“…Traditional contextual models are often doubly manifest (manifest variable, manifest aggregation; see doubly manifest M1 in Figure 1), controlling neither measurement error nor sampling error. Although unreliability has many potential components of error (as emphasized, e.g., in generalizability theory; see Kane, Gillmore, & Crooks, 1976), here we emphasize measurement error (associated with sampling items, a traditional focus of CFA/SEM studies) and sampling error (associated with sampling of persons with L2 groups).…”