02.03.2013 Views

Thinking and Deciding

Thinking and Deciding

Thinking and Deciding

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

322 UTILITY MEASUREMENT<br />

I could continue in this manner, dividing the utility interval in half each time. If<br />

I assigned a utility of 0 to $0 <strong>and</strong> a utility of 8 to $1 million, I would assign a utility<br />

of 4 (halfway between 0 <strong>and</strong> 8) to $x, once I discovered what $x was. I could then<br />

divide the interval between $0 <strong>and</strong> $x in half in the same way, <strong>and</strong> assign this wealth<br />

a utility of 2, <strong>and</strong> so on. In my case, I might assign a utility of 2 to $50,000, since<br />

this would be subjectively halfway between $0 <strong>and</strong> $150,000.<br />

Notice that the bottom <strong>and</strong> top values of the scale could be chosen arbitrarily. To<br />

use utilities in decision analysis, we need only relative values. For example, suppose<br />

that I have $50,000 in savings <strong>and</strong> I want to know whether I ought to invest it in a<br />

real estate deal that has a .5 probability of netting me $1 million <strong>and</strong> a .5 probability<br />

of losing everything. The expected utility of the deal is .5·u($1 million)+.5·u($0).<br />

The way I have scaled utility, this would be .5 · 8+.5 · 0, or 4. This is higher than<br />

the utility of my current wealth, which is 2, so I ought to go for it.<br />

Suppose I had used a different scale, though, so that I had assigned a utility of<br />

100 to $0 <strong>and</strong> a utility of 200 to $1 million. When I cut this interval in half, I would<br />

have discovered that the utility of $150,000 is 150, halfway between 100 <strong>and</strong> 200,<br />

<strong>and</strong> the utility of $50,000 is 125. In this case, the expected utility of the real-estate<br />

deal is 150, <strong>and</strong> the utility of my current wealth would be 125, still less than 150,<br />

<strong>and</strong> by the same proportion of the total range as before.<br />

In order for difference measurement to be used, the utilities of outcomes must<br />

be related by a condition called monotonicity. Consider the following outcomes, laid<br />

out in order of desirability. The letters A through F might st<strong>and</strong> for different amounts<br />

of money, or different jobs.<br />

A B C D E F<br />

Suppose that the difference between A <strong>and</strong> B is subjectively equal to the difference<br />

between D <strong>and</strong> E, <strong>and</strong> the difference between B <strong>and</strong> C is subjectively equal to the<br />

difference between E <strong>and</strong> F. It must be the case, then, that the difference between A<br />

<strong>and</strong> C is also equal to the difference between D <strong>and</strong> F (Krantz, Luce, Suppes, <strong>and</strong><br />

Tversky, 1971, ch. 4). If this monotonicity condition is not met, inconsistencies will<br />

arise in the scale. To see how that might happen, imagine that F is directly above E,<br />

rather than to the right of it. Then the distance between E <strong>and</strong> F could still be the<br />

same as that between B <strong>and</strong> C, yet the distance between A <strong>and</strong> C would be greater<br />

than that between D <strong>and</strong> F. The outcomes would not lie along the same line.<br />

Note that the difference method can be used directly, as in the examples just<br />

given, or it can be used as a check on the direct-rating method. If the rater underst<strong>and</strong>s<br />

that the differences between ratings are supposed to be meaningful, this may<br />

be enough to correct distortion of the scale. Either way, the monotonicity condition<br />

can serve as a check. As yet, no one has asked experimentally whether monotonicity<br />

applies either to direct ratings or to difference measurement. Perhaps this question<br />

has not been asked because no one has thought of a reason why the monotonicity<br />

condition would ever be violated systematically. Checks seem to be a good idea,<br />

however.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!