INTRODUCTION

The total error in a survey can be subdivided into two categories: a) sampling error and b) non-sampling error. The first type of error arises, as explained earlier, due to estimation that is based on a sample (not all the units) which is selected according to a pre-specified sampling design. The errors occurring due to any other reasons, including that which can arise at the time of implementing a sampling design and cause a sample estimate to deviate from the true population parameter, fall in the domain of the non-sampling error.

Unlike the sampling error, which can be estimated when a probability sampling is employed, it is difficult to estimate the extent of non-sampling error in a survey. That is, once it occurs, it is difficult to handle it. But there are ways to avoid the possibility of its occurrence. Also, for certain specific type of non-sampling errors, it is possible to make suitable adjustments at the time of estimation to minimise the effect of such errors. It is suggested that appropriate precautions should be taken at every stage, from the beginning to the end of a survey so as to avoid, as far as possible, the chance of occurrence of non-sampling errors.

This chapter focuses on the areas where the possibility of occurrence of non-sampling error is high, and enumerates how they are likely to occur and the steps that can be taken to minimise their occurrence. There are three areas, during the implementation of a survey design, where the possibility of such errors to occur is maximum. They are:

1. (i) During sample implementation: Error can occur if a sample as selected by a proposed design is not implemented in the field accordingly. This is referred to as the coverage error. Although occurring at a time of sample implementation, the resultant error is not included in sampling error but creates a non-sampling error.

2. (ii) During data collection: Once a sample is selected, responses need to be obtained from the selected individuals. Two types of error can occur at this stage. The first relates to the inability to obtain a correct response from a respondent. This is known as a response error. The other refers to the failure to obtain a response. Two different situations can lead to such a failure. […]