Questions: Between-Subjects Design Implementation and Assignment

Question 1

A researcher studies the effect of sleep deprivation on memory. She matches participants into pairs based on their baseline memory scores, then assigns one member of each pair to the sleep-deprived condition and the other to the control. A colleague argues they should use simple random assignment instead. Why is the colleague correct?

• A. Matching on baseline memory automatically improves statistical power beyond what randomization can achieve
• B. Random assignment controls for all confounding variables simultaneously — including ones the researcher hasn't measured — while matching only controls for the variables explicitly matched on
• C. Matching is only valid in within-subjects designs; between-subjects designs prohibit it
• D. The colleague is wrong; matching is always superior to random assignment for controlling confounds

Answer: (B)

Matching controls for variables you know about and think to measure (here, baseline memory). But it leaves uncontrolled all the other variables that differ between participants — motivation, sleep history, health, personality, etc. Random assignment, by the law of large numbers, equates groups on ALL variables simultaneously, including those the researcher hasn't thought to measure. This is the unique power of randomization: it turns a designed experiment into a causal inference engine.

Question 2

Why do between-subjects designs typically require more participants than within-subjects designs to achieve the same statistical power?

• A. Between-subjects designs use less efficient statistical tests that require larger samples
• B. Between-subjects designs require a separate control group, which doubles the required sample size
• C. Individual differences between participants add error variance to group comparisons, making it harder to detect real treatment effects without a larger sample
• D. Between-subjects designs measure each participant only once, which always reduces reliability

Answer: (C)

In a between-subjects design, the groups differ not only because of the treatment but also because people differ from one another in baseline performance, personality, and countless other variables. This between-person variability is part of the error variance in the statistical test, making real treatment effects harder to distinguish from noise. Within-subjects designs eliminate this source of variance by having each participant serve as their own control. The solution in between-subjects designs is a larger sample, which dilutes individual differences.

Question 3

Random assignment to conditions in a between-subjects experiment controls for confounding variables even when those variables were not measured or anticipated by the researcher.

• T. True
• F. False

Answer: True

This is the central advantage of random assignment. Because participants are assigned to conditions by chance, any variable that might correlate with the outcome — measured or unmeasured, anticipated or not — is equally likely to end up distributed across both groups. Over a large sample, this ensures group equivalence on all characteristics simultaneously. This is why true experiments (with random assignment) permit causal inference in a way that observational studies and quasi-experiments cannot.

Question 4

Between-subjects designs are inherently weaker than within-subjects designs in terms of internal validity when both are properly conducted.

• T. True
• F. False

Answer: False

Internal validity — the degree to which observed differences can be attributed to the independent variable — is determined primarily by random assignment, not by design type. A properly randomized between-subjects design has strong internal validity. Within-subjects designs have different threats (carryover effects, practice effects, order effects) that can compromise internal validity in their own way. The tradeoff is in statistical power and sample size, not in internal validity. Neither design is inherently superior on that dimension.

Question 5

Explain why random assignment is considered more powerful than matching participants on key variables, even when matching is done carefully on variables known to correlate with the outcome.

Model answer: Matching controls only for the variables the researcher explicitly identifies and measures. But countless unmeasured variables — motivation, personality, genetics, health, mood — also affect outcomes. Random assignment, because it assigns by chance, equates groups on ALL such variables simultaneously through probability, including ones the researcher hasn't thought to measure or couldn't measure. Matching also risks creating inadvertent systematic differences on unmatched variables. Random assignment is therefore a more comprehensive and assumption-free method of achieving group equivalence.

The intuition is that matching is a targeted tool (it fixes specific known problems) while randomization is a universal tool (it addresses all problems at once, known and unknown). In practice, researchers sometimes combine them — using stratified randomization to ensure balance on the most important variables while relying on probability to handle the rest.