Public use data files of the Survey of Income and Program Participation (SIPP) are expected to be used by many different researchers, particularly in the areas of Sociology, Economics and public policy. Possible uses of SIPP Micro-Data Files can be broadly classified into two major areas: (1) computation of summary statistics such as means, totals and ratios for states, regions, and subpopulations like Blacks, Hispanics, low income families, etc. (beyond what is published by the Bureau of the Census in SIPP reports), and computation of variances of such statistics for tests of hypotheses and statistical inferences, (2) analytical studies to understand various socioeconomic phenomena such as factors affecting the dependency on welfare, or variables affecting the risk of experiencing an event; e.g., marriage, child birth, divorce or unemployment. These analytical studies will generally involve some form of multivariate analysis and statistical modeling techniques. Data analysts can usually compute summary statistics from micro-data files easily. Estimation of variances and multivariate analysis of survey data are, however, not so easy. It is, therefore, desirable to provide guidelines to users on how to compute sampling errors and conduct a multivariate analysis or statistical modeling. The guidelines should also indicate software packages that are appropriate for computation of variances or multivariate analyses so that users can carry out these complex statistical computations relatively easily.

Exploratory research was undertaken to review general methods of variance estimation from complex surveys and multivariate analysis of survey data, and all available software packages with regard to their suitability for analysis of data from the complex sample design of the SIPP. This research mainly deals with cross - sectional analysis of data. No general purpose software packages are available for longitudinal analysis of data from complex surveys. This report provides guidelines for variance estimation. Multivariate analysis of SIPP data has been discussed in another report by Cbakrabarty (1989).

General methods of variance estimation from complex surveys with their advantages and disadvantages are discussed first. This is followed by a discussion of all available software packages with regard to their statistical methodology, capabilities, computational complexities, and most importantly, their suitability for computing variances from the complex sample design of the SIPP, along with some recommendations.