Generalizability and representativeness: Considerations for internet-based research on substance use behaviors.

Substance use is frequently studied using nonprobability internet-based samples. It is difficult to evaluate the utility of these samples without a clear understanding of two key concepts: generalizability and representativeness. Part 1 of this article (a) offers a particular viewpoint on the distinctions and relations between these two concepts, (b) suggests that purposive (i.e., nonprobability) samples, when used carefully, can be used to construct valid scientific generalizations, and (c) explores some analytical consequences of sampling decisions that change sample heterogeneity. Part 2 of this article explores the overlap between internet-based sampling of substance use behaviors and the concepts discussed in Part 1. Specifically, Part 2 reviews relevant literature and presents example analyses of an internet-based cannabis use data set to highlight (a) strengths and weaknesses of internet-based sampling and (b) how unique elements of a given online platform (e.g., primary motive for visiting the platform) and the substance being studied (e.g., degree of societal stigma) might inform the types of boundaries, caveats, qualifiers, and limitations that are incorporated into a generalization crafted based on the data. (PsycInfo Database Record (c) 2022 APA, all rights reserved).