Scientific understanding of behavior

Research methods are the procedures by which a claim about behavior is tied to evidence another person could inspect. The claim may be about what people do, how often they do it, what tends to travel with it, or what changes it. The method is the part that makes the claim answerable. A vivid story, a confident authority, and a private hunch can all be starting points. None of them is yet a scientific result, because none of them states what was observed, under what conditions, and what would have counted against the claim.

Cozby and Bates treat behavioral science as an empirical enterprise: conclusions rest on evidence collected in a systematic way, and the evidence is public (Cozby & Bates, 2020). Empiricism here means that the court of appeal is observation, not the status of the person who prefers the conclusion. Observation is broader than “I saw it once.” A score on a questionnaire, a count of interruptions in a recorded conversation, a reaction time, and a coded facial movement are all observations once the procedure that produced them is specified. The specification is what later chapters call an operational definition. This chapter stays with the prior question: what job is the observation being asked to do?

A methods course is therefore a course in the jobs of evidence. The same behavior can be described, used to predict a later behavior, tested as an effect of some condition, or placed inside an explanation. Mixing those jobs up is how a careful description gets narrated as a cause, and how a cause that was only shown in one room gets narrated as a law of people in general. The rest of PSY 200 separates those jobs and names the designs that can carry each one.

The standard map, the one Cozby and Bates use, gives behavioral science four goals: description, prediction, determination of cause, and explanation. They are ordered by what they entitle the writer to say. Description entitles a report of what occurred. Prediction entitles a statement that one variable carries information about another. A causal determination entitles a statement that changing one variable changes the other. Explanation entitles a statement of why, in terms of a mechanism that could be wrong.

Description answers “what does this behavior look like, in whom, and how often?” A description can be a frequency, a typical score, a sequence, or a careful account of a single case. Description is not a consolation prize. Most of what a field knows, in the early stage of a question, is descriptive, and a bad description makes every later design pointless because the later design inherits a fuzzy object. When a paper says that a behavior is common, rare, stable across a day, or different in two named groups, and the paper stops there, the paper is doing description. The reader’s job is to notice that it stopped there.

Prediction answers “if we know a person’s standing on one variable, what should we expect on another?” Prediction requires an association. It does not require that anyone manipulated anything. High-school grades predict first-year college grades in a modest way; the grades were not assigned by the researcher. A useful prediction can be real while the causal story is still open. Employers, clinics, and admissions offices live on prediction. The scientific constraint is to say “predicts” when the evidence is an association, and to keep “produces” for a design that can support a cause.

Determination of cause answers “does a change in this condition change the behavior?” The word cause, in this course, is not a metaphysical essence. It is a conclusion that three conditions have been met, taken up in the next section. People often talk as if a cause were simply a strong prediction. A strong prediction can exist because both variables are products of a third condition, or because the outcome came first and the supposed cause is a later echo of it. Causal language is a promotion. The design has to earn it.

Explanation answers “by what process does the cause have its effect?” Two studies can agree that a manipulation changes a behavior and disagree about the mechanism. One lab may claim the change is attention. Another may claim it is arousal. Explanation is a further claim, with its own evidence: measures of the proposed mechanism, designs that block the mechanism and watch the effect shrink, and theories that say what else should happen if the mechanism is real. A theory that only restates the finding (“people do this because they have a tendency to do this”) has not explained anything. The restatement has no extra consequence that could fail.

Methods texts inherit a compact test for a causal claim from the experimental tradition summarized by Cook, Campbell, and later Shadish (Shadish, Cook, & Campbell, 2002). Cozby and Bates teach the same three conditions. First, covariation: the cause and the effect go together. When the cause is present, the effect is more likely, larger, or different in a stated direction. If the supposed cause and the behavior are unrelated in the data, the causal claim is already dead, however attractive the story.

Second, temporal precedence: the cause comes before the effect. A variable measured after the behavior cannot be the cause of that behavior, even if the correlation is large. People rewrite the calendar when a story is flattering. A student who feels confident after a high score may treat the confidence as the cause of the score. The score came first. Temporal order is a fact about the design, not a mood about the variables. If the design does not establish order, the paper does not have a cause.

Third, elimination of alternative explanations: no other plausible variable accounts for the covariation. This is the condition that ordinary life handles worst, because a good story feels like it has already ruled the rivals out. A rival explanation is not an insult. It is another variable that could produce the same pattern. In an experiment the main tool for dealing with rivals is control, especially random assignment, which later chapters treat in detail. In a nonexperimental study the rivals often remain live, and the honest sentence is an association plus a list of the alternatives the design could not close.

All three conditions are required for the causal promotion. A study with covariation and a clean time order can still be wrong about cause if a confound rode along with the manipulation. A study with exquisite control of alternatives can still fail if the manipulation did not actually change the proposed cause, or if the effect was measured before the manipulation. Figure 1.2 is the checklist a reader applies to a paragraph that says “X led to Y,” “X produced Y,” or “the effect of X.”

Basic research tries to understand a process, whether or not a particular institution has asked for the answer. A study of how people update a belief when the evidence is mixed is basic research if the goal is the process of updating. Applied research tries to solve a practical problem in a stated setting: whether a particular instruction reduces errors on a particular task, whether a particular program changes attendance, whether a particular warning changes a health behavior. The distinction is the goal of the study, not the nobility of the topic. Applied work uses the same four goals and the same three causal conditions. A program evaluation that cannot separate the program from the fact that motivated people signed up has a temporal story and a practical hope, and it does not yet have a cause.

The two kinds of work trade findings. A basic result about attention can become the reason an applied study measures attention rather than only measuring the final error. An applied failure can force a basic question that the laboratory version had defined too narrowly. What does not transfer automatically is the conclusion. A process shown with college students, a short session, and a mild manipulation is a finding about that arrangement until someone shows it again under the conditions the application cares about. Generalization is a later chapter because it is a separate argument, with its own evidence.

Program evaluation is applied research aimed at a decision: keep the program, change it, or stop it. The decision-maker often wants a causal answer on a deadline. The method does not become optional because the deadline is real. A description of how many people enrolled is useful and is not an effect. A comparison that lacks a comparison group is a story about time passing. The evaluator’s obligation is to match the strength of the verb to the design that was actually run.

A scientific claim is one that is tied to a procedure and exposed to evidence that could count against it. Falsifiability is the name for that exposure. A sentence that survives every possible observation is not a finding. It is a slogan. “People behave as they do because of many factors” cannot fail. “Under this instruction, errors on this task drop relative to this comparison, and the drop is absent when the instruction is removed” can fail. The second sentence is the kind of object this course studies. Popper’s point, as methods courses use it, is not that scientists enjoy being wrong. It is that a claim with no path to being wrong has left the empirical conversation.

Skepticism in this sense is a rule about evidence, not a personality. The skeptic asks which goal the sentence is attempting, which of the three causal conditions is actually in the design, and what observation would have made the authors withdraw the sentence. Authority is a bad substitute for that check. A famous researcher, a prestigious journal, and a repeated talking point can all be mistaken, and they can also be right. The way to tell is the procedure and the results, which is why the literature search in the next chapter is a method, not a ceremony.

Peer review is one public filter: other researchers read a report before a journal prints it and try to find the failures of design, analysis, and inference. Peer review is not a guarantee. Reviewers miss things, journals prefer some findings over others, and a published paper can still be wrong. The value of review is that the report had to be explicit enough for a skeptical reader to try. A claim that circulates only as a talk, a video, or a private certainty has not yet passed through that filter. It can still be the idea that starts a study. It is not yet the study.

Intuition remains a source of questions. People are good at noticing patterns in their own lives and bad at knowing which rival explanation they have failed to imagine. The scientific attitude keeps the question and replaces the private certainty with a procedure. That is the whole shift this chapter asks for. The next chapter is about where the questions come from and how the existing literature is used so that a new study is an addition rather than a repeat of a paper the researcher never read.