Master of Science ( M.S.) in Applied Statistics

Bivariate Generalized Bernoulli Model for Analyzing Dependence between Antenatal Care and Cesarean Section Births

Wed, 05 Apr 2017 00:00:00 GMT

Bivariate Generalized Bernoulli Model for Analyzing Dependence between Antenatal Care and Cesarean Section Births Shiblee, Shafayatul Islam Cesarean section delivery can prevent maternal and child mortality effectively. However, if there is no medical necessity then it has no benefit. During the past two decades, cesarean delivery has been increasing alarmingly both in developed and developing countries. Along with clinical factors, Number of antenatal visits, Antenatal provider and Place of antenatal care has influence on the increase of cesarean section delivery. In this study, the dependence between antenatal care (i.e. number of antenatal visits, from whom received the antenatal and place of antenatal care) and cesarean section delivery is assessed along with some selected demographic-socioeconomic covariates by implying Generalized Bivariate Bernoulli Model. Bangladesh Demographic and Health Survey, 2014 is used in this study to illustrate the model. In this present study prevalence of cesarean delivery in Bangladesh is 24%, though this rate varies from 12.1% to 34.4% in different divisions. Women with advance age were more likely to have cesarean section (OR = 1.83, p-value = 0.01 and OR = 2.84, p-value = 0.03 respectively for age groups “20-34” and “35-49”). Though the risk was higher only if 4 or more than 4 antenatal visits were made and in case the care was received from qualified doctor then women aged 20-34 years likely to have higher risk of CS (OR = 1.79, p-value = 0.05). Higher risk of CS for this group was also found if antenatal care was received from public or private sector (OR = 1.83, p-value = 0.03) and if the care was received from home or NGO sector then women aged 35-49 years were more likely to have CS compared to those aged <20 years. Overweight and obese women risk of having CS delivery was higher compared to those with BMI level below normal and this higher risk was found if the care was received from qualified doctor and from Public or Private sector. If number of antenatal visits were 4 or more than 4, qualified doctor provided antenatal care and received from public or private sector then women living in Chittagong division were less likely to have cesarean section than those living in Barisal division. Women living in Khulna division were more likely to have CS if number of antenatal visit was no or less than 4 and the care was received from public or private sector. Women living in rural areas and with previous female child were less likely to have CS if antenatal care was received from qualified doctor and private or public sector. Women belonging to the higher economical class exposed with higher risk of CS. In case, if number of antenatal vi visits was no or less than 4 and antenatal care was received from qualified doctor only then women with middle economical class were more likely to have CS otherwise not. Dependence was found between antenatal care and cesarean section, which implies how many antenatal visits were made, motive and influence of doctors and antenatal care providers and the place from where the care was received can instigate which type of delivery will be conducted. Interventionist should take proper steps to evaluate management from where women receive antenatal care, even the quality of antenatal care needs to be reviewed. This thesis submitted in partial fulfillment of the requirements for the degree of Masters of Science in Applied Physics and Electronics of East West University, Dhaka, Bangladesh

Bivariate Generalized Linear Models for Analyzing Survival of Live Births at Neonatal, Infant and under Five Stages

Sun, 12 Jul 2015 00:00:00 GMT

Bivariate Generalized Linear Models for Analyzing Survival of Live Births at Neonatal, Infant and under Five Stages Ahmed, B.M. Rajib Dependence on binary response may pose challenges in analyzing longitudinal data. For the child survival longitudinal data, determinants of child survival for the child death by relationship between rick factor and survival status. In the previous studies mainly marginal analysis was performed for specific birth order or combined analysis all birth orders. However the dependence in the outcome of survival status of the consecutive birth order was not taken into account. This study also show that there is no sex performance of the survival status for first three children at neonatal infant and under 5 stages. It is also found that there is statistically significant dependence of the survival status of the consecutive births. This thesis submitted in partial fulfillment of the requirements for the degree of Master of Science in Applied Statistics of East West University, Dhaka, Bangladesh

Detecting Overdispersion in Count Data: Comparison of Tests

Thu, 13 Aug 2015 00:00:00 GMT

Detecting Overdispersion in Count Data: Comparison of Tests Alam, Nadia When data appear more dispersed than expected under a reference model, the situation is termed as overdispersion. In modelling a count variable in terms of some independent predictor variables, theoretically most established and the simplest available reference model is Poisson regression model. For standard Poisson regression model, variance is equal to mean and there is no extra parameter for dispersion. However, in practi- cal scenario, the estimated variance from data often exceeds the mean and the data is considered to be overdispersed. To solve the overdispersion problem, two common alternative approaches are i) tting a more general parametric distribution ii) having a di erent form mean variance relationship without fully specifying the distribution. Both approaches include parameters for overdispersion to be estimated from data. However, when there is no overdispersion, Poisson regression model is preferred for its simplicity, interpretability and theoretical basis. Therefore, robust test for detecting the signi cance of parameter related to overdispersion is important to use before going for alterative to Poisson regression. In this work, we have investigated tests for detecting overdispersion when Poisson model is used for count data. The tests discussed are derived from partial score and are applica- ble against negative binomial or more generally mixed Poisson alternatives. These tests do not require tting alternative models that incorporate overdispersion to check the ab- sence of overdispersion. Only Poisson model is needed to be tted. Four test statistics are illustrated with their distributional approximations for computing signi cance level. The test statistics have been analyzed and compared based on the assumptions on de- riving the statistics, their limiting distributions and applicability for di erent number of observation in sample. A simulation study was done to check adequacy of distributional assumption for three of them who follow approximately normal distribution. The study involved generating samples of the statistics and proportion of the time each exceeded the standard normal upper 20%, 10%, 5%, and 1% point were tabulated. From the results, the normality assumption of one of the statistics has been observed to be good for large sample size but less accurate for small size. Another one of the statistics has been found to have almost accurate standard normal distribution even for small sample. Some comparisons and recommendations relating to the applicability and assumptions of the statistics are also presented. This thesis submitted in partial fulfillment of the requirements for the degree of Master of Science in Applied Statistics of East West University, Dhaka, Bangladesh