Questions: Cost-Effectiveness Analysis in Policy Research

Question 1

Program A costs $200,000 and produces 50 QALYs. Program B costs $100,000 and produces 20 QALYs. A policymaker funds Program B because it costs less. What error in cost-effectiveness reasoning is this?

• A. No error — lower total cost always indicates better cost-effectiveness
• B. Confusing cost-effectiveness with cost-minimization — Program A produces more benefit per dollar ($4,000/QALY vs. $5,000/QALY)
• C. Failing to apply the willingness-to-pay threshold — without it, neither program can be ranked
• D. Using QALYs inappropriately — cost minimization requires a different benefit metric

Answer: (B)

Cost-effectiveness is measured by the cost-effectiveness ratio: total cost divided by total benefit. Program A: $200,000 / 50 QALYs = $4,000/QALY. Program B: $100,000 / 20 QALYs = $5,000/QALY. Program A is more cost-effective despite costing more in absolute terms. The policymaker is conflating cost minimization with cost-effectiveness — funding the cheaper program without asking how much benefit each dollar buys misses the entire point of CEA.

Question 2

A CEA concludes a new education program costs $8,000 per additional year of schooling completed. A critic argues this finding is meaningless without a willingness-to-pay (WTP) threshold. The critic is:

• A. Wrong — a lower CER is always preferable regardless of WTP
• B. Wrong — WTP thresholds only apply to healthcare CEA, not education
• C. Right — whether $8,000 per year-of-schooling represents good value depends on how much society is willing to spend per unit of that outcome
• D. Partly right — WTP only matters when the program costs more than existing alternatives

Answer: (C)

The CER tells you the price per outcome unit, but not whether that price is worth paying. That judgment requires a WTP threshold — a policy choice about how much society will spend per unit of benefit. $8,000/year-of-schooling could be excellent value (if the threshold is $20,000) or poor value (if it is $3,000). The threshold is 'itself a policy choice, not a mathematical fact.' WTP thresholds apply across all CEA domains, not just healthcare.

Question 3

The choice of benefit metric in a CEA — whether to use QALYs, earnings gains, or crime reduction — embeds value judgments that the analysis itself cannot resolve.

• T. True
• F. False

Answer: True

Defining and measuring 'benefit' requires normative choices. QALY calculations require judgments about health-state quality weights. Educational CEA must decide whether the relevant benefit is earnings, test scores, social mobility, or civic participation. These choices privilege certain values over others. CEA can quantify tradeoffs rigorously once a metric is chosen, but cannot determine which metric is right — that requires ethical and political reasoning outside the model.

Question 4

The most cost-effective intervention is typically the one with the lowest total program cost.

• T. True
• F. False

Answer: False

Cost-effectiveness is measured by the cost-effectiveness ratio (total cost / total benefit), not total cost alone. A more expensive program that produces proportionally more benefit can be more cost-effective than a cheaper one. Cost minimization and cost-effectiveness are fundamentally different objectives. Confusing them — funding the cheapest option regardless of how much benefit it produces — is one of the most common errors in policy evaluation.

Question 5

Why is sensitivity analysis essential to an honest cost-effectiveness analysis, and what does it reveal about the nature of CEA conclusions?

Model answer: CEA conclusions depend on assumptions about cost attribution, benefit measurement, discount rates, and counterfactual comparisons. Sensitivity analysis varies these assumptions to show how much the CER changes — revealing which assumptions are load-bearing. A finding cost-effective only under optimistic assumptions is fundamentally less credible than one that holds across a wide range of reasonable values.

Because CEA typically uses observational data rather than randomized experiments, the causal assumptions in the underlying models do real work. Sensitivity analysis makes those assumptions visible and testable rather than buried. A conclusion that survives many different reasonable assumptions is more robust than one that hinges on a single favorable parameter choice. Sensitivity analysis doesn't undermine CEA — it is what makes it scientifically honest rather than advocacy dressed as analysis.