Topic 11 of 64· Governmental economic appraisal

Cost-Effectiveness Analysis in Government

Cost-effectiveness analysis (CEA) compares the costs of alternative ways of achieving the same outcome, expressed in natural units — cost per rough sleeper housed, cost per pupil brought up to the expected standard, cost per tonne of CO2 abated — without converting the outcome itself into money.

Why it matters

The Green Book treats CEA as the fallback method when social cost-benefit analysis's requirement to monetize every benefit becomes not just difficult but dishonest — where putting a credible price on the outcome would require assumptions nobody actually holds (https://www.gov.uk/government/publications/the-green-book-appraisal-and-evaluation-in-central-government, Chapter 5, on options appraisal where outcomes are not readily monetisable). CEA is the method borrowed most directly from health economics — it is structurally identical to how NICE compares treatments using cost per Quality-Adjusted Life Year — but applied to non-health public programmes: education interventions per pupil-outcome-point, housing programmes per household prevented from homelessness, employment programmes per sustained job outcome.

The reason CEA earns its place alongside SCBA rather than being subsumed by it is that forcing a monetary value onto some outcomes produces a number precise enough to look authoritative and contested enough to be worthless in a public debate — putting a price on "a child reading at the expected standard" invites exactly the kind of challenge that derails a business case at select committee. CEA sidesteps the argument by refusing to have it: it ranks options on cost per unit of the outcome itself, leaving the separate political judgement of whether the outcome is worth pursuing at all to the strategic case.

The maths

Cost-effectiveness ratio (average) = Total cost / Total outcome units achieved

Incremental cost-effectiveness ratio (ICER), comparing option A to option B:
ICER = (Cost_A − Cost_B) / (Outcome_A − Outcome_B)

Procedure:
1. Fix the outcome unit and measurement method across all options being compared.
2. Cost every option on the same basis (see ../green-book-appraisal/, financial case)
   over the same time horizon.
3. Discard dominated options: any option costing more per unit than a cheaper
   alternative achieving the same or better outcome is dropped.
4. Rank remaining options by incremental, not average, cost-effectiveness ratio.

CEA cannot, by itself, say whether a programme is worth funding at all — only which of several approaches to the same goal is cheapest per unit. Deciding whether the goal itself is worth the spend requires either converting back to SCBA (if a credible valuation exists) or a political/ strategic judgement outside the maths. Where outcomes genuinely cannot be reduced to one unit — because a programme produces several outcomes that matter in different ways — use multi-criteria decision analysis instead.

Worked example

Local authority: a council compares three approaches to reducing rough sleeping, each costed over one year against the outcome "individuals moved into settled accommodation for 6+ months":

Option                          Cost       Outcomes achieved   Avg CER
Housing First (intensive)       £900,000   60                  £15,000/outcome
Hostel + move-on support        £600,000   50                  £12,000/outcome
Outreach + private rented sector £350,000  20                  £17,500/outcome

ICER, Hostel vs Outreach:  (600k−350k)/(50−20) = £8,333 per additional outcome
ICER, Housing First vs Hostel: (900k−600k)/(60−50) = £30,000 per additional outcome

Outreach is dominated on average cost by Hostel, but the incremental step from Outreach to Hostel costs only £8,333 per additional person housed — cheap relative to the Housing First step, which costs £30,000 for each additional person beyond what Hostel achieves. A budget-constrained authority scaling up should prefer expanding Hostel before Housing First, even though Housing First looks better on its own average ratio.

National government: a literacy catch-up programme is compared across three delivery models on "cost per pupil reaching age-expected reading standard": one-to-one tutoring (£1,800/pupil), small-group tutoring (£700/pupil), and digital-only intervention (£150/pupil, but only 40% of the outcome rate of small-group tutoring per pupil enrolled once adjusted for engagement drop-off). Once adjusted for actual completion, digital-only costs £375 per pupil reaching standard — still the cheapest, but the CEA cannot say whether the smaller absolute number of pupils helped by digital-only, if delivered at the same budget as small-group, is an acceptable trade against reaching fewer pupils at greater depth; that is a distributional judgement CEA hands back to decision-makers.

Software engineering connection

CEA is the right frame whenever engineering teams evaluate delivery approaches for the same service outcome — cost per successfully verified identity across three identity-verification vendors, cost per case correctly triaged across two casework automation designs, cost per accessibility defect resolved across in-house versus contracted remediation. The discipline it imports directly: define the outcome unit before comparing costs (not "tickets closed" — an output — but "user need actually resolved"), and always compute the incremental ratio between the live system and a proposed replacement, not each system's average cost in isolation. See outcomes vs outputs and cost per outcome.

Pitfalls

  • Comparing average, not incremental, ratios when deciding on an expansion. As the rough sleeping example shows, the option with the best average ratio is not always the cheapest next unit of outcome to buy.
  • Choosing an outcome unit that is really an output. "Referrals made" or "sessions delivered" measure activity, not the outcome the programme exists to produce; CEA on outputs produces a confident-looking number that answers the wrong question.
  • Comparing across genuinely different outcomes. CEA is only valid when every option targets the same outcome measured the same way; comparing "cost per rough sleeper housed" against "cost per care leaver in stable tenancy" needs a generic outcome measure or multi-criteria decision analysis, not CEA.
  • Ignoring outcome durability. A cheaper option that produces outcomes that don't persist (a pupil who regresses after the intervention ends) is not actually more cost-effective once measured over a comparable horizon; match the follow-up period across options being compared.

Sources