Appropriate Methods for Analyzing the Effect of Method Choice on Contraceptive Discontinuation

PDF document icon wp-01-34.pdf — PDF document, 273 kB (280,269 bytes)

Author(s): Steele F, Curtis Sian L

Year: 2001

Contraceptive discontinuation is a topic that has received increasing attention in recent years for several reasons. First, increased attention to the quality of family planning services has lead to interest in outcomes that might be associated with the quality of services. Jain (1989) argues that the quality of services increases contraceptive prevalence through increased adoption of contraceptives but more significantly through improved continuity of use. Although there is relatively little empirical evidence to date to support the theoretical link between the quality of services and contraceptive discontinuation, the interest in contraceptive discontinuation as an outcome associated with the quality of services remains. Second, as contraceptive use rises throughout the world, contraceptive continuation becomes an increasingly important determinant of contraceptive prevalence and unwanted fertility. For example, Blanc and Curtis (1999) find that in 15 countries with Demographic and Health Surveys (DHS) total unwanted fertility rates would be between 44 percent and 81 percent lower in the absence of contraceptive discontinuation and failure. Similarly, in Turkey 78 percent of induced abortions in the three years preceding the 1998 DHS survey were preceded by a contraceptive failure or discontinuation (Senlet et al. 2001). Finally, the increasing availability of contraceptive history data, such as that collected through the DHS program, has provided increased opportunities to study contraceptive discontinuation in more depth and across more countries. Studies of contraceptive discontinuation consistently show contraceptive discontinuation rates vary substantially by method used. Typically, method-specific discontinuation rates are lowest among IUD users and highest among users of condoms and, to a lesser extent, injectables (Ali and Cleland 1995; Blanc and Curtis 1999). Reasons for discontinuation also vary substantially by method; DHS surveys typically find side effects and health concerns are the main reasons for discontinuing hormonal methods and IUDs, inconvenience and partner's disapproval are the main reasons for discontinuing condom use, and contraceptive failure and desire for a more effective method are the main reasons for discontinuing traditional methods. Multivariate analyses of the determinants of contraceptive discontinuation find that the method used is strongly associated with contraceptive discontinuation after controlling for other factors (Curtis and Blanc 1997; Koenig et al. 1997; Steele, Diamond and Wang 1996; Steele and Diamond 1999), although the pattern by method depends on the type of discontinuation being studied. However, as Curtis and Blanc (1997) note, the relationship between the method used and contraceptive discontinuation is complicated by the fact that method choice is determined by women's balancing of a number of factors including ease of continuation, risk of failure, intended length of use, and other characteristics of the woman that might also affect her risk of discontinuing use. These factors could lead to selection of women at high risk of discontinuation to use certain methods, which in turn would lead to bias in the determinants of contraceptive discontinuation. In other words, method choice is potentially an endogenous variable. The direction of the selection effects resulting from method choice is difficult to predict. For example, a woman may choose to use the IUD because she intends to use for a long time and believes the IUD to be an easy method for long-term use because it requires no ongoing action on her part once it is inserted. Such selection may overstate the difference between the discontinuation rates of IUD and other methods. Alternatively, women who have difficulty continuing use of methods may choose the IUD again because of the lack of ongoing action required to continue use. Such selection would lead to understatement of the contraceptive method differentials. We are aware of no studies to date that have explicitly addressed the issue of the potential endogeneity of method choice in the discontinuation process. Most authors have either ignored the problem and used conventional modelling techniques (Curtis and Blanc 1997; Koenig et al. 1997; Pariani et al. 1991; Steele et al. 1996; Steele and Diamond 1999) or have fitted method-specific models (Ali and Cleland 1999; Ping 1995; Steele et al. 1999). The objective of this paper is to explore whether method choice is endogenous in the discontinuation process, and if so, to explore what effect that has on the estimates of method effects on discontinuation. To do this, we utilize multilevel multiprocess models to simultaneously model the processes of contraceptive method choice and contraceptive discontinuation using data from the 1997 Indonesia DHS. We focus on one particular type of contraceptive discontinuation, abandonment of contraceptive use while in need of contraception.