ERIC Number: ED175874
Record Type: Non-Journal
Publication Date: 1975-Jan
Reference Count: N/A
Variance Components for NLS: Partitioning the Design Effect.
Folsom, Ralph E., Jr.
This memorandum demonstrates a variance components methodology for partitioning the overall design effect (D) for a ratio mean into stratification (S), unequal weighting (W), and clustering (C) effects, so that D = WSC. In section 2, a sample selection scheme modeled after the National Longitudinal Study of the High School Class of 1972 (NKS) sample is described. In section 3, variance components are defined which can be used to model the variance of an NLS sample total. A second set of components are defined which parallel the appropriate NLS components with the exception that all effects of unequal weighting are removed. That is, components are estimated as if schools were allocated to final strata strictly proportional to the actual student population per stratum: as if the actual student enrollments were used to assign school size measures; and finally, as if there was no nonresponse. In section 4, sample mean squares are presented for estimating the components defined in section 3. Section 4 translates the components definitions developed for sample totals in to the associated components for a domain mean. In the final section, various functions of variance components and sample size parameters are presented for estimating the effects W,S, and C. (Author/CP)
Publication Type: Reports - Research
Education Level: N/A
Authoring Institution: Research Triangle Inst., Durham, NC.
Identifiers - Assessments and Surveys: National Longitudinal Study of the High School Class of 1972