DRAGENwiki · Article

Empiricism

Sources checked. Clinical review pending.

Empiricism is the idea that knowledge comes from experience. In behavior analysis it works as a daily habit: look carefully, measure, check what you saw against what someone else saw, and let the evidence change your mind. It does not mean that only numbers matter, and it does not mean that science is ever finished or certain.

Empiricism in 60 seconds

  • The idea. In the form philosophers call the Empiricism Thesis, knowledge in a subject area "can only be gained, if at all, by experience." [1]
  • Experience comes in two kinds. Philosophers usually separate sense experience, through the five senses, from reflective experience, our awareness of our own mental operations. [1]
  • Its old rival is rationalism. Most rationalists hold that some concepts or knowledge are gained independently of sense experience. Almost no thinker fits neatly into either camp. [1]
  • Behavior analysis keeps coming back to what can be observed. In 1950, B. F. Skinner wrote that "we must eventually get back to an observable datum." [2]
  • Seeing is not the same as proving. Observers can be biased, and a change can have causes nobody saw. Science controls for this with clear definitions, a second observer, baselines and replication. [3]
  • It deals in likelihood, not certainty. "Science deals with probability, not possibility." [3]
  • At home and school, empiricism means describing what happened, counting before and after a change, and asking "How would we know?" A BCBA reviews every clinical decision.

Jump to: What empiricism means · Labels with care · Empiricism in behavior analysis · Seeing without bias · From observation to experiment · Testing what you try · At home and school · A made-up example · Myths · Words to know · Sources

General information, not medical or clinical advice.

What empiricism means

Empiricism is one side of a long argument in philosophy. The Stanford Encyclopedia of Philosophy describes the argument between rationalism and empiricism as a question of how far we depend on experience to gain knowledge of the world outside our own minds. [1]

Philosophers usually split experience into two kinds. Sense experience comes through the five senses and tells us about things in the world. Reflective experience is our conscious awareness of our own mental operations, and it tells us about our own minds. [1]

Rationalists argue that some of our concepts or knowledge go beyond anything the senses could supply. Most hold that, in significant ways, our concepts and knowledge come from somewhere other than sense experience, and they explain how reason supplies what the senses cannot. [1] Empiricists answer that experience, of one kind or the other, is where knowledge comes from. In the version the encyclopedia calls the Empiricism Thesis, we have no source of knowledge in a given subject area, or for the concepts we use in it, other than experience. The thesis does not promise that we will know anything at all. It says only that knowledge "can only be gained, if at all, by experience." [1]

Two famous images capture the disagreement. John Locke pictured the mind as "a blank slate on which experience writes." Gottfried Wilhelm Leibniz answered with the image of a block of marble whose veins decide which figures a sculptor can carve from it. The encyclopedia's authors side with Leibniz on one point: the mind itself helps decide what it comes to contain. [1]

Empiricism can also lead to doubt. Some empiricists agree with the rationalists that experience cannot give us knowledge of a certain kind, and conclude that we simply do not have that knowledge. David Hume took this line on causation. In the encyclopedia's summary, Hume argued that causation is not actually known, only presupposed, because of "a particular habit of our minds." [1]

In the twentieth century, A. J. Ayer gave the empiricist reply a new form through logical positivism. Under its verification theory of meaning, every cognitively meaningful sentence is one of two kinds. Either it is true just by the meaning of its terms and tells us nothing substantive about the world, or it is open to empirical verification. [1]

Empiricism does not have to cover everything. The encyclopedia notes that a person can be a rationalist about mathematics and an empiricist about some or all of the physical sciences. The two views only collide when they claim the same subject. [1]

Labels with care

History books often sort the philosophers of the seventeenth and eighteenth centuries into two teams. Descartes, Spinoza and Leibniz become the Continental Rationalists; Locke, Hume and Reid become the British Empiricists. The encyclopedia warns that schemes like this should be adopted only "with great caution," because each philosopher's views are more subtle and complex than the labels suggest. [1]

The examples cut both ways. Descartes, usually treated as a representative rationalist, has been argued to have clear empiricist leanings in natural philosophy, where sense experience plays a crucial role. Locke, usually treated as the model empiricist, argued that reason is on an equal footing with experience for some kinds of knowledge, most famously moral truths. [1] The encyclopedia adds that early modern philosophy is far richer than the two-team story suggests. Hobbes cannot be placed cleanly in either camp, the authors note, and neither can Elizabeth of Bohemia, Anne Conway, George Berkeley, Émilie du Châtelet or Mary Shepherd. Outside debates about innate knowledge, they conclude, the distinction should simply be set aside. [1]

That warning is worth carrying into behavior analysis. A label is a summary. It is never the whole story about a thinker, and it is never the whole story about a person.

Empiricism in behavior analysis

In behavior analysis, empiricism shows most clearly in how the field treats observation and theory. A classic statement is Skinner's 1950 paper "Are theories of learning necessary?" [2]

Skinner was careful about what "theory" meant there. The paper did not reject the basic assumptions that any science needs; Skinner's example of one was "That nature is orderly rather than capricious." [2] Its target was narrower: explaining what we observe by pointing to events that happen elsewhere, "at some other level of observation," described and measured in other terms. [2]

Skinner named three kinds of theory in the study of learning that fit this description. [2]

  • Physiological theories explain learning by changes assumed to happen in the nervous system.
  • Mentalistic theories explain it by "mental" events, saying that an organism learns because it "finds something pleasant" or "expects something to happen."
  • A third type explains it by events that are not directly observed. In earlier work, Skinner had suggested reading the letters CNS as the "Conceptual Nervous System" rather than the Central Nervous System.

Skinner's worry was that theories like these can stand in for evidence. When results are messy, a theory can make the underlying process feel smooth and orderly anyway, which Skinner called "the use of theory as a refuge from the data." [2] Explaining behavior by an expectation also leaves a new job undone, because the expectation itself still has to be explained. [2] Skinner's answer was plain: "we must eventually get back to an observable datum." [2] Skinner also argued that research designed around theories is likely to be wasteful, since most theories are eventually overthrown and much of the research that went with them is thrown away. [2]

Two lessons from the paper still shape how behavior analysts look at data.

Find a good measure. Skinner argued that progress in a field "usually waits upon the discovery of a satisfactory dependent variable," and that until one is found, people fall back on theory. In Skinner's view, the theoretical entities of learning theory had mostly stood in for a measure that could be observed directly, and they have little reason to remain once such a measure is found. In Skinner's experiments, that measure was the rate at which a response occurred. [2]

Look at the individual. Skinner's example was latency, the time a bird took to respond after a signal. Averaging many readings, from one bird or many, can create what looks like a steady lengthening of that time, when "the data for an individual organism do not show a continuous process." [2] You can see the same habit in a behavior graph that follows one person's behavior, session by session.

Skinner did not ban theory. The paper's conclusion admits that doing without theories altogether may be too much to expect, and says simply, "Theories are fun." [2] Its argument was that the fastest progress might come from research not designed to test theories: collect data that show orderly change, and relate them to variables you can manipulate, chosen through a common-sense exploration of the field. Skinner even left room for theory of another kind, "a formal representation of the data reduced to a minimal number of terms." [2]

Seeing without bias

Being empirical is not the same as believing whatever you see. In a 2008 paper for practitioners, Matthew P. Normand wrote about the skeptical side of science. Being skeptical, Normand wrote, "does not mean doubting the validity of everything," and it does not mean being cynical. It means "to judge the validity of a claim based on objective empirical evidence." [3] Drawing on the 18th-century philosopher David Hume, Normand added that extraordinary claims can be true, but "the more extraordinary the claim, the more extraordinary the evidence required." [3]

Observation is where bias creeps in. In Normand's words, what scientists say should be controlled by "what is seen, rather than what one hopes to see." [3] And: "Bias in observation cannot be entirely eliminated, but it can be controlled." [3] In basic research, machines often record behavior automatically. In applied work, people usually do the recording, so steps are needed to make sure that a change in the data is a change in the behavior of interest, "not the behavior of the observer." [3] Normand names two such steps: interobserver agreement measures, in which two observers record the same behavior separately and compare, and double-blind procedures, in which the people involved do not know which condition is in effect. [3]

Normand also described a common warning sign of pseudoscience: claims that rely on stories or testimonials rather than on objective experimental evidence. [3]

None of this makes a family's report unimportant. Families see what no one else sees, and their reports are often where the best questions begin. The point is narrower: one unchecked report, from anyone, cannot prove what caused a change. That includes the impressions of the professionals, as the next section shows.

From observation to experiment

Even careful observation has limits. Normand's example: a teacher refers a child who rarely works on assigned tasks. A week of careful records shows the child on task about 30% of the time. A token system is added, with points for every 60 seconds of continuous work that can later be traded for preferred activities or items. A few weeks later the child is on task about 80% of the time, and the teacher reports much better work completion. [3]

Did the token system cause the change? Maybe. But the assignments might have become easier or more interesting at the same time, or the referral itself might have brought the child more attention from adults at school. Normand's verdict: "we cannot know from the information obtained." [3] Success with child after child does not settle it either, because the intervention could keep bringing along some other variable that is the real cause. [3]

The way forward is experiment. In most behavior-analytic designs, behavior is measured repeatedly during a baseline, before anything is changed. The baseline predicts what would happen if the change had no effect. If behavior under the new conditions departs from that prediction, an effect is suggested, and the more the effect can be repeated, through reversals to baseline or multiple-baseline arrangements, the stronger the conclusion. [3] When we can predict how likely a behavior is under certain conditions, and change that likelihood on purpose, we have shown cause and effect. [3]

Science is also built to catch its own mistakes. In Normand's phrase, "although it is fallible, it also is self-correcting." [3] Clear descriptions of procedures let others repeat them at different times and places with different people. At its best, peer review catches flawed studies or wrong conclusions before they spread, and after publication, failed replications and new findings lead to revisions. [3] So changing your mind is not a failure. Normand wrote that science embraces "flip-flopping" as long as the change comes from changes in the evidence rather than from some other influence. [3]

What, then, can we know? "Nothing is known for certain," Normand wrote, but much is known for which the other explanations are so unlikely that they are not worth considering. Knowing, on this view, means describing how strong a prediction we can make. "Science deals with probability, not possibility." [3] And because "Any single study will have limitations," confidence grows when independent investigators, using different sound methods, reach the same conclusion. [3]

Testing what you try

Normand also admitted a practical limit: nobody can check everything personally. Much of what we "know" about unfamiliar fields, from quantum mechanics to car repair, comes from other people describing the evidence for us. So many of us will end up dogmatic in practice, Normand argued, but we should be "carefully dogmatic." At best, we read reviews of the research rather than the research itself, and reviews carry their writers' biases. At worst, we hear about a new diet or therapy from someone already convinced it works, who shows us only the evidence that seems to support it. [3]

For practitioners, Normand's advice was to be a true scientist-practitioner. [3] Normand's example: a family is considering a special diet for their child. A behavior analyst might be able to persuade the family to test it systematically. Everyone agrees in advance on clear, operationally defined behavioral objectives; a good baseline is taken; then the diet is put in place and taken away on a plan over several weeks or months. The parents might even give and hold back the diet on a schedule set in advance, while the behavior analysts do not know which weeks are which until the evaluation is done. Normand noted that this puts the focus on careful evaluation and removes the need for a heated argument about the diet. [3] The choice stays with the family, and any change to a child's diet, medicine or health care belongs with the child's doctor.

The same skepticism applies to a behavior analyst's own work. Normand urged practitioners to ask themselves: "Do you really know that the improvements you see are attributable to your efforts?" Even if they are, which parts of the plan matter most? Could any part be dropped to make treatment easier? [3]

What it means at home and school

These are DRAGEN's plain-language takeaways. They come from the sources only where a source is cited.

  • Describe, don't label. "Pushed the plate away and said no" can be counted and checked. "Was being difficult" cannot. Words like "difficult," "lazy" or "defiant" are conclusions, not observations.
  • Count before you change something. A few days of plain counting gives you a baseline, so you can tell later whether a change made a difference. [3] DRAGEN's free lesson How to count a behavior shows four ways to count.
  • Write the definition down. If two people would count different things, the numbers will not mean much. Sometimes a second person counts at the same time and you compare; that is interobserver agreement. [3] See the lesson How to check your data with interobserver agreement.
  • Ask "How would we know?" When someone recommends a new product, treatment or trick, ask what evidence supports it and how you would tell whether it works for this person. [3]
  • Change one thing at a time when you can. If several things change at once, nobody can tell which one helped. [3]
  • Their words are information too. What a person says about their own experience, including "no," "stop" or "that hurts," counts. Refusal and withdrawn assent are information, never "noncompliance" and never a target to reduce.
  • Safety first, data second. If someone may get hurt, keep everyone safe first. Counting can wait.
  • Bring the data to the team. A BCBA reviews every clinical decision, and medical questions go to a doctor.

A made-up example

Learner G is eight and uses they/them. Mornings at home are hard: many school days end with crying at the front door and a late arrival. A relative says a new morning-routine app fixed everything for their own child. The family wants to try it, and their BCBA suggests a small test first.

  1. Define it. Together they choose what to count: the minutes from the first wake-up call to shoes on, and whether the morning ends with crying at the door.
  2. Count first. For two school weeks, nothing changes. A parent writes down the two numbers each morning. Some mornings take 25 minutes and some take 70, and crying happens on 6 of 10 days.
  3. A second look. On three mornings, a grandparent who lives with the family records the same things separately. Their numbers mostly match, so the family trusts the sheet.
  4. Try it. The app goes in for two weeks, with nothing else changed on purpose.
  5. Listen. On day two, Learner G covers their ears and says the alarm hurts. That is information, not a refusal to cooperate. Learner G picks a softer song for the alarm, and the test goes on.
  6. Look. Mornings now take 25 to 40 minutes, and crying happens on 2 of 10 days.

Is it the app? Maybe. But the BCBA points out that bedtimes also moved earlier during those two weeks. The family can choose: keep the app and keep counting, or, if they want to know more, go back to the old routine for a week while keeping bedtimes the same. Comfort and help stay ready on every hard morning. Learner G's voice stays part of the plan, the BCBA reviews it, and the family makes the final call.

Common myths

  • Myth: Empiricism means only numbers count. Fact: Empiricism is about experience, which philosophers divide into sense experience and reflective experience. [1] Numbers are one careful way to report what was seen. A person's own account of their experience matters too.
  • Myth: If I saw it happen, I know what caused it. Fact: In Normand's token-system example, a big improvement could have come from the intervention or from changes nobody controlled. Only a planned comparison can tell. [3]
  • Myth: Behavior analysts reject all theory. Fact: Skinner's 1950 paper took aim at one kind of theory, explanations at another level of observation. It left room for a compact formal summary of the data and admitted that theories are fun. [2]
  • Myth: Science proves things for certain. Fact: Normand wrote that nothing is known for certain. Science deals in probability, and it corrects itself as evidence comes in. [3]
  • Myth: Changing your mind means you were never scientific. Fact: Science welcomes a change of mind when the evidence changes. [3]
  • Myth: An average describes every person in the group. Fact: Skinner warned that averaging many readings can create the look of a gradual change that one individual's own data do not show. [2]
  • Myth: The old philosophers split into two neat teams. Fact: Almost no author falls neatly into one camp, and the encyclopedia advises using those labels only with great caution. [1]
  • Myth: Being skeptical means being cynical. Fact: Scientific skepticism means judging a claim by objective evidence, not doubting everything. [3]

Words to know

  • Empiricism: the view that, in a given subject area, experience is our only source of knowledge. [1]
  • Rationalism: the view that some concepts or knowledge are gained independently of sense experience. [1]
  • Sense experience: experience through the five senses, which tells us about the world around us. [1]
  • Reflective experience: conscious awareness of our own mental operations. [1]
  • Blank slate: Locke's image of the mind as a surface on which experience writes. [1]
  • Logical positivism: a twentieth-century view; Ayer's version held that every meaningful sentence is either true by the meaning of its terms or open to empirical verification. [1]
  • Observable datum: something that can be observed and measured; Skinner's 1950 paper insists that explanations must come back to one. [2]
  • Dependent variable: the measure watched for change; in Skinner's experiments, the rate of responding. [2]
  • Scientific skepticism: judging a claim by objective empirical evidence, without doubting everything and without cynicism. [3]
  • Observer bias: when what an observer reports is shaped by what they hope to see, rather than by what happened. [3]
  • Interobserver agreement (IOA): a check in which two observers record the same behavior separately and compare their records. [3]
  • Baseline: repeated measurement before any change is made, used to predict what would happen without the change. [3]
  • Replication: repeating an effect or a finding, in the same study or in new ones, to see whether it holds. [3]
  • Anecdote or testimonial: a story or personal endorsement offered as evidence, without a controlled comparison. [3]
  • Scientist-practitioner: a practitioner who evaluates their own work with scientific methods. [3]

Need a PDF?

Create it here, for free: the DRAGEN app. Use it at your next session.

Or print one we already made, also free:

More from DRAGEN

Related DRAGENwiki pages: behaviorism · radical behaviorism · methodological behaviorism · experimental analysis of behavior · seven dimensions of ABA · determinism · applied behavior analysis · measuring behavior · reading a behavior graph · the right to say no

What remains unknown

  • How far experience alone can account for knowledge is still argued. Both sides face open problems: rationalists still owe an account of what intuition is and how it could give us justified beliefs about the world, and the general principles behind the empiricist view, such as Hume's account of ideas and the verification principle, have problems of their own. [1]
  • Normand found very little in the published literature on how to practice skepticism and wished for a fuller functional analysis of skeptical behavior: under what conditions do we call someone skeptical? Once those conditions are known, ways to teach a skeptical repertoire could be tested. Whether Normand's own suggestions help is, in Normand's words, "most certainly an empirical question." [3]
  • Skinner saw no reason in advance why a complete account of learning could not be given without theories set at another level of observation. [2] This page does not settle that question.
  • This page summarizes the main lines. It does not settle these debates.
  • Last checked on November 12, 2026.

Sources

  1. Markie, P., & Folescu, M. (2021). Rationalism vs. empiricism. In The Stanford Encyclopedia of Philosophy (substantive revision of September 2, 2021; first published August 19, 2004). https://plato.stanford.edu/entries/rationalism-empiricism/
  2. Skinner, B. F. (1950). Are theories of learning necessary? Psychological Review, 57, 193–216. Classics in the History of Psychology, York University. https://psychclassics.yorku.ca/Skinner/Theories/
  3. Normand, M. P. (2008). Science, skepticism, and applied behavior analysis. Behavior Analysis in Practice, 1(2), 42–49. https://doi.org/10.1007/BF03391727 (full text: https://pmc.ncbi.nlm.nih.gov/articles/PMC2846586/)

Pages that link here