Within-Subject Designs

Within-subjects design

A design measuring the same participants on a dependent variable multiple times.

Pretest–posttest design

Measuring an outcome before and after an intervention or manipulation to assess change.

Repeated measures design

Exposing the same participants to multiple levels of an IV and measuring outcomes after each exposure.

Longitudinal within-subjects design

Measuring the same people across time to assess within-person change.

Primary reason for within-subjects designs

To directly assess change within people rather than infer change from differences between separate groups.

Within-subjects correlational research

Repeated measurements can assess change and relationships over time without experimentally manipulating variables.

Efficient hypothesis testing

One sample participates in all conditions, reducing the number of separate participants needed compared with equivalent between-subjects comparisons.

Participants as their own controls

Each participant is compared with themselves, directly accounting for stable individual differences.

Power advantage of within-subjects designs

Controlling individual differences reduces error and generally improves the ability to detect effects.

Dependence of within-subjects data

Repeated scores from the same person share information and cannot be treated as unrelated observations.

Main threats in within-subjects designs

Learning, testing, maturation, attrition, and order effects can influence later measurements.

Learning effects

Earlier exposure to a manipulation or study procedure changes responses to later exposures.

Learning-effect example

If Therapy A already reduces depression, Therapy B may appear ineffective because little depression remains to improve.

Testing effect

Being measured changes later responses to the same or related measures.

Maturation effects

Changes over time caused by processes other than the manipulation.

Attrition effects

Participants leaving a study affect the remaining sample, available data, or conclusions.

Attrition in within-subjects designs

Dropout can remove a participant's complete set of observations across conditions; bias is especially concerning when dropout is nonrandom.

Order effects

Changes in responses caused by the sequence in which measures or manipulations are presented.

Practice effects

Changes in behavior or responses due to increasing familiarity with the measures or tasks.

Fatigue effects

Changes or measurement error caused by participants becoming tired or bored during the study.

Carryover effects

Effects of an earlier manipulation persist and influence responses to a later manipulation.

Sensitization effects

Exposure to study materials leads participants to guess the research hypotheses, potentially changing their behavior.

Practice versus sensitization

Practice involves familiarity with the task; sensitization involves figuring out or guessing the study's purpose.

Practice versus carryover

Practice results from task familiarity; carryover results from a previous manipulation continuing to affect a later condition.

Oversampling

Recruiting more participants than the final target sample to compensate for expected attrition.

Planning oversampling

Use dropout rates from similar studies to estimate the initial sample needed.

Oversampling calculation

Initial sample needed = desired final sample ÷ expected retention proportion; 50% retention requires twice the target sample and 25% retention requires four times.

Limitation of oversampling

It helps maintain sample size and power but does not by itself eliminate bias caused by nonrandom dropout.

Control groups in within-subjects research

Comparable participants measured across the same period without the focal intervention help assess background change and differences in dropout.

Control groups and maturation

If an untreated comparison group also changes, some observed change may reflect time-related processes rather than the intervention.

Control groups and attrition

Different dropout rates across groups may indicate that attrition depends on the manipulation or other study features.

Counterbalancing

Presenting measures or manipulations in different orders across participants; full counterbalancing uses every possible order.

Purpose of counterbalancing

To distribute order effects across conditions and allow investigation of whether presentation order changes results.

Number of counterbalancing orders

For n distinct conditions, full counterbalancing requires n! orders; four conditions produce 24 orders.

Limitation of full counterbalancing

The number of possible sequences increases rapidly, requiring more participants and greater logistical effort.

Selective counterbalancing

Counterbalance the most important measures or manipulations when balancing everything is impractical; other measures may still show order effects.

Large samples and order differences

High power can detect very small order differences that may have little practical importance.

Latin square design

A design arranging n conditions across n sequences so that each condition appears once in every presentation position.

Latin square example

For four conditions, sequences ABCD, DABC, CDAB, and BCDA place each condition once in each position.

Latin square versus full counterbalancing

A Latin square balances positions using fewer sequences; full counterbalancing includes all possible sequences.

Latin square limitation

Balancing position does not necessarily balance all sequence-specific or carryover effects.

Effects assessed with Latin square designs

Effects of the manipulation, presentation position, and potentially their interaction.

Analyzing dependent data

Use analyses that explicitly account for repeated observations from the same person rather than ordinary independent-samples tests.

Difference score

A score representing change, commonly calculated as Time 2 minus Time 1.

Independent-samples t-test

Compares two groups of unrelated observations, such as separate treatment and control groups.

Paired-samples t-test

Compares two related sets of scores, such as pretest and posttest measurements from the same participants.

Dependent-samples t-test

Another name for a paired-samples t-test.

Logic of the paired-samples t-test

Calculate each participant's difference score and test whether the population mean difference is zero.

Independence of difference scores

Each person contributes one difference score; these scores can be independent across participants even though each person's original measurements are related.

Repeated measures ANOVA

An analysis testing for differences across related sets of scores from multiple conditions or time points.

Within-subjects ANOVA

Another name for repeated measures ANOVA.

Significant repeated measures ANOVA

Provides evidence that at least one condition or time point differs; planned contrasts or post-hoc comparisons identify where.

Choosing a within-subjects analysis

Use a paired-samples t-test for two related sets of scores; repeated measures ANOVA can compare multiple related conditions or time points.