The Analytic Hierarchy Process (AHP)
Tuesday, November 26, 2013
How do you make a choice in a complex, subjective
situation with more than a few realistic options?
You could sit and think over each option, hoping for divine inspiration
– but you may end up more confused than when you started.
You could leave it to fate – draw straws or pick a number.
Of
course, this won't win you the Decision Maker of the Year award!
An all-too-common strategy is to simply wait out the problem,
doing nothing proactively, until a solution is somehow chosen
for you by circumstances.
None of these approaches are very effective. What you need is
a systematic, organized way to evaluate your choices and figure
out which one offers the best solution to your problem.
As rational beings, we usually like to quantify variables and
options to make objective decisions. However, the problem is that
not all criteria are easy to measure.
So what do you do when you're faced with a decision that needs
significant personal judgment and subjective evaluation? How do
you avoid getting caught in the "thinking over" stage?
And how can you be more objective?
Combining Qualitative and Quantitative
To address this problem, Thomas Saaty created
the Analytic Hierarchy Process (AHP) in the 1970s. This system
is useful because it combines two approaches – the "black
and white" of mathematics, and the subjectivity and intuitiveness
of psychology – to evaluate information and make decisions that
are easy to defend.
Let's look at an (admittedly slightly trivial) example. If you
want to determine the best route to work in the morning, and travel
time is your deciding factor, the decision making process is very
straightforward and simple. You would use each alternative route
for a week, time the commute, and choose the one that's fastest
on average.
If, however, you carpool with other riders, and you have to consider
everyone's priorities, the decision becomes much more complex.
Larry is concerned about his personal safety, because one route
goes through a dangerous part of town. Joanne wants to factor
in a stop at a drive-through coffee shop, so that everyone can
get coffee. Richard points out that Java Jolt is better than Cuppa Jo.
There are several branches of each in the city, with both types
accessible from all routes, although at different distances.
Now you've got tangible and intangible, and quantitative and qualitative,
factors to think about. And you have to consider the different
perspectives and priorities of the various people.
AHP can combine these different types of factors and turn them
into a standardized numerical scale. You can use this to make
your choice objectively, while including all the decision criteria.
AHP Snapshot
Here's a quick overview of the Analytic Hierarchy
Process.
Build your "hierarchy"
Define your goal or objective.
Identify the choices you're considering.
Outline the major factors you'll use to evaluate each option.
Identify criteria (and any subcriteria of these) that you need to consider for each of these major factors. Link these to the major factors (see figure 1 below for an example of this.)
Continue to build a hierarchy of decision criteria until all factors are identified and linked.
Establish your priorities
Using paired comparison , determine your criteria preferences (perhaps A is a little more preferred than B, B is much more preferred than C, and so on).
Rate these preferences from 1-9.
Repeat this for each level in your hierarchy. (The example below will make this clearer.)
Synthesize, or combine, the ratings
Calculate weighted criteria scores that combine all of the ranking data.
Compare the alternatives
Using those combined scores, calculate a final score for each alternative.
Tip 1:
If you're familiar with Paired Comparison Analysis , then this approach might sound familiar. The power of AHP is that it uses paired comparison to determine the relative weights of various criteria, and then it transfers them across each level of criteria to calculate overall weightings. From this you can calculate an objective score for each alternative.
Tip 2:
This is a complicated approach, and one that needs careful
thought and calculation. Only use it if Paired Comparison
Analysis isn't giving you the answer you want, or where
the problem to be solved is complex and significant, and
involves many different subjective factors.
Let's work through our example to determine
which route – A, B, or C – is best for the group of carpoolers.
Step 1: Build Your Hierarchy
Define your goal at the top of the hierarchy:
Find the best route to work.
Identify your first level of evaluation criteria:
Commute time.
Safety.
Drive-through access (to coffee).
Decide if there are second level criteria,
or subcriteria of these, related to any of your Level 1 criteria:
Drive-through access has two subcriteria:
Java Jolt.
Cuppa Jo.
Under each bottom level criterion, write down the alternatives you're considering:
Route A.
Route B.
Route C.
Figure 1: An Example Hierarchy
Step 2: Establish Your Priorities
Have each decision maker rate the relative importance
or preference for each criterion, at each level. Use a paired
comparison approach.
Set up a matrix to compare each criterion to the others. We have three Level 1 criteria, therefore we need a 3x3 matrix.
Criterion
Commute Time
Safety
Drive-through Access
Commute Time
Safety
Drive-through Access
Rank the importance of each criterion relative
to the others, using the scale below.
Relative Importance
Value
Equal importance/quality
1
Somewhat more important/better
3
Definitely more important/better
5
Much more important/better
7
Very much more important/better
9
Note 1:
The numbers 2, 4, 6, and 8 are half way positions
between the values above.
Note 2:
The scale assumes that the ROW (first)
criterion being ranked is of equal or greater importance
than the COLUMN (second) criterion. If you have a pairing
where the row criterion is less important than the column,
use the reciprocal value (1/3, 1/5, 1/7, or 1/9).
Compare the criteria in the columns to the
criteria in the rows.
Commute time is [much MORE important] than safety
Commute time is [somewhat MORE important] than drive-through access
Drive-through access is [extremely MORE important] than safety
Safety is [extremely LESS important] than drive-through access
Safety is [much LESS important] than commute time
Drive-through access is [somewhat LESS important] than commute time
Criterion
Commute Time
Safety
Drive-through Access
Commute Time
1
7
3
Safety
1/7
1
1/9
Drive-through Access
1/3
9
1
Step 3: Calculate the Ratings
Now you have to calculate the overall weighting for each criterion. This is called a "priority vector" (PV) (don't worry about this term – it's not very helpful.)
Add each column in your ratings matrix.
Criterion
Commute Time
Safety
Drive-through Access
Commute Time
1
7
3
Safety
0.14
1
0.11
Drive-through Access
0.33
9
1
TOTAL
1.47
17
4.11
Divide each entry by the total of its column.
Examples:
Commute time ÷ Commute time total = 1/1.47
Safety ÷ Commute time total = 0.14/1.47
Criterion
Commute Time
Safety
Drive-through Access
Commute Time
0.68
0.41
0.73
Safety
0.10
0.06
0.03
Drive-through Access
0.22
0.53
0.24
Notice how the columns add up to approximately 1.0. This is because the weights have now been standardized.
Because the ratings are subjective, we sometimes see inconsistencies. To "smooth" these out, calculate the average of each row. This is the final weight (priority vector) for each criterion.
Criterion
Commute Time
Safety
Drive-through Access
Priority Vector
Commute Time
0.68
0.41
0.73
0.61
Safety
0.10
0.06
0.03
0.06
Drive-through Access
0.22
0.53
0.24
0.33
This weighted score suggests the following:
Commute time represents about 61% of the final decision
Nearness to a coffee drive-through represents about 33% of the decision
Safety represents about 6% of the decision
Level 2 Criteria – If you had no subcriteria,
you could move onto the next step and calculate final scores
for each alternative route. In our example, drive-through access
has another variable to consider – whether the coffee shop on
the route is Java Jolt or Cuppa Jo.
Following the same steps as the Level
1 criteria, start with a 2x2 matrix.
Then add the ratings. For our example,
we'll assume that Java Jolt is somewhat better (3) than
Cuppa Jo.
Brand
Java Jolt
Cuppa Jo
Java Jolt
1
3
Cuppa Jo
0.33
1
Calculate the overall weighting.
Add each column.
Brand
Java Jolt
Cuppa Jo
Java Jolt
1
3
Cuppa Jo
0.33
1
Total
1.33
4
Divide each entry by the column total. Then average each row.
Brand
Java Jolt
Cuppa Jo
Average/Priority Vector
Java Jolt
0.75
0.75
0.75
Cuppa Jo
0.25
0.25
0.25
Access to a Java Jolt
is three times more preferred than access to Cuppa Jo.
Step 4: Compare the Alternatives
Compare each alternative based on the lowest
level in your hierarchy of decision criteria. In our example
(see figure 1), the lowest-level criteria are as follows:
Commute time
Safety
Java Jolt
Cuppa Jo
Note:
The weighting for
drive-through access will be included as 33% of the final
decision. It will be split 75/25 between nearness to Java
Jolt or Cuppa Jo.
By using the lowest level of your hierarchy,
you ensure that all variations of all options are considered.
You can have as many levels of subcriteria
as you need to make a final decision. If Java Jolt and Cuppa
Jo were relatively equal in their preference (score of 3 or
less), then you could further break down the decision into taste,
price, and muffin selection. With each level, the total weight
always adds to about 1.0, and the overall weight is spread up
the hierarchy through each subsequent calculation.
Create a comparison matrix for the first
decision criterion you want to evaluate. We'll use commute time.
Commute Time
Route A
Route B
Route C
Route A
Route B
Route C
Then use the same 1-9 rating scale to determine
how each route compares to the others, based on that decision
criterion. We'll assume the following:
Route A is somewhat faster than Route
B.
Route A is very much faster than Route
C.
Route B is much faster than Route C.
Fill in the matrix, and calculate the priority
vector.
Commute
A
B
C
A
1
3
9
B
0.33
1
7
C
0.11
0.14
1
Total
1.44
4.14
17
Commute
A
B
C
Priority Vector
A
0.69
0.72
0.53
0.65
B
0.23
0.24
0.41
0.29
C
0.08
0.03
0.06
0.06
Repeat this comparison process for each
of the remaining decision criteria.
Which route is safer than the other?
How much closer to a Java Jolt is one route
than the other?
How much closer to a Cuppa Jo is one route
than the other?
Safety
A
B
C
A
1
0.14
0.14
B
7.00
1
0.2
C
7.00
5
1
Total
15
6.14
1.34
Safety
A
B
C
Priority Vector
A
0.07
0.02
0.11
0.07
B
0.47
0.16
0.15
0.26
C
0.47
0.81
0.74
0.68
Access to Java Jolt
A
B
C
A
1
4
7
B
0.25
1
3
C
0.14
0.33
1
Total
1.39
5.33
11
Access to Java Jolt
A
B
C
Priority Vector
A
0.72
0.75
0.64
0.70
B
0.18
0.19
0.27
0.21
C
0.10
0.06
0.09
0.09
Access to Cuppa Jo
A
B
C
A
1
1
0.11
B
1
1
0.2
C
9
5
1
Total
11
7
1.31
Access to Cuppa Jo
A
B
C
Priority Vector
A
0.09
0.14
0.08
0.11
B
0.09
0.14
0.15
0.13
C
0.82
0.71
0.76
0.77
Combine the overall weights, and determine
a value for each route. (See below for this.)
The value for each route is the weighted
sum of all rankings that are associated with it. The priority
values (PVs) for each criterion have been added to the hierarchy
below.
Figure 2: The Example Hierarchy Developed
Calculate Route A's final score:
Commute Time PV (0.61) x Route A's Commute PV (0.65) +
Safety PV (0.06 ) x Route A's Safety PV (0.07) +
Drive-through Access PV (0.33) x Java Jolt PV (0.75) x Route
A's JJ PV (0.7) +
Drive-through Access PV (0.33) x Cuppa Jo PV (0.25) x Route
A's CJ PV (0.11) +
= 0.40 + 0 + 0.17 + 0.01
= 0.58
Complete the calculations for Routes B and C.
Final Scores
Criterion
Route A
Route B
Route C
Commute Time
0.40
0.18
0.04
Safety
0.00
0.02
0.04
Java Jolt
0.17
0.05
0.02
Cuppa Jo
0.01
0.01
0.06
Total
0.58
0.26
0.16
Route A is the clear winner! You can interpret this to mean that Route A meets 58% of all the decision criteria considered. Route B meets only 26% of the criteria, and Route C meets 16%.
Key Points
The Analytic Hierarchy Process can help you
quantify the judgments you use in decision making. When problems
become complex, it's hard to justify and explain all the reasons
why one alternative is better, or more preferable, than another.
With AHP, you calculate weighted scores for each set of criteria
that you consider, and then you use those weights to calculate
a final score for each alternative. The result is an "apples to
apples," quantitative comparison of your choices. Whether you
use this method to make a final choice or as one of many tools
in your decision making process, the results can be remarkably
clear.
Tags:
Decision Making, Skills
situation with more than a few realistic options?
You could sit and think over each option, hoping for divine inspiration
– but you may end up more confused than when you started.
You could leave it to fate – draw straws or pick a number.
Of
course, this won't win you the Decision Maker of the Year award!
An all-too-common strategy is to simply wait out the problem,
doing nothing proactively, until a solution is somehow chosen
for you by circumstances.
None of these approaches are very effective. What you need is
a systematic, organized way to evaluate your choices and figure
out which one offers the best solution to your problem.
As rational beings, we usually like to quantify variables and
options to make objective decisions. However, the problem is that
not all criteria are easy to measure.
So what do you do when you're faced with a decision that needs
significant personal judgment and subjective evaluation? How do
you avoid getting caught in the "thinking over" stage?
And how can you be more objective?
Combining Qualitative and Quantitative
To address this problem, Thomas Saaty created
the Analytic Hierarchy Process (AHP) in the 1970s. This system
is useful because it combines two approaches – the "black
and white" of mathematics, and the subjectivity and intuitiveness
of psychology – to evaluate information and make decisions that
are easy to defend.
Let's look at an (admittedly slightly trivial) example. If you
want to determine the best route to work in the morning, and travel
time is your deciding factor, the decision making process is very
straightforward and simple. You would use each alternative route
for a week, time the commute, and choose the one that's fastest
on average.
If, however, you carpool with other riders, and you have to consider
everyone's priorities, the decision becomes much more complex.
Larry is concerned about his personal safety, because one route
goes through a dangerous part of town. Joanne wants to factor
in a stop at a drive-through coffee shop, so that everyone can
get coffee. Richard points out that Java Jolt is better than Cuppa Jo.
There are several branches of each in the city, with both types
accessible from all routes, although at different distances.
Now you've got tangible and intangible, and quantitative and qualitative,
factors to think about. And you have to consider the different
perspectives and priorities of the various people.
AHP can combine these different types of factors and turn them
into a standardized numerical scale. You can use this to make
your choice objectively, while including all the decision criteria.
AHP Snapshot
Here's a quick overview of the Analytic Hierarchy
Process.
Build your "hierarchy"
Define your goal or objective.
Identify the choices you're considering.
Outline the major factors you'll use to evaluate each option.
Identify criteria (and any subcriteria of these) that you need to consider for each of these major factors. Link these to the major factors (see figure 1 below for an example of this.)
Continue to build a hierarchy of decision criteria until all factors are identified and linked.
Establish your priorities
Using paired comparison , determine your criteria preferences (perhaps A is a little more preferred than B, B is much more preferred than C, and so on).
Rate these preferences from 1-9.
Repeat this for each level in your hierarchy. (The example below will make this clearer.)
Synthesize, or combine, the ratings
Calculate weighted criteria scores that combine all of the ranking data.
Compare the alternatives
Using those combined scores, calculate a final score for each alternative.
Tip 1:
If you're familiar with Paired Comparison Analysis , then this approach might sound familiar. The power of AHP is that it uses paired comparison to determine the relative weights of various criteria, and then it transfers them across each level of criteria to calculate overall weightings. From this you can calculate an objective score for each alternative.
Tip 2:
This is a complicated approach, and one that needs careful
thought and calculation. Only use it if Paired Comparison
Analysis isn't giving you the answer you want, or where
the problem to be solved is complex and significant, and
involves many different subjective factors.
Let's work through our example to determine
which route – A, B, or C – is best for the group of carpoolers.
Step 1: Build Your Hierarchy
Define your goal at the top of the hierarchy:
Find the best route to work.
Identify your first level of evaluation criteria:
Commute time.
Safety.
Drive-through access (to coffee).
Decide if there are second level criteria,
or subcriteria of these, related to any of your Level 1 criteria:
Drive-through access has two subcriteria:
Java Jolt.
Cuppa Jo.
Under each bottom level criterion, write down the alternatives you're considering:
Route A.
Route B.
Route C.
Figure 1: An Example Hierarchy
Step 2: Establish Your Priorities
Have each decision maker rate the relative importance
or preference for each criterion, at each level. Use a paired
comparison approach.
Set up a matrix to compare each criterion to the others. We have three Level 1 criteria, therefore we need a 3x3 matrix.
Criterion
Commute Time
Safety
Drive-through Access
Commute Time
Safety
Drive-through Access
Rank the importance of each criterion relative
to the others, using the scale below.
Relative Importance
Value
Equal importance/quality
1
Somewhat more important/better
3
Definitely more important/better
5
Much more important/better
7
Very much more important/better
9
Note 1:
The numbers 2, 4, 6, and 8 are half way positions
between the values above.
Note 2:
The scale assumes that the ROW (first)
criterion being ranked is of equal or greater importance
than the COLUMN (second) criterion. If you have a pairing
where the row criterion is less important than the column,
use the reciprocal value (1/3, 1/5, 1/7, or 1/9).
Compare the criteria in the columns to the
criteria in the rows.
Commute time is [much MORE important] than safety
Commute time is [somewhat MORE important] than drive-through access
Drive-through access is [extremely MORE important] than safety
Safety is [extremely LESS important] than drive-through access
Safety is [much LESS important] than commute time
Drive-through access is [somewhat LESS important] than commute time
Criterion
Commute Time
Safety
Drive-through Access
Commute Time
1
7
3
Safety
1/7
1
1/9
Drive-through Access
1/3
9
1
Step 3: Calculate the Ratings
Now you have to calculate the overall weighting for each criterion. This is called a "priority vector" (PV) (don't worry about this term – it's not very helpful.)
Add each column in your ratings matrix.
Criterion
Commute Time
Safety
Drive-through Access
Commute Time
1
7
3
Safety
0.14
1
0.11
Drive-through Access
0.33
9
1
TOTAL
1.47
17
4.11
Divide each entry by the total of its column.
Examples:
Commute time ÷ Commute time total = 1/1.47
Safety ÷ Commute time total = 0.14/1.47
Criterion
Commute Time
Safety
Drive-through Access
Commute Time
0.68
0.41
0.73
Safety
0.10
0.06
0.03
Drive-through Access
0.22
0.53
0.24
Notice how the columns add up to approximately 1.0. This is because the weights have now been standardized.
Because the ratings are subjective, we sometimes see inconsistencies. To "smooth" these out, calculate the average of each row. This is the final weight (priority vector) for each criterion.
Criterion
Commute Time
Safety
Drive-through Access
Priority Vector
Commute Time
0.68
0.41
0.73
0.61
Safety
0.10
0.06
0.03
0.06
Drive-through Access
0.22
0.53
0.24
0.33
This weighted score suggests the following:
Commute time represents about 61% of the final decision
Nearness to a coffee drive-through represents about 33% of the decision
Safety represents about 6% of the decision
Level 2 Criteria – If you had no subcriteria,
you could move onto the next step and calculate final scores
for each alternative route. In our example, drive-through access
has another variable to consider – whether the coffee shop on
the route is Java Jolt or Cuppa Jo.
Following the same steps as the Level
1 criteria, start with a 2x2 matrix.
Then add the ratings. For our example,
we'll assume that Java Jolt is somewhat better (3) than
Cuppa Jo.
Brand
Java Jolt
Cuppa Jo
Java Jolt
1
3
Cuppa Jo
0.33
1
Calculate the overall weighting.
Add each column.
Brand
Java Jolt
Cuppa Jo
Java Jolt
1
3
Cuppa Jo
0.33
1
Total
1.33
4
Divide each entry by the column total. Then average each row.
Brand
Java Jolt
Cuppa Jo
Average/Priority Vector
Java Jolt
0.75
0.75
0.75
Cuppa Jo
0.25
0.25
0.25
Access to a Java Jolt
is three times more preferred than access to Cuppa Jo.
Step 4: Compare the Alternatives
Compare each alternative based on the lowest
level in your hierarchy of decision criteria. In our example
(see figure 1), the lowest-level criteria are as follows:
Commute time
Safety
Java Jolt
Cuppa Jo
Note:
The weighting for
drive-through access will be included as 33% of the final
decision. It will be split 75/25 between nearness to Java
Jolt or Cuppa Jo.
By using the lowest level of your hierarchy,
you ensure that all variations of all options are considered.
You can have as many levels of subcriteria
as you need to make a final decision. If Java Jolt and Cuppa
Jo were relatively equal in their preference (score of 3 or
less), then you could further break down the decision into taste,
price, and muffin selection. With each level, the total weight
always adds to about 1.0, and the overall weight is spread up
the hierarchy through each subsequent calculation.
Create a comparison matrix for the first
decision criterion you want to evaluate. We'll use commute time.
Commute Time
Route A
Route B
Route C
Route A
Route B
Route C
Then use the same 1-9 rating scale to determine
how each route compares to the others, based on that decision
criterion. We'll assume the following:
Route A is somewhat faster than Route
B.
Route A is very much faster than Route
C.
Route B is much faster than Route C.
Fill in the matrix, and calculate the priority
vector.
Commute
A
B
C
A
1
3
9
B
0.33
1
7
C
0.11
0.14
1
Total
1.44
4.14
17
Commute
A
B
C
Priority Vector
A
0.69
0.72
0.53
0.65
B
0.23
0.24
0.41
0.29
C
0.08
0.03
0.06
0.06
Repeat this comparison process for each
of the remaining decision criteria.
Which route is safer than the other?
How much closer to a Java Jolt is one route
than the other?
How much closer to a Cuppa Jo is one route
than the other?
Safety
A
B
C
A
1
0.14
0.14
B
7.00
1
0.2
C
7.00
5
1
Total
15
6.14
1.34
Safety
A
B
C
Priority Vector
A
0.07
0.02
0.11
0.07
B
0.47
0.16
0.15
0.26
C
0.47
0.81
0.74
0.68
Access to Java Jolt
A
B
C
A
1
4
7
B
0.25
1
3
C
0.14
0.33
1
Total
1.39
5.33
11
Access to Java Jolt
A
B
C
Priority Vector
A
0.72
0.75
0.64
0.70
B
0.18
0.19
0.27
0.21
C
0.10
0.06
0.09
0.09
Access to Cuppa Jo
A
B
C
A
1
1
0.11
B
1
1
0.2
C
9
5
1
Total
11
7
1.31
Access to Cuppa Jo
A
B
C
Priority Vector
A
0.09
0.14
0.08
0.11
B
0.09
0.14
0.15
0.13
C
0.82
0.71
0.76
0.77
Combine the overall weights, and determine
a value for each route. (See below for this.)
The value for each route is the weighted
sum of all rankings that are associated with it. The priority
values (PVs) for each criterion have been added to the hierarchy
below.
Figure 2: The Example Hierarchy Developed
Calculate Route A's final score:
Commute Time PV (0.61) x Route A's Commute PV (0.65) +
Safety PV (0.06 ) x Route A's Safety PV (0.07) +
Drive-through Access PV (0.33) x Java Jolt PV (0.75) x Route
A's JJ PV (0.7) +
Drive-through Access PV (0.33) x Cuppa Jo PV (0.25) x Route
A's CJ PV (0.11) +
= 0.40 + 0 + 0.17 + 0.01
= 0.58
Complete the calculations for Routes B and C.
Final Scores
Criterion
Route A
Route B
Route C
Commute Time
0.40
0.18
0.04
Safety
0.00
0.02
0.04
Java Jolt
0.17
0.05
0.02
Cuppa Jo
0.01
0.01
0.06
Total
0.58
0.26
0.16
Route A is the clear winner! You can interpret this to mean that Route A meets 58% of all the decision criteria considered. Route B meets only 26% of the criteria, and Route C meets 16%.
Key Points
The Analytic Hierarchy Process can help you
quantify the judgments you use in decision making. When problems
become complex, it's hard to justify and explain all the reasons
why one alternative is better, or more preferable, than another.
With AHP, you calculate weighted scores for each set of criteria
that you consider, and then you use those weights to calculate
a final score for each alternative. The result is an "apples to
apples," quantitative comparison of your choices. Whether you
use this method to make a final choice or as one of many tools
in your decision making process, the results can be remarkably
clear.
