Abstract
Gary Koch has contributed significantly to two major consolidations of statistical practice: the convergence of statistical modeling of discrete and continuous data, and the incorporation of statistical modeling into survey data analysis. This historical review places his 1965–1985 work on categorical data methods in the context of its time and the two other major themes of his research in this period: random effects and mixed model analyses, and regression modeling of sample survey data. Consideration of mixed models and survey data analysis both require a close examination of sources of error in data, from which Koch shaped a pragmatic, highly practitioner and situationally oriented view of the appropriate scope and use of statistical inference in science, along lines of Fisher (1925, 1955) and Cox (1977). This view has informed hundreds of students, many of whom assumed prominent statistical roles in biomedical science.