Jing Yi1, Chang-ting Du, Run-hua Wang, Li Liu. 1. Department of Health Statistics and Infonrmation Management, Public Health School of Chongqing University of Medical Sciences, Chongqing 400016, China.
Abstract
OBJECTIVE: To discuss the application of multiple seasonal autoregressive integrated moving average (ARIMA) predictive model of time series and to establish a predictive incidence model of tuberculosis. METHODS: Parameters of the model were estimated using conditional least squares method according to the data of tuberculosis incidence and the averaged population in a district in Chongqing from 1993 to 2004. In a structure determined according to criteria of residual un-correlation and conclusion, ARIMA predictive model was established and the order of model was confirmed by Akaike's Information Criterion (AIC, for short) and Schwartz's Bayesian Information Criterion (SBC or BIC, for short). RESULTS: There were significant differences of the fitted multiple seasonal moving-average coefficients with the nonseasonal and the seasonal moving-average coefficients being 0.84076 and 0.46602 respectively. The estimated variance was 0.088589, AIC = 19.75979, SBC = 23.28219. Autocorrelation check of residuals of model was white-noise residual. ARIMA(0,1,1)(0,1,1)4NOINT seemed to be the most appropriate model by chi2 test. CONCLUSION: The multiple seasonal ARIMA model can be used to forecast for tuberculosis incidence with high prediction and precision in a short-term.
OBJECTIVE: To discuss the application of multiple seasonal autoregressive integrated moving average (ARIMA) predictive model of time series and to establish a predictive incidence model of tuberculosis. METHODS: Parameters of the model were estimated using conditional least squares method according to the data of tuberculosis incidence and the averaged population in a district in Chongqing from 1993 to 2004. In a structure determined according to criteria of residual un-correlation and conclusion, ARIMA predictive model was established and the order of model was confirmed by Akaike's Information Criterion (AIC, for short) and Schwartz's Bayesian Information Criterion (SBC or BIC, for short). RESULTS: There were significant differences of the fitted multiple seasonal moving-average coefficients with the nonseasonal and the seasonal moving-average coefficients being 0.84076 and 0.46602 respectively. The estimated variance was 0.088589, AIC = 19.75979, SBC = 23.28219. Autocorrelation check of residuals of model was white-noise residual. ARIMA(0,1,1)(0,1,1)4NOINT seemed to be the most appropriate model by chi2 test. CONCLUSION: The multiple seasonal ARIMA model can be used to forecast for tuberculosis incidence with high prediction and precision in a short-term.