Am J Epidemiol. 1984 Nov; 120(5): 777-87.

Loglinear modeling with inexpensive computing equipment.

Koepsell TD.

Loglinear models are finding increasing application in the analysis of data from epidemiologic studies and increasing attention in statistics courses taken by epidemiologists in training. This paper describes a program for microcomputers, written in BASIC, which fits hierarchic loglinear models to categoric data organized into multiway contingency tables of any dimensionality. For each model specified by the user, the program can be directed to calculate and display 1) two goodness-of-fit statistics and their corresponding degrees of freedom and p values; 2) expected cell counts under the fitted model; 3) maximum likelihood estimates of the model's terms and, for saturated models, their standard errors; and 4) for models involving a dichotomous response variable, the corresponding logistic regression model. The program can be run on most microcomputers with at least 16K bytes of random access memory, including some costing less than $100.