Study design
A population based cross-sectional study design was conducted using data analysis from the Ethiopia demographic health survey (EDHS) 2016. EDHS data obtained from the nine regions and two administrative cities were used. The data was collection from January 18-June 27, 2016 (26).
Study area
The study was conducted in Ethiopia (30-40N and 330-480E, situated at the eastern tip of Africa which is located at the horn of Africa and one of the tenth largest countries in Africa. The projections for the 2007 population and housing census estimate the population of nation 108,805,142 in 2018. In the country, there are nine regional states and two city administrations subdivided into 68 zones, 817 districts and 16,253 kebeles. In Ethiopia majority of the population (83.6%) are living in rural areas and the average household size is 4.7 persons. In addition, women in the reproductive ages constitute 24% of the population and 7,685 health posts, 392 hospitals and 3,962 health centers have been giving health care services. In all health facilities, FP service is provided at least five days a week (27, 28).
Data source and extraction
The data for this analysis were extracted from EDHS 2016 and accessed from the Measure DHS website (http://www.dhs program.com). It is a secondary data analysis from nationwide community-based survey. The data sets were downloaded in SPSS format with permission from Measure DHS website (http://www.dhs program.com). Data cleaning and recording were carried out in STATA. The family planning related datasets were joined to Global Positioning System (GPS) coordinates of EDHS using the joining variable as recommended by DHS measure. In the DHS surveys, samples were selected using a stratified, two-stage cluster design, using enumeration areas (EAs) as a primary sampling units and households as the secondary sampling units(26).
Outcome variable
Modern contraceptive utilization is outcome variable for this study. Modern contraceptive use defined as current use of a modern method (pill, intrauterine device, injections, male condom, female sterilization, periodic abstinence, withdrawal, implants/Norplant, lactation amenorrhea, emergency contraception, standard day’s method). It derived from variable recorded as “Current use by method type” in DHS which categorize as “no method”, “using modern method” and “using traditional method”. In this study it categorized as a binary outcome of use of modern method was assigned “yes” (coded as 1) and not using modern method including traditional method was coded as “No” (coded as 0).
Explanatory variables
The predictor variable for modern contraceptive utilization include both individual and community level characteristics.
Individual level variable
Individual level variable include age (coded as 15-19 and 20-24), age of cohabitation (recode as<18, ≥18),education (coded as no education, primary, secondary and higher), wealth quintile (coded as poorest, poorer, middle, richer and richest ), religion (recoded as Orthodox ,Muslim, Protestant and other) ,number of living children (recoded as 0-1, 2-3 and >3), partner education(coded as no education, primary, secondary and higher), respondent working status (code as yes and no) and partner occupation (recode as no, service, agriculture ,manual and other). Knowledge about family planning , heard family planning message which defend as respondent heard about family planning (radio, TV or newspaper/magazine), heard FP in community events ,field workers visit in last 12 month and health facility visit within 12 months (coded as no and yes), perception of distance from health facility (code as no big problem and big program), desire more children (coded as have another ,undecided and no more ), husband desire more children(coded as both want same, husband wants more, husband wants fewer, don’t know) and decisions maker for family planning (recoded as mainly respondents, mainly husband and Joint decisions)
Community level variable
Influence of community level factor on reproductive health particularly on contraceptive behavior was examined by previous studies (18, 29, 30). Community level variables were generated by aggregating the individual characteristics in a cluster since EDHS did not collect data that can directly describe the characteristics of the clusters except place of residence and region. The aggregates were computed using proportion of a given variables subcategory within given cluster. Since the aggregate value for all generated variable was not normally distributed. It was categorized into groups based on the national median values.
Community level of poverty was created by aggregate individual wealth index by cluster. We categorize women living in cluster having less than 50% poor women as low poverty cluster and cluster having greater than 50% of poor women as high poverty cluster.
Community level of health service utilization: the proportion of health service utilization generate by aggregate women who had visited health facility in the past 12 months by cluster. Cluster having <50% visited health facility as low health service utilization and cluster having >50% visited health service as high utilization.
Community distance to a health facility generated by aggregate those report distance to a health facility was a big problem to get medical help, and categorized as big problem and no big problem.
Community women education level was created by aggregate women educational level by cluster. Cluster having less than 50% of educated women categorize as low educated cluster and cluster having greater than 50% of educated women as high educated cluster.
Community mass media exposure was generated by aggregate expose to one of mass media (radio, TV or magazine). Cluster more than 50 % respondent expose to mass media was taken as high exposed cluster otherwise it taken as low expose cluster.
Place of residence (coded as rural and urban) and region that included nine regions and two administrative cities (code as Adiss Ababa, Tigray, Afar, Amhara, Oromia, Somali, Benishangul-Gumuz, SNNPR, Gambela, Harari, Addis Ababa, Dire Dawa) were already considers as community level factor in DHS.
Sample size determination and sampling procedures
Ethiopia demographic and health survey 2016 was done by selecting a total of 18,008 households for sample, of which 17,067 were occupied. Of occupied, 16,650 were successfully interviewed, yielding a response rate of 98%. The total household size was 16,650 and from these 16,583 eligible women was identified for individual interviews. The interview completed with 15,583 women yields a response rate of 95%. 15, 583 women aged between 15-49 years completed the interview and 10,223 (weighted sample) were married women. Among 10,223 (weighted sample) married women ,2298 young married women (weighted sample) age between 15-24 years were included in this study (26). A two-stage samples technique was employed. The stratified based on geographic region and urban/rural areas. In the first stage of selection, the primary sampling units (PSUs) were selected with probability proportional to size (PPS) within each stratum. The PSU forms the survey cluster a total of 645 EAs (202 in urban areas and 443 in rural areas). Then fixed number of 28 households (25-30) per cluster was selected with an equal probability systematic selection from the newly created household listing in the second stage of survey. The overall probability of selection of a household was differed from cluster to cluster (26).
Population and outcome measurement
In EDHS, Women aged 15 to 49 were randomly selected enumeration areas (EAs) were eligible for family planning as part of this those all young married women age 15-24 years were included in this study. Modern contraceptive utilization categorized as modern contraceptive utilization and none modern contraceptive utilization.
Data management, data processing and analysis methods
Sampling weight was applied to an individual interview unit of analysis to adjust for differences in probability of selection and interview between cases in a sample due to design, happenstance or corrections for differential response rates. Since EDHS samples were not self-weighted it need weighted with the available sample weight factor (v005/1,000,000) within the EDHS dataset to minimize the effect of sampling bias. All of these special codes were careful considered when analyzing DHS datasets
Data cleaning done by managing missing values in general were coded 99, 999, 9999, etc. Missing value was managed by excluded missing value from both the numerator and the denominator, by calculate in separate category for percent distribution and leave as it if data is not affected in some variable (26). When the level of missing is relatively large >10 %, the variable should be exclude from regression. However, the bias is always found to be negligible if missing value was less than <10%. The data extraction, descriptive and summary statistics were done by STATA 14. Spatial statics was analyzed by ArcGIS version 10.3 and Sat Scan™ version 9.6 software.
Spatial analysis of unmet need for family planning
Spatial autocorrelation
Moran's I is one of spatial autocorrelation method that assess the extent of clustering of modern contraceptive utilization in the regions. Moran's I test statistic computed to test the null hypothesis, no significant clustering of modern contraceptive utilization in the entire study region (31).
Getis OrdGi* statistic (Hot spot analysis)
Hotspot statistic was computed to measure how spatial autocorrelation varies over the study location by calculating Gi* statistics for each area. The Z-score is computed to determine the statistical significance clustering of modern contraceptive utilization, and the p-value computed for the significance. If the z-score is between −1.96 and +1.96, the p-value would be larger than 0.05, and could not reject the null hypothesis; the pattern exhibited could very likely be the result of random spatial processes. If the z-score falls outside the range, the observed spatial pattern is probably too unusual to be the result of random chance, and the p-value would be small to reflect this. Therefore, it is possible to reject the null hypothesis and proceed with figuring out what might be causing the statistically significant spatial pattern in the data. So high Gi* indicates “hotspot” whereas low Gi* means a “cold spot” (32, 33).
Spatial scan statistic
Spatial scan statistic is based on Bernoulli model which applied by Kuldorff methods using the SaTScan™ software to analyze the purely spatial and clusters of modern contraceptive utilization. A Bernoulli-based model was used in which events at particular places analyzed if married women were modern contraceptive utilization or not represented by a 0/1. A spatial scan statistic used a scan window (the population at risk) in the shape of a circle, which moves across the study region. The size of the scan window was adjusted to scan for small clusters up to 50 %. It also used to examine a large number of distinct geographical windows to test for the presence of modern contraceptive utilization. For each window Monte Carlo simulation used to test the null hypothesis that there was no statistically cluster of modern contraceptive utilization cases within the window.
The cluster with the greatest maximum likelihood ratio was considered as the primary cluster of modern contraceptive utilization. Other statistically clusters that did not overlap with the primary cluster were identified as secondary clusters of modern contraceptive utilization, and ranked according to their likelihood ratio test statistic. (31, 34)
Multilevel logistic regression analysis
The determinants of contraceptive utilization were identified by using multilevel logistic regression model. Those variables with P-value < 0.2 in bi-variable logistic regression model were entered into multivariable logistic regression model to measure the effect of each variable after adjusting for the effect of other variables. Variables with p-value < 0.05 were considered as statistically significant to identify independent factors for modern contraceptive utilization. To demonstrate the importance of the community level and individual level component multilevel analysis was used candidate variables p-value less than 0.05. were entered into the model and will be checked fitted for models(35).
Multilevel analysis is appropriate to measure DHS data since it is hierarchical data therefore two stage multilevel analyses was used to explore factors affect modern contraceptive utilization at individual and community level factor (36). Four models were considered in the multilevel analysis; model one empty without explanatory variable that specified only the random intercept and it presents the total variance in modern contraceptive utilization among clusters, model two adjusted for individual variable, model three for adjusted community level variable and model four both adjusted individual and community level variable. The association was measured by odd ratio of individual level variable and community level to identify factors that associate with contraceptive utilization. Measurement of variation was identifying using interclass correlation (ICC) and proportional change in variance (PCV). The model for fitness diagnostics was select by using Deviance Information Criteria (DIC) or Alkaile information criteria (AIC). Model with lowest AIC and highest log likelihood test was selected which better explain modern contraceptive utilization (36, 37).
Multicollinearity checked by using Variance Inflation Factor (VIFs) to identify correlations between variables, and determines the strength of the relationships. It examined instability of effect size of predictors as the result of high collinearity among the factors. VIF of different covariates was also assessed with cutoff value <10. If VIF of variable was <10 it would be included in the analysis (35). .
Ethical consideration
The data was accessed by registration on the DHS website (www.dhsprogram.com) and getting approval from the measure DHS. Prior to the actual interview, informed consent was obtained from the participants, their guardian or household heads. Data was used only for the purpose of statistical reporting and analysis, and for the proposed research project. The data treated as confidential, and no effort should be made to identify any household or individual respondent interviewed in the survey. Ethical clearance was obtained from the institutional ethical review board of the Institute of Ethiopia public health association, Ethiopia.