class Recall
Last we looked at:
Interpersonal strategies; and Extrasensory perception (ESP)
These extended examples showed how to apply some of what we have learned in critical thinking to practical cases. These are examples of determining what to believe and perhaps what to value. Today we extend this practical approach to consider how to make decisions.
Decision making
What we decide depends on what we believe (discussed earlier) as well as what we desire. So we need to look at some procedures that:
help you determine what you value; and can help you decide what to do.
In examining these, we have to account for uncertainty.
When making decisions, it is wise to take into account how likely a result is given a particular action that we might decide to make.
Deciding what we value
Seldom easy. Especially if we have to assign fairly precise relations between things we might value.
E.g. Exactly how much more do you like chocolate than Jell-O?
Some procedures can be used with just a ranking.
E.g. I like Jell-O more than chocolate.
But, the better ones often demand precise weighted ratings, i.e., a preference scaling.
Deciding what we value
N.B.: Even if we aren't sure of the exact scaling, the more precise procedures might be useful since:
1. 2.
They make clear what the options are; and We can consider assignments of the value of an outcome to be variables E.g. If I like chocolate anywhere between 1 and 3 times more than Jell-O, I should not have dessert.
Deciding what we value
Two notes:
1.
2.
Preference scalings are arbitrary scales (the actual numbers dont matter, just the ratios) Notably, even when we have precise values, decision making can be very difficult (even for experts): http://www.gladwell.com/2006/2006_05_29_a_ga me.html
Framing the problem
Framing: identifying a set of possible actions. You should identify an exhaustive set of mutually exclusive actions.
Exhaustive: at least one of the actions in the set will be done and no possible actions are omitted. Mutually exclusive: no more than one can be done at the same time. Sets are often huge: 10 pants, socks, shirts = 10^3 Math can be subtle (McKay is wrong about courses) Can be hard: e.g. do this for getting to school.
This is obviously an ideal
Framing the problem
It is seldom easy to determine what the exhaustive set of actions is, and often difficult (though less so) to know which actions are mutually exclusive. Often when making a decision, we are fairly lax about ensuring that we have an exhaustive and mutually exclusive set actions (too much effort) But for important decisions it's useful to know the ideal constraints on making good decisions
Decisions with known outcomes
Even when there is no uncertainty, it won't always be obvious which of several possible actions is the best one. Usually, this is because we have some difficulty assigning value to the various outcomes, or there are too many to consider. Well consider three ways of making decisions in these cases:
Satisficing (sub-optimal) Elimination by aspects (sub-optimal) Multi-attribute utility theory (MAUT; optimal)
Satisficing
Herbert Simon suggested that people often make decisions by taking the first action that has an acceptable outcome (minimum threshold). This is both beneficial and problematic.
It can be beneficial because it will save a lot of time in making decisions. It can be problematic because, as we discussed in Gilovich, "optional stopping" can bias us in favour of findings evidence that supports a conclusion we already favour.
We are unlikely to find the optimal solution.
Elimination by aspects
A heuristic in which you:
First rank the importance of properties of the outcomes Then you eliminate all possibilities that are unacceptable with respect to the most important property, then the second most important property, and so on Until there's only one option left
Simple method to follow Not optimal since one option might satisfy only the first property while another may not satisfy the first property but satisfy all the others
Human decision making
Note that both Elimination by Aspects and Satisficing are descriptions of standard rules people tend to employ for decision making. Both are also sub-optimal. So, they seem fine for unimportant (quick) decisions But in some contexts, optimality (or near optimality) might be most important
Multi-attribute utility theory (MAUT)
MAUT is optimal but is more difficult to apply. You first construct a preference scaling in which you rate each attribute of the possible outcomes. Then you judge how well each outcome meets each attribute (by using a preference scaling). The final "decision score" is calculated by multiplying each attribute weight by its satisfaction for a particular outcome and summing the result.
MAUT (cont.)
Here is an example (p. 206: 1b):
Attribute weight: 5 cost: 3 Italia 10 1 Ralston 6 7 9 Schwein 1 10 1
Maintenance: 2 3
Italia: (5*10) + (3*1) + (2*3) = 59 Ralston: (5*6) + (3*7) +(2*9) = 69 Schwein: (5*1) + (3*10) + (2*1) = 37
Italia and Ralston are close. In any case, the Schwein should be eliminated.
MAUT (cont.)
Depending on how close the final scores are, we may or may not be confident about the result. In any case is it a useful process that lets you determine what attributes are important for your decision-making. Having gone through this process, you have compiled all of (or at least many of) the relevant considerations. This is difficult to do simply "in your head."
Elimination
We can often minimize the set of possibilities by eliminating some of them. There are two kinds of elimination to consider:
1. 2.
the elimination of unsatisfactory choices; and elimination of dominated actions.
For 1. we can establish a level of minimal acceptability and eliminate any choices that fall below that level. In many cases, each option will have some unacceptable traits so you can't eliminate any.
Elimination
For 2., an action is dominated if every possible outcome of that action is equal to (and one is less than) some other action in the set of possibilities. So, if two options depend on the same external variable, then although in one case a certain option might be more beneficial than another in a different case, it may be dominated by the option over the range of the variable (e.g. crops).
Elimination
For example, suppose I am determined to eat the food with the fewest calories.
Chocolate has between 400 and 600 calories Jell-O has between 50 and 100 calories Crackers has between 25 and 90 calories
Chocolate is dominated by both Jell-O and crackers. So, I can eliminate it from consideration. But, neither Jell-O nor crackers dominates the other
Maximin and maximax
Maximin means maximizing the minimum possible value of the outcome (risk averse). Maximax means maximizing the maximum possible value of the outcome (risk takers).
E.g., you can either choose the option that gives you $100 if you lose and $500 if you win, or the option where you get $0 if you lose and $1000 if you win.
Maximin results in the first option, whereas maximax results in the second. But, both ignore the probabilities (and show why doing so is a bad idea.
Decisions under uncertainty
As mentioned in chapter 2, in many cases it is difficult to know whether a success condition will be reached. Similarly, we often don't know exactly what the true result of taking some action will be. So we need a way of balancing the probabilities of the possible outcomes and the values (often called "utilities") of the possible outcomes.
The expected utility method
Optimal but the most difficult to employ. You need to identify probabilities and determine a preference scale. Often only have rough estimates of probability.
But, even rough estimates can be useful.
The total probability assigned to possible outcomes will be equal to 1
Why? Because the set of events is exhaustive and mutually exclusive
The expected utility method
The preference scale ranks items by their relative utility (i.e. usefulness or desirableness).
So something that is twice as useful as something else will receive twice the value on the preference scale.
Applying the expected utility method is much like applying MAUT
but it is more general (MAUT is a special case) its the best method we consider
The expected utility method
Example (3.a2-d2, p. 224):
weather int. veg ...