REJOINDER.

The article presents a reply to a paper titled "Predicting Passenger Miles Flown," by sociologist Hornell Hart's, published in the February 1, 1949 issue of the journal "American Sociological Review." The article states that the discussion in the paper by Hart falls into two parts, the fitting of curve to existing data and the extrapolation of those curves for prediction purposes. The author of the article is not as enthusiastic about the goodness of fitting curves to existing time series, as Hart appears to be. According to him, a curve can be fitted so well to a time series that it passes through every point. The fit is perfect and the correlation is 1. But the predictive value of such a perfect fit through extrapolation would be worse in general than a line with a poorer fit. In the analysis of time series for prediction purposes, the procedure should be to separate the trend from the cyclical or other fluctuation around the trend. Two series may have the same identical trend but in one case fluctuations around it may be twice as great as in the other and hence the fit twice as bad. The trend, though, is not unrelated to the fit, for if the trend is a curve and a straight line is used to measure the trend, the fit will be bad.