Study Design, and Population
Cross-sectional data from Performance Monitoring for Action Ethiopia (PMA Ethiopia) which was collected in the month of Nov 2021 survey were used for this analysis. PMA Ethiop collects nationally representative cross sectional and longitudinal data on maternal, new born health along with Family planning, women empowerment/norms and other relevant reproductive health indicators. It was carried out by Addis Ababa University's School of Public Health in collaboration with the Ethiopian Public Health Association and Ethiopian Statistical services with financial support from Melinda Gates Institute for Population and technical support from Reproductive Health (Johns Hopkins Bloomberg School of Public Health), JHSPH.
Women in reproductive age groups who are married and/or cohabitated and who responded to the female questionnaire constitute the study's source population. The analysis is restricted to women who are currently using and/or recently (within the last two years) used modern contraceptive methods. A total of 2446 (weighted count) women were included in this analysis.
Sample size and Selection Techniques
The overall weighted count for this analysis 2446 was found adequate to generate reasonably unbiased estimate. Cell sample size adequacy was checked using chi square test.
The PMA-Ethiopia cross sectional and longitudinal survey was conducted among all women aged 15 and 49 residing in the selected households between. PMA Ethiopia is using a two-stage stratified sampling. A complete census was carried out in the chosen enumeration areas, after which 35 households per enumeration area were chosen using random number generator. After the household survey, all women of reproductive age were questioned.
Under the framework for the 2019 Ethiopia Population and Housing Census (PHC), which was conducted by the Ethiopia Central Statistics Agency, the primary sample units, or enumeration areas (EAs) were selected. As observations are picked using a method other than simple random sampling, the sample data is neither uniform nor randomly distributed. This method, commonly referred to as complex survey sampling, involves a range of selection probabilities at several stages. Each person's weight is inversely correlated with the chance of selection. Estimate makes use of sampling weights that are included with the survey data rather than a straightforward random sample weight. Using independent selection in each sample stratum and a probability proportional to EA size, a total of 243 EAs were selected in the first stage. In the second round of selection, 35 HHs per cluster were randomly selected using a random number generator program from the newly created household listing. The interview was open to all females between the ages of 15 and 49 who were either permanent member of the selected HH or visitors who slept there the night before the survey. All of the information on sample design and selection techniques is included in the protocol of PMA Ethiopia (44).
Study Variables
Dependent Variable
“Women alone contraceptive decision making” was the study's outcome variable. The dependent variable question ‘who made the final decision about what method you got?’, with five response categories was dichotomized for analysis purposes into "important others = 0" (for married/cohabitated reproductive age women who reported that the decision on their FP use was made mainly by provider, partner, you and provider, you and partner and other) and "you alone = 1" (for married/cohabitated reproductive age women who reported that the decision on their contraceptive use was made only by themselves), (Table 1 below) (45).
Table 1
Description of the dependent variable
FP Decision Making
|
Variable
|
Question & Responses
|
Categories
|
Item
|
Response
|
FPDM
|
who made final decision current/recent method
|
You alone = 1
|
1 = You alone
|
Provider = 2
|
0 = Important others
|
Partner = 3
|
You and Provider = 4
|
You and Partner = 5
|
Other = 96
|
Independent Variable
Potential confounders: Individual women characteristics, husband sociodemographic characteristics, contraceptive method related characteristics and group or enumeration area related variables (both integral and derived) were considered in this analysis.
Composite variables were created for contraceptive knowledge, contraceptive mass to mass media exposure, experiencing IPV, husband forced pregnancy.
“Contraceptive knowledge” was generated by sum up responses to the nine contraceptive knowledge questions and further categorizes into three groups; 1= ‘poor knowledge’ if respondent heard of 1–3 contraceptive methods, ‘moderate knowledge’= 2 if they knew 4–6 contraceptive methods and 3= ‘good knowledge’ if respondents heard of 7 to 9 contraceptives.
“Contraceptive exposure to -mass media” was formed from the variables (watching tv, listening to radio, and reading the newspaper and social media about contraceptives methods). As a result, women who watch TV, listen to the radio, or read on social media or newspaper about FP at least once were classified as having exposure to mass media (coded = Yes "1"), whereas those who did not do any of those things were classified as not having exposure to the media (coded = No "0") (45).
“Husband/Partner forced to became pregnant” variable was created by sum up three variables and categorized 0= ‘not force’ if none and 1= ‘forced pregnancy’ if whether the respondents reported that her husband/partner forced her by treated by will have a baby with other women, will leave her, and forced her to get pregnant (45).
Analysis
Two data sets, namely, household, and female respondent were used for this study. STATA v16 was used for this analysis. Frequencies and percentages were computed to characterize the study population. Chi-square test statistics was computed to see the overall association/relationship of the independent variables with the two categories of women alone decision making. And it is also used to check cell sample size adequacy.
Frequency was run for every variable to check item nonresponse rate and don’t know response which were later excluded from the analysis. Following these variables were recoded to create biological plausible categories. This is followed by checking distribution of the variable using mean and proportion whenever appropriate categories were merged to make cell sample size adequacy.
Multicollinearity was checked and no strong multicollinearity was detected except between EA level and HH level wealth index variables, parity and marriage duration variables, partner discussion before contraceptive use and partner knows contraceptive use. The correlation coefficient for these variables was 0.8611, 0.7297, and 0.7023 hence EA level wealth, parity, discussion before use and partner know excluded from the final model.
Multilevel binary logistics regression was used to identify important factors of women alone contraceptive use decision making. At bivariate analysis a p value cut of 0.25 was used to select candidate variable for multilevel multivariable logistics regression analysis (46). Results were presented in the form of percentage, and odds ratio with 95% CI. Significance was declared at a significance level of 0.05.
Four models were run; the first was the intercept only model in which no factors is included following which intra cluster correlation coefficient (ICC) is calculated to check the level of clustering observation among enumeration areas (EAs). The clustering was found to be 32.2% which is far beyond the conventional cut of point, 0.5 for the fulfillment of independent observation assumption, hence, supports the use of multilevel logistics regression. In the second model individual level variables were included while in the third model only enumeration area level variables were included. In the final model both individual and enumeration area level independent variables were included. For each model ICC, Akaike and Bayesian information criteria (AIC and BIC) along with log likelihood was calculated to check for model fitness. Based on the result, the final model with lower AIC and BIC along with higher likelihood was selected as best fitted model from which the adjusted odds ratio was computed and reported. Percentage change variation (PCV) was calculated except for the null model as the intercept only model was used as a reference.
Data Quality Management and Control
In PMA Ethiopia survey, data were collected by well experienced PMA filed staff, resident enumerators using smart phones by customized Open Data Kit (ODK) system called PMA collect which facilitates real time data generation and timely feedback. Standard piloted questionnaires prepared in three local languages (Amharic, Afan Oromo, and Tigrigna). Weekly error progress report and response, close follow up during listing, householder, and female questioner data collection along with 10% re_interview and random spot checkups were conducted by field supervisors.
PMA Ethiopia data have been cleaned for public use to ensure its appropriateness for this analysis, data response completeness was checked, item response rate and necessary measure was taken. Data cleaning and quality before conducting different analyses techniques was be employed in this study to exclude the missing values in each variable.
Ethical Consideration
This study involved a secondary analysis of de_identified data from the PMA Ethiopia. The PMA Ethiopia survey was conducted strictly under the ethical rules and regulations of world health organization and IIRB of Ethiopian Health and Nutrition Research Institute (EHNRI). Informed consent was obtained from respondents during the data collection process of PMA Ethiopia on data collection on Oct 2021. PMA surrey has been also conducted after obtained ethical approval from Bloomberg School of Public Health at Johns Hopkins University in Baltimore, USA.