Psychology of Learning

Module 02: Research Methods

Part 2: Research Variables & Experimental Control

Looking Back

In Part 1, we established the scientific foundations of learning research, exploring how science differs from other ways of knowing through its emphasis on empiricism and self-correction. We examined the goals of science (description, explanation, prediction, and control), the assumptions underlying scientific psychology (determinism, discoverability, and parsimony), and surveyed classic research equipment. Now we turn to the logic of experimental research: How do researchers isolate the effects of specific variables while controlling for everything else? What makes an experiment convincing evidence of cause and effect?

The Goal: Causal Inference

The ultimate goal of experimental research is causal inference—determining that changes in one variable cause changes in another variable. Simply observing that two variables are related (correlated) isn’t enough. Correlation does not imply causation, as you’ve probably heard many times. But why not? And how do experiments allow us to make causal claims when correlational studies don’t? (Shadish, Cook, & Campbell, 2002)

The answer lies in experimental control. By systematically manipulating one variable while holding everything else constant, researchers can isolate that variable’s causal effect. This requires carefully identifying and managing different types of variables in the research design.

Independent & Dependent Variables

The independent variable (IV) is the stimulus or aspect of the environment that the experimenter manipulates to determine its influence on behavior. It’s called “independent” because the researcher controls it independently—deciding what values or conditions to create (Kerlinger & Lee, 2000).

The dependent variable (DV) is the response or behavior that the experimenter measures. It’s called “dependent” because we expect it to depend on the independent variable. When the IV has a significant effect, changes in the DV are directly related to the manipulation of the IV (Kerlinger & Lee, 2000).

A simple way to remember the distinction: Changes in the IV cause changes in the DV. The IV is what you manipulate; the DV is what you measure. In learning research, the IV is typically some aspect of the learning procedure or environment, and the DV is some measure of learning or performance.

A Concrete Example: Caffeine & Learning

Let’s make this concrete with an example. Suppose Nancy, a student researcher, wants to find out how caffeine affects learning in rats. (This hypothetical example is based on actual research by Cathey, Smith, & Davis, 1993.) In experimental terminology, caffeine would be the independent variable, and learning would be the dependent variable.

Why is caffeine the IV and learning the DV rather than vice versa? Because Nancy manipulates caffeine (some rats get it, others don’t) and measures learning (how quickly rats acquire a new behavior). She’s asking: Does manipulating caffeine (IV) cause changes in learning (DV)?

Operational Definitions

Operational definitions specify how variables are produced and measured in a particular study. They define concepts in terms of the specific operations used to manipulate or measure them (Bridgman, 1927). Operational definitions are crucial for replication and clear communication of research methods.

For Nancy’s caffeine study, she needs operational definitions for both variables:

Caffeine (IV): Nancy operationally defines caffeine as .17 mg/ml (.017%) of caffeine in drinking water throughout gestation, weaning, and early adulthood (the time of testing). Note the specificity—not just “caffeine” but exactly what concentration, how delivered, and for how long.

Learning (DV): Nancy operationally defines learning as how long it takes rats to learn to bar press. Further, she operationally defines “learn to bar press” as a rat making 10 bar presses within a 30-second period. Again, complete specificity eliminates ambiguity.

Without operational definitions, terms like “caffeine” and “learning” remain vague. Different researchers might use different concentrations, delivery methods, or learning criteria, making results incomparable. Operational definitions ensure everyone knows exactly what was done (Underwood, 1957).

Extraneous Variables: The Problem of Confounds

Extraneous variables are undesired variables that may operate to influence the DV and, thus, potentially invalidate an experiment. In Nancy’s study, any variable that might affect learning other than caffeine is an extraneous variable. Examples include age, hunger, previous learning experiences, time of day, housing conditions, and many others (Campbell & Stanley, 1963).

The experimenter’s task is to eliminate or control extraneous variables. Why? Because if an extraneous variable affects the DV, we can’t be certain whether changes in the DV resulted from the IV or from the extraneous variable. This creates ambiguity about causation.

The Deadly Threat: Confounding Variables

A confounding variable (or confound) is an extraneous variable that varies systematically with the IV. Confounds are far worse than ordinary extraneous variables—they destroy causal inference completely (Shadish, Cook, & Campbell, 2002).

Here’s why confounds are “deadly” to experiments: Imagine Nancy tested her experimental group (caffeine rats) in the morning and her control group (no-caffeine rats) in the afternoon. Now time of day is confounded with caffeine. If the groups differ in learning, we have no way to know whether the difference resulted from caffeine, time of day, or both. The IV (caffeine) and the confound (time of day) are hopelessly entangled.

Whenever a confound is present, causal inference becomes impossible. We lose the ability to make statements about causality. This is why researchers are obsessive about eliminating confounds through proper experimental design (Campbell & Stanley, 1963).

Control Techniques: Random Assignment

Random assignment is the most powerful technique for controlling extraneous variables. It means each participant has an equal chance of being assigned to any group in the experiment. This control technique ensures that groups are equivalent at the start of the experiment (Fisher, 1935).

Random assignment doesn’t eliminate extraneous variables—rats in Nancy’s study will still vary in age, previous experience, and other characteristics. But random assignment ensures these variables are distributed equally across groups. The caffeine group and no-caffeine group will have similar average ages, similar experiences, etc. Any differences that exist are due to chance, not systematic bias.

This is why random assignment is so powerful: it controls for variables we know about and those we don’t. Even if Nancy forgot to consider some extraneous variable (say, genetic differences between rats), random assignment will distribute those differences roughly equally across groups (Fisher, 1935).

Experimental & Control Groups

In a two-group design, the experimental group receives the treatment being studied (in Nancy’s case, caffeine), while the control group does not receive the treatment. The control group provides a baseline for comparison—what happens without the treatment? (Campbell & Stanley, 1963)

Nancy decides to have one group of rats that receives caffeine (experimental group) and one group that does not receive caffeine (control group). This is the simplest experimental design: two groups, randomly assigned, differing only in whether they receive the treatment.

Because Nancy is interested in caffeine’s effects before and after gestation, she needs to expose rats to caffeine before they’re born. She gets four female rats and randomly assigns two to the experimental group and two to the control group. She supplies them with the appropriate water and then breeds the females. When the rat pups are born, she’ll have two litters in the experimental group and two litters in the control group.

Independent Groups vs. Related Groups

Independent groups are groups of research participants formed by random assignment. Each participant appears in only one group, and participants in different groups are unrelated (Fisher, 1935).

Related groups are groups of research participants that are related in some way—through repeated measures, matched pairs, or natural pairs. These designs have advantages for certain research questions but require different statistical analyses (Keppel & Wickens, 2004).

Repeated measures: The same participants are tested in multiple conditions. For example, testing each rat’s learning with caffeine and without caffeine (in counterbalanced order). This controls for individual differences perfectly—each participant serves as their own control.

Matched pairs: Participants are measured and equated on some variable before the experiment, then one member of each pair is assigned to each group. For example, measuring each rat’s baseline activity level, then assigning matched rats to different groups.

Natural pairs: Participants who are naturally related (siblings, littermates). Nancy’s design doesn’t use natural pairs because all pups from each litter are in the same group.

Nancy’s design uses independent groups. Because she randomly assigned the mother rats to conditions, the pups are essentially randomly assigned too. The two groups aren’t related through repeated measures, matching, or natural pairs.

Levels of the Independent Variable

Levels are the differing amounts or types of an IV used in an experiment (also called treatment conditions). Nancy’s first experiment has two levels: caffeine versus no caffeine. But researchers often use more than two levels to understand dose-response relationships or compare multiple conditions (Keppel & Wickens, 2004).

After running her first experiment, Nancy finds no effect of caffeine on learning. She’s surprised because background research suggested caffeine increases arousal and task performance. Reviewing the literature, she notices researchers have used varying caffeine concentrations. Perhaps her concentration (.017%) was too low to affect learning?

Nancy decides to conduct a second experiment using multiple caffeine concentrations. She’ll compare three levels: .017% (her original concentration), .034% (twice as much), and .05% (three times as much). Now her IV has three levels instead of two. This multiple-group design can reveal whether caffeine effects depend on dosage.

Internal & External Validity

Internal validity refers to the extent to which an experiment demonstrates a causal relationship between the IV and DV, free from confounds and alternative explanations. High internal validity means we’re confident the IV caused the DV changes (Campbell & Stanley, 1963).

Threats to internal validity include confounding variables, selection bias (non-random assignment), history (events occurring during the study), maturation (developmental changes), testing effects (practice from repeated measurement), instrumentation (measurement changes), and regression to the mean (Campbell & Stanley, 1963). Good experimental design minimizes these threats.

External validity refers to the extent to which research findings generalize beyond the specific study—to other populations, settings, and times. High external validity means the findings apply broadly, not just to the specific circumstances tested (Campbell & Stanley, 1963).

Often there’s tension between internal and external validity. Highly controlled laboratory studies maximize internal validity but may sacrifice external validity. Field studies in natural settings increase external validity but often reduce control, potentially threatening internal validity. Researchers must balance these competing goals (Berkowitz & Donnerstein, 1982).

Additional Research Equipment

Beyond the equipment covered in Part 1, learning researchers use several other specialized apparatus:

Lashley Jumping Stand: Developed by Karl Lashley to study discrimination learning. Rats stand on a platform facing two stimulus cards (e.g., circle vs. square). Jumping at the correct stimulus opens a door to food; jumping at the incorrect stimulus results in bumping into a locked door and falling into a net. This apparatus measures how quickly animals learn to discriminate between stimuli (Lashley, 1930).

Motor Skills Apparatus: Mirror-tracing apparatus requires participants to trace designs while viewing their hand only in a mirror—a challenging motor coordination task. The pursuit rotor requires maintaining contact between a handheld stylus and a small moving target on a rotating disk. These apparatus measure motor skill acquisition and demonstrate learning curves (Ammons, 1947).

Verbal Learning Equipment: Hermann Ebbinghaus pioneered the study of human memory using nonsense syllables (CVC trigrams: consonant-vowel-consonant combinations like “ZOK” or “BEM”). These stimuli minimize previous learning, allowing study of “pure” memory processes. The memory drum presents stimuli one at a time through a window at controlled rates, enabling systematic study of memorization (Ebbinghaus, 1885/1913).

Looking Forward

We’ve covered the fundamental concepts of research variables and experimental control—independent and dependent variables, the distinction between extraneous and confounding variables, and techniques like random assignment that allow causal inference. In Part 3, we’ll examine specific experimental designs in detail—two-group designs, multiple-group designs, and factorial designs—along with considerations about sample size, power, and effect size.