SAMPLING & NON-SAMPLING ERRORS

1. SAMPLING ERROR

The difference between the estimate derived from the sample (i.e. the statistic) and the true population value (i.e. the parameter) is technically called the sampling error. For example,
Sampling error = X -U
Sampling error arises due to the fact that a sample cannot exactly represent the pop, even if it is drawn in a correct manner

2. NON-SAMPLING ERROR

Besides sampling errors, there are certain errors which are not attributable to sampling but arise in the process of data
collection, even if a complete count is carried out.
Main sources of non sampling errors are:
· The defect in the sampling frame.
· Faulty reporting of facts due to personal preferences.
· Negligence or indifference of the investigators
· Non-response to mail questionnaires.
These (non-sampling) errors can be avoided through
· Following up the non-response,
· Proper training of the investigators.
· Correct manipulation of the collected information,
Let us now consider exactly what is meant by ‘sampling error’: We can say that there are two types of non-response ---
partial non-response and total non-response. ‘Partial non-response’ implies that the respondent refuses to answer some of the questions. On the other hand, ‘total non-response’ implies that the respondent refuses to answer any of the questions. Of course, the problem of late returns and non-response of the kind that I have just mentioned occurs in the case of HUMAN populations. Although refusal of sample units to cooperate is encountered in interview surveys, it is far more of a problem in mail surveys. It is not uncommon to find the response rate to mail questionnaires as low as 15 or 20%.The provision of INFORMATION ABOUT THE PURPOSE OF THE SURVEY helps in stimulating interest, thus increasing the chances of greater response. Particularly if it can be shown that the work will be to the ADVANTAGE of the respondent IN THE LONG RUN.
Similarly, the respondent will be encouraged to reply if a pre-paid and addressed ENVELOPE is sent out with the questionnaire. But in spite of these ways of reducing non-response, we are bound to have some amount of non-response.
Hence, a decision has to be taken about how many RECALLS should be made.
The term ‘recall’ implies that we approach the respondent more than once in order to persuade him to respond to our queries.
Another point worth considering is:
How long the process of data collection should be continued? Obviously, no such process can be carried out for an indefinite period of time! In fact, the longer the time period over which the survey is conducted, the greater will be the potential VARIATIONS in attitudes and opinions of the respondents. Hence, a well-defined cut-off date generally needs to be established. Let us now look at the various ways in which we can select a sample from our population. We begin by looking at the difference between non-random and RANDOM sampling. First of all, what do we mean by nonrandom sampling?