Is Contraceptive Prevalence Declining in India?

This study investigates the quality of the National Family Health Survey-4 data on contraceptive use by estimating investigator-induced bias. An outlier-bound approach was used to detect investigator bias, and contraceptive use was re-estimated adjusting for the bias in six study states. The findings suggest investigator bias at two levels: over-reporting of women as “never users” of contraception and a tendency to report sterilised women as current non-users and as cases of hysterectomy. Re-estimation of contraceptive use confirmed a declining trend in contraceptive prevalence in four study states. While the effect of the bias was moderate at the state level, it can potentially distort district-level estimates to a great extent.

Among the four rounds of India’s National Family Health Survey (NFHS), equivalent to the demographic and health surveys conducted around the world, the last round conducted in 2015–16 was unlike the others in terms of its scale of operation and complexity. The popularity and credibility of the NFHS in providing state-level estimates on a variety of demographic and health issues in the earlier rounds led to an increase in the expectations from the future rounds of the survey. In a large country like India, with 30 states and five union territories (as of 2016), the sample size covered in the third round of the survey had crossed 1,00,000 households. For NFHS-4, the demand was even higher to provide district-level estimates of the basic demographic parameters. In effect, the sample size to realise this demand across 640 districts surged to more than 6,00,000 households—a sixfold increase from the previous round (IIPS and ICF 2017). With such a multifold increase in sample size, the obvious concern was about the increased non-sampling errors. Another aspect equally important in influencing non-sampling errors was the length of the questionnaire. An examination of the questionnaire used in successive rounds of the survey revealed that the number of issues covered in the survey had been expanding, though the count of the exact number of questions asked may not have increased significantly (Rajan and James 2008).

However, these two aspects—introducing a number of questions for a fresh area of inquiry and extending an ongoing one with the same number of questions—are not exactly comparable even in terms of the time and quality of engagement required with a respondent. The role of a respondent, who shares their experiences, and an interviewer, who facilitates extracting this information are of the utmost importance in maintaining the quality of data in any survey. As both appear rather tangentially to this entire endeavour, they should at least be “comfortable,” if not “happy” with their participation. The less we recognise this, greater the possibility of introducing bias in the data.