unmanipulated phage would only result in the trivial statement that the labeled phage were radioactive. In other words, controls would be superfluous since a control group would not be able to show an effect. The same is true of Millikan's experiment in Example 6.5, where the studied effect could not have been produced without manipulating the electric field. There was no need for controls in this experiment but, as we have seen, other types of precaution were needed to interpret the results correctly. To mention one example, he eventually used oil drops instead of water to eliminate the problem with evaporation. Not all experiments need a control group but all experimenters need to be vigilant against the effects of background factors.
Finally, in some of the experiments there are random elements that may somewhat dim the effects. In such cases it is important to use sufficient sample sizes to discern the effects and to use appropriate mathematical tools to make the effects visible. We have seen examples of this both in Mendel's experiments and in the widowbird study. It is especially common in biology that the analysis of experiments relies on appropriate use of statistics, but statistics knowledge is important in all data analysis. For this reason, we are going to spend the next few chapters looking at statistical techniques.
Before that, we will round this chapter off with a number of exercises to reflect further on our tour of the gallery: