Professional Documents
Culture Documents
References:
Benjamin & Cornell, ‘Probability, Statistics & Decision Making for Civil Engineers’.
Chapter 5. McGraw Hill
Problem:
We have a decision maker who has a set of strategies that may be followed and
an exhaustive set of mutually exclusive states of the world which may confront
the chosen strategy. We assume that we can calculate the value to the decision-
maker of the outcome of any particular strategy with any particular state of the
world.
Only one state of the world can actually occur and we sometimes assume that we
know the probability of occurrence of each.
We want to find the ‘best’ strategy to follow in the face of the uncertainty of the
problem.
For any one criterion, one strategy will be the best or several will be equally
superior to the rest.
The different criteria are likely to give different ‘best’ strategies and no criterion is
guaranteed to give the strategy that is the best choice for the state of the world
that actually occurs.
We will, however, have studied the information of the problem and have made
our decision in a rational and sensible manner.
STRATEGIES
• All alternatives should be considered and discarded for good reason rather
than being forgotten.
• These are future happenings over which the decision maker has no
control.
PROBABILITY ESTIMATES
Statistical Probabilities
Subjective Probabilities
Experience and training could lead us to believe that some SOWs are more likely
than others for the problem in hand.
It is important that ∑p=1 since something must happen and the SOW list is
exhaustive. Also, only one SOW will occur in practice since the SOWs are
mutually exclusive.
If all the strategies are evaluated for every possible state of the world, the results
can be displayed as a Pay-off Matrix. This will show the outcome of each
strategy i for each possible SOW j.
In the payoff matrix, each cell i,j represents the utility of strategy i in SOW j and
is denoted by U(i,j).
EXAMPLE
From our knowledge of the effect of the chemical, we can split the possible
concentrations into classes. Say:
<0.3%
>2.0%
Assume that from previous work, we know that pipe type 1 costs 10 units per
installation and has a life of
Pipe 2 costs 20 per installation and has a life expectancy of 20 yrs for
concentrations <2.0% but will require heavy remedial work at a cost of 10 to
give the 20yr life for higher concentrations.
Use pipe 1
Use pipe 2
<0.3%
>2.0%
Having used our engineering knowledge to give us the strategies and states of
the world, we then use simple ‘engineering economics’ to evaluate each cell of
the payoff matrix and thereby give the bulk information required to evaluate the
strategies.
There is always a decision to be made and the payoff matrix acts as a basis for
this. However, the choice between the strategies is not usually straight-forward
even after the information has been produced and ordered.
Consider as an example the choice of the length of piles to order for a particular
job.
It has been estimated that the depth to set (D) of the piles will be 20m, 25m, or
30m.
Piles cost £1 per metre and have a scrap value of £0.5 per metre. The contractor
is paid £1.2 per metre of pile installed.
If the pile is too long, the surplus is burned off and sold for scrap. If the pile is
too short it is lengthened with a weld costing £3.
In general we should ask ourselves what would we like our chosen strategy to do.
A common answer is
Maximize profit
For strategy i, there will be some SOW j for which the payoff U(i,j) is as small as
or smaller than that of all the other SOWs.
Rationally we can now choose the strategy with the best worst outcome.
Therefore the best worst is from strategy 1 and we will choose to order 20m
piles.
For strategy i, there will be some SOW j for which the payoff U(i,j) is as large as
or larger than that of all the other SOWs.
Rationally (??) we can now choose the strategy with the best best outcome.
Therefore the best best is from strategy and we will choose to order m piles.
This seeks to give some weight to the best outcome for a strategy as well as a
the worst one in order to give a more ‘balanced’ view of the strategy value.
where we can set m to suit our level of optimism or conversely, we can see what
level of optimism a decision implies.
It can be seen that the Hurwitz value alters linearly with the measure of optimism
and a simple way of plotting the graph can be developed from this.
Draw a vertical axis line at either end of the optimism line. The one on the left
represents utter pessimism and the other complete optimism.
For each strategy in turn, plot the worst possible outcome on the pessimistic
‘axis’ and the best on the ‘optimistic’ axis.
People often review decisions with perfect hindsight and full knowledge of what
SOW did in fact occur. (Or at least they like to think that they do).
The wrong strategy is sometimes employed for the particular SOW which existed.
In this case we can make some attempt to measure the regret which we
incurred by our choice of strategy.
This is most easily done in terms of the difference between the outcome of
strategy i in SOW j (the one we chose ) and the outcome of strategy k with SOW
j ( where strategy k is that strategy which would give the best outcome for
SOW j ).
This can be calculated for every combination of strategy and SOW and the result
is the regret equivalent of the payoff matrix.
The sensible decision is then to choose the strategy with the Minimax regret. i.e.
the strategy which gives the minimum maximum regret or that which is the Best
Worst option.
In the example of the choice of pile lengths, the regret matrix would be:
Expectation Measures
The criteria so far described are likely to be unduly affected by extreme and
unlikely states of the world and may therefore perhaps not give sensible answers
in all situations.
These are used to value the strategies in terms of their long term average
performance if the same decision were to be made many times, independently.
This is done in spite of the fact that the decision is, in reality, usually only made
once and is rarely made a large number of times.
In the piling example, say that the probabilities of the depth of set of the piles
are believed to be:
This gives
UTILITY
People tend to react to small gains and losses of similar size with changes in
feeling of ‘well being’ which are about equal in size but of opposite sign.
People tend to be more moved to disgust for large losses than they are moved to
elation by gains of equal size.
Utility measures try to use these subjective reactions to get a more satisfactory
of an outcome and a strategy.
• Principle of Indifference
When a decision-maker cannot choose between two courses of action he is
said to be indifferent between them and they are of equal value in the
current value system.
• Standard Gamble
A gamble with only win or lose outcomes, known prizes and probability of
winning can be altered at will.
We have to evaluate the utility for a particular decision-maker for the outcomes
assessed for a particular problem.
Method
1. The decision-maker chooses the best and worst outcomes and values the best
at 100 and the worst at 0.
2. Take the best and worse outcomes and construct a standard gamble with
these as win and lose prizes.
3. Take each outcome in turn and adjust the win/lose probabilities until the
decision-maker is indifferent between the outcome and the gamble.
4. If the ‘win’ indifference probability is p(n) for outcome n then the utility (u) of
outcome n is given by