INTRODUCTION

This chapter presents case studies on both survey and evaluation designs. Two case studies, one on a sampling design, namely, that for the NFHS in India and another on an evaluation design conducted in Lucknow, India, in which the authors were involved, have been discussed in greater detail. The purpose has been to demonstrate the steps involved in implementing the designs in the field. In addition, sampling designs for a few other large-scale surveys have also been summarized. Similarly, one application for each of the different evaluation designs has been included. The studies are presented in two separate parts.

PART I: SAMPLE SURVEY DESIGNS

NATIONAL FAMILY HEALTH SURVEYS, INDIA (NFHS, INDIA)

NFHS are the Indian editions of the DHS conducted in over 70 countries around the world. Three rounds of NFHS have already been conducted in the country: 1992–93, 1998–99 and 2005–06. The coverage and parameters estimated in the three rounds of the survey have been summarized in Table 10.1.

SAMPLING DESIGN OF NFHS

The overall structure of NFHS design is similar to that adopted in other such surveys globally. A household is considered as the basic unit for sample selection. Since there was no viable frame available for selection of households directly, it was necessary to select a sample in multiple stages.

10.3.1 Sample size

A general approach was to have 4000 completed interviews with ever married women in a state with a population of 25 million or more, 3000 for those having a population between 2 and 25 million, and 1500 for states with population of 2 million or less. This is because a sample of 1500 households was assumed to be adequate in terms of providing estimates for most of the demographic and maternal and child health parameters being studied.

Allocation of sample size for each state was not in proportion to their population. The slightly higher sample size for the larger states was mainly to facilitate understanding the socio-economic differentials in a parameter. Having larger sample size for the bigger states also made sense because it reduces, to some extent, the banal effect of the disproportional allocation of sample while obtaining the national level estimates.

The estimated sample size was, however, revised whenever necessary. For example, HIV/ AIDS was estimated in NFHS-3, and it required a larger sample size than was considered in the earlier rounds.