Please use this identifier to cite or link to this item: https://www.um.edu.mt/library/oar/handle/123456789/93724
Title: Equivalence of loglinear models and logistic regression models for binary data related to the survival of neonates
Authors: Schembri, Elaine (2003)
Keywords: Operations research
Newborn infants -- Malta
Binary control systems
Issue Date: 2003
Citation: Schembri, E. (2003). Equivalence of loglinear models and logistic regression models for binary data related to the survival of neonates (Bachelor's dissertation).
Abstract: This dissertation deals with the analysis of binary categorical data using legit and loglinear models. It is a proven fact that for binary data the two models are equivalent. The aim of this dissertation is to illustrate this equivalence for the binary data obtained from Maltese Government demographic records dated 1980 from the National Statistical Office (NSO) concerning the survival of neonates by type of maternity (single or multiple) and age of mother. The response variable in our data, the state of the neonate, is binary and is affected by two other factors - the age and the maternity of the mother - each of which has two levels. This data can be modelled using either logistic regression models, having the form log (\pi/l-\pi\pi1) =\pi1. or loglinear models having the form log(µ;)= pi1; . The former 1-;r l has a logit link function and the latter has a log link function ni the linear predictor. Logistic regression models are probabilistic regression models whereas loglinear models are models for count data. 11 Part of the discussion concerns obtaining the model of best fit. The parsimonious logistic regression model is logit ;r ( x) = -4.679 + 2.698 M and the parsimonious loglinear model is log µpi1 = 3.525-0.5783A +4.679S-1.071M-0.4025AM-2.698SM where A, S and M represent Age, State and Maternity respectively. It was found that Maternity is the most important factor as it is the one which influences considerably the state of the neonate. Finally we compare the parameter estimates for the corresponding logit and loglinear models and show that they are equivalent. It is more sensible to use logit models directly when there is a single binary response variable and loglinear models when there are at least two response variables.
Description: B.SC.(HONS)STATS.&OP.RESEARCH
URI: https://www.um.edu.mt/library/oar/handle/123456789/93724
Appears in Collections:Dissertations - FacSci - 1965-2014
Dissertations - FacSciSOR - 2000-2014

Files in This Item:
File Description SizeFormat 
BSC(HONS)STATISTICS_Schembri_Elaine_2003.pdf
  Restricted Access
7.25 MBAdobe PDFView/Open Request a copy


Items in OAR@UM are protected by copyright, with all rights reserved, unless otherwise indicated.