Introduction to Experiments

For readings on statistics and experimental design, see Mike’s Biostatistics Book, Chapter 2.4 and Chapter 5.

Background

Science is a diverse discipline, from the broad fields of biology, chemistry, and physics, with each having multiple sub-disciplines. For biology, a possible, though not exhaustive listing may include (Table 1).

Table 1. List of disciplines in biology research.

modified from “Careers in Biology” list UNC-Pembrook. Besides “biology,” what do these sub-disciplines have in common?

The “scientific method”

While the tools and traditions in each field may not overlap, the need for proper data collection and experimental design unite all of these disciplines. We call this the Scientific Method, which, in a standardized version that continues to appear in some texts would seem to flow logically from observations of some phenomenon in nature, asking questions about how the phenomenon might occur, which leads to creation of testable hypotheses (questions that lead to predictions), and thus to a formal process called an experiment designed to test the hypothesis. Thus, the hypothesis is then evaluated relative to the collected data, and, at last, the scientist evaluates whether the hypothesis is falsified, or if not, remains as a potential explanation for the observations. See Chapter 2.5 in Mike’s Biostatistics Book for more on the interplay between The Scientific Method and data analysis.

This form of the scientific method is better termed the hypothetico-deductive approach. To some, a hypothesis is only useful if we can show it to be, in practice, “falsifiable,” a term we owe to Popper, a philosopher. Platt (1964) wrote a very powerful and influential paper on the need to design experiments such that “strong inference” could be applied to distinguish between alternative explanations or models (see also Fudge 2014). While these discussions are helpful and the epistemology of science can help you be a better scientist, I don’t think anyone would hold this view as the only way scientists work, but it is tempting to the view the process that because of the accumulation of facts and inductive logic, eventually one finds the truth. Many authors have written on this subject, but one of my favorites comes from Richard Lewontin (1974):

But this textbook myth has no congruence with reality. Long before there is any direct evidence, scientific workers have brought to the issue deep-seated prejudices; the more important the issue and the more ambiguous the evidence, the more important are the prejudices, and the greater the likelihood that two diametrically opposed and irreconcilable schools will appear. Even when seemingly incontrovertible evidence appears to decide the matter, the conflict is not necessarily resolved, for a slight redefinition of the issues results in a continuation of the struggle (p. 29).

Lewontin was writing about the competing hypotheses on genetic variation in populations: The classical view, which held that at each locus there would be a “wild type,” and that most individuals would have that allele. Thus, under the classical view, genetic variation within populations would be small, but between populations there would be greater diversity. In contrast is the balanced view, which in its extreme form held that individuals would be heterozygous  at virtually all genetic loci. The classical view comes from the writings of Mueller (1950); the balanced view comes from Dobzhansky (1955). Under the balanced view, the implication for genetic variation is opposite to that of the classical view: within populations individuals differ, so that no single allele could be considered the wild type, and variation between populations would be small.

Lewontin also wrote that even the definitive experiments about genetic variation in populations only solves the problem in a limited way – we may not have been asking the right questions about the problem. Lewontin, too, is exaggerating here; surely not all scientific questions are bogged down by competing schools of thought, loaded with long-held prejudices brought about by working with different organisms or systems.

Good experimentation, accompanied by improvements in technology, can go a long way towards finding something we recognize as the truth about the world.

Research design types

A common curriculum in biology at the university level includes extensive emphasis on experiments, from demonstrations run by the instructor or advanced student to hands-on approaches. Students may engage in “Canned experiments,” aka “Verification experiments,” as in the outcomes are known, or they may be “open-ended” in which case, the outcomes are unknown, eg, CUREs or Course-based Undergraduate Research Experiences. Both have their place. The advantages of canned experiments is that it allows students to focus on methods and techniques, which yield verified results. Canned experiments also allow development of critical thinking skills — identifying evidence of support or refute hypothesis. These exercises also have the advantage of being contained; CUREs necessarily contain significant unknowns, from methods to techniques to data quality that must be addressed before analysis of results and tests of hypotheses can be engaged.

At Chaminade, biology majors are required to take a biostatistics course. Biostatistics is intended to teach proper design of experiments, in addition to application of statistical methods to data analysis. Since your introduction to science laboratory courses at Chaminade University, you have had “hands-on” experience with experiments, but perhaps not as much attention has been paid to experimental principles and the nature of research. At the least, you should now be able to distinguish observational studies from experimental studies. Observational studies are those in which nature selects the treatments and we measure and record the outcomes to identify patterns. For example, during an outbreak of flu virus (influenza) in a population, we generally see more adverse effects (mortality) in the very young and the very old. In 2014, influenza caused 4.7 deaths for every 100,000 infants under the age age of one year, but less than 3 deaths per 100,000 other ages groups up to age 45-54 years. While clearly not an experiment, these patterns are used to guide decisions about who should receive immunizations against the flu virus (e.g., 2016-17 CDC GuidelinesLinks to an external site.).

Models, Systems Biology, empirical, statistical

Hypotheses and tests

Role of statistics

Experiments

• Platt (1964)

Data, observations, variables

• Data types
• Observations and measurement
• independent, dependent variables
• treatments, controls (negative, positive)

Probability, process vs observation uncertainty

Distributions for statistics

References

Dobzhansky, T. (1955). A review of some fundamental concepts and problems of population genetics. In Cold Spring Harbor Symposia on Quantitative Biology (Vol. 20, pp. 1-15). New York: Cold Spring Harbor Laboratory Press.

Fudge, D. S. (2014). Fifty years of J. R. Platt’s strong inference. Journal of Experimental Biology 217:1202-1204

Lewontin, R. C. (1974). The genetic basis of evolutionary change. New York: Columbia University Press (XXV in the Columbia Biological Series).

Mueller, H. J. (1950). Our load of mutations. Am J Hum Genet. 2(2): 111–176.

Platt, J. R. (1964). Strong inference. Science 146(3642):347-353 (link to pdf)

Quinn, G.P., Keough, M.J. (2002). Experimental design and data analysis for biologists. New York: Cambridge University Press.