BuzzerBeater Forums

Here are some results on modelling a few of the formulas behind BuzzerBeater that we as users did not fully understand until now. In particular these results concern:
1) DMI
2) Game Shape change
3) Enthusiasm
4) Salary (not much new here, but the current estimates had not found their way to the forum somehow)

I present some information on these topics below, but if you're not into formulas I suggest that you just play around with the Calculators sheet in the new version 2.0 (or any future version) of the coachParrot tool (119279.157) that tries to turn all of this into something useful and was published at the same time as this thread.

1) DMI
Relevance: The Deliberately Meaningless Index is not so meaningless after all. As it turns out it is a function of the virtual current salary (i.e. salary based on current skills, if a seasonal update occurred right now. For DMI the salary deflation coefficient is not included here), game shape, stamina and free throws. Because of how the formula looks like, it can be useful in three distinct ways:
a) For players at game shape 7 or 8 and whose skills are known, DMI can be used to give a sublevel estimate for game shape
b) For players at other levels of game shape, and in particular game shape 9, DMI can be used to give a more precise estimate of virtual current salary than what is possible with the skill-based salary estimators we have had so far.
c) For players at game shape 9 in two successive weeks, we can now get a clear reading on the salary effect of a week's training, and will be able in the future to use this to get much more precise training speed analysis than what we have until now.
Formula: There is an exponential decay of the DMI in game shapes below 9 (in addition to a small signomial 'garbage' term I need), multiplied with a linear weight of stamina, multiplied with a linear weight of free throw.There is also a constant term.
DMI=100*int(VirtualSalary^(1/SalaryDeflator)*(0.1+0.76*exp(-1.13567746*(9-GS))+0.015780656*GS^0.9)*(1-0.035*(10-ST))*(1+0.018*(FT-1))/10)
Diagrams: DMI_model_vs_observations_at_ST10.PNG (http://img163.imageshack.us/img163/3784/dmimodelvsobserva...)This diagram shows how DMI falls in line with game shape, comparing the observed data points and predictions according to the formula. Values are for ST 10, FT 1 and minimum sublevels of GS.
Method: Because of the strong dependence on game shape, valuable data have been those from players at game shape 9.0 (GS 9 has no sublevels) and at game shape 7.0, which is what all players have in the second week of the season. Also stamina 10.0 (no sublevels either). For other levels of GS and ST, it was a matter of collecting data and more data, to find minima at low sublevels at all of GS, ST and FT. In the end I used some 800 observations collected off the transfer list. An early part of the observations was gathered with the help of the community in the 'little DMI research project' thread (114729.1)
Quality of the model: This formula explains almost all observed data points at GS 9.0 and GS 7.0 and varying ST and FT. It also explains the DMI of all estimates of data at ST 10.0 .