You are on page 1of 17

Class #4 Introduction to Sensitivity Analysis

4.2 Sensitivity Analysis – Defined


The OED defines sensitivity as:
Sensitivity The degree to which a device, test, or procedure responds to small amounts of or slight changes
in that to which it is designed to respond.
On page 96 of the Stevenson, Ozgur, and Nsakanda textbook, the purpose of sensitivity analysis is. . .
Sensitivity Analysis . . . to explore the effects of potential changes in parameters of LP models on the
optimal solution.
Stated another way, this is what-if analysis. If you recall the Style and Comfort Furniture Manufacturing
Company example from last time, we used up all 800 machine hours but still had an extra 136 labour hours
available. What if we can get one more hour (a small change) of machine time – what would the impact
be on profit (our objective function)? What if our $180,000 production budget is reduced by 10%; what
would the impact be on profit? What if we loosen (or drop) the requirement to produce 30 love seats, what
would the impact be on profit? The Solver Sensitivity Report gives us the data we need to answer these
questions and a whole lot more!

4.3 Kelson Sporting Goods


Kelson makes 2 different types of baseball gloves: a regular glove and a catcher’s mitt. The firm has 900
hours of production time available in its cutting and sewing department, 300 hours available in its finishing
department, and 100 hours available in its packaging and shipping department. The production time re-
quirements and the profit contribution for each product are shown in the table below. Assume the company
is interested in maximizing the total profit contribution.

Model Cut & Sew Finishing Pack & Ship Profit/glove


Regular 1.0 (hr) 0.5 0.125 $5
Catcher 1.5 0.333 0.25 $8

4.3.1 LP Formulation – Verbal


This is the part our brain does – parse the problem and pull out the important bits of information:
1. Decision variables

• number of regular gloves to be produced


• number of catcher’s mitts to be produced

2. Objective function

• maximize the total profit contribution.

3. Constraints

• cutting & sewing: 900 hours available


• finishing: 300 hours available
• packaging & shipping: 100 hours available

Ah! There are only two decision variables so we can represent this problem graphically!

BU 275 Business Decision Models Page 53 of 130


Class #4 Introduction to Sensitivity Analysis

4.3.2 LP Formulation – Mathematical


Its not much of a stretch to go from the verbal description above to a formal linear program. Normally I’d
use x1 and x2 for the number of regular gloves and catcher’s mitts respectively, but for our purposes today,
I’ll be using R and C for the decision variables. Thus the problem can be stated as:

Let R = number of regular gloves, and


Let C = number of catcher’s mitts produced.

MAX Z = $5R + 8C Objective Function


s.t.
1R + 1.5C ≤ 900 Cut & Sew
0.5R + 0.333C ≤ 300 Finishing
0.125R + 0.25C ≤ 100 Pack & Ship
R, C ≥ 0 Non-negativity.

4.3.3 Graphical Solution


Refer to Figure 4.3 for a preliminary graphical solution. Preliminary in the sense that all of the constraint
lines are drawn and the feasible region is shown. An arbitrary objective function line (at Z = $2, 000) is
also shown, but we still have to “push” our objective function line out as far as it will go before leaving the
feasible region.

Figure 4.3: Partial graphical solution to Kelson’s problem

We can see that the corner point at the intersection of the Finishing and Packing & Shipping constraints
is the optimal solution. In order to get the optimal solution, we need to solve the two equations simultane-
ously to figure out the coordinates of the intersection.

BU 275 Business Decision Models Page 54 of 130


Class #4 Introduction to Sensitivity Analysis

0.5R + 0.333C = 300, and


0.125R + 0.25C = 100.

The Optimal Solution (OS) is R = 500, C = 150 and the Objective Function Value (OFV) at the op-
timal solution is Z = $5(500) + 8(150) = $3, 700. Refer to the 01-KelsonGraphical.xls spreadsheet
available at MyLS.

4.3.4 Spreadsheet Solution


Refer to Figure 4.4 for the spreadsheet set-up of the problem (with the Solver solution in place). We will
spend the bulk of our time today going over the details of the Answer Report (Figure 4.5) and Sensitivity
Report (Figure 4.6) which Solver produces for us on new tabs in our Excel workbook. This spreadsheet,
02-KelsonLP.xls, is available at MyLS.

Figure 4.4: Spreadsheet set-up of Kelson’s problem

BU 275 Business Decision Models Page 55 of 130


Class #4 Introduction to Sensitivity Analysis

4.4 Answer Report


After running Solver, it gives you the option of producing various reports. The Answer Report for the
Kelson problem is shown in Figure 4.5. Working top to bottom and left to right. . .

Figure 4.5: Answer Report (available after running Solver)

Target Cell (Max) refers to the cell which contains your objective function. The objective (Max or Min)
is displayed in the title. You can ignore the original values; the important bit of information here is
Final Value which is our Objective Function Value (OFV) at optimality.

Adjustable Cells refers to the spreadsheet cells which contain our decision variables. Again, you can
ignore the original values; the Final Values represent the Optimal Solution (OS).

Constraints refers to the constraint set entered into Solver. Cell Value is the value of the Left Hand Side
(LHS) of the constraint at optimality. Status (Binding or Not Binding) has the same meaning as
discussed in Class #02 (i.e. whether or not the constraint is part of the optimal solution or not).
The last column, Slack, is not the same as the notion of slack and surplus variables discussed earlier.
If a constraint is binding, it will have a slack of zero. E.g. we’ve used up 300 hours of Finishing
Department time; there is no slack left in the system. If a constraint is non-binding, the slack is
whatever is left over between the original RHS value and the LHS value. Thus, for the Cut & Sew
constraint, we used up 725 of 900 hours, so we have a slack of 175 hours.

4.4.1 Ch Ch Ch Changes. . .
Sensitivity analysis (also called “post-optimality analysis”) is used to determine how the optimal solution
is affected by changes, within specified ranges, in the objective function coefficients, and the constraint
right-hand side (RHS) values. Ask yourself, HOW and WHY might these parameters change? Sensitivity
analysis is important to the manager who must operate in a dynamic environment with imprecise estimates
of the coefficients. And, as mentioned above, sensitivity analysis allows the manager to ask certain what-if
questions about the problem.

BU 275 Business Decision Models Page 56 of 130


Class #4 Introduction to Sensitivity Analysis

4.5 Sensitivity Report


The Sensitivity Report produced by Solver for the Kelson problem is shown in Figure 4.6. You’ll notice
that the report is divided into two sections. The first, the Adjustable Cells section deals with the decision
variables and the objective function, while the Constraints section deals with the constraints. We’ll get
into these sections in greater detail next.

Figure 4.6: Sensitivity Report (available after running Solver)

4.5.1 Adjustable Cells


The Cell, Name, and Final Value columns have the same meaning as in the Answer Report, namely where
you’ll find the Optimal Solution. Working left to right, here’s what the remaining columns are all about:
Reduced Cost The reduced cost for a decision variable whose value is 0 in the optimal solution is the
amount the variable’s objective function coefficient would have to improve (increase for maximization
problems, decrease for minimization problems) before this variable could assume a positive value (i.e.
a non-zero value). The reduced cost for a decision variable with a positive value is 0. If reduced cost
is zero for a variable which takes the value zero at optimality, this indicates that there are alternate
optimal solutions. More detail & examples momentarily.
Objective Coefficient are the current objective function coefficients associated with each decision vari-
able, i.e. $5.00 profit per Regular glove, $8.00 for a catcher’s mitt. While the Answer Report gives
us the OFV at optimality (i.e. $3,700), the Sensitivity Report does not. However, the Sensitivity
Report does provide us with sufficient information to calculate the objective function value. How?
Allowable Increase represents the amount the objective function coefficient for the decision variable can
increase from its current value and we still have the same optimal solution. E.g. the profit on a
catcher’s mitt (currently $8.00) could increase by $2.00 to $10.00 and we’d still produce the same
number each type of glove.
Allowable Decrease represents the amount the objective function coefficient for the decision variable
can decrease from its current value and we still have the same optimal solution. E.g. the profit on
a regular glove (currently $5.00) could decrease by $1.00 to $4.00 and we’d still produce the same
number each type of glove.

BU 275 Business Decision Models Page 57 of 130


Class #4 Introduction to Sensitivity Analysis

Reduced Cost
In the Kelson example, the reduced cost for each of the products is zero. That is, we don’t have to improve
the profit on either product before we have a non-zero number of products being produced – that’s already
happening (R = 500, C = 150). If you recall the Farmer Dave minimization problem from Class #02 we
had a situation where there is a reduced cost. When we solved the problem graphically, we found that the
optimal solution was 0 bags of SuperGro fertilizer and 8 bags of CropQuik fertilizer. Why? Well, the cost
of SuperGro was $6.00 per bag while the cost of $CropQuik was $3.00 per bag. By how much would we
have to reduce the cost of a bag of SuperGro before we start buying a non-zero number of bags? Solving
the problem using Excel gives us the answer in the Sensitivity Report:

Ah! The cost needs be reduced from 6.00 $/bag by at least $2.00 before we’ll buy any SuperGro. If we
re-solve the problem with a cost of 4.00 $/bag for SuperGro, we are now going to need 4.8 bags of SuperGro
and 1.6 bags of CropQuik to meet the nutrient requirements of Farmer Dave’s field. The reduced cost of the
decision variables is now zero. Why? Because we don’t need to improve the objective function coefficient
any more to have some of each product (i.e. a non-zero amount) in our optimal solution.

If you think back to the graphical solution of this problem, we are moving from the corner point (0, 8) to
the corner point at (4.8, 1.6). Its also worth mentioning that the new objective function Z = 4x1 + 3x2 is
parallel to the Phosphate constraint (4x1 + 3x2 ≥ 24) so (0, 8) and (4.8, 1.6) (and every point in between) is
a cost minimizing solution (Z = $24). Another way to think about reduced cost is “what is the impairment
to the solution by including a unit of this decision variable?” Right now the minimum cost is $24.00; what
if Farmer Dave buys one bag of SuperGro? We can model this in the LP by adding x1 ≥ 1 as a constraint.
Solving the new LP gives a minimum cost of $26.00; $2.00 more than it was before. One last point to make:
if the reduced cost is zero for a variable which takes the value zero at optimality, this indicates that there
are alternate optimal solutions. This idea of reduced cost leads into the introduction of some terminology
to refer to our decision variables:

Basic Decision Variable is a decision variable which takes a non-zero value in the optimal solution.

Non-Basic Decision Variable is a decision variable which takes a value of zero in the optimal solution.

In the first Farmer Dave solution, x1 , the number of bags of SuperGro is a non-basic decision variable, while
x2 , the number of bags of CropQuik, is a basic decision variable. If you’re asked to change a non-basic
decision variable into a basic decision variable, the reduced cost is the piece of information required to do
this.

BU 275 Business Decision Models Page 58 of 130


Class #4 Introduction to Sensitivity Analysis

Allowable Increase / Allowable Decrease; a.k.a. Range of Optimality


The range of optimality of an objective function coefficient is found by determining an interval for the
coefficient in which the original optimal solution remains optimal while keeping all other data of the prob-
lem constant. Note: The objective function value (OFV) may change in this range. Graphically, the limits
of a range of optimality are found by changing the slope of the objective function line within the limits of
the slopes of the binding constraint lines.

Recall: the slope of a line in standard form: a1 x1 + a2 x2 = b is: −a1 /a2 .

Working our Kelson Sporting Goods example. . .


The slope of the objective function line, Z = 5R + 8C is −5/8.
The slope of the binding Finishing constraint is 0.5R + 0.333C = 300, or −(1/2) ÷ (1/3) = −3/2. And,
The slope of binding Packing & Shipping constraint is 0.125R + 0.25C = 100, or −(1/8) ÷ (1/4) = −1/2.

What range of values can the profit on R, regular gloves, take? We can see that:
−3/2 ≤ -5/8 ≤ −1/2.
If we substitute c1 for the profit for a regular glove, we can solve for the range as follows:
−3/2 ≤ −c1 /8 ≤ −1/2.
Multiply through by −8 and we get:
12 ≥ c1 ≥ 4 or if you prefer: 4 ≤ c1 ≤ 12. c1 = 5 currently, thus, our allowable increase is 7 (from 5 up
to 12) and our allowable decrease is 1 (from 5 down to 4). Look familiar? You’ll find these values in the
“Adjustable Cells” cells section of the Sensitivity Report from above.

Apply the same logic to the coefficient for C, catcher’s mitts:


−3/2 ≤ -5/8 ≤ −1/2 (from above).
Substitute c2 for the profit for a catcher’s mitt and we can solve for the range as follows:
−3/2 ≤ −5/c2 ≤ −1/2
Inverting gives us:
−2/3 ≥ −c2 /5 ≥ −2/1
Multiply through by -5 yields:
10/3 ≤ c2 ≤ 10. c2 = 8 currently, thus, our allowable increase is 2 (from 8 up to 10) and our allowable
decrease is 4 32 (i.e. from 8 down to 3 13 ). Again, you’ve seen these numbers in the Sensitivity Report, now
you know where they come from. What do they mean?
If the change in the objective function coefficients is within allowable limits, the current solution will
remain optimal. Furthermore, the new objective function value (OFV) can be calculated by substituting
the optimal solution in the new objective function. This allows us to answer what-if questions like: “What
if the profit per regular glove is $7.00? What is the optimal solution and total profit?”
To answer this question, look at the magnitude of the change: an increase of 2 in the objective function
coefficient of R. The allowable increase is 7, so we are within the range of optimality, thus the optimal
solution will remain R = 500, C = 150. The new objective function value will be: Z = 7(500) + 8(150) =
4, 700. Hopefully this is what you expect – an extra $2 on 500 gloves means an extra $1,000 to our old
OFV of $3,700.
If the magnitude of the change is outside the range of optimality, the problem has to be re-solved.

BU 275 Business Decision Models Page 59 of 130


Class #4 Introduction to Sensitivity Analysis

Refer to Figure 4.7 for the Range of Optimality (yellow shaded area). The interactive spreadsheet
03-KelsonObjFn.xls available at MyLS lets you play around with the objective function coefficients so
you can see the impact on the objective function within the range of optimality.

Figure 4.7: Range of Optimality for Kelson Problem (graphical representation)

4.5.2 Sensitivity Report – Adjustable Cells – SUMMARY


If the change in objective function coefficient is outside allowable increase / decrease, the problem needs to
be re-solved on the computer and we cannot say anything about the objective function value or the optimal
solution. If, however, the changes are within the allowable range, then. . .

Sensitivity Tree: Branch ¬ Basic Variable


If the objective function coefficient of a basic variable changes within allowable limits, then:

• The optimal solution (OS) is unchanged,

• The objective function value (OFV) changes. The new OFV can be determined by substituting the
OS into the new objective function.

• The reduced cost for a basic variable is always zero.

BU 275 Business Decision Models Page 60 of 130


Class #4 Introduction to Sensitivity Analysis

Sensitivity Tree: Branch ­ Non-Basic Variable


If the objective function coefficient of a non-basic variable changes within allowable limits then the optimal
solution is unchanged, and the objective function value is unchanged. For non-basic variables, the reduced
cost is usually non-zero; if the reduced cost is zero, it indicates the existence of an alternate optimal solution.
Problem type (Max or Min) matters!
• For a maximization problem:
– The reduced cost is negative or zero.
– The allowable increase = −1 × reduced cost.
– The allowable decrease = infinite.
• For a minimization problem:
– The reduced cost is positive or zero.
– The allowable increase = infinite.
– The allowable decrease = reduced cost.

Exam Formula Sheet


Here’s how I would summarize all of this information for my exam cheat sheet – of course, how you choose
to do it for your own cheat sheet is up to you!

Change in objective function coefficient value. Outside allowable increase/decrease? Resolve, else:
Basic Variable (Xi > 0; RC = 0 in OS) Non-basic Variable (Xi = 0; RC 6= 0 in OS)
OS is unchanged, OFV change (use OS w/ new Ci ) OS/OFV same. MAX: [∞, −RC], MIN: [RC, ∞]

4.5.3 Constraints
Next we’ll walk through the Constraints portion of the Sensitivity Report. Cell refers to the Excel spread-
sheet cell where you have the formula for the left-hand-side (LHS) of the constraint. Name is generated by
Excel by searching left and up from the cell reference for a text string.
Final Value is the calculated value of the LHS of the constraint using the optimal solution values from
the Adjustable Cells section. E.g. we used 725 hours of Cutting & Sewing time.
Shadow Price will be discussed in much greater detail in a moment, but we’ll define it this way: “A shadow
price for a right hand side (RHS) value (or resource limit) is the amount the objective function will
change per unit increase in the right hand side value of a constraint.” For example, if we can get one
more hour of time in the Finishing department, we can increase our profit by $3.00.
Constraint R.H. Side is the right hand side (RHS) value of our constraint that we entered into the
spreadsheet.
Allowable Increase is the upper bound for increases to the RHS value which allows us to use the current
shadow price in calculations. Note: If you see 1E+30 in Excel, this is the spreadsheet’s way of saying
“infinity” (1E+30 = 1 × 1030 ).
Allowable Decrease is the lower bound for decreases to the RHS value which allows us to use the current
shadow price in calculations. Changes to the RHS values outside the allowable increase or decrease
ranges require the problem to be re-solved.

BU 275 Business Decision Models Page 61 of 130


Class #4 Introduction to Sensitivity Analysis

Shadow Price – Defined


Shadow prices are associated with changes in the RHS value of a constraint.

• Again: A shadow price for a right hand side value (or resource limit) is the amount the objective
function will change per unit increase in the right hand side value of a constraint.

• Graphically, a shadow price is determined by adding +1 to the right hand side value in question and
then resolving for the optimal solution in terms of the same two binding constraints.

• The shadow price is equal to the difference in the values of the objective functions between the new
and original problems.

• The shadow price for a non-binding constraint is 0.

Much like our decision variables fell into two categories of basic and non-basic, we can divide our constraints
into binding and non-binding and examine the shadow price in each case:

Shadow Price – Non-Binding Constraints


If we re-solve the Kelson LP with a one unit increase in the RHS of the Cutting & Sewing constraint (i.e
from 900 hours to 901 hours; 1x1 + 1.5x2 ≤ 901 ) we will get the same OS and OFV as before. Thus
its shadow price is zero (no change in OFV). Stated another way, a unit increase in the RHS of the non-
binding Cutting & Sewing constraint does not effect the feasible region, nor does it effect the optimal
solution. Again, its effect on the OFV is 0.

Shadow Price – Binding Constraints


In the case of a binding constraint, when the RHS of the constraint is increased by one unit, the feasible
region changes and optimal solution changes as well. In the case of the Finishing constraint we increase
the number of hours by 1 (from 300 to 301) and resolve the problem. Our optimal solution will be at the
intersection of: 0.5R + 0.33C = 301 and 0.125R + 0.25C = 100, which gives us a new optimal solution
of R = 503, C = 148.5 and a new OFV of $3,703. Thus, the new OFV of 3703 is an increase of 3 over the
previous OFV of 3700, thus the shadow price for the RHS of the Finishing constraint is 3.

You can apply the same logic to the Packing & Shipping constraint. Our optimal solution will be at
the intersection of: 0.5R + 0.33C = 300 and 0.125R + 0.25C = 101 , which gives us a new optimal solution
of R = 496, C = 156 and a new OFV of $3,728. Thus the shadow price for the RHS of the Packing &
Shipping constraint is 28. Both these constraints are binding, hence they have non-zero shadow prices.

Interpretation If we can get an extra hour of time in the Finishing department, the profit would increase
by 3 units. Likewise, if we lose an hour of Finishing time, the profit would decrease by 3 units. However,
if we get an extra hour of packing, profit would increase by 28 units! We can say that profit is more
sensitive to changes in the Packing & Shipping constraint than in the Finishing constraint.

Shadow Price – Allowable Increase / Allowable Decrease; a.k.a. Range of Feasibility


Much like we had the Range of Optimality for our decision variables, we have a Range of Feasibility for the
RHS values of our constraints. As you might guess, this ties into the allowable increase / decrease values
from the Sensitivity Report. But I’m getting ahead of myself.

BU 275 Business Decision Models Page 62 of 130


Class #4 Introduction to Sensitivity Analysis

Range of Feasibility The range of feasibility for a change in the right hand side value is the range of
values for this coefficient in which the original shadow price remains constant. Graphically, the range of
feasibility is determined by finding the values of a right hand side coefficient such that the same two lines
that determined the original optimal solution continue to determine the optimal solution for the problem.

Range of Feasibility – Non-Binding Constraints


We’ll turn our attention first to the non-binding Cutting & Sewing constraint. Because it is non-binding,
it is not part of the current optimal solution. Changing the RHS of the constraint will move it parallel to
itself, toward or away from, its current location. How far can we move the constraint in or out before the
shadow price is no longer in effect? Stated another way, when would this become a binding constraint (and
lose its non-binding status)?

Allowable Increase: We started with 900 hours in this department, but we only use up 725 to produce
the optimal number of regular gloves & catcher’s mitts. Adding one more hour, ten more hours, a thousand
more hours, or an infinite number of hours to the RHS of this constraint won’t change the feasible region
at all, hence the Allowable Increase of infinity (1E+30) on the Sensitivity Report.

Allowable Decrease: Again, we started with 900 hours in this department and we’ve only used up 725
hours, thus we have an extra 175 hours in the department. We can decrease the RHS value from 900 by
175 to 725 and still satisfy this constraint. If we do so of course, it will become a binding constraint, but we
haven’t changed the feasible region at all. If we reduce it further we will have to resolve the problem (since
the feasible region would now be smaller than before). The Allowable Decrease then is 900 − 725 = 175 as
we see in the Sensitivity Report.
If the RHS of a non-binding constraint is changed within allowable limits, the OS (R = 500, C = 150)
and OFV (Z = 3, 700) are unchanged.

Range of Feasibility – Binding Constraints


Refer to Figure 4.8 as we look next at the Finishing constraint (0.5R + 0.333C ≤ 300). We see that it
can be pushed out to the right until the point (800,0). Any further movement to the right will mean that
Finishing is no longer a binding constraint (the Packing & Shipping constraint, plus the R-axis become the
binding constraints). Solve for the new RHS value of the constraint:
0.5R + 0.33C = RHSnew at the point (800,0), gives:
0.5(800) + 0.33(0) = RHSnew , thus RHSnew = 400.
Thus our Allowable Increase is RHSnew −RHSold = 400−300 = 100. This is the value that Solver reported
in the Sensitivity Report.
Do the same thing by pushing the constraint the other way (left) until Finishing and the Packing &
Shipping constraints aren’t the only binding constraints forming the optimal solution. This occurs at the
point (0,400), when these two constraints and the vertical axis form the optimal solution. Any further
movement to the left will alter the shape of the feasible region. Solve for the new RHS value of the con-
straint:
0.5R + 0.33C = RHSnew at the point (0,400), gives:
0.5(0) + 0.33(400) = RHSnew , thus RHSnew = 133.33.
Thus our Allowable Decrease is RHSold −RHSnew = 300−133.33 = 166.67. This is the value that Solver re-
ported in the Sensitivity Report. I leave it as an exercise for the reader to try the same thing with the “Pack-
ing & Shipping” constraint. Check out the spreadsheets 04-KelsonShdwPr.xls and 05-KelsonRngFeas.xls
at MyLS to give you the visualization of what’s going on with graphical sensitivity analysis.

BU 275 Business Decision Models Page 63 of 130


Class #4 Introduction to Sensitivity Analysis

Figure 4.8: Kelson: Finishing constraint – range of feasibility

4.5.4 Sensitivity Report – Constraints – SUMMARY


If the change in constraint RHS values is outside allowable increase / decrease, the problem needs to be
re-solved on the computer and we cannot say anything about the objective function value or the optimal
solution. As long as the change in RHS of the binding constraints is within the allowable limits, the current
binding constraints will remain binding. The new OFV will change by an amount equal to shadow price
multiplied by the change in RHS. The new OS, however, can be determined only by re-solving.

Tightening or Relaxing Constraints Tightening a constraint means to make it more restrictive; i.e.
decreasing the RHS of a ≤ constraint, or increasing the RHS of a ≥ constraint. This does a couple of
things: it compresses the feasible region and may make solution worse (e.g. if it is a binding constraint that
is being tightened). On the other hand, relaxing a constraint means to make it less restrictive. This has
the effect of expanding the feasible region and may make solution better (or may make it possible where it
was previously infeasible).

Sensitivity Tree: Branch ® Binding Constraints


If the RHS of a binding constraint changes within allowable limits, then:

• The optimal solution changes,

• The objective function value changes, moreover:

• OF Vnew = OF Vold + ∆RHS × Shadow P rice.

Problem type and constraint type matter! The shadow price for a maximization problem is:

BU 275 Business Decision Models Page 64 of 130


Class #4 Introduction to Sensitivity Analysis

• Positive for a ≤ type constraint, and

• Negative for a ≥ type constraint.

The shadow price for a minimization problem is:

• Positive for a ≥ type constraint, and

• Negative for a ≤ type constraint.

Sensitivity Tree: Branch ¯ Non-Binding Constraints


If the RHS of a non-binding constraint changes within allowable limits, then:

• The optimal solution (OS) is unchanged, and

• The objective function value (OFV) is unchanged.

• Remember: the shadow price for a non-binding constraint is always zero.

As above, the direction of the inequality is important:


Non-binding constraint of ≤ type:

1. Allowable Increase = infinity

2. Allowable Decrease = RHSF inal V alue (RHS − LHS)

Non-binding constraint of ≥ type:

• Allowable Increase = F inal V alue − RHS (LHS − RHS)

• Allowable Decrease = infinity

Exam Formula Sheet


Here’s how I would boil down the last couple pages for an exam formula sheet. Note: this mathematical
notation is meaningful to me; it might not be to you. Find your own way of capturing these ideas for your
own exam ‘cheat sheet’.

Change in RHS value. Outside allowable increase/decrease? Resolve, else:


Binding Constraint (LHS = RHS; SP 6= 0) Non-Binding Constraint (LHS 6= RHS; SP = 0)
OS changes but OF Vnew = OF Vold + ∆RHS × SP OS/OFV same. ≤ type: [RHS − LHS, ∞]
MAX: SP +ve (≤), SP -ve (≥); opposite for MIN ≥ type: [∞, LHS − RHS]
I’ll use the  icon below for other ‘things to add to your cheat sheet’. . .

4.5.5 Multiple Changes – The 100% Rule


Simultaneous changes in values (Objective Function coefficients OR constraint RHS values) will not change
decision variables in optimal solution as long as the sum of percentages of change divided by corresponding
maximum allowable change does not exceed 100%. Stated another way:
X Changei
< 100%
M AX Allowed
All Changes

BU 275 Business Decision Models Page 65 of 130


Class #4 Introduction to Sensitivity Analysis

100% Rule for Objective Function Coefficients. . . For all objective function coefficients that are
changed, sum the percentage of allowable increases and allowable decreases. If the sum of the percentages
is less than or equal to 100%, the optimal solution does not change.

100% Rule for Constraints. . . For all right-hand sides that are changed, sum the percentage of al-
lowable increases and allowable decreases. If the sum of the percentages is less than or equal to 100%, the
shadow prices do not change.
 100% rule – multiple changes to ObjFn coeffs OR constraint RHSs. Σchange/allowed ≤ 1, else re-solve.

4.6 But Wait! There’s more. . .


The most interesting parameter changes to consider are changes in the objective function coefficients and
changes in RHS values. It may be the case that other changes are being made. Here is a run-down of those
possible changes and their impact on the the current optimal solution:

Sensitivity Tree: Branch ° Adding a New Constraint If you add a new constraint and the current
optimal solution satisfies the new constraint, then the optimal solution is unchanged, and correspondingly
the objective function value is unchanged. If, however, you add a new constraint and the current optimal
solution violates it, then we will need to re-solve the problem with the new constraint included. This of
course will change the OS and OFV.
 Add constraint – if current OS satisfies constraint OK (OS/OFV same), else re-solve.

Sensitivity Tree: Branch ± Deleting an Existing Constraint If we try to remove an existing


binding constraint, then both the optimal solution and OFV will change and we will have to re-solve the
problem on a computer to obtain the new values. If, however, we delete a non-binding constraint, even
though it may effect the feasible region, both the optimal solution and OFV will be unchanged.
 Delete constraint – if binding, resolve. If non-binding no change to OS/OFV.

Change to LHS Coefficients of a Constraint Again, no sensitivity analysis can be done using the
Excel Solver Sensitivity Report; the problem must be re-solved.

Adding a New Decision Variable No sensitivity analysis can be done using the Excel Solver Sensitiv-
ity Report; the problem must be re-solved.

 Change to LHS coefficients, add DV – re-solve.

4.7 Sensitivity Tree


You may have been wondering what is this Sensitivity Tree and branches (e.g. ¬) stuff all about? Well, its
just a mnemonic device to help you decide how to approach sensitivity analysis questions. You can choose
to use it (or not). See Figure 4.9 for the details.
The tree is straightforward to use. When confronted with a sensitivity analysis question, ask yourself:
“What am I being asked to consider changes to?” The branch you choose to go down next will depend on
your answer to that question. If there is no answer to the question (e.g. adding a new decision variable)
then the answer to the sensitivity analysis question is automatically “re-solve the problem.”

BU 275 Business Decision Models Page 66 of 130


Class #4 Introduction to Sensitivity Analysis

Sensitivity Tree

Changes to: Changes to:


Add or Delete
Objective Function Constraint RHS
a constraint
Coefficient(s) values

Basic Non-Basic Binding Non-Binding Add A Delete A


Variable Variable Constraint Constraint Constraint Constraint
1 2 3 4 5 6

Figure 4.9: Sensitivity Tree

• Are you being asked to consider changes to the objective function coefficients? In that case, proceed
down the left branch of the tree. Is the variable a basic variable or not? One of the branches, ¬ or
­, follows next. The notes (above) tell you how to address this situation.

• Are you being asked to consider changes to the RHS values of constraints? In this case, follow the
tree straight down. You’ll choose between branches ® or ¯ depending on whether the constraint in
question is binding (or not). Again, these notes provide guidance.

• Lastly, are you being asked to add or delete a constraint? Follow the tree down the right hand side.
In this case, you have one of the ° or ± branches to follow next. The impact of adding or deleting
was discussed on the previous page.

We’ll explore each of these changes (and more!) as we work through the Kelson Sporting Goods sensi-
tivity analysis questions which follow.

BU 275 Business Decision Models Page 67 of 130


Class #4 Introduction to Sensitivity Analysis

4.8 Kelson’s Example – Ten Sensitivity Questions


OK. Time to put our new found knowledge of Sensitivity Analysis to work. Remember, the Sensitivity
Report allows us to answer what-if questions without re-solving the problem. What if. . .

1. . . . the profit contribution of a regular glove is $7 instead of $5? What if it is $4?

2. We can get 10 more hours in the Finishing Department. Does this increase our profit? If yes, by how
much? If not, why not?

3. What is the maximum amount we should pay for obtaining an extra hour in Cutting and Sewing?

4. If we can get 3 more hours in Packing & Shipping or 20 hours in Finishing, which is better?

5. If the profit of a regular glove is increased to $6 and the profit of a catcher’s mitt is reduced to $7,
what is the OS and OFV?

BU 275 Business Decision Models Page 68 of 130


Class #4 Introduction to Sensitivity Analysis

6. If the available hours in cutting and sewing is reduced by 50 hours and the available hours in finishing
is increased by 50 hours, what happens to the OS and OFV?

7. If management decides to impose a new constraint that at least a total of 600 gloves (includes both
types of gloves) should be made, what is the new optimal solution and OFV?

8. If there is a new constraint which says that number of regular gloves produced should be at least 4
times as many as catcher’s mitts, what is the new optimal solution and OFV?

9. If management wants to introduce a new product, baseball bats, which yields a profit of $10 per bat,
what is the new optimal solution and OFV? Each baseball bat requires 2 hours in Cutting, 2 hours
in Finishing and 15 minutes in Packing.

10. If we can have as many hours of Finishing as we want at no extra cost, what is the new optimal
solution and OFV?

BU 275 Business Decision Models Page 69 of 130

You might also like