About StochSD StochSD version:
using Insight Maker version:

StochSD is an extension to Insight Maker for Stocastic System Dynamics models. StochSD has the following facilities for IM models:

Optim - A simplex optimizer.

Sensi - A sensitivity analyzer.

StatRes - For running a stochastic IM model N times, performing statistical analysis and presentation.

Parmest - Parameter estimation for stochastic IM models.

Send Seed Help By the Send Seed device you will make the study (including many simulations) of a stochastic model reproducible.

Checking the Send Seed box opens a sub-frame that contains two text fields. In the first, named Seed-of-Seeds, you can specify a single integer number (default 123) that will generate a seed (Actual Seed) for each simulation run. The Actual Seed is sent to the model before a new simulation run.

Confidence Level Help STATISTICS

After N simulations statistics can be calculated:

Avg = M = ∑(X₁...X_N)/N.
StdDev. = s = √∑[(X_i-M)(X_i-M)] / √(N-1).

Choose 1-sided or 2-sided conf intervals.
Confidence limits for the mean are set by 1-α (e.g. 90%, 95%, etc.)
The corresponding λ (= λ_α or λ_α/2) is then calculated (see Figure).

Confidence limits around a mean estimate are calculated using
the Normal approximation. (This requires at least N=20 simulations.)

The Confidence Interval is then:

1-sided
C.I. = x < M + λ_α*s/√N Or x > M - λ_α*s/√N

2-sided
C.I. = (M - λ_α/2*s/√N , M + λ_α/2*s/√N)

(Note: You can reduce C.I. by increasing the number of simulations.)

SkipOnCondition Help Checking the SkipOnCondition box enables you to use an auxiliary with the name “SkipOnCondition” that must be included as an auxiliary in the model. If this auxiliary is TRUE at the end of a simulation run, the whole simulation run will be skipped and replaced by a new run. This is handy if you for some reason want to exclude certain simulation runs, e.g. where nothing of interest happens. The number of skipped runs will also be shown.

The auxiliary named SkipOnCondition can e.g. have the condition:

  [Quantity] < 0.5

SkipOnCondition Help Checking the SkipOnCondition box enables you to use an auxiliary with the name “SkipOnCondition” that must be included as an auxiliary in the model. If this auxiliary is TRUE at the end of a simulation run the whole estimation round will be skipped and replaced by a new esimation round. This is handy if you for some reason want to exclude certain simulation runs, e.g. where nothing of interest happens. The number of skipped estimation rounds will also be shown.

The auxiliary named SkipOnCondition can e.g. have the condition:

  [Quantity] < 0.5

StatRes is a tool for collection of end-point results from N simulations
and for statistical analysis of these results.

Quantities to be studied (e.g. Stocks, Flows, Variables, Differences and
cumulated quantities) have to be modeled so their
resulting values can be transferred to StatRes after each run.

StatRes will calculate: Average, Standard Deviation, Confidence Interval,
Min, Max, and Percentiles.
The Confidence Level for the Conf. Int. can be specified as well as 1 or 2-sided intervals.
(Default is 95% and 2-sided interval.)
The Percentile can also be specified. (Default: 50% (Median).)

To use the StatRes module you have to:
1. Build or load a StochSD model.
2. Specify and Add the names of the quantities to be studied.
[3. Specify the Level for the Confidence Interval & 1 or 2-sided.]
[4. Specify the Percentile.]
5. Specify the number of Requested Simulations (Default=100).
6. Press the Run button.

( NOTE: Only write in Yellow fields.)

During a parameter estimation round, the parameters in the model are tuned during a large number of simulation runs. It is then important that the stochastic model behaves the same during all simulations in this round. This is accomplished by using the same Seed-of-Seeds in the model.

For the next parameter estimation round a new Actual Seed is required. This means that we for N parameter estimation rounds will need N Actual Seeds. ParmVar therefore has its own random number generator (RNG1) that delivers these N Actual Seeds to the model's RNG2. ParmVar’s generator is initiated by a single Seed-of-Seeds (default value 123).

During a Deterministic Sensitivity Analysis, the effect on an Objective function, V, from a studied parameter, P, is studied. For this we need two simulation runs to obtain V(P) and V(P+ΔP).

During a Stochastic Sensitivity Analysis, it is important to lock the Actual Seed for the two runs. However, the sensitivity is dependent on the chosen seed. Therefore, you should in the stochastic case do several runs to see how the sensitivity changes randomly. You can do this by repeatedly press the Run-button. For each new pair of runs a new Actual Seed is required. This is provided by Sensi’s own RNG that delivers a new Actual Seed when you press the Run button – See Figure.

When the Sensitivity analysis is repeated N times (for a stochastic model) Sensi’s random number generator (RNG1) delivers N seeds to the models RNG2. Sensi’s generator is initiated by a ‘Seed-of-Seeds’ (default value 123).