Calibration REP file stats Autocor1 - Autocor4

[mike]

When doing a calibration run, the REP file presents the following statistics;

Autocor1, Autocor2, Autocor3 and Autocor4

The other statistics presented in the REP file are self-explanatory but I would not
want to assume these are one thing when in fact they may be something else.

I could not find any discussion of these statistics in the Vensim documentation.
Could anyone provide a definition of precisely what these statistics represent ?

Guidance appreciated.

[bob@vensim.com]

These are the first, second, third and fourth order autocorrelation of the residuals. (the correlation of error and error[-1], error and error[-2] and so on). If the data are not uniform in time or have missing values these are not that meaningful except that, for an ideal model, all should be 0 (range is -1 to 1).

[mike]

Thank you Bob

OK, so

for a model with time step = 1 week and data every 4 weeks ; where the model's calibration analysis covers the 25 datum point observations when FINAL TIME = 100

autocor1 is the correlation of error[t] vs error[t-4]
autocor2 is the correlation of error[t] vs error[t-8]
autocor3 is the correlation of error[t] vs error[t-12]
autocor4 is the correlation of error[t] vs error[t-16]

over the available observations in each case

[bob@vensim.com]

Yes you have it.

I just checked, if an observation is missing the number of data points used to compute this is decreased. Say you have data a 0,4,8,16, 20 then autocor1 would computeusing 0/4, 4/8 and 16/20 so the results still make sense, just use fewer points.