Outcome of interest | Group | Estimated effect of the intervention | ||||
Not listened to soap opera | Listened to soap opera | Unadjusted analysis | Including potential confounders in regression model† | Using propensity scores approach‡ | Using ‘listens to the radio weekly’ as an instrument | |
Current use of a modern contraceptive method | 31% | 43% | +12% | +6.2% | +9.2% | +8.5% |
↵* The data on which these analyses are based are downloadable on request from http://www.measuredhs.com/
↵† Probit regression, a statistical approach from the same family as logistic and linear regression that are more commonly used in public health, was used to estimate risk differences as presented in the table. The principles we discuss apply equally to other regression methods. Variables included in the model as confounders were: woman's age, woman's educational level, household asset index, religion, lives in rural area, visited by family-planning worker in past 12 months, currently employed, number of living children, husband's education, watches television weekly, listens to radio weekly.
↵‡ No information provided on variables used to calculate propensity score.