CHAPTER 8: What-If Analysis for Linear Programming

Wyndor (Before What-If Analysis)

Using the Spreadsheet to do Sensitivity Analysis

The profit per door has been revised from $300 to

$200. No change occurs in the optimal solution.

The profit per door has been revised from $300 to

$200. No change occurs in the optimal solution.

The profit per door has been revised from $300 to

$1,000.
The optimal solution changes.

Setting up a Parameter Analysis Report

 Select D4 and choose Optimization under the Parameters menu.

Under the Reports menu on the Analytic Solver tab, choose

Optimization>Parameter Analysis.
Specify which parameters to vary and which results to show.

The Parameter Analysis Report

Using the Sensitivity Report to Find the Allowable Range

Variable Cells
Graphical Insight into the Allowable Range

The two dashed lines that pass through the solid

constraint boundary lines are the objective
function lines when (the unit profit for doors)
is at an endpoint of its allowable range,

Using the Spreadsheet to do Sensitivity Analysis

The profit per door has been revised from $300 to $450.
The profit per window has been revised from $500 to $400.
No change occurs in the optimal solution.
The profit per door has been revised from $300 to $600.
The profit per window has been revised from $500 to $300.
The optimal solution changes.

Setting up a Two-Way Parameter Analysis Report

 Select C4 and choose Optimization under the Parameters menu. (Unit Profit per Door to be varied from $300 to $600)
 Select D4 and choose Optimization under the Parameters menu. (Unit Profit per Window to be varied from $100 to $500)

Setting up a Two-Way Parameter Analysis Report

Two-Way Parameter Analysis Reports

 Total Profit:
  Doors Produced:

 Windows Produced:

The 100 Percent Rule

 The 100 Percent Rule for Simultaneous Changes in Objective Function Coefficients
 If simultaneous changes are made in the coefficients of the objective function, calculate for each change the percentage of
the allowable change (increase or decrease) for that coefficient to remain within its allowable range. If the sum of the
percentage changes does not exceed 100 percent, the original optimal solution definitely will still be optimal. (If the sum
does exceed 100 percent, then we cannot be sure.)

Graphical Insight into 100 Percent Rule

The estimates of the unit profits for doors and windows change to

which lies at the edge of what is allowed by the 100 percent rule.
 When the estimates of the unit profits for doors and windows change to

(half their original values), the graphical method shows that the optimal solution
still is (D, W) = (2, 6) even though the 100 percent rule says that the optimal
solution might change.

Using the Spreadsheet to do Sensitivity Analysis

The hours available in plant 2 have

been increased from 12 to 13.

The total profit increases by $150

per week.

The hours available in plant 2 have

been further increased from 13 to
18.
The total profit increases by $750
per week ($150 per hour added in
plant 2).

The hours available in plant 2 have

been further increased from 18 to
20.
The total profit does not increase
any further.

Parameter Analysis Report for Hours Available in Plant 2

Using the Sensitivity Report

Graphical Interpretation of the Allowable Range

Using the Spreadsheet to do Sensitivity Analysis

 One available hour in plant 3 has
been shifted to plant 2.
The total profit increases by $50 per
week.

Spreadsheet to Generate a Parameter Analysis Report

By inserting a formula into cell G8

that keeps the total number of hours
available in plant 2 and 3 equal to
30, it will be possible to generate a
one-way parameter analysis report
that shows the effect of shifting
more and more of the hours from
plant 3 to plant 2.

Parameter Analysis Report

This parameter analysis report shows the effect of shifting more and more of the hours available from plant 3 to plant 2 for the
Wyndor problem.

The 100 Percent Rule

 The 100 Percent Rule for Simultaneous Changes in Right-Hand Sides
 The shadow prices remain valid for predicting the effect of simultaneously changing the right-hand sides of some of
the functional constraints as long as the changes are not too large. To check whether the changes are small enough,
calculate for each change the percentage of the allowable change (decrease or increase) for that right-hand side to
remain within its allowable range. If the sum of the percentage changes does not exceed 100 percent, the shadow
prices definitely will still be valid. (If the sum does exceed 100 percent, then we cannot be sure.)
Robust Optimization
 It is often not possible to estimate parameters with as much accuracy as desired.
 The seriousness of the consequences depends somewhat on whether there is any latitude in the functional constraint.
 A soft constraint is a constraint that can be violated a little bit without serious complications.
 A hard constraint is a constraint that must be satisfied.
 The goal of robust optimization is to find a solution for the model that is guaranteed to remain feasible and near optimal for
all plausible combinations of the actual values of the parameters. Such a solution is called a robust solution.

Robust Optimization with Independent Parameters

 Assumptions:
 Each parameter has a range of uncertainty surrounding its estimated value.
 This parameter can take any value between the minimum and maximum specified by this range of uncertainty.
  This value is uninfluenced by the values taken on by the other parameters.
 All the functional constraints are in either ≥ or ≤ form.
 To guarantee that the solution will remain feasible regardless of the values taken on by the parameters, assign the most
conservative value within the range of uncertainty to each uncertain parameter, and then find the optimal solution for the
revised problem

Example:
What happens if the production rates of doors and windows at plant 3 are uncertain?
production hours required per door in plant 3
production hours required per window in plant 3

The initial estimates of could be off by as much as half an hour.

Range of uncertainty for
Range of uncertainty for

What is the most conservative estimate for within the range of uncertainty?

Solution to Example

The General Procedure for Robust Optimization

 For each functional constraint in ≤ form, use the maximum value for each uncertain coefficient on the left of the ≤ and the
minimum value for an uncertain right-hand-side.
 For each functional constraint in ≥ form, use the minimum value for each uncertain coefficient on the left of the ≥ and the
maximum value for an uncertain right-hand-side.
 For an objective function in maximization form, use the minimum value of each uncertain coefficient.
 For an objective function in minimization form, use the maximum value for each uncertain coefficient.

Chance Constraints with Analytic Solver

 Two shortcomings of robust optimization are
 it can be extremely conservative.
 it can be difficult to accurately identify an upper and lower bound for an uncertain parameter.
 A chance constraint only requires that the original constraint will be satisfied with some very high probability while leaving
a small chance that it will be violated a little bit.
 Chance constraints are useful when dealing with soft constraints.
 Analytic Solver includes the ability to define uncertain parameters and also can include chance constraints.

Example:
What happens if the production rates of doors and windows at plant 3 are uncertain?
production hours required per door in plant 3
production hours required per window in plant 3

The initial estimates of could be off by as much as half an hour.

 Range of uncertainty for
Range of uncertainty for

 Assume any value within these ranges is equally likely.

 The Analytic Solver function PsiUniform (Min, Max) is used to specify that a value should follow the uniform
distribution of the range from Min to Max.

Solution to Example (Before Optimizing)

The Chance Constraint

Auto Adjust Chance Constraints

Setting the Auto Adjust Chance Constraints option to True results in a better solution while still meeting the requirements
of the chance constraint.

Solution to Example (After Optimizing)

 Compare to the robust optimization solution which yields a more conservative solution (0.857 doors and 6 windows) and a
profit of $3,257.

Chance Constraints with the Normal Distribution

The robust optimization procedure discussed in the text requires that the uncertain parameters have a specified range of
uncertainty (a minimum and a maximum).
 The uniform distribution meets this criterion.
 Many other distributions do not. For example, the normal distribution has long tails with no minimum and no maximum.
Chance constraints can still be used even for distributions that do not have a specified range of uncertainty.
Example:
Suppose the hours used per door and window follow the normal distribution, with a standard deviation of 0.5 hours around their
original estimates of 3 hours and 2 hours, respectively.

Solution to Example with Normal Distributions