NCJ Number
86079
Date Published
1981
Length
128 pages
Annotation
This research addressed the classic problem of inferring cell enries in contingency tables from marginals.
Abstract
The research centered on problems in which a given population is partitioned into three subgroups at two points in time, where the aggregate size of those subgroups is known at both times, but the movement of the elements is not known. The approach used in the research is based on the premise that to know what did happen in the quasi-experimental design expressible as a 3x3 table, the investigator must know what could have happened. The research extends the Davis and Duncan approach of finding extreme cell values from 2x2 tables to 3x3 tables and handles multioperative constraints by using a program that enumerates all possible solutions and examines each to identify the smallest largest values for each cell. By tallying all cell values, an empirically derived probability distribution underlying the cell that may incorporate inequality relations is mapped. The development of the suggested techniques is illustrated in papers that deal with (1) estimating juvenile recidivism by cross-level inference, (2) estimating cell entries in contingency tables, (3) estimating individual level change in public opinion from aggregate data, (4) estimating the degree of mobilization and conversion in the 1890's (the nature of political change in one critical election), and (5) evaluation with sparse nominal data (the case of differential compliance with the 55 mph limit). The concluding section provides an illustrative FORTRAN program embodying the research technique used in the research. For individual entries, see NCJ 86080-81.