Time Series Analysis Using Autoregressive Integrated Moving Average (ARIMA) Models

Most methods of defining a statistical relationship between variables require that errors in prediction not be correlated. That is, knowledge of the error in one instance should not give information about the likely error in the next measurement. Real data frequently fail this requirement. If a Durbin-Watson statistic reveals that there is autocorrelation of sequential data points, analysis of variance and regression results will be invalid and possibly misleading. Such data sets may be analyzed by time series methodologies such as autoregressive integrated moving average (ARIMA) modeling. This method is demonstrated by an example from a public policy intervention.