Ethiopia demographic health survey (EDHS) 2016 was cross-sectional study design and used multistage sampling technique. It includes all regions and two administrative cities. EDHS study period from January 18, 2016 to June 27, 2016 (19).
Ethiopia was study area since EDHS was national representative data. Ethiopia located in at horn of Africa. Ethiopia second most populous country most that estimated about 108,805,142 in 2018 (20) . (21). In country 24% of population constitute reproductive ages constitute and family planning service provide 24hour.
Data source and extraction
The data for this analysis were extracted from the 2016 Ethiopian Demographic and Health Survey (EDHS). The data sets were downloaded in STATA format with permission from the Measure DHS website (http://www.dhs program.com). After understanding the detailed data sets, further data cleaning and recording were carried out. Required data sets were joined to Global Positioning System (GPS) coordinates of EDHS.
Sample size determination and sampling procedures
Ethiopia demographic and health survey 2016 collect 588 adolescent married women (weighted sample) age between 15-19 years(19). Stratified and two-stage sampling technique was used. In first stage primary sampling unit selected and in second stage household was selected with different probability of selection. (19).
Data management, data processing and analysis methods
Data was downloaded from measure DHS website in STATA format. Data management and cleaning was done using STATA. All needed variable was selected and recoded in convenient way. After managing adolescent related data, sampling weight was employed to produce the proper representation of family planning information in analysis of data. STATA 14 used for data extraction, descriptive and summary statistics and impact assessment model. The descriptive statistics report summarizes data and respondent characteristics while inferential data analysis using chi square tests report relationship of characteristic. Impact of mass media exposure on modern contraceptive utilization is analyzed by propensity score matching and recursive biprobit probit model regression.
We use PSM and recursive biprobit probit model to asses impact (22, 23). Recursive biprobit probit model was used for addressing the issue of endogeneity(24).
Propensity Score Matching
Treatment effect refers to the causal effect of a given treatment or intervention on an outcome variable of interest. Treatment effect is defined for each individual unit in terms of two "potential outcomes." Mass media is treatment variable and modern contraceptive (Y1) the potential outcome or no modern contraceptive (Y0) counterfactual for that subject. Those expose to mass media would observe Y1, so Y0 would be the counterfactual outcome for that subject. The "treatment effect" is the difference between these two potential outcomes. However, this individual-level treatment effect is unobservable because individual units can only receive the treatment or the control, but not both.
Propensity score matching clarify the effects of campaigns on contraceptive use. Propensity score matching allows the researcher to “create” an unexposed, matched control group that is statistically equivalent. The main attribute of the matching procedure is the creation of the conditions of randomized experiment in order to evaluate a causal effect as in a controlled experiment (25-27).
The impact of mass media on modern contraceptive utilization is estimated using the endogenous regression model where model can be used to compare observed and counterfactual mass media. PSM estimated in Stata using command psmatch2. The psmatch2 implements full Mahalanobis matching and a variety of propensity score matching methods to adjust for pre-treatment observable differences between a groups of treated and a group of untreated. It used to estimate ATT, ATU and ATE (28-30).
Recursive Bivariate Probit
Recursive Bivariate Probit (RBP) process the resulting Bivariate Probit correlation parameter as a recursive model of simultaneous equations. It is important because of handling of endogenous binary repressor on two binary outcome variables. It also estimates maximum-likelihood two-equation probit models by allow to two binary outcome variables correlated with each other. The bivariate probit model is frequently used to determine factor, impact and test endogenous of treatment variable(24, 31).
The exogeneity of exposure to mass media family planning message is test in recursive bivariate probit model by 𝜌. The Wald test, which is reflected by statistical significance of was used to determine whether the models would be best estimated jointly in a recursive manner of not. Likelihood-ratio test is available after bivariate probit to detect the presence of endogeneity and rule out using a simple single-stage binary probit(32, 33).
The outcome variable modern contraceptive recorded in measured DHS as modern method, traditional and no method. It categorize as “modern method” and “no modern method”. Traditional method user consider as none modern method user. It classified as “yes” (coded as 1) and “no” (no coded as 0). Explanatory variables
The predictor variable for modern contraceptive utilization was taken from previous studies. It consider as control variable to determine effect of mass media. Variable include in study were education, wealth quintile , respondent religion, living children, partner education(coded as no education, educated ) ,working status of respondent , desire more children , visit to health facility and exposed to mass media by radio, TV and magazine message .
Endogenous treatment variable
Mass media exposure to family planning message is considers as treatment variable. It was generated by aggregating family planning message exposed to one of media television, radio and magazine (coded as yes and no).
After request EDHS Program for dataset, permission was granted to download and use the data for this study and data were used solely for the purpose of the current study. The consent was obtained from participant before study. Individual identification was not used to treat data confidential. Further the study approved by institutional ethical review board of Dessie Health Science Collage (IRB), Ethiopia.
Patient and public involvement (PPI)
Since the analysis was based on secondary analysis of EDHS 2016 there was no Patient and public involvement (PPI). But during data collection by measure DHS PPI participation was considered.