Contraceptive Stories

We hope this brief analysis of results from our Cluster Prediction Model app helps spark investigation into a potential question of your own!

From previous studies on the demographic variables that determine contraceptive use, we know that having less living children (with the exception of having no children), being married, and having a higher education level are all variables generally associated with higher levels of contraceptive use (Lunani et al).

But contraceptive use is not always constant. Rather, it changes over time, and those changes are important to analyze and understand. We might wonder: what factors point to a woman who will start, discontinue, or maintain use of contraceptives? By looking at contraceptive use and efficacy (how effective a method is with perfect use) data over time in conjunction with demographic data, we can start to explore different trends and stories that may help answer that question.

Like Lunani et al, we use DHS data below, filtered by age range, to examine why Kenyan women who consistently don't use contraceptives or use low efficacy contraceptives rather than starting high efficacy contraceptives. This is an example of a potential analysis of the results from our Cluster Prediction Model app.

From Figure 1.1, we see the results of the clustering, or grouping, algorithm on the contraceptive use over time of Kenyan women from ages 25 to 30. Here, the red cluster A represents the typical contraceptive use over time of women who don't use contraceptives or use low efficacy contraceptives. The green cluster B represents women who consistently use high efficacy contraceptives, and the blue cluster C represents women who switch from not using contraceptives to starting a high efficacy method. Finally, the purple cluster D represents women who discontinue contraceptive use.

We are specifically interested in the demographic variables that set cluster A (consistent nonuse or low efficacy use) apart from cluster B (consistent high efficacy use) and the other clusters.

Let's first look at parity, or the number of times a woman has given birth. From the segmented bar plot in Figure 1.2, women that fit the profile of cluster A tend to have given birth more times by age 30 compared to women from other clusters. Also, the women in cluster A are notably more likely to have a parity of 2 or 3 (majority outcome) than women in the other three clusters. This contrast is especially marked for the women in cluster B, who tend to have given birth only once if at all. If we assume parity to be a proxy for living children, these findings agree with the inverse correlation between number of living children and contraceptive use shown by Lunani et al.

The next factor to examine is marital status. While we do see a greater percentage of women that fit the profile of cluster B are married at age 30 than for other clusters in Figure 1.3, which agrees again with the analysis of Lunani et al, the distribution of marital status is not that different between the clusters.

Instead, we can look at Figure 1.4 for the distribution of highest education level attained by cluster to paint a clearer picture. Here, we see that, while most other clusters have a similar distribution with roughly three-fourths having at least completed primary schooling by age 30, cluster A does not show that distribution. The women in cluster A are significantly more likely to have no education at all, with only one-fourth having at least completed primary schooling by age 30.

This analysis matches what we might expect. Women with little or no education may belong to socioeconomic classes among which there is not much knowledge of contraceptives or contraceptives are inaccessible for some other reason. When these women marry, the lack of contraceptive use may lead to a greater number of pregnancies and births.

Consider the exception mentioned in the study — women with no children were the least likely to use contraceptives. That seems counterintuitive based on our findings above, but if we look at Kenyan women from ages 15 to 20 years old, the data tells a surprising story.

Similar to above, we see in Figure 2.1 the results of the clustering algorithm on the contraceptive use over time of Kenyan women from ages 15 to 20. Here, the red cluster A again represents the typical contraceptive use over time of women who consistently don't use contraceptives or use low efficacy contraceptives. However, the meaning of the other clusters differs from before. Both the green cluster B and the blue cluster C represent women who switch from not using contraceptives to starting a high efficacy method. Lastly, the purple cluster D represents women who have relatively consistent high efficacy contraceptive use (despite the notably larger confidence intervals).

Based on these cluster profiles, we are specifically interested in the demographic variables that set cluster A (consistent nonuse or low efficacy use) apart from the other clusters.

Let's look at parity again. From the segmented bar plot in Figure 2.2, women that fit the profile of cluster A tend to have given birth fewer times by age 20 compared to women from other clusters. Also, the women in cluster A are notably more likely to have a parity of 0 (majority outcome) than women in the other three clusters. If we again assume parity to be a proxy for living children, these findings agree with the exception found by Lunani et al, where women with no living children have the lowest contraceptive use.

The next factor to examine is marital status. In Figure 2.3, we see that the women in cluster A are somewhat more likely to not be married or to have never been in a union at age 20 than the women from the other clusters. The difference is especially stark compared to women in cluster B, who are married or have already been married (divorced, widowed, etc.) much more often.

Lastly, we can look at Figure 2.4 for the distribution of highest education level attained by cluster. Here, we see that, while most clusters have generally similar distributions for education levels at or above primary schooling, a greater proportion of the women in cluster A have no education at all by age 20 than in other clusters. On the other hand, the proportion of women who have an education level above primary school is slightly greater in cluster A than in other clusters.

This analysis paints a slightly different picture than the first. While there is evidence for the story we traced earlier — that women whose lack of knowledge or access to contraceptives may result in less family planning ability and a greater number of pregnancies — in this data there is also support for a less clear story. For girls entering adulthood, we see that some may not use contraceptives simply because there is no need. They might stay in school longer and delay marriage; without a partner or the expectation of childbirth, they will be less likely to become pregnant.

By distinguishing among different demographic profiles for women and their associated contraceptive use over time, we hope to contribute to the understanding of who is at risk of discontinuing contraceptives and who might be open to starting contraceptives.

References:

Lunani, Laura L et al. “Prevalence and Factors Associated with Contraceptive Use Among Kenyan Women Aged 15-49

Years.” AIDS and behavior vol. 22, Suppl 1 (2018): 125-130. doi:10.1007/s10461-018-2203-5