You are on page 1of 5

11

Tradeoffs and Concessions


In preparing for negotiations, either bilateral or multilateral, each
side should try to sort out its own preferences. Bargainers are continually asked during negotiations whether they prefer one constellation of outcomes to another: Would they rather end up with this or
that? Not only must they decide what they ultimately want, but
they also must determine what they would be willing to give up in
order to achieve their goal. How can a negotiator assess the values
of various tradeoffs, and what effect do these values have on the
dynamics of negotiations?
Suppose that you are the administrator of the Environmental Protection Agency and that you must choose between Policy A and
Policy B. Your staff has prepared atable listing the attributes that
are of concern to you (some involvng economc efficiency, some
economic equity, some health indices, some environmental indices,
some political indices) and has evaluated the two policies on these
aUributes. A is better than B on some aUributes and worse on
others. How can you think systematically about such composite sets
of evaluations? This issue arises not only in negotiations, but more
broadly in decision and policy making.
The problem is mind-boggling in its complexity, but formal analysis can help bring some order to the morass. One approach is to try
to generate scoring systems that assign points to various levels
within each attribute and that quantify tradeoffs between issues.
This is not easily done, but values can be probed by observing preferences between simple hypothetical choice s for which aH but two
or three attributes have identical scores, and then by invoking some
intuitively plausible consistency requirements. Most decision and
policy makers are skeptical and suspicious of this whole approach.
They just don't see the need for formalization, believing that the

148

decision maker can simply make a subjective choice among the real
alternatives when they are presented at the time of the decision.
But now let' s change the setting. Suppose that you as the EPA administrator have to give instructions to a representative who must
negotiate a complex contract with industry representatives. Several
issues are involved and compromises wiU have to be made during
negotiations. What' s more, you must handle dozens of these same
kinds of negotiations simultaneously. At this point, the desire to establish the equivalent of a formal scoring system becomes more
compelling: without it, the representative would be at sea, with no
way of knowing how to make tradeoffs between ssues, and you
would not be able to delegate your authority.

THE ADDITIVE MODEL


Assume that prior to its negotiations with AMPO, City listed the ten
issues to be discussed and the possible levels on eaeh of the issues.
The City negotiators were concerned about money, real and perceived security of its citizens, security of the police, symbolic consequenees with possible ramifications for other wage negotiations,
poltical image, and so on. Suppose that they started out monetizing
various issues, sueh as starting salaries, maximum salaries, vacations, creation of the rank of corporal, number of sergeants; but that
they found it hard to put a price tag on the reinstatement of suspended officers (there was a principIe at stake), on two-man patrols
(lives were at stake), on the Polce Review Board (justice and alienation were at stake), on the police commissioner (the mayor's job
may have been at stake). How eould they put a dollar figure on what
happened to the Police Review Board? One way to do this would
be to imagine a situation in which everything was settled except
the issues of the Police Review Board and the stalting salary level.
The negotiators could then decide how they would be willing to
trade one against the other-in effeet, aeting as if they were plaeing a monetary value on various Police Review Board options. It' s
the structure of the problem situation that essentially forces this
evaluation.
When we turn our aUention to other applieations (such as international treaty negotiations), reducing everything to money may not
be convenient or appealing. Some abstract scoring system may be

150/ TWO PARTIES, MANY ISSUES

easier to work with. In the case of AMPO versus City we could have
evaluated City's reactions for nonmonetary issues in terms ol'
equivalent salary concessions, and thereby monetized these nonmonetary concems. This might, in fact, have been the more "natural" approach. But the introduction of abstract scores for City
served a useful purpose: they will be easier and more comfortable
to handle when we de al with subsequent examples like the Panama
Canal Treaty and the Camp David negotiations.
In the laboratory experiment, we assumed that City and AMPO
assigned a specific point score to each outcome level on each issue
and then added these to get an entire contract evaluation.We'll call
this an additve scoring system-although there was one small deviation from this system. Remember that if City held AMPO to zero
additional vacation days for all officers, City achieved a boIius of 10
points. In this case we simply could not add up City' s score for
these two issues. The bonus introduced what is known as an interaction effect between the vacation issues. If we combined thetwo
separate vacation issues into a single composite issue, then we
would have strict additivity among the nine resulting issues.
Considering just two issues-starting salary and number of sergeants-suppose that the other seven issues (treating vacations as a
composite issue) are already fixed. We're now investigating tradeoffs between starting salary and sergeants only. In the scoring system we are using, notice that any tradeoff comparisons between
levels on these two issues do not depend on the levels of the remaining seven issues: the tradeoffs between starting salaries and
sergeants can be said to be preferentially independent of the levels
of the remaining issues. Indeed, it can easily be seen that with an
additive scoring system, the tradeoffs between the levels of any two
issues are preferentially independent of the levels of the remaining
issues. It can also be seen (but not so easily!) that the converse is
true: if there are more thantwo issues, and ifthe tradeoffs between

the levels on any two issues are preferentially independent of the


remaining issues, then an additive scoring system is appropriate.
Let' s look at one particular technique for obtaining scores for the
additive case, using a fictitious situation that is just complicated
enough to illustrate the complexities 1 wish to address. Suppose
that you, the manager of an expanding business, are entering into
negotiations with a building contractor for the construction of a fac-

TRADEOFFS AND CONCESSIONS / 151

tOly. You are concemed about three factors: cost, time to completion, and quality. From preliminary discussions you limit the
ranges of these factors to, respectively, $3.0-4.5 million, 250-400
days, and a "best" value of 1 to a "worst" value of 5 (on an ordinal
scale). You would most prefer a cost of $3.0 million, a time of 250
days, and the best quality (an index of 1). But you realize that it' s
highly unlikely you will be able to negotiate such a de al.
Assume that your tradeoffs between the levels of any two factors,
keeping the level of the third factor fixed, do not depend on the
level of this third factor. For example, your tradeoffs between cost
and time do not depend on quality, as long as the level of quality is
he Id fixed. So it' s legitimate in this case for you to seek an additive
scoring system. You agree for nOlmalization purposes to give the
best contract ($3.0 million, 250 days, quality 1) a score of 100 points
and the worst contract ($4.5 million, 400 days, quality 5) a score of
zero points. This is like an exam with three questions, in which the
scorer must decide how much weight should be given to each question and how many points should be given to each paItially correct
answer. You decide to score individual factors in the same way
(100 = best, O = worst), and to combine the scores with proportional weights that sum to 1. For example, suppose that you give a
weight of.5 to factor e (cost), a weight of.3 to factor T (time), and a
weight of .2 to factor Q (quality). Suppose that the internal component scoring is as shown in Figure 25. A contract that gives you $4
million, 350 days, and quality 2 would then receive-multiplying
weight times score for each factor-a total score of (.5 x 50) + (.3 x
25) + (.2 x 80), or 48.5 points.
How should you detennine the weights of the factors (reflecting
the impoItance of each) and the component scoring within each factor? Following are some observations that should provide insights
into these questions. 1
StaIting from the worst case ($4.5 million, 400 days, quality 5), if
you have the choice of improving one factor from the worst to the
best level, let' s suppose that you would most prefer to improve the
cost factor first, the time factor second, and the quality factor third.
This reflects the ordinal ranking of the weights. Suppose, fmthermore, that you would be indifferent between improving the cost
1. For a systematic discussion, see Keeney and Raiffa (1976).

TRADEOFFS AND CONCESSlONS / 153

factor alone and improving both the time and quality factor together,
reflecting the fact that.5 = .3 + .2. Using the exam analogy: getting
the cost question perfect and the other questions completely wrong
would be as desirable as getting the cost question completely wrong
and the others completely right.
The cost jactor. The more you spend, the more important it is
that you save a given increment of money. Reducing your costs
from $4.5 million to $4.0 million is just as important as reducing
your costs from $4.0 million to $3.0 million, which accounts for the
shape of your cost function.
The time jactor. Improving the time value from 400 days downward is not very important at first, but improvements become more
important as the value goes from 350 days to 300; the re afte 1', the
value of time reductions decreases (which accounts for the shape of
your time function).
The quality jactor. Going from one quality index to the next is
approximately worth the same as moving between any other two indices, except that quality index 2 is doser in value to quality index
1 than to quality index 3.
Roughly, the way to go about constructing any scoring system is
to formulate sorne rough guidelines, and then to tune the system by
manipulating numbers and curves and by testing the implied results. There are fancier and more systematic methods, but the task
should be approached in the same way one would grade an exam
with several questions. If you are sol el y responsible for giving a
grade and you don't have to explain your grading to anyone else or
to the student, then you might want merely to respond intuitively
and impressionistically to the entire exam. But if you want someone
el se to do the grading for you, then sorne scoring system, even if it
is not perfect, can be a great help. A case can also be made for
adopting sorne formal system of grading even if you are not accountable to anyone and do not plan to delegate authority to an
agent. A formal scheme of your own devising might help you decide how to grade each question separately and how to combine the
scores of different questions.
An additive scoring system sometimes falls far short of what is
reasonable. This may be a result of the interdependence between
factors, an extreme example being the case where preference ranking of levels within one factor depends on the level of another fac-

Score
(J

oCIl

....
o

S'
~
:3 p
(1)
oq'

...::r-

o'

::l

'"

CIl

o.....,
en Q..

2SO
...,
~

'!j
~.

'"

:::

Score

iil

....
8

to

S11

to:>o

rJJ
(')

::l,
::1

OQ
(JO

'<
(JO

.....

(!)

:3
...O'

(1)

oq'
::r-

...,.

::l
3

<b

pj

'<

g:.

'"

'"
<:>

'"

(JJ

S'

Q..

c..,

to

(JJ

'"o
.:..

Ol

......sr'

....,

(!)
(')

'"

......

~
oq'
(1)

:r

::l

><

to ~
....
'<

....
8

to

Q..
<b

o
.....,
..o
::

Score

aJ

(JJ

~
.:..

I
",'o

to:>

'"

TRADEOFFS AND CONCESSIONS

/155

154/ TWO PARTIES, MANY ISSUES

toro For example, the better the military defenses of an ally of Country X, the better offCountry X will be; however, X's preferences for
the ally's military defenses (the more the better) might reverse (to
the less the better) if the level of their friendship slips below sorne
critical point.
Factors may also be interdependent when there is a need for balance or equity. Suppose that you are a negotiator, acting in a benevolent way so as to favor two groups (A and B) internal to your side.
For any contract you negotiate, you are primarily concerned with
the benefits to groups A and B. For political reasons you must make
sure, however, that the benefits to A are commensurate with those
accruing to B. The value of an increase in benefits to A may depend
critically on the level ofbenefits to B; indeed, ifbenefits to B are at
a very low level, the increase in already high benefits to A may be
deemed undesirable. An additive scheme that scores the benefits
independently for A and for B and adds these together misses the
need for balance.
In cases such as these, a nonadditive scoring system can be used.
Nonadditive systems are not too difficult for current state-of-the-art
measurement, but they are too difficult and too involved to be di scussed heTe. Suffice it to say that often there may be many factors
under consideration, but only a few will be interdependent; negotiators can derive advantage from grouping them together and treating them as one composite factor in an othelwise additive scheme.
. VALUE AND UTILITY FUNCTIONS
Researchers sometimes distinguish between a value scoring
scheme and a utility scoring scheme (see Keeney and Raiffa, 1976),
but this distinction is not standard. In the case involving cost, time,
and quality, the scoring system, as we have seen, allows you to assign an overall numerical value to any contracto The scoring system
has been tuned in such a way that contracts with higher scores are
prefe rred. No uncertainties are involved. Such a system can be
caBed a value scoring system.
Now suppose that you must decide between a compromise contract ($4 million, 350 days, quality 2) and a gamble in which, witb
equal probability, you could end up wth the best contract ($3 million, 250 days, quality 1) or with the worst contraet ($4.5 million,

400 days, quality 5). The value scores of the best and worst eontracts are, respectively, 100 and O, and therefore the gamble has an
expeeted value return of -50. But regardless of what the numbers
imply, you might strongly prefer the certainty of the contract with a
score of 48.5 to the uncertainty of the gamble with the higher expected score of 50. This is not surprising, beeause the scoring system was constructed on the basis of nongambling tradeoff options:
the derived numbers do not reflect any attitudes toward risk. Here
is where the advantages of utility scoring become apparent. Such
techniques enable one to find suitable scoring procedures that not
only reflect preferences under certainty, but that appropriately use
expected utility calculations as guidelines forchoices between lotteries with well-specified probabilities.2
In negotiations, probabilities may become relevant in several
ways. The consequences associatred with an agreed-upon final contract might involve uncertainties not under the control of the negotiators. Differences in probability assessments might be exploited
in terms of contingency contracts. But even in idealized cases
where there are no external uncertainties outside the control of the
negotiators, each negotiator is uncertain about what his adversary
ultimately will do. Should Steve hold out for $350,000 in the Elmtree House sale, instead of settling for $300,OOO? Should a union,
which can secure a given contract from management, refuse to
accept the contract and submit to the uncertainties of voluntary
arbitration?
A well-developed theory of utility analysis has been devised to
handle both uncertainties and multiple attributes, but the theory,
while operational, is not easy to use and requires a level of coherency that few individuals, and still fewer groups, achieve. Most
people, even in simple risky situations, don't behave the way the
theory of utility would have them behave. There are a few researchers who prefer to trust the recommendations of formal utility
analysis rather than their own intuition, even though this behavior
would not occur without the existence ofthe theory. A larger num2. Many analysts as sume that a value scoring system-designed for tradeo!l's
under certainty-can also be used for probabilistic choice (using expected values).
Such an assumption is wrong theoretically, but as 1 become more experienced 1
gain more tolerance for these analytical simplifications. This is, 1 believe, a relatively
benign mistake in practice (see Bell and Raiffa, 1980).

156/ TWO PARTIES, MANY ISSUES

ber of analysts who understand the theory simply don't trust it; tht"
point to examples of situations (the Allais Paradox, the Ellsht~'l
Paradox, the experimental results of Kahneman and Tversky f' ~:,:;,
which they, even knowing the theory, would deliberately act out ui
accord with it. Sorne are probably confused and will eventually s.c;lC
the value of utility analysis. Sorne are not confused, but have d(,t,;"l>
psycholo'gical concerns; they may anticpate that a given act mig.tJ&1
lead to an unfortunate outcome, which will result in persistel4
deeply felt pangs of regret. Such psychlogical concems are usuaH:.
not accommodated in applications of the theory of utility, but ,,;
principIe they could be-with further complexities in the theOf)
Even though you, as a negotiator, might want to act reflectivel~.
coherently, and rationally, your adversaries in alllikelihood will Jtl
with very limited rationality. Don't be naive and expect them to l)t>,
have like you may want to behave. However, if they are prone ~.
the gambler's fallacy; to an excessive zeal for certainty, to an tol'
cessive avoidance of potential ex post regret, to misperceptions >:ti
small probabilities (one could come up with a litany of so-call<:'>:
nonrational, descriptive behaviors), then you might be able to t"L'
ploit such behavior in negotiations.

60

400

rn

>.

~ 350

e
t.::
~

300

250 I

3.0

85

70

150

30

3.5
4.0
4.5
Cost (in millions of dollars)

ngure 26. Iso-value curves for Mr. Hee (solid lines) and Ms. Shee (brokn lines). Her direction of preference is southwest; his is northeast.

lOOr
TRADEOFFS WITH TWO CONTINUOUS ISSUES
Mr. Hee and Ms. Shee are negotiating over two continuous issllt,,,.
cost and time. The ranges under discussion are $3.0-4.5 millic.,l,
and 250-400 days. He wants high dollars and high days; she want.,
low dollars apd low days. Figure 26 indicates by means of indifTc:::,z
ence (iso-value) curves their respective tradeoffs. He, for exampit:
deems contracts V, Q, and P equally desirable, and thus they are (~.t
his same iso-value curve; he prefers contract R to any of the vallii:~
equivalent contracts V, Q, and P, and hence R is on a higher SoYvalue curve. He wants to go northeasterly; she southwesterly. Let'!
suppose that they have tentatively settled on a contract agreeme.,OOi
of $4.0 million and 275 days, which is depicted as point Pan':,
which is scored 20 for him and 50 for her (see Figure 27).
Notice from Figure 26 that if the final contract were to be mO\'6:::
from point P along Mr. Hee' s iso-value contour (along the are PQ\'\
3. See Raiffa (1968) and Kahneman and Tversky (1979).

~
~

75

...c:
Vl
..;
~

..

50

Vl

25

25

50

75

100

Score for Mr. Hee


figure 27. The efficient frontier. (Point P represents theioint scores for a

You might also like