Menu
Menu
Menu
Menu
The national health insurance was established to increase access to health care services and the maternal component was later introduced to improve the health outcomes of both mother and child. The main objectives of this study are to investigate the factors that affect neonatal deaths as well as examine the effect of the Ghana Health Insurance on neonatal deaths in Ghana. Using the most recent round of the Ghana Demographic and Health Survey, the study estimates the probit model with interaction effects to account for the heterogeneity in outcomes. Additionally, the study employs the propensity score matching approach to account for the possible endogeneity in the insurance enrolment decision. Results from the estimations, after controlling for relevant individual and household characteristics suggest that the national health insurance significantly reduces the likelihood of neonatal deaths. Estimates remain consistent even after more robust estimators are employed. Estimates from the interaction between place of residence and health insurance indicate that health insurance beneficiaries who reside in urban areas are at a higher risk of neonatal deaths compared to other women. Access to medical facilities proxied by distance to the nearest health post emerged as an important predictor of neonatal death. The study also suggests significant regional differences in neonatal deaths. We, therefore, conclude that the national health insurance may have the potential to substantially improve the health outcomes of neonates and have policy implications for increasing coverage to more mothers and their neonates, as well as coverage in critical neonatal services and drugs.
The study employs two empirical techniques for each of the research objectives. The first objective of examining factors that influence neonatal deaths makes use of the probit model due to the binary nature of the dependent variable. The form of the probit model estimated is as follows: Where Neonatal is a binary variable equal to 1 if the child died in the neonate stage and zero otherwise. NHIS is the variable of interest which captures whether or not the individual (mother) has a valid health insurance card or not. Socioeconomic is a vector that captures mother’s demographic and socio economic characteristics such as age, education, wealth, marital status as well as region and area of residence. Child is a vector of child specific characteristics such as gender and birth order. The coefficient, α merely measures the probability of neonatal mortality if a mother has a valid health insurance and not necessarily the impact of the health insurance on neonatal death as this would provide biased measure due to the potential endogeneity in the health insurance uptake decision. According to Berk [35] and Nichols [36], selection bias occurs when relevant covariates, whether observable or unobservable are omitted in the analysis. This omission creates a situation where the explanatory variables are correlated with the residuals, thereby producing biased estimates and consequently affecting the reliance of the probit estimates for causal inferences. For example, the probit estimates may be overstating the effects of health insurance on the probability of neonatal mortality even after controlling for other relevant covariates. In the NHIS enrolment decision and as argued by Wang et al., [37] in a related study, women who enroll in the NHIS may differ significantly from women who do not. For example, women who are at a higher risk of losing their neonates may be more likely to enroll on the health insurance scheme. Women’s enrolment on the scheme is therefore not random. For example, some unobserved characteristics of women (perhaps specific health conditions) that make then more risky to lose their neonates also makes them risk averse. These may motivate them to enroll on the scheme. The unobserved heterogeneity in the characteristics of women in the sample may lead to unobserved selection bias. Also, evidence from health care utilization studies, such as Arthur [38] and Dixon et al. [39] have indicated the rural-urban dichotomy in the utilization of maternal health care. As a result of this dichotomy, this study attempts to test the heterogeneity in the risk of neonatal death, by taking an interaction between the NHIS and whether or not the individual lives in the rural or urban areas. The endogeneity problem is addressed by employing the Propensity Score Matching technique which is discussed below. The Propensity Score Matching technique developed by Rosenbaum and Rubin [40] has been described as an alternative to obtaining unbiased estimates in assessing program effects [41, 42]. In this technique propensity scores which are defined as the probability of assignment to the treated group, conditional on observed covariates are estimated. This balancing score is estimated based on a logit or probit regression where the treated and control subjects are then grouped based on similar propensity scores. The propensity scores then allows for the estimation of the average treatment on the treated [43]. This precisely allows for measurement the effect of the intervention or treatment. Despite the advantage of being able to directly estimate the treatment or program effect, the propensity score matching technique makes an assumption that unobservable differences does not exist between the treated and control groups [44] and as such does not balance on the unobserved characteristics. Table 3 below provides details of the balancing process for the study variables . The results of the overidentification test also suggest that both control and treatment groups are balanced. Also, the balance plots in Fig. 1 below indicate that the control and treatment groups are fairly similar. Propensity Score Balance Summary Balance plots of control and treated samples Following [6, 45, 46] the paper estimates the propensity scores on which women in the sample are matched into and put in two groups-women with valid health insurance and women without valid health insurance. The estimation adopted a maximum of two matches (based on the psmatch option). This means that for each score, a maximum of two matches are considered. The balancing test2results indicate that the treatment and control groups are very similar. In the matching, model let ED = 1 represents a woman who has a valid national health insurance and ED = 0 represent a woman with no valid national health insurance. The treatment effect of valid health insurance is represented by TREAT for the individual women written as: In this context, Y i(1) represents the risk of neonatal death if the mother has a valid national health insurance and Y i(0) represents the risk of neonatal death if the mother did not have a valid national health insurance. In this paper, the average treatment effect on the treated (ATET)3 is estimated. The ATET evaluates the outcomes for those who received the treatment. In this case the, ATET estimates the risk of neonatal death for those who had a valid national health insurance. This is represented by the equation as: Given that the ATET directly focuses on the actual treatment participants, it evaluates precisely the gain from a program and therefore it can help determine whether or not the program or treatment was successful [47]. To check for sensitivity of results to different estimators, the paper employs other treatment effects estimators namely regression adjustment, inverse probability weights (ipw) and inverse probability weights with regression adjustment (ipwra).4 These three estimators model for the non-random treatment assignment in different ways. Regression adjustment accounts for the non-random assignment by modeling the outcome (neonatal deaths in this case), ipw models the treatment assignment process and not specify a model for the outcome. The IPWRA estimator accounts for the non-randomness in the treatment assignment by modelling both the outcome and the treatment. The estimator uses the ipw weights to estimate corrected regression coefficients that are then used to perform the regression adjustment. The IPWRA estimator is characterized by the double-robust property which ensures consistent treatment effects. All three estimators pose the question “how would the outcomes (neonatal deaths) have changed if the mothers who had valid health insurance did not have” or “how would the outcomes have changed if the mothers who did not have valid health insurance ensured that they had valid health insurance?” The difference in these two counterfactual outcomes, also called potential outcomes precisely give the actual effect of the health insurance on neonatal deaths.
N/A
