Back to browse results
Correcting Estimates of HIV Prevalence Due to Survey Non-Participation in India Using Heckman Selection Model
Authors: Shrikant Singh, Swati Srivastava, and Ashish Kumar Upadhyay
Source: Demography India, 44(1&2): 17-30
Topic(s): HIV/AIDS
Non-response bias
Country: Asia
Published: JAN 2015
Abstract: Using the data from the third round of National Family Health Survey and Heckman Selection Model this paper aims to determine the estimates of HIV prevalence in India due to survey non-participation. Interviewer ID was taken as the selection variable, which affects the survey participation but did not affect HIV status independently. Study also compared the estimates of Heckman selection model to conventional imputation model. It has been found that prevalence of HIV is greater among men (0.77; 95% CI= (0.71-0.83)) and women (0.42; 95% CI= (0.39-0.45)), who did not participate in the survey as compare to those who participated in HIV test (0.35 for men & 0.22 for women). Thus, the national estimate for men and women derived from selection model was higher than the unadjusted imputation method. Results of this study demonstrate that the selection variable was significantly associated with the HIV status of the men and women. Further, this study shows the significant association between the survey participation and the HIV status of those who has been interviewed but did not consent to the HIV test, which clarifies that the sample selection led to substantial underestimation of the national HIV prevalence in men and women. Therefore, a valid and efficient way to provide the estimate of HIV prevalence is to incorporate the Heckman selection model instead of the conventional method to provide an estimate of the national prevalence.