May 17, 2000 Sampling Why sample? Types of sampling methods Probability Non-probability Sampling distributions Purposes of Today s Class Define generalizability and its relation to different sampling strategies Define basic elements of sampling planning Distinguish major types of sampling strategies Explain concept of sampling distribution Evaluate quality of samples Why Sample? Everyday errors in reasoning Generalizability Sample generalizability Cross-population generalizability Why not take a census? Efficiency Validity When is sampling unnecessary? Basic Terminology Elements units of analysis
Population aggregate of elements that forms focus of study From which we sample To which we generalize Sample subset of population selected for study Generalizability What can we say about those we didn t study? Representativeness Sample Quality Sampling error any difference between characteristics of sample and population Key questions to evaluate: What is the population? How was the sample drawn? Is the sample representative? Types of Sampling Probability No systematic bias Laws of chance known probabilities Nonprobability Bias unknown Decreased generalizability Appropriate under some circumstances
Probability Sampling Sampling error due to chance Key factors Sample size matters Homogeneity of population matters Proportion of population doesn t matter Lessons About Sample Quality How well is the population defined? How were cases selected from population? Unbiased Depend on chance How was the sample actually obtained? Nonresponse Are the conclusions limited to the original population? Probability Sampling Simple random sampling Strictly chance Sampling frame
Systematic random sampling Sampling interval Periodicity Stratified random sampling Ensures subpopulations represented Sampling frame divided on key independent variables Proportionate v. Disproportionate Cluster sampling No sampling frame required Useful for dispersed populations Cluster natural grouping of elements Multistage Sampling Distributions Theoretical distribution of a statistic across infinite number of samples Each sample is one of infinite possibilities Value of each statistic varies from sample to sample Mean of many samples approaches true mean
Normal distribution Predictable proportion of cases in certain ranges Statistical inference Sample Size Amount of sampling error Larger samples have more compact distributions Heterogeneity of population Number of subgoups Independent variables Strength of relationships among variables Nonprobability Sampling Availability sampling Haphazard Quota sampling Predefined proportions of subpopulations Purposive sampling Seek elements for particular needs Snowball sampling Social networks, hard-to-find populations
Uses of Nonprobability Sampling When probability sampling impossible Field conditions Document the bias When focus on cultural data, not population parameters Individual attributes require probability sampling Cultural data require experts, key informants Combining Methods (More PR) Exploratory Purposive cluster sampling Maximize heterogeneity Cultural data Explanatory Multistage probability cluster sampling Census blocks > households > individuals Individual data