Whenever I present comparative results between the NFS and CPS, data will be for currently married women aged between 15 and 59 to make
TABLE 3.1: PERCENTAGE DISTRIBUTION OF CURRENTLY MARRIED WOMEN AGED 15-49 BY TYPES OF QUESTION USED
3. A male child in almost all agrarian societies, (where infant deaths are very frequent), is considered as a
3.4 RESPONSE ERRORS RELATING TO CONTRACEPTION
Almost all of the sample surveys which collected information regarding knowledge and practice of contraception were not accompanied by evaluation of their reliability, indicating that the survey planners ignored the importance of response error. The lack of knowledge on reliability of these data raises doubts. As Mauldin(1965:98) argues, '.... in the sensitive area of fertility behavior, will people tell about their behavior, and will they tell the truth? How useful are such surveys?'. Hauser(1967:403) pointed out that KAP Surveys neglected to include adequate efforts to study the reliability and validity of their data. Hauser added that these surveys ignored the literature on response errors.
In the sixties and seventies, a few researchers attempted to assess the quality of their data in terms of reliability and validity. Some of the most notable were Westoff, Potter and Sagi(1961:52-69), Freedman and Takeshita(1969:405), Brackbill(1974: 261-266), Mukherjee(1975:127-142), Knodel and Piampiti(1977:55-66), Coombs(1977:218-232) and Ryder(1979:115-128). Ihe Indonesia Fertility Survey vhich was carried out in cooperation with the WFS was accompained by a reliability survey which, however, did not include contraception aspects (MacDonald et jil., 1978). Under the WFS programme, the Response Errors Project conducted a study to (1) provide guidance to users of substantive data; (2) provide the basis of improved survey design; and (3) satisfy methodological research objectives (O 'Muircheartaigh,1982). Ihe study was designed to be carried out in three stages: main interview, re-interview and reconciliation interview and was conducted in Lesotho, Peru, Turkey and the Dominican Republic. Ihe report is yet to come out.
The response reliability studies mentioned above were based on the test-retest design. Ihis type of study has several drawbacks. Knodel et al.(1977:55) noted that the long period between interviews increased the chances of genuine change occurring in attitudes and in some behavioural characteristics, such as contraceptive use. It may also increase the chances of sample loss due to death or migration. When there is a very short interval between interviews, the respondents may refuse to co-operate. Another problem is the error that occurs in recording, editing, coding, punching and matching records from each round, and also from interviewer bias.
In short, the finding of these studies was that there was greater reliability for factual data than for data relating to knowledge of contraceptive methods. Mukherjee(1975:140) concluded, based on his findings, that unreliability of responses to attitudinal questions was not due to (1) complex and ambiguous items, (2) inconsistent items, (3) mechanical errors in reading, marking and scoring, (4) inadequate and non-uniform instructions, and (5) changes in attitudes. The possible reasons suggested were situational factors, particularly (1) the effect of the interviewer, (2) the presence of other people at the interview, (3) the nature and wording of the questions asked, (4) the degree of rapport established before the interview, (5) the willingness on the part of the interviwee to continue the interview, and (6) response according to perception of the interviewer's attitude and v iews.
A group of anthropologists conducted a survey[12] a few months after the NFS. The authors concluded, based on their study, that the knowledge of contraception was grossly underreported. They argued that the underreporting was due to linguistic unintelligibility, i.e. the language used in the NFS was primarily a highly literate variety, and to the circumstances of the interview: a stranger (interviewer) asking sensitive questions in the presence of the respondents' kinsmen
[12] Their primary aim was to cross-check the validity and accuracy of survey research results. A questionnaire was compiled that consisted of questions from four different major Nepalese Surveys, i.e. surveys on household economics, health and family planning. They were administered to 76 people in three different villages with which authors were very familiar. Cross-checking was carried out with in depth interview and based on authors' past knowledge on particular individuals. Tests of the intelligibility of sample questions and vocabulary items were conducted among 64 respondents.
and neighbours (Campbell et al., 1979:8).
In the CPS, every effort was made to improve the language of questions by conducting a pre-test and a language seminar attended by linguists. Interviewers were drawn from various places in the country so that there would be at least one local person in a team.
In the following analysis, the CPS(1981) information concerning the knowledge and use of family planning methods will be assessed based on the effects of situational variables. More specifically, level of knowledge and use of contraception will be compared by the sex of interviewers, the extent of reliability and degree of co-operativeness as seen by interviewers, and presence of other persons during the interview.
3.4.1 RESPONSE RELIABILITY
Every interviewer was asked to scale the overall co-operativeness shown by respondents during the interview. Interviewers were also asked to give their opinion concerning the reliability of the respondent1s responses with respect to knowledge and use of contraception. Scales run from one to nine. One indicates that a response is unreliable or the respondent is unco-operative, nine indicates that a response is reliable or the respondent is co-operative during the interview. I imagine that the level of co-operativeness or reliability as seen by individual interviewers may also be subject to a chance effect. More specifically, interviewers may mark three instead of four or vice versa without it meaning any real difference. If we try to differentiate reliability level between three or four, we may not get meaningful information. However, the
broad scale will give some meaningful information. Th u s , I will discuss my findings based on two broad categories i.e. below average
(one to four) and above average (five to nine) .
Table 3.5 indicates that more than four-fifths of the respondents are above average in the overall co-operativeness scale in all three geographical regions. However, there are six percent more respondents in the Hills categorized within 'above average' in the overall co-operativeness scale than in the other regions. It is hypothesized
TABLE 3.5: PERCENTAGE DISTRIBUTION OF RESPONDENTS BY THE LEVEL