[Arena] How to define statistical model in order to use it as a factor?

[j_d]

Hi,

I want to perform using Arena package factorial simulation experiment with statistical model (probability density function with given parameters) of a parameter treated as factor.

There is no problem to define factors in Process Analyzer when factor are constant values. But what about the situation when a factor should be defined as following:

factor level factor value
(-) Unif(8; 12)
(+) Unif(10; 15)

Any help?

Regards,

Jack
Poland

[ewilliams]

In your situation, I would begin with sensitivity analysis. Suppose the "-1" level of a factor, as you suggest, is UNIV(8,12). I would run the model with the factor set at 8, 9, 10, 11, 12, for example. Is there much difference among the resulting performance metrics? -- that is, you are assessing the sensitivity of the model to the low(er) level of this factor. If there is, you have already unearthed valuable information meriting further detailed examination. If not, do the same at the upper level of the factor. If again sensitivity is low, proceed with Process Analyzer at the mean (or median) factor levels at low and high (unless the sensitivity is SO low you can ignore this factor henceforth). Supposing you find high sensitivity (e.g., among the input levels 8 through 12), I suggest using not Process Analyzer, but rather a back-end heavy-duty statistical package to conduct ANOVA and DOE. (There are many such; Minitab, JMP, SPSS, and SAS, among others, come to mind.) Posters who recommend use of Process Analyzer throughout for this situation are welcome to amend these suggestions.

[j_d]

Dear ewilliams,

Thank you very much for your answer and suggestions.

But I will return to my question - is there any hint to implement such situation via Arena + PAN?

Regards,

Jack

[ewilliams]

To my knowledge (and I have just checked the "standard" text Simulation with Arena, 5th edition, authors Kelton, Sadowski, and Swets, only constants, not expressions for randomness (e.g., UNIF(x,y)) can be entered in the Process Analyzer. Other suggestions on this thread topic are welcome.

[Hirotaro]

I believe that Williams is right. Therfore, maybe you should consider disinging your experiment in the Arena not in PAN. Why it is so important for you to use PAN?

Pozdrawiam

[j_d]

I think I've found the way to do it.

In order to use statistical models (like UNIF or NORM) as factors in PAN, one should code it in the Arena model and then use these coded values in PAN.

Here's an example:

(-) Unif(8; 12) -> code: 1
(+) Unif(10; 15) -> code: 2

In Arena we define the parameter (i.e. delay in "Process" module) using "Expression" option as following:

(procTimeCode == 1)*UNIF(8, 12)+(procTimeCode == 2) * UNIF(10, 15)

where procTimeCode is the Variable created using "Assign" module.

That way we can use procTimeCode as "Control" in PAN.

In PAN the controls can't be of negative values, so the codes have to be 0 or positive (that's a pity, because coding them as -1 and +1 would be smarter).

Regards,

Jack
Poland