How Science Works: Testing Ideas, Designing Experiments, and Catching Our Own Biases
by Professor Stephen T. Abedon Steve Abedon, Ph.D., has taught well over 100 individual classes and ~10 individual courses (including introductory biology, microbiology, physiology, biochemistry, etc.) over his 30+ years as a college professor. He has over 150 publications, including multiple monographs, and is ranked in the top 1 or 2 percent among researchers of all disciplines worldwide. His research focuses on bacteriophage biology and phage therapy. (abedon.1@osu.edu)
phage.org | phage-therapy.org | biologyaspoetry.org | abedon.phage.org | google scholar | publications
Jump to: 📋 Overview | 👀 Observation | 🦢 Testable? | 🎯 Hypothesis to Prediction | ⚗️ Design an Experiment | 🕵️ Spot the Flaw | 📚 Theory, Law, Fact | 🧬 How and Why | 🗂️ Kinds of Research | 🕰️ A Short History | 🎥 Videos | 📖 Terms | 🧮 More Calculators
biologyaspoetry.org/laboratories/science/ · Abedon’s Books
How can I improve this page? contact: science@biologyaspoetry.org
Science is a way of discovery, which happens to represent the most efficient manner we know of to come to better understand the workings of the physical universe. The word comes from the Latin scire, to know, and it is used in two senses: a body of knowledge about the world, and the methods used to build and test that knowledge. This laboratory is about the second sense. Science, in that sense, is about how answers are found rather than which answers are found.
What makes a claim scientific is not who makes it or how confidently. It is that the claim is tested against observation (what counts as an observation has a tab of its own, Observation), in ways that other people can check, and that it is given up, or revised, when the tests go against it. Everything else on this page follows from that: how a guess becomes a prediction, how an experiment is built so that its result means something, how the mind fools itself, and how a community of scientists comes to agree.
The scientific method as it is usually taught has six steps:
Richard Feynman put it in three: guess a law, compute what the guess implies, and compare those consequences with experiment. If they disagree with experiment, the guess is wrong, however beautiful the guess and however clever the person who made it. The Hypothesis to Prediction tab works through his three steps.
Real science is messier than either list. Experiments fail, steps are repeated or rearranged, observations make no sense until other observations fill the gaps, and answering one question raises several more. It is also rarely one person's work. Louis Pasteur's famous flasks came after a century of experiments by others, were sharpened by critics who doubted him, and were refined by others after him (the Design an Experiment tab tells the story). What stays constant is the core: ideas tested against evidence, by people willing to be shown wrong.
The Royal Society, founded in London in 1660, took as its motto nullius in verba, usually translated as “take nobody's word for it.” A claim about nature should be something anyone can test and reproduce, not something believed because of who said it. Its members demonstrated experiments in front of one another, and from 1665 reported them in the Philosophical Transactions, the oldest scientific journal still published. (They called themselves natural philosophers. The word “scientist” was coined by William Whewell in the 1830s.)
The same idea runs through every video on this page. Neil deGrasse Tyson tells Richard Dawkins that science never requires belief, only a judgment of how compelling the evidence is. Feynman says that it makes no difference how smart the person who made a guess is, or what his name is. Science works as a community because no one, the person who had the idea least of all, is trusted to check it alone.
Science can tell us whether a mammoth could be cloned, but not whether one should be. It can say which wavelengths of light a painting reflects, but not whether the painting is beautiful. This is an old distinction, usually credited to David Hume as the is–ought problem: facts about how things are do not, by themselves, tell us how things ought to be.
That does not put whole subjects off limits. Love and jealousy are studied by psychologists and neuroscientists, and questions of value are often sharpened by facts: whether animals should be used in experiments depends in part on whether the animals feel pain, which is a question science can investigate. Science informs such questions without settling them on its own.
Your answers are kept as you go. In most browsers they are still here if you leave and come back on the same device, though a private window, a different device, or cleared browser data will start you fresh. When you are done, download a summary to hand in or keep. It lists your answers and, for each, whether it was right the first time.
If science tests claims against observation, the next question is what an observation is. In science, an observation is information gathered directly from the world, through the senses or through instruments, and recorded. The word means both the act of observing and the record it leaves: the broth turned cloudy; the dough ball floated after twelve minutes; the plate shows 47 plaques.
Observations are qualitative when they describe a quality without numbers (the broth is cloudy, the colony is yellow) and quantitative when they count or measure (47 plaques, 37.5 degrees Celsius). A measurement compares something with an agreed standard, a meter or a degree or a second, which is what lets one person's measurement be compared with another's.
That comparison is the heart of the matter. What makes an observation count in science is not that someone saw something but that anyone looking in the same way, under the same conditions, would see the same thing. An observation only one person could ever make, however sincere, cannot test anything. Observations are meant to be repeatable, which is nullius in verba again: take nobody's word for it, look for yourself.
Our senses are narrow and easily fooled. We see a thin band of the electromagnetic spectrum, the colors of the rainbow, as Richard Dawkins and Neil deGrasse Tyson discuss in their conversation, and optical illusions show how readily the eye can be misled even within that band. Instruments extend the senses. Microscopes and telescopes show what is too small or too far to see, and thermometers and clocks measure what we can only feel. Other instruments detect what no sense can: radio waves, ionizing radiation, magnetic fields.
Instruments bring problems of their own. Every measurement carries some uncertainty, which is why a careful observation reports not just a number but how far it might be off. Many instruments also process their data before anyone looks at it, so the line between observing and interpreting blurs; a computer-enhanced image is already partly an interpretation. Scientists keep the raw data for that reason, and journals set rules on how images may be processed.
Observing can also change what is observed. Checking a tire's pressure lets a little air out; a cold thermometer cools the water it is dipped in. Good instruments make such effects too small to matter, but they never make them zero.
Karl Popper liked to begin a lecture by telling his students to take out paper and pencil, observe carefully, and write down what they observed. They asked, of course, what he wanted them to observe. His point, made in Conjectures and Refutations (1963), was that observation is always selective. It needs a question, an interest, a point of view; without one, there is nothing in particular to see. The traditional picture of science as gathering observations with no preconceived notions and letting hypotheses emerge from them gets the order backward.
The philosopher Norwood Russell Hanson, in Patterns of Discovery (1958), called observations theory-laden: what we see depends in part on what we already know and expect. A microbiologist looking down a microscope sees cocci in chains where a beginner sees dots. Perception is not a camera. The brain selects from what reaches the senses and fills in the rest from what it expects, and memory reconstructs rather than replays, which is why two honest witnesses to the same event can remember it differently.
In Berlin in the early 1900s, a horse called Clever Hans seemed able to add, subtract, and tell time, tapping out answers with his hoof. Experts who examined him found no trickery, and there was none: his owner believed in him sincerely. In 1907 the psychologist Oskar Pfungst showed that Hans got the answer right only when the questioner knew it and Hans could see the questioner. People unconsciously tensed as the count approached the right number and relaxed when it arrived, and Hans stopped tapping when they relaxed. The horse was observing very well; the people were not.
Scientists are no more immune. In a 1963 experiment, Robert Rosenthal and Kermit Fode gave students ordinary laboratory rats, telling some students that their rats had been bred to be bright at running mazes and others that theirs were dull. The rats labeled bright did better, though the labels had been assigned at random. The expectations of the observers had found their way into the results.
The defenses are the ones that run through this page:
There are facts, what is actually so, and there is what we observe. The two are not the same, and observations vary in how far they can be trusted to report the facts. Even observations of the world in front of us can mislead, as Clever Hans shows. And people usually limit the word observation to what we notice with our senses, our instruments, or our experiments. But we also observe in other ways, and each tells us something new. The further an observation stands from the world itself, though, the less reliable, on average, it is as a guide to the world.
Each step down the list observes less of the world and more of our ideas about it. That is no reason to dismiss the lower steps; models, thought experiments, and insight are where many of science's best hypotheses come from. The philosopher Hans Reichenbach (1938) distinguished the context of discovery, where an idea comes from, from the context of justification, how it is shown to be right. A beautiful model output or a vivid insight can feel like evidence about the world when it is evidence about our ideas.
That is why we do experiments: to constrain our observations, so as to better appreciate what reality is telling us, and to bring observation into line with fact. Controls rule out explanations other than the one being tested; randomizing and blinding keep the observer's expectations out; measurement against a standard lets others check the result; and repetition, by others and by other methods, shows whether an observation holds. Every kind of observation above is eventually brought back to the world this way. Le Verrier's planet was not found until Johann Gottfried Galle turned a telescope to the spot and saw it. Galileo did not rest on his stones; he timed balls rolling down inclined planes. And Wallas's last stage is verification.
For each, decide whether what is being observed is the world itself (directly or with instruments), the world in aggregate (a pattern in many cases), a model, or what we already know (a thought experiment or an insight).
An observation records what was seen or measured. An inference is a conclusion drawn from observations, which may be right or wrong. Sort each statement.
Does the observation describe a quality, or does it count or measure?
As a young man in Vienna after the First World War, Karl Popper noticed a difference between two kinds of theory that were then making a great stir. The psychological theories of Sigmund Freud and Alfred Adler, and Karl Marx's theory of history, seemed to be confirmed by everything: whatever a person did, or whatever happened, the theory could explain it after the fact. Albert Einstein's general theory of relativity was different. Before the solar eclipse of 1919, it predicted how far the light of stars passing close to the Sun would be bent. Had the measurements come out otherwise, the theory would have been in serious trouble.
Popper concluded that what makes a theory scientific is not how much evidence seems to support it but that it makes risky predictions: predictions that rule things out, and so could turn out wrong. A theory that explains every possible outcome forbids nothing, and so tells us nothing. In his short version, science disconfirms, while pseudoscience confirms. Popper set this out in Logik der Forschung (1934; in English as The Logic of Scientific Discovery, 1959) and, most readably, in the essay “Science: Conjectures and Refutations,” in his 1963 book Conjectures and Refutations.
The white swans make the point. No number of white swans can prove that all swans are white, but a single black swan proves the claim false, and in 1697 Europeans did find black swans, in Australia. Feynman makes the same point about physics: a theory is never proved right, only not yet proved wrong. Newton's law of gravitation survived every test put to it for more than a century and a half, until a slight discrepancy in the orbit of Mercury showed that it was not the whole story.
Feynman's example of a theory that cannot fail: a man hates his mother, and the explanation is that she did not love him enough as a child. Investigate, and it turns out that she loved him a great deal. Then the explanation becomes that she loved him too much. A theory that can explain either result explains neither. The cure, Feynman said, is to state ahead of time exactly how much love counts as too little and how much as too much, and then the theory can be tested.
Feynman also pointed out that science deals in what is more likely and less likely, not in what can be proved impossible. Asked whether he could prove that flying saucers were impossible, he could not. He judged it far more likely that the reports came from the known irrationality of people on Earth than from the unknown rationality of visitors from elsewhere.
For each claim, decide whether some observation could show it to be wrong (it is testable), whether no observation could (it is untestable as stated), or whether it is a question of value, which no observation could settle on its own.
Popper also said that once a theory fails a test, a scientist must be willing to give it up. But not every rescue of a theory in trouble is cheating. When Uranus strayed from the orbit predicted by Newton's laws, astronomers proposed an unseen planet farther out rather than abandoning Newton, and in 1846 Neptune was found close to where the calculations put it. The difference between that and special pleading is that the rescue made a new prediction that could itself be tested. A rescue that cannot be tested, such as a Dad caught putting presents under the tree who explains that Santa came earlier and he is just helping, protects the theory from evidence altogether.
For each, decide whether the rescue could itself be tested, or whether it is an excuse that no evidence could touch. A testable rescue can still fail, and failing is how science moves on.
Popper's test is widely used, but philosophers have not stopped arguing about it. Thomas Kuhn, in The Structure of Scientific Revolutions (1962), and Imre Lakatos, among others, argued that working scientists do not, and should not, abandon a theory at its first failed prediction, since a failure can just as well lie in the experiment or in some background assumption. Falsifiability survives as a useful test for telling science from pseudoscience, not as a settled definition of science.
Experiments can be wrong too. In 2011 the OPERA experiment reported neutrinos that seemed to travel faster than light. Rather than discard relativity, physicists checked the experiment, and in 2012 the result was traced largely to a badly attached fiber-optic cable. Disagreement with experiment is decisive only once the experiment has survived the same scrutiny as the theory.
Feynman's three steps separate two things that are easily run together. A hypothesis is a proposed explanation: water trapped inside a hard-boiled egg boils in a microwave oven, and the pressure has nowhere to go. A prediction is what must be observed if the hypothesis is true: an egg microwaved in its shell will burst. The prediction is the part that meets the world, and it is chosen so that a different result would count against the hypothesis.
A hypothesis is often called an educated guess. What makes a guess scientific, though, is not how educated it is but that it can be tested. Feynman insisted that it does not matter where a guess comes from, only that it be definite enough to compute consequences from, and that those consequences agree with experiment.
The egg also shows how testing leads somewhere unexpected. If trapped steam is the whole story, a peeled egg should be safe. In 2018, researchers reported that peeled hard-boiled eggs can burst too, often after they come out of the oven, because the yolk itself traps pockets of superheated water. The prediction failed, and the hypothesis had to change. (Please do not test this at home.)
For each hypothesis, first choose the prediction that follows from it. Then choose the result that would count against it. Some hypotheses here turned out right and some did not.
By the early 1800s, microscopes had shown that microbes were everywhere, and many people believed that they arose spontaneously from rotting food and flesh. Others thought that microbes were carried on dust in the air and landed in broth or wounds. In the early 1860s Louis Pasteur put the question to a test. He boiled broth in flasks whose necks he had drawn out into a long, curving S-shape: a swan-neck flask. Air could still reach the broth, but dust settled in the bend of the neck. The broth stayed clear. When he tipped a flask so that broth ran into the bend and back, microbes soon grew.
His critics were doing science too. Félix Pouchet argued that if microbes floated everywhere in the air, the air would be almost as dense as iron, and others asked whether boiling had destroyed some life-giving property of the broth. The tipped flask answered the second objection: the boiled broth could still support growth once dust reached it.
Pasteur was not the first to try. Lazzaro Spallanzani in the 1760s and Theodor Schwann in the 1830s had tested spontaneous generation with sealed or heated flasks, and their results had been disputed. Nor was the debate over after Pasteur. In the 1870s Henry Charlton Bastian found that some boiled broths, especially infusions of hay, still grew microbes. The answer came from Ferdinand Cohn's discovery of bacterial endospores, which survive boiling, and from John Tyndall's repeated heating, which kills the spores as they germinate. Pasteur's design could not have caught the spores; a community working over decades did.
The independent variable is what the experimenter changes, so that its effects can be seen: in Pasteur's flasks, whether dust from the air can reach the broth. The dependent variable is the outcome measured: whether microbes grow. Controlled variables are everything kept the same so that they cannot explain a difference: the same broth, boiled the same way, in the same room.
A control is a comparison that shows what happens without the treatment, or shows that the setup can respond at all. A negative control is expected to show no effect; a positive control is expected to show one. Pasteur's tipped flask was a positive control: without it, a clear broth could have meant that boiling had ruined the broth rather than that dust carries microbes.
Controls can fail in their own ways. A negative control that receives some of the treatment by accident, a culture splashed from the treated plate next to it, say, will narrow or erase a real difference. An experimenter who then concludes that the treatment had no effect has drawn a conclusion from inappropriately different circumstances: the comparison was never between treated and untreated.
In 1747, the ship's surgeon James Lind gave twelve sailors with scurvy six different remedies, two sailors to each. The two given oranges and lemons recovered far faster than the rest. Lind's trial is one of the first controlled experiments in medicine. Modern trials add safeguards against the experimenter's own bias:
Some things cannot be put in a flask. Nobody can build a second Earth to test how carbon dioxide affects climate, so much of what is known comes from observational science: measuring what is already happening, or has happened, and testing hypotheses against those records. Experiments still play their part. In 1856 Eunice Newton Foote placed glass cylinders of carbon dioxide and of ordinary air in sunlight and found that the carbon dioxide warmed much more.
Observational studies of people face a special difficulty: confounding variables. Suppose people who drink more alcohol have more heart disease. They may also eat differently, exercise differently, or smoke more, and any of those could be responsible. A correlation between two things does not show that one causes the other. Good observational studies measure and adjust for confounding variables, but no one can adjust for a confounder no one has thought of, which is why conclusions rest on many studies of different kinds.
For each experiment, choose the independent variable, the dependent variable, and the control.
Did the scientists change something and watch what happened, or observe what was already happening?
In a famous scene from Monty Python and the Holy Grail (1975), villagers drag a woman before Sir Bedevere, insisting that she is a witch. The scene is a mock trial with the form of science and none of its substance, and nearly every flaw on this tab is in it:
The scene is on the Videos tab.
The historian Niall Ferguson, in The Ascent of Money (2008, pages 345–346), lists ten biases that make human beings less than effective economic creatures. Many, perhaps all, of them also explain why, without training and discipline, we do not tend to make effective scientists. Indeed, the whole point of science, really, is to understand the world better by striving to reduce such biases, in our own perspectives and in our collective ones. Ferguson's list follows the sections of a 2008 chapter by Eliezer Yudkowsky, “Cognitive Biases Potentially Affecting Judgment of Global Risks,” in Global Catastrophic Risks, edited by Nick Bostrom and Milan Ćirković. Most of the biases were first described by psychologists, credited in the last column.
These biases are not stupidity. They are mental shortcuts that let us decide quickly and usually well, and they fail in predictable ways when a question is too big, too complex, or too unfamiliar for intuition. Even the Earth-centered universe was not simply a failure of intuition: if the Earth moved, the nearer stars should seem to shift against the farther ones over a year, and no such shift was seen until Friedrich Bessel measured it in 1838.
| Bias | What it does | How it misleads in science | First described by |
|---|---|---|---|
| Availability bias | Judging by what comes easily to mind rather than by the data needed. | A vivid recent result, or a dramatic case, outweighs a dull but larger body of evidence. | Amos Tversky and Daniel Kahneman, 1973 |
| Hindsight bias | Judging an outcome, once known, to have been more predictable than it was. | After a result is in, the hypothesis it supports seems to have been obvious all along. | Baruch Fischhoff, 1975 |
| The problem of induction | Drawing general rules from too little information. | All the swans so far are white, so all swans are white. | David Hume, 1700s; black swans, Nassim Nicholas Taleb, 2007 |
| The conjunction (and disjunction) fallacy | Overestimating that a chain of likely events will all happen, and underestimating that at least one of several unlikely ones will. | An experiment that needs seven steps to go right, each with a 90 percent chance, succeeds less than half the time. | Amos Tversky and Daniel Kahneman, 1983 |
| Confirmation bias | Looking for evidence that supports a first hypothesis rather than for evidence that would disprove it. | The opposite of Popper's advice: testing only where the hypothesis is expected to pass. | Peter Wason, 1960 |
| Contamination effects | Letting irrelevant but nearby information shape a judgment. | The first number seen, or the last case discussed, anchors an estimate that should not depend on it. | Amos Tversky and Daniel Kahneman, 1974 (anchoring) |
| The affect heuristic | Letting feelings about something, liking or disliking it, color judgments of its risks and benefits. | A favored hypothesis seems both more likely and more important than the evidence warrants. | Paul Slovic and colleagues, 2000 |
| Scope neglect | Failing to scale concern to the size of what is at stake. | A harm a hundred times larger is treated as only a little worse. | William Desvousges and colleagues, early 1990s |
| Overconfidence in calibration | Setting the range around an estimate too narrowly, and mistaking the best case for the most probable. | Error bars too small; a schedule that assumes nothing will go wrong. | Studies of calibration from the late 1960s on |
| Bystander apathy | Leaving responsibility to others when in a crowd. | Everyone assumes someone else has checked the data, the analysis, or the reference. | John Darley and Bibb Latané, 1968 |
Asked whether cell phones cause cancer, Neil deGrasse Tyson pointed out that every measurement has some uncertainty, so studies of an effect that does not exist will now and then come out positive by chance alone. If the positive studies are published and the negative ones are not, the published record will show an effect that is not there. The psychologist Robert Rosenthal called this, in 1979, the file drawer problem: the negative results stay in the file drawer.
Try it. Each run below is twenty studies of an effect that does not exist. A study counts as “finding” the effect when chance alone gives it a result that would happen less than one time in twenty (the usual threshold, p < 0.05).
On average about one study in twenty comes out positive, but in any one run there may be none, or three. Real effects, by contrast, show up again when the study is repeated, by other people, in other places. That is why replication matters, and why a single surprising result is a reason to look again rather than a conclusion. Tyson also said that when one thing really causes another, the signal is huge. That holds for large effects, which need no statistics to demonstrate and are easily repeated. Small effects are another matter. They need statistics to demonstrate, a study of a real small effect will not always produce evidence in its support, and they are particularly susceptible to confounding by small or biased samples.
Each case has one main thing wrong with it. Name it.
In everyday speech a theory is a hunch, which is why people say a scientific idea is “just a theory.” In science the word means close to the opposite: a theory is about as well supported as an explanation gets. Theories and laws also do different jobs. A law describes what happens; a theory explains why. A theory does not graduate into a law however much evidence it gathers, and the theory of gravity sits alongside the law of gravitation rather than below it.
Scientists also work with models: drawings, physical models, and mathematical or computer models that represent how something works. No model is a perfect copy of reality, and that is not their purpose. The statistician George E. P. Box put it as “all models are wrong, but some are useful.”
Scientific consensus is not a vote or an opinion. It is the point at which an idea has been tested from so many angles, by so many people, that one explanation is far more likely than any other. There is no fixed threshold: figures like “99 percent agree” signal only that the bar has been cleared, and they usually come from counting published papers that take a position, or from surveys of the scientists who work in a field, not from polling all scientists.
Consensus is newer, and harder won, than it can look. The idea of atoms goes back to the ancient Greeks, and John Dalton gave it a chemical footing in the early 1800s. Yet as late as 1900 respected scientists such as Wilhelm Ostwald and Ernst Mach doubted that atoms were real. Their doubts gave way after Einstein's 1905 analysis of the jiggling of small particles in water (Brownian motion) and Jean Perrin's measurements confirming it a few years later. Ostwald accepted atoms in 1909: a skeptic persuaded by evidence.
Within a consensus, questions stay open at the edges. That vitamin C prevents scurvy and vitamin D prevents rickets is settled; whether chocolate raises or lowers the risk of cancer is not, because diets cannot be controlled for long, people differ in a thousand other ways, and small effects are hard to see. Debate about such details does not overturn the core, and overturning the core would take overwhelming evidence.
A consensus can even slip backward when its evidence is poorly understood. James Lind's 1747 trial showed that citrus cures scurvy, and the Royal Navy came to issue lemon juice. Later the navy switched to West Indian limes, which carry far less vitamin C, and juice that had been boiled and stored lost most of what it had. Polar expeditions of the late 1800s suffered scurvy again, and some physicians came to doubt citrus altogether. The cause, vitamin C, was not identified until the 1930s.
Which of these has broad scientific consensus behind it, and which is still being worked out?
Ask why a bird migrates, and there are two quite different answers. One is about how: lengthening days set off hormonal changes that make the bird restless and send it north. The other is about why in the evolutionary sense: birds that migrated to breed where food was plentiful left more offspring than birds that stayed. The evolutionary biologist Ernst Mayr called the first kind of explanation proximate and the second ultimate, in “Cause and Effect in Biology,” Science 134:1501–1506 (1961). Both are needed, and they do not compete: a full answer to the bird's migration includes both.
The ethologist Niko Tinbergen divided each kind in two, in “On aims and methods of ethology” (1963). His four questions are still the standard checklist for explaining any trait:
| About the individual (proximate) | About the lineage (ultimate) | |
|---|---|---|
| How it works now | Mechanism: the physiology, molecules, and signals involved | Function: how the trait helps its bearers survive and reproduce |
| How it came to be | Development: how the trait forms as the individual grows | Evolutionary history: how the trait arose and changed over generations |
One trap to avoid: ultimate explanations are not purposes. “Giraffes grew long necks in order to reach high leaves” is the classic wrong answer. The ultimate explanation is about which ancestors left more offspring, not about anything a giraffe was aiming for. And the distinction has critics too: Kevin Laland and colleagues argued in Science in 2011 that development and behavior can feed back on evolution, so that the two kinds of cause are not fully separate.
Reductionism means different things, and it helps to keep them apart. As a method, it is the strategy of explaining a system by taking it apart and studying its pieces, and much of molecular biology was built this way, a good deal of it on bacteriophages. As a claim about what exists, it says that a living thing is made of nothing but its molecules, with no extra vital substance, which few biologists dispute. The strong claim, that biology's explanations can all be rewritten in the terms of chemistry and physics, is where the arguments are.
The counterweight is emergence: properties that appear at one level of organization and are not usefully explained at a lower one. The physicist Philip W. Anderson made the case in “More is different,” Science 177:393–396 (1972): each level of organization brings new laws, even though nothing at that level breaks the laws of the levels below. Mayr argued on similar grounds that biology cannot be reduced to physics, since ultimate explanations are historical and no amount of chemistry recovers them. Richard Dawkins called his own position hierarchical reductionism, in The Blind Watchmaker (1986): explain each level by the one just below it, not by leaping straight to atoms. The philosopher Daniel Dennett called the leap greedy reductionism, in Darwin's Dangerous Idea (1995).
Bacteriophages show both sides. They were central to molecular biology's reductionist triumphs, from the 1952 experiment of Alfred Hershey and Martha Chase showing that phage genes are made of DNA, to the cracking of the genetic code. Phage ecology, by contrast, asks about outcomes such as the rise and fall of phage and bacterial populations, or why one length of infection pays better than another, which make sense only at the level of populations. The opposite of taking apart is putting together, synthesis, which has a tab of its own: Kinds of Research.
For each explanation, choose which of Tinbergen's four questions it answers. Watch for one that is not really an explanation at all.
For each explanation, choose the level of organization at which it works. The last two ask which of several true-sounding answers is at the right level.
The word “theoretical” does double duty across the sciences. In physics, theory usually means building the mathematical framework that makes predictions, often well ahead of experiment: Paul Dirac's equations predicted the positron before anyone saw one, and the Higgs boson, predicted in 1964, was found in 2012. Connecting theory to data already in hand has a name of its own in physics, phenomenology. In biology, the closest match is mathematical theory in the tradition of Ronald Fisher, J. B. S. Haldane, and Sewall Wright, who worked out what Mendelian genetics implies for evolution. Yet “theoretical” in everyday biological use often stretches to cover conceptual and synthetic work as well.
In Greek, analysis means breaking apart and synthesis putting together, and the two were treated as complementary halves of method long before anyone spoke of reductionism. Synthesis has a strong claim to being research, and biology supplies the best evidence. The central framework of evolutionary biology is called the Modern Synthesis, after Julian Huxley's Evolution: The Modern Synthesis (1942), which joined Mendelian genetics to natural selection; Ernst Mayr was one of its architects. Charles Darwin's On the Origin of Species (1859), which he called one long argument, was itself built in large part from other people's observations.
Synthesis also has institutions. The U.S. National Science Foundation founded the National Center for Ecological Analysis and Synthesis in 1995 to support it in ecology, and in medicine, systematic reviews and meta-analyses, which combine the results of many studies into one, sit at the top of the usual ranking of evidence.
In Scholarship Reconsidered (1990), Ernest L. Boyer of the Carnegie Foundation for the Advancement of Teaching argued that universities had come to count only discovery as research. He proposed four kinds of scholarship: discovery, integration, application, and teaching. Integration, making connections across findings and fields, is his name for synthesis.
Which kind of research is each?
People have studied nature systematically for thousands of years, in ancient India, Greece, and China, among the Maya, whose astronomers tracked the Sun, Moon, and planets precisely, and in Indigenous traditions of close observation the world over. The method this page describes, with its controlled experiments, open reporting, and community checking, was assembled more recently and piece by piece. Some of the pieces:
For each pair, choose the one that happened first. The answers are in the timeline above, if you need them.
Those marked core are the place to start; additional ones fill in background or add perspective. The notes say what to watch for, and where a video gets something wrong. Where only part of a video is worth watching, the player starts and stops at that part.