"Sid" <[EMAIL PROTECTED]> wrote in message news:<[EMAIL PROTECTED]>... > Hi, > > We can look at the performance figures for any sport, team or player. > > What I want to do is produce an weighted average for this figure. Older > figures should have a lower weighting, scaled by two variables - both the > number of games ago this performance was achieved AND the time since that > performance. > > For example a figure of 65 achieved 3 games and 28 days ago would be worth > less than a figure of 65 achieved 3 games and 16 days ago, which in turn > would be worth less than a figure of 65 achieved 1 game and 16 days ago. > > Any ideas, formulas, references would be appreciated. > > Thanks
Sid, My take on this is that your problem is similar to the supply chain problem that arises when demand does not occur every day or every period but rather with uneven intervals. Consider for example the following history of the amount of gasoline that I put into my 2003 Lexus . today 5 gallons number of days since last purchase = 2 2 days prior 12 gallons number of days since last purchase = 6 6 days prior 3 gallons number of days since last purchase = 1 ordered in time from oldest to newest ... demand interval y(1)= 12 x(1)=6 y(2)= 5 x(2)=2 In your case the score is the thing you are trying to predict based upon lnowledge of the number of days and interval is the number of days before the next occurance. IMHO game number is a RED HERRING and is not needed for your analysis but rather just muddles up your thinking. Dependence on previous games scores is captured (if significant !) in the memory of scores i.e.the history of scores. FreeFore has an option to develop the predictive equation (weights) between score and interval capturing all needed memory such or auto-projective structure such as exponential smoothing , ARIMA structure in general and even develops this in a Robust Way via Intervention Detection ( Pulses , Level Shifts ). This is called a Transfer Function and has regression and exponential smoothing and a zillion more models as particular cases. For example an expected score might depend on the number of days between the last game and the next game AND the number of days in the previous interval AND for example some adaptation ( read weights ! ) based upon a previous score(s). What I believe you want to do ( but you didn't say it ! ) is to predict the next score taking into account the expected interval before the next game for purposes of fun and profit!. Some freeware software to play with http://www.autobox.com/freef.exe or visit TUCOWS or DOWNLOAD.COM and search. If I can help .... be glad to ... Dave Reilly Automatic Forecasting Systems 215-675-0652 http://www.autobox.com . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
