From the total of 5,651 reproductive age (15-49years) women population extracted from the EDHS 2016 dataset, 44.11% had an intention to use contraceptive methods. The mean number of children ever born to women in the survey was 4.5±2.9, while the average age at first cohabitation was 16.9±3.99, and the mean ideal number of children reported 4.77±2.0. We used a weighted frequency for the included women at the community level, and the larger numbers were from Oromo (43.78%), Amhara (19.51%), and 19.09% SNNP regions. Greater than three-fourth (88.22%) of women were from rural-based residences. At the individual level, 2/3rd (68%) of women were illiterate (no education), and 1/4th learned at least primary school. Due to this, 18.34%, 11.09%, & 2.36% of them only heard family planning information on radio, TV, and newspaper/magazines in sequence. Their exposure to service providers was 26.14%, and 43.66% were only visited health facilities or met with community health workers respectively in the last 12 months. Indeed only 12.32% of them have ever terminated a pregnancy. We also described women in terms of other socio-demographic variables and found Muslims and Orthodox Christians as larger religions, 27.68% currently working women, 45.61% & 34.48% poor and rich women respectively; 51.61% of women live with illiterate husbands and age group with higher women were 25-29, 30-34, & 35-39 and 19.98%, 19.49%, & 17% respectively; (Table 1).
We analyzed the spatial data based on the coordinates captured during the data collection. There we found significant clustering that went well with the statistical analyses we applied. We presented the results in table 2 and fig. 1-4. We applied hotspot and cold spot analysis to show areas with higher and lower future use (fig.1), Ordinary Kriging interpolation applied if these areas are more closely (fig.2), and spatial autocorrelation if the closeness is significant (fig.3). And SaTscan statistics for the number of clusters (fig.4) applied to each of them held significantly. Depending on table 2, there were five most likely clusters observed, among which three (cluster 1-3) showed significant association with intention to use contraceptive methods. Cluster one included Gambella, Benishangul, Addis Ababa, and some parts of SNNP and Amhara at (10.298371 N, 34.649187 E) coordinates with 588.81 km radius, cluster two contained the whole Tigray, and almost full Amhara regions at (12.669915 N, 36.775082 E) with the radius of 335.28 km, and cluster three contained the whole SNNPR, some part of Oromia, Gambella, Addis, and Benishangule at (6.934084 N, 36.520510 E) with the radius of 308.29 km. and relative risk of clusters(1-3)1.88, 1.83, & 1.50 respectively; (table 2 and fig.4).
Individual and Group Level Analyses
Since we used the data collected on various clusters, we decided to check for the clustering effect among sampling units. It showed dissimilarities, and a larger ICC confirmed variation among clusters. We handled these variations via multilevel logistic regression analyses. We built four models to account for the inter-cluster differences(Table 3 & 4). The initial (null) model was the model without predictors (intercept only model) followed by the I-III model for individual and community level predictors. At individual level, compared to age group 15-19, women in age group of 25-29,30-34,35-39,40-44, 45-49 had 58%, 72%, 90%, 97%, & 99% reduced intention to use contraceptive methods with AOR of 0.42(0.3-0.7), 0.28(0.17-0.5), 0.1(0.05-0.18), 0.03(0.02-0.06), & (0.010(0.004-0.02) respectively. Reproductive age women with primary education had a 1.4 likelihood contraceptive use intention with an AOR of 1.4(1.04-1.8) relative to none educated women. As the number of ever born children increased, the women showed intention for contraceptive use, which increased with an AOR of 1.1(1-1.2); otherwise, women with a larger ideal number of children had 11% reduced intention to use contraceptive methods with an AOR of 0.89(0.84-0.94). The women who obtained information to use contraceptives at health facilities showed 1.6 times more intention with an AOR of 1.6(1.3-21). As age at first cohabitation increased, intention to use contraceptive methods raised 1.04 times with an AOR of 1.04(1-1.06).
At the community level, compared to the women in Tigray, women in Afar, Oromia, Somali, Benishangul, SNNPR, Harari, Addis Ababa, and Dire Dawa had 89%, 61%, 96%, 58%, 64%, 85%, 66%, & 80% reduced intention to use modern contraceptive methods with AOR of 0.11(0.06-0.2), 0.39(0.23-0.64), 0.04(0.02-0.09), 0.42(0.25-0.72), 0.36(0.2-0.67), 0.15(0.08-0.28), 0.34(0.2-0.62), & 0.2(0.11-0.36) respectively; (Table 3).
Initially, the null model had 28% intention of contraceptive use appeared only due to the variation among clusters and the remaining left for individual. We sequentially developed models to handle such variations, the inter-cluster variation reduced by 1/4th to be only (7%). The PCV for intention to use contraceptive methods in the initial model escalated to 2.9 times more variable due to the clustering effect. We described the full modeling procedures to handle variation in table 4 below. The variance, ICC, media odds ratio, and deviance decreased, while log-likelihood ratio and proportional change in variances increased and showed good fitness.