Why Utility Numbers Are Not Automatically Probabilities
A utility value of 0.7 can look like a probability, and a value of \(-4\) can look like a loss. Neither interpretation follows from the number alone. In consumer theory, utility numbers often serve a more limited purpose: they represent the ordering of bundles from less preferred to more preferred. When utility is ordinal, the ranking matters; the particular numerical labels do not.
This distinction is important when comparing utility functions, interpreting negative values, or checking whether a transformation preserves preferences. A utility scale can be shifted up or down, or stretched, while representing exactly the same ranking. Consequently, a utility value does not have to be positive, lie between zero and one, or have a meaningful zero point.
The convention that larger utility corresponds to a more preferred bundle is used here. The numerical value assigned to any one bundle is not, by itself, a measure of happiness, a probability, or a monetary amount. In particular, utility \(0\) does not generally mean “no utility,” and utility \(1\) does not generally mark a maximum.
Positive Affine Transformations Preserve Rankings
A positive affine transformation changes a utility function by multiplying it by a positive constant and adding a constant. Let \(a>0\) and let \(b\) be any real number. If \(u\) represents preferences, then \(v(A)=a u(A)+b\) represents the same preferences. The positive multiplier can stretch or compress the scale, while the added constant shifts it.
The equivalence holds because adding the same \(b\) to both sides does not change an inequality, and multiplying both sides by a positive \(a\) does not reverse it. The same reasoning preserves strict rankings and ties. Thus \(u\) and \(v\) assign the same ordering to every pair of bundles, even though their utility values may be quite different.
A transformation must be increasing, not merely different from the original function. If \(a<0\), multiplying by \(a\) reverses every strict inequality: a bundle with higher original utility receives lower transformed utility. Such a transformation represents the reverse ranking, rather than the same preference ordering, unless all relevant bundles are tied.
Worked Examples: Reading and Transforming Utility Scales
Worked Example: Shifting Negative Utility Values
A consumer ranks three service bundles using \(u\): Basic has utility \(-3\), Standard has utility \(1\), and Extended has utility \(4\). A report uses the transformed scale \(v=2u+6\). Compute each transformed value, state the ranking, and decide whether the negative original value signals a bad or unacceptable bundle.
Method. Apply the same transformation to every bundle, then compare the resulting values. This avoids treating the location of zero as an economic threshold.
For Basic, \(u=-3\), so \(v=2(-3)+6=-6+6=0\). For Standard, \(u=1\), so \(v=2(1)+6=2+6=8\). For Extended, \(u=4\), so \(v=2(4)+6=8+6=14\).
The transformed ranking is Extended (\(14\)) above Standard (\(8\)) above Basic (\(0\)). The original ranking was also Extended (\(4\)) above Standard (\(1\)) above Basic (\(-3\)). Since the multiplier \(2\) is positive, the ranking is preserved.
The negative utility of Basic does not show that it is unacceptable or worse than receiving nothing. No “nothing” bundle or meaningful zero threshold was specified. The calculation checks out: each transformed value equals twice the original value plus six, and all three values retain their original order.
Worked Example: Values Below Zero and Above One in One Representation
For bundles of two goods, a consumer’s utility representation is \(u(x,y)=2x+y\). Consider \(A=(1,2)\), \(B=(3,1)\), and \(C=(5,2)\). A second analyst uses \(w=2u-10\). Calculate both utility scales, rank the bundles, and explain whether \(w\) being negative or greater than one conflicts with ordinal utility.
Method. Evaluate the original function at each bundle before applying the affine transformation. This avoids applying the transformation to only one component of a bundle.
For \(A=(1,2)\), \(u(A)=2(1)+2=4\), so \(w(A)=2(4)-10=8-10=-2\). For \(B=(3,1)\), \(u(B)=2(3)+1=7\), so \(w(B)=2(7)-10=14-10=4\). For \(C=(5,2)\), \(u(C)=2(5)+2=12\), so \(w(C)=2(12)-10=24-10=14\).
On the original scale, \(12>7>4\), so \(C\) is preferred to \(B\), which is preferred to \(A\). On the transformed scale, \(14>4>-2\), giving the same ranking. The values include one below zero and two above one, but this presents no conflict: \(w\) is an ordinal representation, not a probability scale.
The calculation checks out bundle by bundle, and \(w=2u-10\) has a positive multiplier. The key trap is to assume that a utility value must be between zero and one because probabilities lie in that interval. Utility and probability are different concepts.
Worked Example: A Nonlinear Transformation Changes Gaps, Not Rankings
Three research plans receive ordinal utility values \(u(A)=-2\), \(u(B)=1\), and \(u(C)=3\). An analyst replaces each value by \(z=u^3\). Calculate the new values and rankings. Then compare the numerical gaps between adjacent plans on the two scales.
Method. Apply the transformation to each utility value and check that the transformation is strictly increasing over the real numbers. This distinguishes preservation of order from preservation of utility differences.
The transformed values are \(z(A)=(-2)^3=-8\), \(z(B)=1^3=1\), and \(z(C)=3^3=27\). The original ranking is \(C>B>A\), since \(3>1>-2\). The transformed ranking is also \(C>B>A\), since \(27>1>-8\). The function \(g(u)=u^3\) is strictly increasing, so it preserves the order of every pair of real utility values.
In the original scale, the gap from \(A\) to \(B\) is \(1-(-2)=3\); the gap from \(B\) to \(C\) is \(3-1=2\). In the transformed scale, the corresponding gaps are \(1-(-8)=9\) and \(27-1=26\). Their relative sizes have changed: the first gap was larger on the original scale, but the second is larger on the transformed scale.
Thus the ranking is unchanged, but the numerical differences do not have a stable ordinal interpretation. The calculations verify both the preserved ordering and the changed gaps. A common error is to infer that \(C\) is “13 times as preferred” as \(A\) because \(27/(-8)\) has some numerical value; an ordinal representation gives no such ratio meaning.
Worked Example: Normalizing a Finite List to the Interval from Zero to One
An analyst records utilities \(u(P)=-5\), \(u(Q)=0\), \(u(R)=7\), and \(u(S)=15\) for four available bundles. The analyst wants the lowest listed value to become zero and the highest to become one. Find a positive affine transformation that does this, transform all four values, and state what this normalization does—and does not—establish.
Method. Subtract the minimum listed utility, then divide by the difference between the maximum and minimum. This maps the endpoints of this particular list to zero and one while preserving its ranking.
The minimum is \(-5\), the maximum is \(15\), and the range is \(15-(-5)=20\). Therefore, use \(v=(u-(-5))/20=(u+5)/20\). For \(P\), \(v(P)=(-5+5)/20=0/20=0\). For \(Q\), \(v(Q)=(0+5)/20=5/20=0.25\). For \(R\), \(v(R)=(7+5)/20=12/20=0.60\). For \(S\), \(v(S)=(15+5)/20=20/20=1\).
The normalized ranking is \(S>R>Q>P\), matching \(15>7>0>-5\). The transformation has multiplier \(1/20>0\), so the ordering is preserved. The calculation checks that the listed minimum maps to zero and the listed maximum maps to one.
This normalization places these four listed values between zero and one. It does not show that utility is inherently bounded, that zero means no satisfaction, or that one is a universal maximum. A different collection of bundles could have values outside this normalized interval; the endpoints were chosen for this list.
What Ordinal Utility Does and Does Not Compare
If two utility functions are related by a strictly increasing transformation, they convey the same preference ordering, but not the same numerical scale. The safe comparison is whether they rank bundles in the same order and preserve indifference. The numerical size of a gap, the ratio of two values, and the sign of a value can change under transformations that leave preferences intact.
| Utility feature | Ordinal interpretation | Why caution is needed |
|---|---|---|
| Ranking | Higher utility means a more preferred bundle under the stated convention. | Rankings are the information the representation is designed to capture. |
| Sign | A negative value is simply a value below the chosen zero. | A constant shift can change the sign without changing preferences. |
| Difference | A subtraction gives a numerical difference on a particular scale. | A nonlinear increasing transformation can change the relative sizes of gaps. |
| Ratio | A ratio can be computed if the denominator is nonzero. | Ordinality does not make that ratio a meaningful measure of preference intensity. |
| Boundaries | A chosen normalization can place selected values in a chosen interval. | The bounds are properties of the chosen scale or list, not necessarily of preferences. |
A common exam trap is to treat a utility value as if it had an absolute interpretation. For example, suppose \(u(A)=2\) and \(u(B)=1\). It is valid to conclude that \(A\) ranks above \(B\) under this representation. It is not valid, from ordinal information alone, to conclude that \(A\) is “twice as good” as \(B\), or that the gain from \(B\) to \(A\) is twice some other gain.
Another trap is to use a negative multiplier and claim the ranking is preserved. If \(u(A)>u(B)\), then \(-u(A)<-u(B)\); the inequality reverses. By contrast, adding a constant or multiplying by a positive constant does not reverse it. When given two utility functions, compare their rankings directly or identify the transformation and check that it is strictly increasing.
For each relevant pair of bundles, determine which has higher utility, or whether the values tie.
A positive affine transformation has the form \(v=au+b\) with \(a>0\). More generally, check whether \(v=g(u)\) is strictly increasing over the values being compared.
Confirm that every strict ranking and tie is preserved. Do not infer preference intensity from changed gaps or ratios.
Check Your Understanding
For each problem, focus on the ranking and distinguish it from the numerical scale.
- A consumer assigns utilities \(u(M)=-4\), \(u(N)=2\), and \(u(O)=6\), then uses \(v=3u+1\). Find the transformed values and state the ranking. (Answer: \(v(M)=-11\), \(v(N)=7\), \(v(O)=19\); \(O>N>M\).)
- Utilities for bundles \(A\) and \(B\) are \(u(A)=5\) and \(u(B)=2\). An analyst proposes \(w=-2u+4\). Calculate both transformed values and say whether \(w\) represents the same ranking. (Answer: \(w(A)=-6\), \(w(B)=0\); no, it reverses the ranking, so it does not represent the same preferences.)
- A set of three plans has utility values \(-3\), \(1\), and \(4\). Apply \(g(u)=u^3\), give the transformed values and ranking, and state whether the gaps are preserved. (Answer: \(-27\), \(1\), and \(64\), in the same increasing order; the gaps are not preserved.)
- For four bundles, utilities are \(-2\), \(3\), \(8\), and \(18\). Find a positive affine transformation that maps the lowest value to zero and the highest to one, then transform the two middle values. (Answer: \(v=(u+2)/20\); the middle values become \(0.25\) and \(0.50\), and the endpoints become \(0\) and \(1\).)
- Two bundles have utility values \(u(C)=-1\) and \(u(D)=3\). A report claims that \(D\) is “three times as preferred” because \(3/(-1)=-3\), and then shifts all values upward by five. Explain what can be concluded after the shift. (Answer: The ratio does not measure preference intensity; after the shift the values are \(4\) and \(8\), so \(D>C\) remains the ranking, but the ratio \(8/4\) is not an ordinal measure either.)