Independent and dependent variables

A variable is any property of a person, a situation, or a behavior that can vary and that a study assigns values to. Age varies. Whether a person was given the instruction varies. The number of errors varies. A constant does not. If every participant in a study is a first-year student at one college, “year in school” is not a variable in that study, however important it is in life. The study cannot say what year in school does, because year in school never changed. Naming the variables is the first act of design, because the sentence the study will be allowed to utter is a sentence about those variables and no others.

The previous posted chapters separated a construct from its operational definition. This chapter uses that distinction and does not reteach it. Stress is a construct. A score on a particular scale, a heart-rate change, or a count of sick days is an operational definition. When this chapter says “the independent variable,” it means the variable as it was actually manipulated or selected in the study, which is always some operationalization. Two studies that both claim to manipulate stress and that use different operations may be manipulating different things. The variable name on the slide is not the procedure.

Variables also differ in the kind of value they take. A categorical variable sorts cases into kinds: condition A or B, yes or no, diagnosis present or absent. A quantitative variable assigns an ordered numerical amount: a count, a duration, a rating treated as a scale. The kind of value constrains the picture and the later statistic. It does not, by itself, say which variable is the cause. That job belongs to the role the variable plays in the design.

The independent variable is the variable the researcher treats as the proposed cause. In an experiment the researcher manipulates it: participants are assigned to one level or another. A level is one value of that variable. A study of sleep with conditions of 4 hours, 6 hours, and 8 hours has one independent variable with three levels. The dependent variable is the measured outcome, the variable that is predicted to depend on the independent variable. Errors, reaction time, a symptom score, a choice: if it is the behavior the researcher is trying to explain or to move, it is the dependent variable in that study.

A participant variable (also called a subject variable) is a property of the person that the researcher measures or selects rather than assigns: age, gender, a personality score, a diagnosis, years of training. Participant variables are often placed in the same sentence as independent variables, and the grammar hides a difference in license. The researcher did not assign the person to be forty. Groups that differ on a participant variable may differ on a crowd of other things that came along with it. A comparison of “experts and novices” can show a difference. It cannot, by itself, show that the training caused the difference, because the people who sought the training were not a random half of one pool. Cozby and Bates mark this by reserving the clean causal reading for manipulated variables.

In nonexperimental work the language slips. Researchers sometimes call a predictor the independent variable even when it was not manipulated. The slip is understandable and it is expensive. This course will call a manipulated proposed cause an independent variable, a measured proposed cause a predictor, and the outcome a dependent variable in both cases. The words then keep the design visible. A reader who sees “predictor” knows the time order and the alternatives may still be open. A reader who sees “independent variable” in an experiment knows assignment was the researcher’s act.

A situational variable is a feature of the setting: noise, instructions, the presence of another person, the wording of a question, the time of day if the researcher sets it. A response variable is a feature of the behavior: a choice, a latency, a rating, a physiological change recorded as an outcome. Most simple experiments are a situational independent variable and a response dependent variable. The sentence is then “this setting difference was followed by this behavior difference.” That sentence is still not automatic causation. It is the shape of the variables. Confounds, taken up later, are other situational variables that changed at the same time.

A mediating variable is the proposed process through which an independent variable affects a dependent variable. Sleep loss raises errors because attention lapses. Attention, if measured, is the mediator in that claim. A mediator is not a third topic. It is a claim about the path. Showing that the independent variable changes the mediator, and that the mediator accounts for part of the change in the outcome, is how a study moves from “cause” toward “explanation,” which was the fourth goal in the opening chapter. Naming a mediator in the introduction without measuring it leaves the explanation as a story beside the result.

A moderating variable is a variable that changes the size or direction of the effect. The instruction helps novices and does nothing for experts. Expertise moderates the effect of the instruction. A moderator is not a mediator. The mediator is the path. The moderator is the condition under which the path is strong, weak, or reversed. Mixing the two words is how a discussion section becomes fog. If the effect depends on who the participant is, the paper has a moderator. If the effect happens because something in the middle changed, the paper is claiming a mediator, and it owes a measure of that middle.

A study can have more than one independent variable. That is a factorial design, and it has its own chapter. A study can have more than one dependent variable. Each additional outcome is another claim, and another opportunity to notice a difference by chance. The methods point in this chapter is modest and prior: write the list before the session. Every variable that will be used in a conclusion has a name, a role, and an operational definition. Variables that were collected “just in case” and then searched until one of them moved are not innocent extra description. They change the meaning of the result. The inference chapter will say why. The cure starts here, with a list that was fixed when the question was fixed.

Control variables are the ones the researcher holds constant so they cannot be the alternative explanation. If every session is run in the same room, by the same script, at the same time of day, those features are constants. They cannot explain a difference between conditions, because they did not differ. They also cannot be generalized. A result from one room is a result from one room until someone varies the room on purpose. Holding a variable constant buys internal clarity and spends external reach. That trade is honest only if the paper says what was held still.

The exercise this chapter asks of a reader is grammatical and strict. Take a published sentence that says one thing affected another. Underline the proposed cause and write whether it was assigned or only measured. Underline the outcome and write the operation that produced the score. If a third word is doing secret work (“especially for women,” “because they were anxious”), label it moderator or mediator and ask whether it was measured. A paper that cannot survive that underlining does not yet have variables. It has a narrative that hopes the variables are obvious. They are not obvious. They are the study.