Chapter -4

ANYLISIS OF SIMULATION OUTPUT

In the previous chapters we have learn how to produce a model of a system, how to simulateand
produce a computer program for simulating the system such as predator prey system. By which
of this program, we can performed the desired experiment on the system being studied and can
analyze the output. In those programs, we have a freedom to specify initial condition, run length
or replication runs in order to analyze the output simulation.
A large body statistical method has been developing over the years to analyze results in science,
engineering and other fields where experimental observation are made. So, because of
experimental measure of system of simulation for these statistical methods can be adapted to
simulation results to analyze.
The newly developing statistical methodology concerns;
1) To ensure that the statistical estimates are consistent, meaning that as the sample size
increases the estimate tends to true value.
2) To control biasing in measure of both new values of variance. Bias causes the distinction
of estimate to differ significantly from the true population statistics, even though
theestimate may be consistent.
3) To develop sequential testing methods, to determine how long a simulation should be run
in order to obtain a confidence in its return.
ESTIMATION METHOD
- Statistical methods are commonly used on random variable. Usually a random variable is
drawn from an infinite population with a finite mean ‘µ’ and finite variance ‘σ2’.
- These random variable are independently and identically distributed (i.e. iidvariables ).
Let,xi=iid random variables. (i=1, 2,..…., n), then according central limit theorem and applying
transformation, approximate normal variance,

Where = sample mean

1
- It can be shown to be a consistent estimator for the mean of population from which the
sample is done.
- Since the sample mean is some of random variable, it is itself a random variable. So a
confidence interval about its computed value needs to be established.
- The probability density function on the standard normal variable (Z) is shown in figure
(1).

0 +
Fig. (1)
- The integral formed – to µ is the probability that Z is less than or equal to µ and is denoted
by ϕ (u).
- Let us consider the value of u (uα/2) such that ϕ(u) = 1-α/2 where α = some constant < 1.
- Then probability of Z for Z>uα/2= α/2.
- Probability of Z for (-uα/2≤ Z ≤ +uα/2) = 1-α.
- In terms of sample mean µ, the probability statement can be written as
- Prob

Here the constant is a confidence level (usually expressed in %) and the interval
is the confidence interval.

Estimation population variance s2 (not which is actual population variance) is given by:

Replacing by , then the estimated variance s2, the confidence interval for is given
by

2
SIMULATION RUN STATISTICS
On every simulation run, some statistic are measure based on some assumption ; for example –
on establishing confidence interval it is assume that the observation are mutually independent
and distinction from which they are drawn is stationary. But many statistics are interest in a
simulation don’t meet this condition.
Let us illustrate the problems that arise in measuring statistic from simulation run with the
example of single server system.
Consider occurrence of arrivals has a Poisson distribution.
- The service time has an exponential distribution.
- The queuing discipline is FIFO
- The inter-arrival time is distributed exponentially
- System has a single server.
Then in a simulation run, the simplest way to estimate the mean waiting is to accumulate the
waiting time of n successive entities and dividing it by ‘n’. This gives sample mean denoted
by .
If (i = 1, 2, 3, ……….,n) are the individual waiting times, then

The 1st problem, here is that, the waiting times measure this way are not independent because
whenever a waiting line forms, the waiting time of each entity on the line depends upon the
waiting time of its predecessor (i.e. the entities are auto co-related ).
The usual formula for estimating the mean value of the distribution remains on satisfactory
estimate for the mean of auto co-related data. However the variance of auto-correlated data is not

related to the population variance by simple expression as occurs for independent data.

The 2nd problem is that the distribution may not be stationary; it is because a simulation run is
started with the system in some initial state, frequently the idle state, in which no service is being
given and no entities are waiting, thus the early arrivals have a more probability of obtaining
service quickly. So a sample mean that includes the early arrivals will be biased. As the length of

3
simulation run extended and the sample size increases, the effect of bias will be minimum. This
is shown in below fig.

Steady State Mean

Expected mean
wait time Expected value of sample mean

0
500 1000 1500 2000 No. of samples

REPLICATION OF RUNS
One problem in measuring the statistic in the simulation run is that the results are dependent. But
it is required, in simulation, to get the independent result. The one way of obtaining independent
result is to repeat the simulation.
Repeating the experiment with different random numbers for the same sample size ‘n’ gives a set
of independent determination of sample mean
Even though the distribution of sample mean depends upon the degree of auto correlation, this
independent determination of sample mean can be used estimate the variance of distribution.
Suppose,
- Experiment is repeated p-times with independent random numbers.
- observation of run. Then estimates for:

- Sample mean

- Then estimate for variance for run is given by

4
Now, combining the results of p independent measurement gives the following estimate for the
mean waiting time , and variance :

Here, the value of is an estimate for mean waiting time and the value of can be used to
establish a confidence of intervals.

ELIMINATION OF INITIAL BIAS

There two general approaches that can be used to remove the initial bias;
1) The system can be started in more representative states rather than in the empty state.
2) The first part of the simulation run can be ignored.

- In the first approach, it is necessary to know the steady state distinction for the system
and we then select the initial state distinction. In the study of simulation, particularly
existing system, there may be information available on the expected condition which
makes it feasible to select better initial condition and thus eliminating the initial bias.
- The second approach that is used to remove the initial bias is the most common approach.
In this method, the initial section of the run which has highly bias (simulation) result is
eliminated. First, the run is started from an idle state and stopped after a certain period of
time (the time at which the bias is satisfactory). The entities existing in the system at that
are left as they are and this point is the point of restart for other repeating simulation run.
Then the run is restarted with statistics being gathered from the point of restarted. These
approaches have following difficulties:
1) No simple rules can be given to decide how long an interval should be eliminated. For
this we have to use some pilot run starting from the ideal state to judge how long the
initial bias remains. These can be done by plotting the measured, statistics against the run
length.

5
2) Another disadvantage of eliminating the first part of simulation run is that the estimate of
variance will be based on less information affecting the establishing of confidence limit.
These will then cause to increase in confidence internal size.