Interaction Effects on Prediction of Children Weight at School Entry Using Model Averaging
Model selection introduce uncertainty to the model building process, therefore model averaging was introduced as an alternative to overcome the problem of underestimate of standards error in model selection. This research also focused on using selection criteria between Corrected Akaike's Information Criteria (AICC) and Bayesian Information Criteria (BIC) as weight for model averaging when involving interaction effects. Mean squared error of prediction (MSE(P)) was used in order to determine the best model for model averaging. Gateshead Millennium Study (GMS) data on children weight used to illustrate the comparison between AICC and BIC. The results showed that model selection criterion AICC performs better than BIC when there are small sample and large number of parameters included in the model. The presence of interaction variable in the model is not significant compared to the main factor variables due to the lower coefficient value of interaction variables. In conclusion, interaction variables give less information to the model as it coefficient value is lower than main factor.