Model-Building of Multiple Binary Logit using Model Averaging


  • Siti Aisyah Mohd Padzil
  • Khuneswari Gopal Pillay
  • Rohayu Mohd Salleh





AICc, BIC, Model Averaging, Model-building, Multiple Binary Logit, UGIB.


Many researchers had been carried out on the study of statistical modelling, making it easier for new researchers in many sectors (social sciences, economics, medical, and etc.) to obtain knowledge in order to ease their research study. Nevertheless, there is still no agreed guidelines in obtaining the best model for multiple binary logit (MBL) using model averaging (MA). This research will demonstrate the proper guidelines to obtain best MBL model by using MA. Upper Gastrointestinal Bleed (UGIB) data were studied to illustrate the process of model-building using the proposed guidelines. This study will pinpoint the factors with high possibility leading to mortality of UGIB patients using obtained best model. Corrected Akaike Information Criteria (AICc) and Bayesian Information Criteria (BIC) were used to compute the weights in model averaging method. The performance of the models was computed by using Root mean square error (RMSE) and mean absolute error (MAE). Model obtained by using BIC weights showed a better performance since the RMSE and MAE values are lower compared to model obtained using AICc weights. The factors that affects the survivability of UGIB patients are shock score, comorbidity and rebleed. In conclusion, model-building of multiple binary logit using model averaging showed a better performance when using BIC.


[1] Akaike H (1978), A Bayesian Analysis of the Minimum AIC Procedure. Annal of the Statistical Mathematics 30(A), 9-14.

[2] Buckland TS, Burnham KP & Austin NH (1997), Model Selection: An integral part of inference. Biometrics 53(2). 603-618.

[3] Burnham KP & Anderson DR, Model Selection and Multimodel Inference: A practical Information-theoretic Approach. 2nd Ed. NewYork: Springer-Verlag, (2002).

[4] Chai T & Drexler RR (2014), Root Mean Square Error or Mean Absolute Error? Arguments against avoiding RMSE in the literatures. Geoscientific Model Development 7, 1247-1250.

[5] Chiu PWY, Ng EKW, Cheung FKY, Chan FKL, Leung WK, Wu JCY, Wong VWS, Yung MY, Tsoi K, Lau JYW, Sung JJY & Chung SSC (2009). Predicting Mortality in Patients with Bleeding Peptic Ulcers after Therapeutic Endoscopy. Clinical Gastroenterology and Hepatology 7, 311–316.

[6] Claeskens G & Hjort NL, Model Selection and Model Averaging. United Kingdom: University Press, Cambridge, (2008).

[7] Forster MR (2001), The new science of simplicity. Simplicity, inference and modelling 76(2), 83-117.

[8] Giombini G & Szroeter J (2007), Quasi Akaike and Quasi Schwarz criteria for Model Selection: A Suprising consistency result. Economic Letters 95, 259-266.

[9] Hurvich CM & Tsai CL (1989), Regression and time series model selection in small samples. Biometrika 76, 297-307.

[10] Jairath V, Rehal S, Logan R, Kahan B, Hearnshaw S, Stanworth S & Travis S (1993), Acute variceal haemorrhage in the United Kingdom: Patient characteristics, management and outcomes in a nationwide audit. Digestive and Liver Disease 46, 419–426.

[11] Kutner MH, Nachtsheim CJ & Neter J, Applied Linear Regression Models. 4th edition. Singapore: McGraw-Hill Inc, (2008).

[12] Mehdiyev N, Enke D, Fettke P & Loos P (2016), Evaluating Forecasting Methods by Considering Different Accuracy Measures. Procedia Computer Science 95, 264-271.

[13] Noraini A, Zainodin HJ & Rick LB (2013), Risk factor determination on UGIB patients in Kota Kinabalu, Sabah, Malaysia. Medical Sciences 13(7), 526-536.

[14] Osmun WE, Kim GP & Harrison ER (2015), Patients with multiple comorbidities: Simple teaching strategy. Can. Fam. Physician. 61(4), 378–379

[15] Posada D & Buckley TR (2004), Model Selection and Model Averaging in Phylogenetics: Advantages of Akaike Information Criterion and Bayesian Approaches Over Likelihood Ratio Tests. Systematic Biology 53(5), 793-808.

[16] Rockall TA, Logan RFA, Devlin HB, Northfield TC & the steering committee and members of the National Audit of Acute Upper Gastrointestinal Haemorrhage (1996), Risk assessment after acute upper gastrointestinal haemorrhage. Gut 38, 316-321.

[17] Schomaker M, Wan ATK & Heumann C (2010), Frequentist model averaging with missing observations. Computational Statistics and Data Analysis 54, 336-3347.

[18] Schwarz G (1978). Estimating the dimension of a model. Annals of Statistics 6, 461–464.

[19] Vreeburg EM, Terwee CB, Snel P, Rauws EAJ, Bartelsman JFWM, Meulen JHP & Tytgat GNJ (1999), Validation of the Rockall risk scoring system in upper gastrointestinal bleeding. Gut 44, 331-335.

[20] Zainodin HJ & Khuneswari GP (2007), Model-Building Approach in Multiple Binary Logit Model using Coronary Heart Disease. Malaysian Journal of Mathematical Sciences 4(1), 107-133.

View Full Article: