Can \(p\)-values be meaningfully interpreted without random sampling? (Q2191897)

The main focus of the paper is on `\(p\)-values' and inductive statistical inference relating to estimation and testing of hypotheses about unknown population parameters based on samples drawn from the population. In economics and social sciences, on certain occasions, random sampling is not adhered to and non random selection using observational methods such as convenience sampling, purposive sampling, voluntary selection among others is followed. The authors note that the assumptions and conditions for using \(p\)-values for inference are not usually satisfied and caution that fallacious conclusions may result. They consider, in detail, simple random sampling, stratified sampling and cluster sampling techniques and suggest the adjustments for calculating meaningful standard errors and \(p\)-values. The use of propensity scores is also discussed. Furthermore, they also study the situations leading to a 100\% sampling such as analyzing aggregate economic data. Finally, the authors refer to a host of recent publications criticizing the use of \(p\)-values incorrectly. There is a long list of useful references at the end of the paper.