In: Statistics and Probability
Describe 3 possible sources of non-sampling error in this survey
Following three Non-Sampling errors.
1) Specification error
Ans:- This occurs when the concept implied by the question is different from the underlying construct that should be measured. A simple question such as how many children does a person have can be subject to different interpretation in some cultures. In households with extended family member's biological children may not be distinguished from children of brothers or sisters in the same household. In a disability survey, a general question asking people whether or not they have a disability can be subject to different interpretations depending on the severity of the impairment or the respondent's perception of disability.
2) Coverage or Frame error
Ans:- In most area surveys primary sampling units comprise clusters of geographic units generally called enumeration areas (EAs). It is not uncommon that the demarcation of EAs is not properly carried out during census mapping. Thus households may be omitted or duplicated in the second stage frame. For example, in Zambia some urban areas were hurriedly sketched during the 2000 Population and Housing mapping exercise. When such urban area units are used, without updating, during the survey, exclusion of sample units in some EAs and duplication of units in other EAs are highly probable.
- Frame imperfections can bias the estimates in the following ways:- If units are not represented in the frame but should have been part of the frame, this results in zero probability of selection for those units committed from the frame. On the other hand if some units are duplicated, this results in over coverage with such units having larger probabilities of selection.
3) Nonresponse
Ans:- Nonresponse refers to the failure to measure some of the sample units. Thus failure to obtain observations on some units selected for the sample. It is instructive to think of the sample population as split into two strata, one consisting of all sample units for which measurements can be obtained and the second for which no measurement could be obtained.
In most cases, nonresponse is not evenly spread the sample units but is heavily concentrated among subgroups. As a result of differential nonresponse, the distribution of the achieved sample across the subgroups will deviate from that of the selected sample. This deviation is likely to give rise to nonresponse bias if the survey variables are also related to the subgroups.
This is the three possible sources of non-sampling errors.