Designing Studies: every key term you need (+ practice quiz)
53 flashcard terms for AP Statistics Unit 1, written to match the course framework. Read them here, drill them as flashcards, or take the 29-question quiz. Free, no account needed.
Random assignment to treatment/control. Eliminates bias, balances confounding variables, enables causal inference.
Replication
Repeat experiment with new subjects. Confirms findings, identifies if effect reproducible and generalizable.
Response Bias
Respondents answer inaccurately: social desirability, misunderstanding. Reduced with careful wording, anonymity.
Non-response Bias
People who don't respond differ from responders. High non-response can severely distort results.
Measurement Error
Difference between true value and measured value. From instrument error, observer error, rounding.
Validity
Does instrument measure what it claims? Valid test measures the construct of interest accurately.
Reliability
Does instrument measure consistently? Reliable test gives same result when repeated (if trait unchanged).
Unit 1 Key Ideas
Distinguish population/sample, parameter/statistic, descriptive/inferential, observational/experimental. Sampling methods and biases determine data quality. Causation requires randomization.
Undercoverage
Some groups in the population are left out of the sampling frame entirely (e.g., landline phone surveys miss cell-only households). A form of selection bias that no sample size can fix.
Voluntary Response Sample
People choose themselves, usually those with strong opinions. Online polls and call-in surveys are classic examples; results overstate extreme views.
Wording Effect
Leading, loaded, or confusing question wording shifts responses. A source of response bias distinct from who is sampled.
Strata vs Clusters
Strata are internally homogeneous groups, sample some units from EVERY stratum. Clusters are internally heterogeneous mini-populations, sample ALL units in SOME clusters.
Multistage Sampling
Combining methods in stages, e.g., randomly choose school districts (clusters), then stratify by grade within chosen districts, then SRS students.
Group experimental units into blocks of similar units (by a variable expected to affect the response), then randomize treatments within each block. Reduces variability so treatment effects are easier to detect.
Matched Pairs Design
Special block design with blocks of size 2 (or one subject receiving both treatments in random order). Each pair's difference removes subject-to-subject variability.
Completely Randomized Design
All experimental units are assigned to treatments purely at random with no blocking. Simplest design; relies on randomization alone to balance lurking variables.
Control Group
Group receiving no treatment, a placebo, or the standard treatment. Provides a baseline so the treatment effect can be separated from other changes over time.
Factor and Levels
A factor is an explanatory variable manipulated in an experiment; its levels are the specific values used. Treatments are the combinations of factor levels.
Statistically Significant
An observed effect so large it would rarely occur by chance alone under random assignment. Says nothing about practical importance.
Lack of Realism
Experimental conditions or subjects that don't resemble the real setting limit how far results generalize, even when the experiment is internally valid.
Scope of Inference
Random selection allows generalization to the population; random assignment allows causal claims. Only both together give both.
Experimental Units vs Subjects
Experimental units are the smallest entities to which treatments are randomly assigned; when they are people they are called subjects.
Double-Blind Necessity
When the response is subjective (pain, mood, ratings), the person measuring the response must be blinded too, or their expectations bias the measurements.
Census
Attempts to measure every unit of the population. Expensive, slow, and still subject to nonresponse and measurement error, so it is not automatically better than a good sample.
Random Digit Table Use
Label units with equal-length numbers, read the table in consecutive groups of that many digits, skip repeats and out-of-range labels, stop when the sample is filled.
Nonresponse vs Voluntary Response
Nonresponse: chosen units fail to answer. Voluntary response: units choose themselves into the sample. Both bias results, but they arise at different steps.
Bias vs Variability of a Design
Bias is systematic error in a fixed direction; variability is scatter across repeated samples. Larger samples reduce variability but do NOT reduce bias.
Ethical Requirements
Institutional review board approval, informed consent, and confidentiality of individual data are required for studies with human subjects.
Retrospective vs Prospective
Retrospective observational studies look back at existing records; prospective studies follow subjects forward in time. Both are observational, neither proves causation.
Placebo vs Active Control
When withholding treatment is unethical, compare the new treatment with the current standard treatment instead of a placebo.