Collecting data and study design
Sampling methods, experiment structure, and the bias terms that decide whether a conclusion generalizes.
38 terms
BlindingBlinding keeps people in an experiment from knowing which treatment a subject received, so their expectations cannot bias the response or its measurement.BlockingBlocking groups similar experimental units together before random assignment, then randomizes the treatments within each block to sharpen the comparison.CensusA census collects data from every individual in a population, so the value it produces is the parameter itself rather than an estimate of it.Cluster sampleA cluster sample divides the population into groups called clusters, randomly selects whole clusters, and includes every individual in the chosen clusters.Completely randomized designIn a completely randomized design, all experimental units are assigned to treatments entirely at random, with no blocking or prior grouping.Confounding variableA confounding variable is associated with both the explanatory variable and the response, so its effect and the explanatory variable's cannot be told apart.Control groupA control group is the group in an experiment that supplies the baseline for comparison, receiving no treatment, a placebo, or the current standard treatment.Convenience sampleA convenience sample includes whichever individuals are easiest to reach, rather than selecting them through any random process.Double-blindA double-blind experiment hides the treatment assignment from both the subjects and the people who interact with them or measure the response.ExperimentAn experiment imposes treatments on subjects and compares their responses, supporting a cause-and-effect conclusion when treatments are assigned at random.Experimental unitAn experimental unit is the individual or object a treatment is assigned to; when the units are people they are called subjects or participants.Factor (in an experiment)A factor is an explanatory variable in an experiment whose categories, called levels, are imposed on the experimental units by the researcher.GeneralizabilityGeneralizability is whether a study's results extend to a larger population, and it depends on how the sample was selected, not on random assignment.Matched pairs designA matched pairs design compares two treatments within pairs of similar units, or one unit measured twice, then analyzes the difference inside each pair.Multistage sampleA multistage sample is selected in two or more stages, sampling large groups first and then sampling units within the groups that were chosen.Nonresponse biasNonresponse bias occurs when people chosen for the sample do not respond, and those who do respond differ systematically from those who do not.Observational studyAn observational study measures individuals without assigning treatments, so it can reveal associations but cannot by itself establish cause and effect.ParameterA parameter is a fixed numerical value that describes a feature of an entire population, such as its true mean or proportion.PlaceboA placebo is an inactive treatment given so that the control group's experience matches the treatment group's in every respect except the active ingredient.PopulationA population is the entire group of individuals or objects you want to study and draw conclusions about.Question wording biasQuestion wording bias occurs when a confusing or leading survey question pushes responses away from the true value in one direction.Random assignmentRandom assignment lets a chance device decide which experimental unit receives which treatment, which is what licenses a cause-and-effect conclusion.Random digit tableA random digit table is a printed list in which every position is equally likely to hold any digit 0 through 9, independently of every other position.Randomized block designA randomized block design first groups experimental units into blocks that are alike, then randomly assigns every treatment within each block.ReplicationReplication means assigning each treatment to more than one experimental unit, so treatment effects can be told apart from unit-to-unit variation.Response biasResponse bias occurs when respondents give systematically inaccurate answers, for example due to confusing question wording or pressure to answer a certain way.SampleA sample is the subset of a population that you actually collect data from in order to estimate something about the whole population.Sampling frameA sampling frame is the actual list of individuals a sample is drawn from; when it misses part of the population, undercoverage results.Sampling variabilitySampling variability is the natural variation in a statistic from one random sample to the next, and it is variation rather than bias.Scope of inferenceScope of inference is how far a study's conclusions reach: random assignment permits causal claims, and random selection permits claims about the population.Selection biasSelection bias is bias created by how the sample was chosen, leaving the individuals studied systematically different from the population.Simple random sampleA simple random sample is selected so that every possible sample of the given size has an equal chance of being the one chosen.StatisticA statistic is a numerical value computed from sample data, used to estimate a corresponding population parameter.Stratified random sampleA stratified random sample splits the population into similar groups called strata, then takes a separate simple random sample from each stratum.Systematic sampleA systematic sample orders the population, picks a random starting point, then selects every kth individual from that point onward.TreatmentA treatment is the specific condition applied to an experimental unit, made up of one level of each factor whose effect the experiment compares.UndercoverageUndercoverage occurs when the list a sample is drawn from leaves out part of the population, so those individuals have no chance of being selected.Voluntary response sampleA voluntary response sample is made up of people who choose to respond on their own, such as to an open online poll or a call-in survey.