2027 Public Health PhD Research: Distinguishing Within-Person and Between-Person Associations
Co-Founder and Chief Data Scientist
A study may find that people who report more stress also report worse sleep. That finding does not necessarily mean a person sleeps worse when their own stress rises. For public health PhD applicants and graduate methods learners, this distinction can change the research question, analysis plan, and interpretation of results. This guide explains how to separate within-person from between-person associations, what each can and cannot establish, and how to judge whether doctoral training will prepare you to study them responsibly.
Key Things You Should Know
- Within-person associations compare a person's observations over time; between-person associations compare people with different typical exposure levels. They can differ in size or direction.
- Estimating a within-person association requires repeated observations for individuals; at least two observations per person are needed to observe individual change, though more may be needed for a credible analysis.
- Person-mean centering, fixed-effects models, and related longitudinal methods help separate the questions, but none makes an observational association automatically causal.
What are within-person and between-person associations in public health research?
A between-person association asks whether people with different typical exposure levels also differ in an outcome. A within-person association asks whether an individual's outcome differs at times when their exposure departs from their own usual level. The unit of comparison - not merely the subject of the study - defines the distinction.
Consider a hypothetical study collecting weekly stress and sleep reports. Comparing average sleep among people who typically report high versus low stress is a between-person analysis. Comparing one participant's sleep in higher-stress weeks with that same participant's sleep in lower-stress weeks is a within-person analysis. These are different questions even when they use the same records.
Timing also matters: stress measured after a poor night's sleep cannot straightforwardly explain that earlier sleep. Students building their foundation through online public health undergraduate education options should look for training in study design and measurement before interpreting repeated-measures results.
Why does distinguishing these associations matter for public health PhD research?
A pooled association can combine both sources of variation and answer neither question cleanly. For example, people living in neighborhoods with more tree cover might have better average mental health, while a temporary change in an individual's exposure to green space might have little measurable association with their short-term symptoms. The example is hypothetical; neither pattern establishes what would happen after an intervention.
The distinction determines what a dissertation can claim. A policy that changes neighborhood conditions concerns a different intervention, timescale, and set of confounders from one that changes a person's daily behavior. Within-person comparisons remove confounding by characteristics that do not change during observation, but changing work conditions, illness, season, and reverse causation may remain. Between-person analyses can address enduring differences but are especially vulnerable to differences in the people being compared.
Before committing to doctoral research, write the proposed comparison in plain language: Whose exposure changes, over what interval, and whose outcome is compared? If the aim is simply to conduct routine program monitoring rather than develop an original research agenda, short online public health degree options or targeted evaluation training may be a more proportionate starting point than a PhD.
How do public health researchers analyze within-person and between-person effects?
Researchers first define the population, exposure, outcome, measurement schedule, and target comparison. Repeated observations must be linked to the same person, and the exposure must vary within enough individuals to support a within-person estimate. The table distinguishes common approaches by the question they can answer; it is not a substitute for checking assumptions.
| Approach | Within-person information | Between-person information | Important limitation |
| Pooled regression without person-level separation | Mixed into the estimate | Mixed into the estimate | The coefficient may not represent either association clearly. |
| Person-mean-centered, or within-between, model | Estimated from deviations from each person's mean | Estimated from differences in person means | Requires a suitable model for timing, confounding, and correlated observations. |
| Individual fixed-effects model | Estimated from changes within people | Time-invariant person differences are not separately estimated | Does not eliminate time-varying confounding or measurement error. |
In a within-between model, the analyst includes both a person's average exposure and each observation's deviation from that average. The two coefficients describe distinct associations, provided the model and data support their interpretation. A fixed-effects model instead uses within-person variation while accounting for stable individual characteristics.
Next, check whether exposure precedes outcome, whether people with missing visits differ systematically, whether measurement error obscures change, and whether results depend on model choices. For a causal claim, also assess time-varying confounding and whether prior exposure affects later confounders; a standard fixed-effects model may be inadequate. Complex survey weights or clustered sampling require design-appropriate analysis. Learners who need these skills but not a dissertation can compare methods depth in low-cost public health master's degrees against focused longitudinal-analysis coursework.
Useful preparation joins epidemiologic study design with regression, longitudinal data analysis, causal inference, and research ethics. Students should be able to explain the estimand, organize person-period data, inspect missingness, write reproducible code, and communicate why an estimate may not be causal. An introductory statistics course alone rarely demonstrates readiness to make these judgments independently.
Sequence matters more than collecting course titles: establish probability and regression fundamentals, practice with repeated-measures data, then study longitudinal models and causal assumptions with faculty feedback. For applicants still completing undergraduate prerequisites, undergraduate public health programs with faster course cycles may help with scheduling, but confirm that the available statistics and programming courses provide the depth your intended PhD program expects.
Requirements vary by program. Applicants generally need to document prior academic preparation, research interests, and their ability to undertake supervised scholarship; individual schools set their own degree, coursework, testing, writing-sample, recommendation, and application rules. An MPH, publication, GRE score, or named software package is not a universal requirement.
For this research topic, persuasive evidence of readiness might be a carefully explained analysis of longitudinal data, a research assistantship, or a proposal that identifies the comparison, measurement schedule, and plausible biases. Ask prospective programs whether relevant faculty are accepting students, what methods courses are accessible, and whether dissertation data access depends on a grant or external agreement. Applicants who first need general population-health preparation can review an MPH admissions accessibility guide, while checking each program's actual prerequisites rather than treating an MPH as an automatic PhD gateway.
Completion time depends on required coursework, qualifying milestones, data access, analysis, publication expectations, and dissertation approval. A longitudinal project can be delayed if follow-up data are not yet available or permission to use restricted records takes longer than expected. Request the program's published completion information rather than assuming an advertised minimum applies to your project.
Compare net financial exposure, not tuition alone: ask about stipend and tuition-remission terms, health insurance, fees, summer coverage, teaching or research duties, funding renewal, and charges after standard support ends. No program-wide price or funding guarantee applies to every applicant. Online epidemiology doctoral funding options can help frame funding questions, but verify each offer in writing and compare the time required for paid assistantship work with the time available for dissertation research.
Delivery mode alone does not establish rigor. Compare the research-methods sequence, access to suitable faculty supervision, opportunities for sustained feedback, data infrastructure, dissertation standards, and actual scheduling obligations. An online student may still need live seminars, campus visits, secure-data access, or regular meetings during work hours.
Ask how remote students obtain help with model diagnostics, reproducible workflows, ethics review, and stalled projects - and whether the proposed supervisor can work with the data you hope to study. Comparing inexpensive online MPH study pathways may be useful if your immediate need is foundational training, but an online MPH's flexibility or price says nothing by itself about an online PhD's research supervision.
A public health PhD is primarily preparation to produce original research; a DrPH is generally oriented toward advanced public health practice and leadership. Either degree may use longitudinal data, but the expected contribution differs. A PhD dissertation might investigate how analytic choices change the interpretation of repeated stress measures. A DrPH project might use appropriate analysis to improve a health department's existing service, subject to organizational approvals.
Choose by the work you want to do, not by which degree sounds more advanced. If your goal is to lead an agency evaluation, check whether a practice-focused program supports that project and your professional background. If your goal is to develop and publish methodological or substantive research under faculty mentorship, examine PhD supervision and data access. Matching public health graduate degrees to careers can also help determine whether a master's-level route already meets your goal.
Relevant roles include university research and teaching positions, epidemiology or biostatistics research roles, and evaluation or research leadership in government, health systems, and nonprofit organizations. The work may involve designing longitudinal studies, supervising analysts, securing funding, reviewing ethical safeguards, and explaining findings to decision-makers. A PhD can prepare someone for this work, but hiring also depends on publications, methods expertise, subject-area experience, and the employer's requirements.
There is no single reliable "public health PhD salary." U.S. Bureau of Labor Statistics occupational wages describe jobs such as epidemiologist or statistician, not graduates of a particular degree, and academic pay structures differ from government and industry roles. Use the wage data for the specific occupation and location you are considering, then examine actual job postings for required skills and credentials. For an adjacent credential comparison, understanding salary variation after an MPH reinforces why a degree-wide earnings figure is a weak basis for a doctoral investment.
How can you choose an accredited public health PhD program?
Verify the university's institutional accreditation and separately check any claimed Council on Education for Public Health accreditation, including the accredited school, unit, or program and whether the degree you are considering falls within its scope. Accreditation does not establish that a particular advisor, dataset, or longitudinal-methods course will be available when you enroll.
For this topic, the strongest fit is a program that can support the specific comparison you intend to study. Review faculty work, methods-course access, dissertation expectations, funding conditions, and policies for obtaining and protecting data. Ask a prospective advisor how students distinguish within-person from between-person questions, who reviews their analysis plans, and what happens if an anticipated dataset or grant becomes unavailable. If you are comparing a master's route first, university awards for online public health graduate education are a separate financial consideration - not evidence of PhD funding or research fit.
Other Things You Should Know About Public Health
Usually not if it measures each person only once. Repeated measurements of the relevant exposure and outcome are needed to compare a person's observations across occasions.
Yes. Differences between people's usual circumstances can produce one pattern, while changes within a person produce another. Report and interpret the estimates separately rather than averaging them into one story.
No, but irregular spacing can change what an average and a time-related coefficient mean. Record observation dates and consider whether the model accounts adequately for unequal intervals.
The responsible institution makes that determination under its policies and applicable rules. Consult its research-oversight office before using participant records; data-use and community permissions may also be required.
- Methods for time-varying exposure related problems in... - PMC https://pmc.ncbi.nlm.nih.gov/articles/PMC5814826/
- Principles of Epidemiology | Lesson 3 - Section 6 https://archive.cdc.gov/www_cdc_gov/csels/dsepd/ss1978/lesson3/section6.html
- Causal inference and within/between person comparisons | columbia.edu https://statmodeling.stat.columbia.edu/2019/12/15/causal-inference-and-within-between-person-comparisons/
- How to explore within-person and between... - Springer Nature https://link.springer.com/article/10.3758/s13428-022-01898-1
- Thinking within-persons: Using unit fixed-effects models to describe causal mechanisms - ScienceDirect https://www.sciencedirect.com/science/article/pii/S2590260121000333
- Regression to the Mean and Change Score Analysis | virginia.edu https://library.virginia.edu/data/articles/regression-mean-and-change-score-analysis
- Methods of Adjusting for Exposure-affected Time-varying Confounding | unsw.edu.au https://unsworks.unsw.edu.au/entities/publication/2653ac06-6cfc-4c1d-b6c7-1c627c60fbe6
- Design choices for observational studies of the effect of exposure on disease incidence | BMJ Open https://bmjopen.bmj.com/content/9/12/e031031
- Multi-Level Modeling | Columbia University Mailman School of Public Health https://www.publichealth.columbia.edu/research/population-health-methods/multi-level-modeling
