“…His conclusion was that whether the object was to predict a daily, a monthly or an annual mean temperature, an average taken over the most recent 15 to 25 years was better than one taken over any longer or shorter period. Craddock and Grimmer (1960), working independently, by trial and error, came to a very similar conclusion for annual mean temperatures. They also tried using a weighted average, with weights decreasing in geometrical progression into the past, but found this generally inferior to the unweighted average.…”