C Kinds of Science: Objectives and Methods

You can distinguish among kinds of science not only by their subject matter or methods, but also by their ultimate objectives. I mentioned (Chapter 2.G.) that many philosophers have difficulty in understanding why scientists hold Popper's and John Platt's programs in such high regard, given that scientists generally pay almost no attention to any philosophical teachings.

4. C.1 Basic Science Versus Applied Science

Many discussions of modern science either take it for granted that we're talking about basic research science, or they blur—even reject (Chapter 10.A)—the dis­tinction between basic and applied science. The express purpose of basic science is to discover Truths about nature, solely for the sake of understanding it. Naturally, basic scientists and society, especially research funders, hope that basic science discoveries will translate into solutions for societally significant problems, but concrete payoffs are not its immediate goal.

Applied science is directed toward concrete payoffs; it seeks to solve specific problems, often in the service of technology development paid for by commer­cial enterprises or governments. Despite the fact that applied science typically relies on or extends the findings of basic science, it is not primarily dedicated to advancing knowledge about the world per se.

Of course, it may be difficult or impossible to say precisely when basic turns into applied science. Scientists working on practical problems do make ground­breaking basic discoveries, as Arno Penzias and Robert Wilson did when their efforts to modify a radio telescope accidentally led them to discover Cosmic Microwave Background radiation²⁶ thereby gaining evidence for the Big Bang (and eventually winning a Nobel Prize to boot). Conversely, we support basic cancer research because we hope that it will lead to a cure for cancer. To a large extent, though, it makes sense to consider basic and applied science to be dif­ferent kinds of science.

4. C.2 Confirmatory Versus Exploratory

Unfortunately, the ill-defined “confirmatory-exploratory” terminology²⁷ has come into vogue as another way of distinguishing among different kinds of sci­ence. The underlying impulse is not bad: it can be useful to differentiate between large-scale, rigorous studies that are intended to produce definitive evidence for or against an important prediction, and small-scale studies, where you’re just trying to get a foothold in a new area of research.

You might do a “confirmatory” study ifyou had a lot of data and wanted to test a prediction of a well-developed and -corroborated hypothesis. An ideal confirm­atory study would have large sample sizes, be tightly controlled, double-blinded, preregistered, and have impeccable statistical design. So far so good. However, some confirmatory studies are also called “hypothesis-testing” studies, while others do not test an explicit hypotheses. A drug company might do a confirma­tory study to test a prediction, such as, “drug X will reduce the frequency of heart attacks,” if it hopes to take X to the marketplace and treat patients, not because it cares about the hypothesis underlying X’s cellular mechanism of action per se. Its main objective is to establish whether or not X works so it can either continue to its program on X or cut its losses and move on to a new project. Calling them “hypothesis-testing” creates a misunderstanding that is compounded by the am­biguity of the term “confirm” (Chapter 1L.). A confirmatory study could quite unexpectedly reveal that drug X is worthless or even harmful.

A neutral and descriptive name (e.g., “decision-making” studies) would call attention to the defining characteristics of confirmatory studies, which is to de­cide something once and for all.

“Exploratory” on the other hand, has been applied to preliminary pilot­level studies that are sometimes also called “hypothesis-generating” or even “pre-hypothesis” studies, and, in this context, “exploratory” seems apropos. Unfortunately, “exploratory” is sometimes applied to small-s cale rigorously conducted hypothesis-testing studies, which is not at all helpful.

You can probably already guess that the biggest problem with the “confirmatory-exploratory” terminology is that it suggests a dichotomy that en­tirely omits the vast middle ground between primitive exploratory studies and well-advanced, dedicated decision-making investigations.

4. C.3 Scale: Big Science/Small Science, Big Data/Little Data

If you search for information about the status of the hypothesis in today's bio­medical sciences, you'll eventually encounter pundits who proclaim that science no longer “needs” the hypothesis and that the successes of Discovery Science, Systems Biology, or Big Data prove it. In the next sections, I'll review definitions and point out how they fit into the discussion about hypothesis-based and non­hypothesis-based science.

Big Science differs from Small Science²⁸ by its scale: Big Science projects can have industrial-sized magnitude, multisite collaboration, thousands of researchers, and budgets and administrative structures to match. Big Science takes advantage of automated analytic devices to collect massive amounts of data and relies heavily on computer-based algorithms for analysis. It strives for “throughput,” the ability to obtain and process information quickly and ef­ficiently. But, beyond its scale, Big Science is characterized by its large fund of accepted general principles and methods, as well as by the “invisible colleges of researchers”²⁹ who interact mainly with each other. Big Science is mature; there is wide agreement within a Big Science community about the major scientific problems that it must solve.

Small Science³⁰ labs tend to be self-contained and have bare-bones admin­istrative support and budgets. Small Science is typified by its multiplicity of re­search topics and experimental methods.

If you were to change perspectives and look only at the sizes of the datasets (a “dataset” is a collection of data about a specific topic) that scientists deal with, rather than the size of the laboratories or extent of their collaborations, you might see science as divided into Big Data and Little Data investigations. If you were to plot the size of the dataset that a given laboratory works with against the numbers of laboratories working with that dataset, you'd get a relationship that resembles a falling exponential function (Figure 4.1). There are only a few labs working with huge datasets (on the left of the x-axis) and increasingly many labs working with smaller and smaller ones as you move to the right of the axis. This plot, said to illustrate the “long tail”³¹ of science, gives a global impression of how the entire scientific enterprise sorts itself out on these dimensions. How does the information in the long tail line up with the other classifications? Big Data and

Figure 4.1 The “long-tail” of science. First named by Chris Anderson (quoted by C. L. Borgman, see Note 31) the graph shows that as the number of laboratories grows, the size of the datasets that each laboratory uses and generates decreases. Only a relatively few large labs are dealing with the extremely large datasets that characterize Big Data. Small Science is carried out in the long tail of the distribution and generates the large majority of scientific findings today.

Little Data are only “awkwardly analogous”³² to Big Science and Small Science. While all Big Science projects make use of Big Data, some Small Science labs do as well.

“Little Data” is a logically necessary though rarely used term for the raw mate­rial and final output of Small Science. Little Data are generated by the materials, methods of data collection, and analyses that most Small Science projects em­ploy; small datasets with relatively few subjects and variables, and people, rather than machines, to do much of the work. Small Science using Little Data methods is not easily “scalable,” meaning that you can't readily increase the rate of progress from the start of data collection to the generation of conclusions. For instance, for the past 50 years or so the premier way to measure the electrical properties of neurons in the brain has been for a single investigator to insert a fine glass probe with a submicroscopic tip into a single neuron in an in vitro slice of brain tissue. The method yields extremely high-resolution information; it is also a pit­ifully slow method, and so far no one has come up with a better one. Single-cell recording like this represents a typical Small Science/Little Data scalability bot­tleneck. Despite such drawbacks, Little Data projects make up the long tail of the distribution and represent the principal sources of research information in many fields, including biology and biomedical science, neuroscience, and psychology. And, even today, Little Data and Small Science generate the great majority, per­haps 85%,³³ of all scientific knowledge.

4.