2027 Public Health PhD Research: Checking Identifiability Before Making a Causal Claim
Co-Founder and Chief Data Scientist
A public health study can produce a precise estimate without establishing what would happen if a policy or intervention changed. For prospective PhD researchers, the crucial step comes earlier: deciding whether the causal effect is identifiable from the proposed design, data, and assumptions.
This guide explains that decision using a hypothetical policy study, then shows how to assess doctoral coursework, supervision, and data access. It is for research-oriented applicants and graduate methods learners who want to make defensible causal claims - not merely run more sophisticated models.
Key Things You Should Know
- Identifiability comes before estimation: define the population, intervention, comparison, outcome, and timing, then determine whether the available data and defensible assumptions can connect observations to that causal effect.
- Three checks are indispensable in many observational studies: whether relevant confounding is addressed, comparable exposure groups exist, and the intervention and outcome are defined consistently. No statistical significance test establishes these conditions.
- Choose PhD training around the research you can actually do: assess methods teaching, faculty supervision, governed data access, and funding terms before treating a dissertation idea as feasible.
What is identifiability in public health causal research?
Identifiability means that a specified causal effect can, under stated assumptions, be expressed using information in the observed data. Estimation comes afterward: it uses a sample and an analytical method to calculate that effect. A larger dataset or a more flexible model may improve estimation, but neither repairs a causal effect that the study design cannot identify.
Consider an illustrative question: among residents of a city, what would be the difference in asthma-related emergency visits over the next year if a proposed traffic-emissions policy were implemented versus not implemented? The researcher must specify who is included, what counts as implementation, the comparison condition, when follow-up starts, and how visits are measured. If the policy begins alongside other changes that cannot be separated with available data, the policy's effect may not be identifiable from that design.
A PhD is not necessary simply to learn this distinction. Someone deciding whether to pursue a full research degree can compare the time and cost of doctoral study with targeted methods courses and public health degree affordability and completion planning. Doctoral study makes more sense when the goal is to develop and lead an ongoing research program.
How does a Public Health PhD prepare researchers to make causal claims?
A research-focused Public Health PhD can provide sustained practice in turning a substantive problem into a defensible estimand: the exact causal quantity a study aims to learn. Students may develop study protocols, scrutinize causal diagrams, work with restricted data under approved conditions, receive methodological criticism, and defend the limits of dissertation findings. The degree does not make every observational result causal; its value is training researchers to recognize when the claim must be narrowed or withheld.
The strongest fit depends on supervision as much as course titles. Ask whether prospective advisors study comparable interventions and populations, which methods their groups use, how doctoral students receive feedback, and whether a feasible dissertation dataset exists. An advisor's interest is not a promise of supervision, funding, data access, or authorship. When comparing public health doctorate pathways for advanced study, distinguish a research-focused PhD from a practice-focused DrPH: both can use evidence, but their typical scholarly and applied aims differ.
What assumptions must researchers check before claiming causation?
Assumptions depend on the design. For a conventional observational comparison, the central question is whether observed groups can stand in for the outcomes those same people would have experienced under a different exposure. The table distinguishes common assumptions from what the data alone cannot settle.
| Assumption | Why it matters | Important limit |
| Exchangeability | After appropriate adjustment, comparison groups differ in outcome risk because of the exposure rather than remaining common causes. | Measured balance cannot rule out unmeasured confounding. |
| Positivity | Relevant population groups have a chance of receiving each exposure being compared. | A model cannot recover an unsupported comparison where an exposure never occurs. |
| Consistency and a well-defined intervention | The recorded exposure corresponds to the intervention whose effect is being claimed. | Different versions of a policy may have different effects. |
| Limited interference, when required | One person's exposure does not change another person's outcome in a way the analysis ignores. | Spillovers are especially plausible for infectious disease and place-based policies. |
Researchers must also assess whether exposure, outcome, and covariates are measured appropriately; whether missingness or loss to follow-up distorts the comparison; and whether the target population matches the people studied. Random assignment can strengthen identification, but nonadherence, spillovers, and missing outcomes still matter. Difference-in-differences, instrumental-variable, and regression-discontinuity designs replace some assumptions with other demanding ones; none is a general cure for confounding.
Applicants filling a statistics or epidemiology gap before advanced study may find intensive online public health bachelor's classes worth comparing, but should first check whether a specific doctoral program requires undergraduate credit, graduate-level preparation, or demonstrated research experience.
Look beyond a course labeled "causal inference." A useful sequence begins with epidemiologic study design and biostatistics, then develops causal diagrams, counterfactual reasoning, longitudinal data analysis, and design-specific methods. Advanced work may address time-varying confounding, missing data, measurement error, sensitivity analysis, and reproducible computing. A dissertation seminar should give students opportunities to defend an estimand and assumptions before presenting results.
Course catalogs alone do not establish depth or availability. Request recent syllabi and ask when courses run, whether prerequisites create delays, who teaches them, and whether students apply the methods to research under supervision. For applicants considering a master's degree first, online MPH programs with CEPH accreditation can be a starting point for comparison - not evidence that any particular MPH offers doctoral-level causal methods. A research-oriented MS or focused graduate coursework may fit a substantial methods gap better.
Confounding occurs when a cause of the exposure also affects the outcome, creating a misleading comparison. Selection bias can arise when inclusion in the analysis depends on factors related to both exposure and outcome. In the illustrative traffic-policy study, neighborhood resources might influence both where a policy is adopted and residents' asthma risks. Restricting analysis to people who remain in a health system could introduce a separate bias if remaining depends on exposure and illness.
Researchers begin with subject-matter knowledge and the study's timeline, not an automated list of statistically significant covariates. A causal diagram can make proposed relationships explicit and reveal why adjusting for a variable affected by both exposure and outcome - a collider - may create bias. They then examine recruitment, exclusions, follow-up, missing data, and whether each comparison group contains people with similar pre-exposure characteristics. Sensitivity analyses can show how conclusions change under alternative assumptions, but cannot prove that all bias is gone.
If the immediate goal is to strengthen applied epidemiology skills rather than produce independent doctoral research, inexpensive online MPH study pathways may warrant comparison. Check the actual methods curriculum and supervised analysis opportunities; a lower advertised price does not establish causal-inference depth.
Requirements vary by institution and field. Programs commonly evaluate prior academic work, quantitative preparation, research interests, recommendations, and a statement of purpose. Some specify prerequisite courses, tests, or a prior master's degree; others do not. Eligibility to apply is different from evidence that an applicant is ready to formulate a feasible dissertation question.
For causal research, useful evidence of readiness includes a methods paper, thesis, research-assistant work, or an analysis that explains design choices and limitations. Applicants from computing, social science, clinical, or other backgrounds can be competitive when they show how their preparation transfers and identify gaps honestly. Ask admissions staff which prerequisites are mandatory, and ask potential supervisors whether the proposed question aligns with available mentorship and realistically obtainable data.
Where a master's-level bridge is appropriate, investigate conditional public health master's eligibility rather than assuming provisional admission to a PhD is available.
A PhD usually extends beyond coursework into examinations or other advancement milestones, a dissertation proposal, research, and defense. The time required depends heavily on program rules, study intensity, advisor availability, approvals, and data access. A promising question can stall if a partner cannot release records or a planned design requires data that were never collected.
Compare net cost, not advertised tuition alone: include fees, required enrollment after coursework, health coverage, travel, computing, and foregone earnings. A funding offer needs close reading. Confirm its covered terms, renewal conditions, teaching or research duties, summer support, and what happens if dissertation work continues beyond funded years. For a related methods-focused comparison, epidemiology doctorate net tuition planning can help frame questions, but only a current written offer establishes an individual student's charges and support.
They can be, if the program provides rigorous methods instruction, sustained supervision, research-group participation, and feasible access to appropriate data and computing resources. Delivery format cannot substitute for those conditions. An online student may still face scheduled seminars, residencies, collaborative meetings, research duties, and fixed dissertation milestones.
Before enrolling, request the current doctoral handbook and ask how often students meet supervisors, how methods feedback works remotely, whether required courses are offered reliably, and who can approve data use. Check whether the dissertation can proceed with data the student is authorized to access; a public dataset may be suitable for some questions but not the proposed estimand. Compare budget-friendly online public health PhD options on these research conditions as well as price. "Online" is a delivery description, not a guarantee of flexible scheduling or a viable dissertation.
Graduates may pursue faculty research, epidemiology, health-services research, policy evaluation, or methodological roles in universities, government, health systems, and other research organizations. Responsibilities can include designing studies, securing and governing data, supervising analyses, publishing findings, and explaining uncertainty to decision-makers. Hiring also depends on research record, specialty, computing skills, and experience; the degree alone does not qualify someone for every senior role.
Salary comparisons should use the actual occupation and employer setting, not a purported "PhD salary." For example, the US Bureau of Labor Statistics' epidemiologist occupation includes workers with different degrees and experience levels, so its occupation-wide median is not a forecast for a new PhD graduate. Before accepting a costly offer, compare advertised pay for the roles you want with your likely funding and earnings during study.
If your target is applied public health work rather than independent research, assess whether an MPH plus experience is a more direct route; reducing MPH tuition with institutional awards may be relevant to that alternative, subject to each school's award rules.
How can students choose an accredited Public Health PhD program?
First, verify the university's institutional accreditation and the exact scope of any Council on Education for Public Health (CEPH) accreditation. A school or unit may hold CEPH accreditation without every degree being covered in the way an applicant assumes. Confirm the status and degree coverage with the accreditor and institution rather than relying on a general marketing statement.
Then test research fit: read the current handbook, inspect recent faculty work, identify methods courses and their teaching schedule, and ask how students obtain dissertation data, approvals, and backup supervision if plans change. Compare written funding terms and progression policies alongside those research conditions. If the work you want requires strong public health practice preparation but not a dissertation, comparing online MPH admission requirements may help assess an alternative entry point; easier admission is not evidence of better causal-research training.
Other Things You Should Know About Public Health
No. Statistical significance addresses uncertainty within an analysis; it does not establish that the design identifies the claimed effect or eliminates bias.
No. A well-chosen negative control may reveal a problem with an explanation or design. A reassuring result cannot rule out every unmeasured source of bias.
When feasible, recording the question, design, and planned analyses in advance helps distinguish prespecified tests from later exploration. Registration does not make weak assumptions valid.
State which assumption or data requirement is unsupported, describe what the analysis can establish instead, and avoid presenting an association as the intervention's effect.
- Causal diagrams to evaluate sources of bias - NCBI https://www.ncbi.nlm.nih.gov/books/NBK612863/
- Sensitivity analysis for causality in observational studies for regulatory science - PMC https://pmc.ncbi.nlm.nih.gov/articles/PMC10877517/
- Chapter 6: Methods to address bias and confounding - ENCePP https://encepp.europa.eu/encepp-toolkit/methodological-guide/chapter-6-methods-address-bias-and-confounding_en
- Introduction to Fundamental Concepts in Causal Inference and ML Approaches of Causal Inference | purdue.edu https://www.stat.purdue.edu/~sabbaghi/teaching/causalMLNovelSettings/Introduction%20-%20Causal%20Inference%20and%20ML.html
- Chapter 9 Causal assumptions | ox.ac.uk https://www.stats.ox.ac.uk/~evans/APTS/causassmp.html
- Using causal inference to advance public health and medicine | Biostatistics https://www.biostat.washington.edu/news/stories/using-causal-inference-advance-public-health-and-medicine
- Difference-in-Difference Estimation | Columbia University Mailman School of Public Health https://www.publichealth.columbia.edu/research/population-health-methods/difference-difference-estimation
- 10 Causal estimands | r-causal.org https://www.r-causal.org/chapters/10-estimands
- 10 Types of Treatment Effect You Should Know About | egap.org https://methods.egap.org/guides/research-questions/effect-types_en.html
