2027 Distinguishing Relative Risk From Risk Difference in Program Evaluation for MPH and MS Students
Co-Founder and Chief Data Scientist
A program can increase a favorable outcome by 50% and still change outcomes for only a small share of participants. For MPH and MS students evaluating public health programs, that distinction affects how results are reported, compared, and used to allocate resources. This guide explains relative risk and risk difference using one hypothetical evaluation, then shows how study design, uncertainty, and bias affect what either measure can support. You will be able to choose the measure that fits your evaluation question and explain its limits to a practice or research audience.
Key Things You Should Know
- Relative risk divides one group's outcome risk by another's; risk difference subtracts the two risks. Their null values are 1 and 0, respectively.
- In an illustrative evaluation with favorable-outcome risks of 12% and 8%, relative risk is 1.5, while risk difference is 4 percentage points - 40 additional outcomes per 1,000 people under the same conditions.
- Report both when feasible: relative risk describes proportional change, while risk difference better conveys the absolute scale of a program's possible benefit or harm. Neither establishes causation without a credible design.
What are relative risk and risk difference in program evaluation?
Both measures compare the risk - the proportion of people who experience a specified outcome during a specified period - in two groups. In a program evaluation, the groups might be people offered a program and people receiving usual services. The outcome could be beneficial, such as completing a screening, or harmful, such as developing an infection. Define the eligible population, outcome, comparison, and follow-up period before calculating either measure.
Relative risk (RR), also called a risk ratio, expresses proportional comparison: program-group risk divided by comparison-group risk. Risk difference (RD) expresses absolute comparison: program-group risk minus comparison-group risk. With the same outcome and group order, RR above 1 and positive RD both indicate a higher observed risk in the program group. Whether that is desirable depends on the outcome.
These measures are useful beyond graduate study. Students comparing public health undergraduate program comparisons can look for a foundation in population denominators and study design; the more demanding task in MPH or MS work is deciding whether a comparison supports a program-effect claim.
How do you calculate relative risk and risk difference?
Consider this illustrative, not empirical evaluation: 120 of 1,000 eligible people offered a screening-navigation program complete screening within six months, compared with 80 of 1,000 eligible people receiving usual services. The observed risks are 120 ÷ 1,000 = 0.12 and 80 ÷ 1,000 = 0.08.
RR = 0.12 ÷ 0.08 = 1.5. The program group's observed screening-completion risk is 1.5 times the comparison group's risk. RD = 0.12 − 0.08 = 0.04, or 4 percentage points. That is an observed difference of 40 completions per 1,000 eligible people, not a claim that the program caused those additional completions.
The comparison below keeps the outcome and denominators fixed, so only the interpretation changes.
| Measure | Calculation in the illustrative example | What it communicates |
| Relative risk | 0.12 ÷ 0.08 = 1.5 | Proportional outcome frequency between groups |
| Risk difference | 0.12 − 0.08 = 0.04 | Absolute difference of 4 percentage points |
Keep the units straight: "50% higher relative risk" and "4 percentage points higher risk" describe the same illustrative comparison, but "50 percentage points higher" does not. If quantitative preparation is an admissions concern, investigate the specific school's requirements and any public health master's provisional admission pathways rather than assuming one introductory calculation satisfies a methods prerequisite.
When should MPH and MS students use relative risk versus risk difference?
Use RR when the decision concerns proportional differences across populations or when comparing associations across settings with different baseline risks. Use RD when a health department or community partner needs to estimate the scale of an outcome difference for its eligible population. For a grant report or evaluation brief, showing both usually prevents a large-sounding relative result from obscuring a modest absolute change.
Consider two communities with the same RR but different starting risks. If the comparison-group risk is low, the absolute difference may be small; if it is high, the same proportional increase may affect substantially more people. A program-budget decision therefore needs baseline risk and the number of eligible people, not RR alone. Conversely, RD alone can conceal a substantial proportional change in an uncommon but important outcome.
Neither measure replaces a question about feasibility, equity, cost, or unwanted effects. Students choosing a degree to perform this work should compare the methods instruction and supervised evaluation opportunities they need, not treat addressing GPA concerns in an MPH application as a substitute for building quantitative readiness.
Wording determines what decision-makers hear. In the illustrative screening example, "50% higher completion risk" emphasizes proportional change; "4 additional completions per 100 eligible people" conveys absolute scale. Both should be paired with the underlying group risks and the six-month follow-up period. Avoid calling either result a "reduction in disease" when the measured outcome is screening completion.
For an adverse outcome, the direction of benefit reverses. If a program group has lower infection risk than a comparison group, RR below 1 and negative RD can indicate a favorable observed difference. State the outcome and group order explicitly rather than relying on "higher" or "lower" to imply good or bad.
If - and only if - a credible design supports a causal interpretation, an absolute risk reduction can inform a number needed to treat or an analogous program-planning quantity. With an observational comparison subject to selection bias, dividing 1 by an RD does not turn an association into a causal estimate. Similarly, salary planning for MPH graduates is a separate career decision; a strong evaluation finding does not establish an individual graduate's earnings or job prospects.
Both measures require comparable definitions of eligibility, outcome, follow-up, and group assignment. A randomized evaluation can strengthen a causal interpretation if assignment is maintained and outcomes are measured comparably. In a cohort or nonrandomized program evaluation, participants may differ from nonparticipants before the program begins; even precisely calculated RR and RD may then reflect those differences.
Check whether each person was at risk at the start, whether follow-up was long enough and similar across groups, and whether missing outcomes or migration between groups could distort the denominators. With substantially unequal follow-up, a simple cumulative risk comparison may be inappropriate; an incidence-rate analysis or time-to-event approach may better match the data. A case-control study generally cannot calculate population risks directly from its sampled case and control counts, so its usual association measure is an odds ratio.
An MS student planning advanced design work should assess access to appropriate data and research supervision before comparing public health research doctorate tuition options. More schooling does not fix an evaluation whose comparison group or outcome data cannot answer the question.
A point estimate is incomplete without uncertainty. Report a confidence interval for RR and RD using an analysis appropriate to the design. For RR, the null value is 1; for RD, it is 0. A conventional two-sided 95% confidence interval that includes the relevant null value does not exclude no difference at the corresponding significance level. An interval that excludes it still does not rule out bias or prove causation.
Interpret the interval's range, not just whether it crosses a threshold. A wide RD interval may encompass both a practically useful benefit and a meaningful harm - important information for a partner deciding whether to expand a program. Small samples, rare outcomes, clustering by clinic, and complex survey sampling can change interval calculations; standard independent-person formulas may understate uncertainty when participants share sites or sampling structures.
Students considering further methodological research can compare financial support for online epidemiology doctoral study alongside supervision and statistical training. A doctorate is not necessary simply to learn to report an interval responsibly.
Confounding occurs when a factor associated with program participation and the outcome distorts the comparison. If people entering a screening-navigation program already have better access to clinics, an unadjusted RR or RD may partly reflect access rather than the program. Consider design-stage solutions such as randomization or a defensible comparison group, then appropriate adjustment for measured pre-program factors. Adjustment cannot guarantee removal of unmeasured confounding.
Effect modification means the comparison differs across relevant groups - for example, across communities with different starting access. Report subgroup-specific absolute and relative results when prespecified, sufficiently supported by data, and useful to partners. A similar RR across groups need not imply a similar RD because baseline risks differ. Conversely, different RRs do not necessarily imply different absolute effects. Avoid treating noisy subgroup contrasts as established inequities or program successes.
For students who want to investigate these questions as original research, a public health doctoral admissions requirements guide is only an initial check; research fit, design training, and feasible data access matter more to the proposed analysis than an admissions label.
Begin with the decision the evaluation must support: whether to continue a program, adapt it, test it more rigorously, or investigate unequal reach. Then define the eligible population and a meaningful outcome window. Calculate group risks with transparent numerators and denominators, present RR and RD with uncertainty, and explain why the comparison is - or is not - credible. Pair outcome measures with implementation evidence, such as who was actually offered services, so a weak outcome does not automatically get mistaken for an ineffective program design.
For an applied assignment, a practical reporting sequence is:
- Specify the program offer, comparison condition, eligible population, outcome, and follow-up period before examining results.
- Audit missing records, unequal follow-up, and whether people were counted in the intended groups.
- Report both group risks, RR, RD, intervals, and any justified adjusted or subgroup estimates.
- State design limitations and the decision each estimate can reasonably inform; obtain required site, data-use, privacy, and institutional review guidance before using restricted records.
RD can help frame a possible impact at a known scale, but neither measure by itself establishes cost-effectiveness. When education costs are part of a student's decision, comparing public health master's costs and pay requires separate, verified tuition and employment evidence rather than extrapolating from an evaluation exercise.
An introductory epidemiology course should establish populations at risk, incidence, study designs, and interpretation of ratios and differences. Biostatistics adds sampling uncertainty and confidence intervals. Evaluation or causal-inference coursework helps students define an estimand - the specific comparison they intend to estimate - and examine selection, confounding, missing data, and implementation. Course titles alone do not establish depth; inspect syllabi and assignments.
A practice-oriented MPH student may prioritize an evaluation project that requires a defensible denominator and a brief for a community partner. A research-oriented MS student may need deeper work in regression, longitudinal analysis, reproducible computing, and study design. Both benefit from supervised feedback on whether a numeric estimate actually answers a public health question. A targeted methods course and a supervised project may be more direct than another degree for someone who already has the necessary professional foundation.
If cost limits an MPH student's choices, investigate university-funded aid for online MPH programs while checking whether the funded option actually offers the needed quantitative coursework and project supervision.
How should students choose an accredited program for epidemiologic training?
Start with the exact degree and curriculum, not the word "epidemiology" in marketing materials. Confirm the institution's recognized institutional accreditation, then verify whether the specific public health school, unit, or program - and the degree under consideration - falls within Council on Education for Public Health (CEPH) accreditation scope. Institutional accreditation and CEPH accreditation answer different questions; neither alone proves advanced methods instruction.
Ask for recent syllabi, the sequence of epidemiology and biostatistics courses, prerequisites, available electives, software and reproducibility expectations, and examples of supervised evaluation or research work. Check whether a practicum, applied practice experience, or thesis is required for that specific degree and whether access to suitable data and mentorship is realistic. Compare total costs, scheduling, and project obligations as well as tuition: a cheaper format is not a bargain if it cannot support the training goal.
For a cost-constrained practice pathway, low-tuition online MPH options can provide a starting shortlist, but verify each program's current charges, accreditation scope, course availability, and live-session or fieldwork demands directly. Choose the program - or a narrower methods course - that gives you credible practice in choosing, calculating, and explaining the right risk comparison.
Other Things You Should Know About Public Health
No. Risks cannot be negative, so their ratio cannot be negative. A risk difference can be negative because it subtracts one group's risk from another's.
A conventional RR cannot be calculated by dividing by zero. Report the group counts and risks, seek statistical guidance on an appropriate interval or model, and avoid presenting an arbitrary substitute as an ordinary RR.
Not directly. RR and RD compare the risks of a defined event. For a continuous outcome such as a blood-pressure reading, consider an appropriate difference in means or another measure matched to the research question.
Only if a composite outcome was defined in advance and its components make sense together. Otherwise, report outcomes separately so a change in one does not conceal a different pattern in another.
- CDC Approach to Program Evaluation https://www.cdc.gov/evaluation/php/about/index.html
- Relative risk, relative and absolute risk reduction... - NCBI https://www.ncbi.nlm.nih.gov/books/NBK63647/
- Impact of your results: Beyond the relative risk - PMC https://pmc.ncbi.nlm.nih.gov/articles/PMC6178613/
- A Simple Guide to Effect Size Measures | nih.gov https://pubmed.ncbi.nlm.nih.gov/36951858/
- Principles of Epidemiology | Lesson 3 - Section 5 https://archive.cdc.gov/www_cdc_gov/csels/dsepd/ss1978/lesson3/section5.html
- Uncertainty and how we measure it for our surveys | Office for National Statistics https://www.ons.gov.uk/methodology/methodologytopicsandstatisticalconcepts/uncertaintyandhowwemeasureit
- Relative Extent, Relative Risk and Attributable Risk | US EPA https://www.epa.gov/national-aquatic-resource-surveys/relative-extent-relative-risk-and-attributable-risk
- Competing Risk Analysis | Columbia University Mailman School of Public Health https://www.publichealth.columbia.edu/research/population-health-methods/competing-risk-analysis
