Benefit-Cost Analysis
A public project is justified when every dollar spent returns at least a dollar to the public.
Before this Lesson
- Equivalent Annual Worth
- Probability · Mathematics and Statistics
A town can build a levee for $2,000,000. It costs $40,000 a year to maintain, lasts 30 years, and will never earn a cent. At 6%, its annual worth is −$185,200, and by the test you already know, that's a flat no.
But the people behind the levee would stop losing, on average, $250,000 a year to floods. That money never touches the town's books, so the test can't see it. A public project needs a test that can.
A/P spreads a lump at time 0 across a life. P/A collapses a yearly row into a lump one period before its first payment, and P/F moves a lump back to time 0. For more, see Equivalent Annual Worth.
Whose money is it?
A private firm weighs its own money against its own money. A public agency spends the public's money so the public ends up better off, and the gain lands in other people's pockets, where the agency's books will never show it. So the question becomes whether the public gets more than the agency spends. Answering it starts with sorting every item by who feels it.
| Item | Who feels it | Examples |
|---|---|---|
| Benefit | The public, for the better | Time saved, damage avoided |
| Disbenefit | The public, for the worse | Noise, lost farmland |
| Cost | The agency | First cost, O&M |
The noise from a new highway is a real loss to the homes beside it, but it never shows up on the state's invoice. That's what separates a disbenefit from a cost.
A dollar back for every dollar
The Handbook's Benefit-Cost Analysis entry, in its Engineering Economics section, takes two lines. The benefits of a project should exceed the estimated costs :
The two forms always agree. is the benefit-cost ratio, the dollars of public benefit per dollar spent, and a project at or above 1 is justified. Those dollars fall at different times, though, so the topic's habit applies: draw the diagram, then move everything to one point in time.
One way is to turn everything into a yearly row. The $250,000 already is one, and so is the O&M. Only the first cost needs moving, with A/P at 0.0726 from the Handbook's Interest Rate Tables:
The other way is to bring everything back to time 0, with P/A at 13.7648:
They agree, and not by luck. Going from one basis to the other multiplies the top and the bottom by the same P/A, which cancels. So use whichever basis the data make easier, but put both sides on it. Present-worth benefits over annual costs give , which would make any project look like a miracle.
Where disbenefits and O&M go
Now say the levee also swallows some farmland, costing the public $30,000 a year: a disbenefit, . The Handbook mentions neither disbenefits nor O&M, so where they go is a convention, and a question usually states it. Do what it says. When it says nothing, the usual placement subtracts disbenefits from benefits and adds O&M to costs, as in the first row here, where $145,200 is the annualized first cost.
| On top | On the bottom | B/C |
|---|---|---|
| B − D = $220,000 | $145,200 + O&M = $185,200 | 1.19 |
| B = $250,000 | $145,200 + O&M + D = $215,200 | 1.16 |
| B − D − O&M = $180,000 | $145,200 | 1.24 |
The ratio moves, so look at what doesn't. In every row, the money left over is . The verdict can't change either, because wherever sits, you're making the same statement:
Moving an item across the fraction bar changes the ratio, but never which side of 1 it lands on.
Running the test in reverse
What's the most the town could pay for this levee? Leave the farmland aside for now. The limit is the first cost that brings the ratio down to exactly 1, where benefits just cover costs, so set them equal and solve:
The O&M stays in. Leave it out, and you'd get $3,443,526, a price at which the real ratio is only 0.86. Or hold the first cost fixed, and the smallest annual benefit is just itself, $185,200. If the damage estimate falls below that, the levee stops being justified.
Benefits that start late
A levee protects nobody until it's finished. Say construction takes 3 years, with the $2,000,000 treated as spent at time 0, and the 30 years of benefits and O&M run from year 4 through year 33.
P/A knows only that there are 30 payments, so , not 33, and it drops its lump one period before the first payment. That's year 3, so the lump has 3 periods left to travel:
The O&M runs on the same schedule, so it takes the same two moves and comes to $462,277. The first cost is already at time 0 and takes none:
Starting the same benefits three years late took the ratio from 1.35 down to 1.17.
A number for a gamble
So far, the $250,000 has been handed to you, and it's an odd number. Floods don't deliver $250,000 a year. Some years they deliver nothing, and some years ruin. Getting one figure out of that takes expected value.
An expected value weights each outcome by its probability and adds them up. The Handbook's Expected Values section writes it as . For more, see Probability.
Start with a simpler gamble. A contractor spends $15,000 preparing a bid. One time in four it wins, and the job nets $185,000 after the bid cost. Otherwise, the $15,000 is gone. The Handbook's Engineering Economics section prints expected value beside its Economic Decision Trees, where the s are chance outcomes, not costs:
A loss goes in with a minus sign, so . Bidding pays in the long run, though no single bid ever ends at +$35,000. To choose among options, compute each one's expected value, and take the higher expected profit or the lower expected cost.
A hundred years on one river
Flood data comes as a table: each flood size, its damage, and its probability in any given year. Here's the river without the levee, with each year landing in exactly one row.
| Outcome | Annual probability | Damage |
|---|---|---|
| No flood | 0.80 | $0 |
| Minor flood | 0.14 | $500,000 |
| Moderate flood | 0.05 | $2,000,000 |
| Major flood | 0.01 | $14,000,000 |
That's $31,000,000 of damage per century, or $310,000 a year: the expected annual damage. The Handbook says nothing about flood damage, and it doesn't have to, because it's the same sum:
Three things about that number:
- The no-flood row counts. Probabilities must sum to 1, because they cover every year. A question may list only the floods, which leaves the rest as no-flood years. They add $0, but they're still years, so never divide by the floods' combined 0.20.
- It isn't the likeliest outcome. The most likely year costs nothing, and no year ever costs $310,000. It's a long-run weighted average, and it needn't match any outcome.
- The rarest flood carries the most. Probability times consequence is one way to state risk, which the Handbook's Safety and Prevention section writes as Risk = Hazard × Probability. The three floods carry $70,000, $100,000, and $140,000 a year, so the once-a-century flood is the biggest risk on the river.
The benefit is the damage you avoid
A levee doesn't stop floods. It changes the damage column. Say this one holds back minor and moderate floods completely, and a major flood overtops it and does $6,000,000 of damage. The new expected damage is a year, and the benefit is the difference:
There's the number from the top of the page. It's damage that doesn't happen, so it goes on top as a benefit, however much the word "damage" sounds like a cost. The $60,000 goes nowhere: the levee doesn't prevent it, so it isn't a benefit, and the town doesn't pay it, so it isn't a cost.
Highway safety works the same way: the reduction in yearly crash cost is the benefit. Where crash counts come from is a question for Crash Rates and Countermeasures; here, the savings arrive in dollars.
How much protection to buy
The town can do more than choose between that levee and nothing. A lower levee costs less and leaves more damage behind. For each level, annualize the protection (the first cost through A/P, plus O&M) and add the expected damage still left. The lowest total wins.
| Level | Protection cost | Expected damage | Total |
|---|---|---|---|
| None | $0 | $310,000 | $310,000 |
| Low | $60,000 | $190,000 | $250,000 |
| Medium | $120,000 | $110,000 | $230,000 |
| High | $185,200 | $60,000 | $245,200 |
Read it one step at a time. Going from nothing to Low spends $60,000 a year and removes $120,000 of damage. Low to Medium spends $60,000 and removes $80,000. Medium to High spends $65,200 and removes only $50,000. Protection is worth buying only while each added level removes more expected damage than it costs, and that stops at Medium. So the High levee, the one from the start of this page, has a healthy ratio of 1.35 and still isn't the one to build.
The highest ratio doesn't win
The town can build only one of these levees, which makes them mutually exclusive. Now look at them as ratios.
| Level | Cost, C | Benefit, B | B/C | ΔB/ΔC |
|---|---|---|---|---|
| Low | $60,000 | $120,000 | 2.00 | 2.00 |
| Medium | $120,000 | $200,000 | 1.67 | 1.33 |
| High | $185,200 | $250,000 | 1.35 | 0.77 |
Every level clears 1, and the best ratio belongs to Low. If ratios picked winners, the town would build Low and lose $20,000 a year against Medium. A ratio measures how hard each dollar works, not how many dollars are working. Would you rather turn $1 into $3, or $1,000 into $2,000?
So test each extra step on its own: does the extra benefit cover the extra cost? That's the incremental ratio, and it's those same steps again.
On a plot of benefit against cost, each step's ratio is its slope.
High's overall line still climbs faster than , but its last step doesn't: the good early dollars are carrying the bad late ones. Here's the procedure, with everything on one shared basis:
- Order the alternatives by cost, cheapest first.
- Start from doing nothing. The cheapest one's own ratio is its step up. If that's below 1, drop it and try the next.
- From the one still standing, find to the next costlier one. At 1 or more, the costlier one takes over. Below 1, drop it. Whatever stands at the end gets built.
The Handbook prints only the single-project rule, so a question may state this one or leave it to you. It's the same-footing idea from Comparing Alternatives, applied to the money that differs.