2027 Public Health PhD Research: Separating Mediation From Effect Modification
Co-Founder and Chief Data Scientist
A public health dissertation can reach very different conclusions depending on whether a variable explains how an exposure affects health or identifies for whom its effect differs. The distinction matters when researchers study interventions, structural conditions, and health inequities - and when prospective PhD students assess whether a program can support that work. This guide explains the two questions, the assumptions behind common analyses, and the coursework, supervision, funding, and data access to examine before committing to a doctoral project.
Key Things You Should Know
- Mediation asks how an effect occurs; effect modification asks whether an effect differs across defined groups or conditions. A variable can have both roles, but each requires a separate question and analysis.
- Establish timing before modeling: specify the exposure, outcome, proposed mediator or modifier, comparison, and target population. An interaction term alone does not establish causal effect modification, and an adjusted regression coefficient alone does not establish mediation.
- Judge PhD fit by research feasibility: advanced causal-inference training, suitable mentorship, appropriately timed data, ethical access, and dependable funding matter more than a course title or delivery format.
What is the difference between mediation and effect modification in public health research?
Mediation concerns a pathway: exposure influences an intermediate variable, which in turn influences an outcome. Effect modification concerns differences in an exposure's effect across levels of another variable. These are different causal questions, not competing labels for any variable that appears in a regression.
Consider an illustrative study of a housing intervention and childhood asthma symptoms. Indoor air quality might be a mediator if the intervention changes air quality and that change affects symptoms. Baseline building condition might be an effect modifier if the intervention's effect on symptoms differs between buildings with different starting conditions. The researcher must specify when each was measured and what intervention and comparison the effect refers to.
The table separates the questions a dissertation proposal would need to answer. It also shows why the same statistical adjustment cannot serve both purposes without a defined estimand.
| Research feature | Mediation | Effect modification |
| Core question | Through which specified pathway might an effect operate? | Does the effect differ across specified groups or conditions? |
| Variable's usual timing | After the exposure and before the outcome | Often measured before the exposure, especially for decisions about targeting an intervention |
| Typical result | An estimated direct or indirect effect, under stated assumptions | Group-specific effects or an effect contrast between groups |
| Common error | Calling attenuation after adjustment proof of a pathway | Calling any significant interaction a meaningful causal difference |
Both analyses depend on the effect scale. Effects can differ across groups on a risk-difference scale without differing on a risk-ratio scale. Students choosing among specialization choices in public health graduate study should look beyond an epidemiology label to whether training covers causal estimands and interpretation on relevant scales.
How do public health PhD researchers identify mediation and effect modification?
Researchers begin with a public health decision, not a software command. For the housing example, one decision might be whether improving indoor air quality is a plausible route to better health; another might be whether the intervention should be adapted for buildings in different initial conditions.
A defensible identification plan makes the following choices before inspecting subgroup results:
- Define the target population, exposure or intervention, comparator, outcome, follow-up period, and effect scale.
- Draw a causal diagram that distinguishes pre-exposure confounders from variables caused by the exposure.
- For mediation, justify the proposed exposure-to-mediator-to-outcome sequence and assess confounding of each relevant relationship.
- For effect modification, prespecify the groups or conditions, estimate effects within them, and compare those effects with uncertainty intervals.
- Check whether measurement, missingness, sample size, and data-use permissions can support the planned analysis.
A modifier need not cause the outcome to be useful for describing heterogeneity. A mediator, by contrast, requires a credible intermediate role; measuring exposure, mediator, and outcome at one time point usually makes that sequence difficult to establish. A causal diagram can also reveal exposure-induced confounders of the mediator-outcome relationship, which invalidate some simple mediation approaches.
For an applicant still building basic study-design and statistics skills, short online public health degree options may be worth comparing with targeted methods courses. Neither is a substitute for supervised causal research, but either may clarify whether doctoral-level work is the right next step.
Which research methods and statistical techniques analyze mediation and effect modification?
For effect modification, researchers commonly estimate exposure effects within prespecified strata or fit a model with an exposure-by-modifier interaction. They should report the effects themselves - not only the interaction's p-value - and choose a measure that matches the decision, such as an absolute risk difference when planning services for a population.
For mediation, traditional regression approaches can be useful descriptively but often rely on restrictive assumptions, particularly when exposure and mediator interact. Counterfactual approaches define effects such as natural or interventional direct and indirect effects; techniques may include g-computation, inverse-probability weighting, or related estimators. The choice depends on the estimand, data structure, assumptions, and available expertise. Randomizing an exposure does not automatically remove mediator-outcome confounding.
Both questions require attention to selection, measurement error, missing data, model specification, and uncertainty. Repeated mediator and outcome measurements may call for longitudinal methods rather than one baseline-adjusted regression. In complex surveys, design weights and clustering can affect estimates and standard errors. Sensitivity analyses are particularly valuable for mediation because unmeasured mediator-outcome confounding can change a causal interpretation.
A research-oriented master's program with a strong thesis and advanced methods may be a more direct next step than a PhD for someone whose immediate goal is to learn these tools. When comparing that option, public health master's university award comparisons can inform affordability, but applicants still need to verify actual methods courses and supervised research opportunities.
Look for a sequence rather than one course promising to cover every method. A suitable pathway usually builds from epidemiologic study design and regression into causal diagrams, potential outcomes, longitudinal methods, advanced biostatistics, and seminars where students critique published analyses. Computing, reproducible workflows, and research ethics should accompany the statistical work.
Course titles do not establish depth. Ask for recent syllabi: Do students define estimands, analyze interactions on more than one scale, examine mediation assumptions, and conduct sensitivity analyses? Can they take relevant biostatistics courses outside the department? Most importantly, can an advisor and committee supervise the proposed methods with the data the student can actually access?
Applicants lacking introductory preparation can compare prerequisite courses with budget-friendly online public health degrees. An additional full degree makes sense only if its broader curriculum serves their goals; a statistics sequence and supervised research project may address the gap more efficiently.
Requirements vary by university. Programs may assess prior quantitative coursework, research experience, statements of scholarly interest, recommendations, and fit with available faculty. Some admit applicants with a bachelor's degree; others expect graduate preparation. An MPH, publication, thesis, GRE score, or commitment from a prospective advisor should not be assumed universally required.
For this research area, readiness is more specific than eligibility. A strong application can show how an applicant framed an exposure-outcome question, handled data responsibly, assessed alternative explanations, and communicated uncertainty. Work in mathematics, computing, clinical research, social science, or community health may be relevant when the applicant can explain its connection to epidemiologic research.
Check each program's current prerequisite and testing rules before investing in additional coursework. Guidance on public health master's admission workarounds for GPA gaps addresses a different degree level, but its emphasis on documenting later academic performance may help applicants decide what evidence to assemble; it does not override PhD admissions policies.
A PhD generally spans multiple years of coursework, qualifying or candidacy milestones, research, and dissertation writing. The relevant timeline is the program's recent completion experience - not its shortest advertised pathway. Access delays, a change in research question, or an analysis requiring new data can extend a dissertation.
Cost is not simply published tuition. Compare the written offer's tuition coverage, fees, stipend, health insurance, summer support, renewal rules, research or teaching duties, and any unfunded period after guaranteed support ends. Include living costs and forgone earnings. Funding tied to a particular grant or advisor can be consequential if the dissertation requires time or data beyond that project.
When weighing tuition for a public health doctorate, distinguish a funded research PhD from a tuition-charging professional doctorate. Before accepting an offer, request written terms and ask current students how funding and supervision work through dissertation completion.
They can be suitable when remote students receive sustained research supervision, access to advanced courses and secure computing, and realistic opportunities for scholarly collaboration. Online delivery by itself does not establish any of those conditions. A program may also require live seminars, campus visits, exams, or in-person research activity despite advertising online coursework.
Ask how often students meet advisors, whether methods courses are available remotely, how restricted datasets can be accessed, and what happens if an advisor leaves. Researchers using community or health-system data also need to establish who controls access and whether remote work is permitted under the data-use agreement.
A review of public health doctoral programs and professional goals can help generate options, but confirm that each option is actually a research PhD rather than a practice-focused DrPH. If the immediate objective is learning one technique, a supervised methods course may be less costly and more appropriate than an online doctorate.
CEPH accreditation concerns the public health school or program within its stated scope; it is distinct from an institution's accreditation. Do not infer that every doctorate at an accredited university is covered. Check CEPH's current listing and the university's description of the exact epidemiology PhD, including which school awards it and whether that degree falls within the accredited unit's scope.
Accreditation is a quality-assurance consideration, not evidence that a prospective advisor has capacity, that mediation methods are taught at dissertation depth, or that funding is secure. Review those issues separately. When comparing accessible public health doctoral programs, verify degree type and accreditation scope directly rather than treating a ranking or admissions description as proof of either.
Researchers use these skills in academic epidemiology, government research, health systems, research institutes, and organizations evaluating interventions or policy. Responsibilities may include specifying causal questions, designing analyses, managing data and approvals, checking assumptions, writing reports, and explaining why a result should - or should not - guide a population-health decision.
A PhD is particularly relevant to roles centered on independent study design and original methods-intensive research. Analyst or applied evaluation roles may instead value an MS or MPH plus strong quantitative projects and supervised experience. Job titles alone are unreliable: compare postings' actual expectations for study leadership, programming, publication, stakeholder communication, and subject-matter knowledge.
Readers comparing a doctorate with practice-oriented work can use the public health master's compensation outlook as context for an alternative pathway, not as a prediction of what either degree will earn.
What salaries and job outlook can public health PhD graduates expect?
There is no single PhD-specific salary or employment forecast for this research specialty. The US Bureau of Labor Statistics reports occupation-wide wages and projections for categories such as epidemiologists and statisticians; those categories include workers with different degrees, experience levels, employers, and responsibilities. A faculty position, government epidemiology role, and private-sector research position should not be valued using one occupational median.
For a sound comparison, consult the latest Bureau of Labor Statistics Occupational Outlook Handbook entry matching the tasks you want, note its wage reference year and projection period, and compare it with actual postings in your location. Then weigh prospective compensation against the written funding offer, time to completion, and the career you could pursue without a PhD. Public health master's cost and completion planning offers a useful alternative-pathway comparison, but neither degree's return should be inferred from occupation-wide wages alone.
Other Things You Should Know About Public Health
Yes. A variable could lie on one causal pathway and also alter another effect. Define each question separately, check its timing, and use methods appropriate to each estimand rather than assigning the variable one permanent label.
Usually, researchers should instead specify the social or structural processes they mean to study, such as differential access to resources. Treating race itself as a manipulable biological mechanism can misstate the causal question and obscure racism's effects.
Not by itself. Examine group-specific absolute effects, precision, potential bias, feasibility, and community priorities. A small or poorly measured subgroup difference may not support a different policy.
When feasible, documenting the question, effect scale, variables, and planned analyses before examining results helps distinguish confirmatory work from exploratory findings. Follow the study's applicable protocol, funder, and institutional requirements; clearly report later changes.
- Clarifying causal mediation analysis for the applied researcher | nih.gov https://pmc.ncbi.nlm.nih.gov/articles/PMC8496983/
- Chapter 7: Effect modification and interaction - ENCePP https://encepp.europa.eu/encepp-toolkit/methodological-guide/chapter-7-effect-modification-and-interaction_en
- Interactions in statistical models: Three things to know - Duncan https://besjournals.onlinelibrary.wiley.com/doi/10.1111/2041-210X.13714
- 8 Bias, Confounding, Random Error, & Effect Modification | psu.edu https://online.stat.psu.edu/stat507/Lesson08.html
- Best (but oft-forgotten) practices: mediation analysis | sciencedirect.com https://www.sciencedirect.com/science/article/pii/S0002916522048985
- Interventional effects for mediation analysis with multiple mediators -ORCA | cardiff.ac.uk https://orca.cardiff.ac.uk/id/eprint/106065/
- Overview of mediation methods: from estimands to estimations | oup.com https://academic.oup.com/exposome/article/6/1/osag010/8520904
- Using mediators to understand effect modification and interaction - Journal of Clinical Epidemiology https://www.jclinepi.com/article/S0895-4356%2823%2900238-X/fulltext
