2027 Comparing Administrative Records and Survey Data for a Public Health Project
Co-Founder and Chief Data Scientist
A public health project can reach different conclusions depending on whether it counts service encounters, asks people about their experiences, or combines both. Administrative records and surveys are built for different purposes, so the better source depends on the population, outcome, and decision you need to study. This guide is for students and practitioners planning an evaluation or research project. It explains how to compare the sources, check their limitations, protect participants, and choose an analysis that the available data can support.
Key Things You Should Know
- Use administrative records to study documented events among people captured by a system; use surveys when the question requires experiences, behaviors, or information the system does not record.
- Define the target population and denominator first. A complete set of clinic records is not necessarily representative of everyone who needed care, while a survey sample requires attention to selection and nonresponse.
- Combining sources can answer richer questions, but linkage requires permission, reliable matching, and an analysis of who could not be matched.
What are administrative records and survey data in public health research?
Administrative records are information collected primarily to run a service or fulfill an operational requirement. Examples include insurance claims, hospital discharge records, immunization registries, and program enrollment files.
Their units may be people, visits, claims, or transactions; those units are not interchangeable. A claim documents a billed service, for example, not necessarily every instance of illness or need for care.
Survey data come from questions asked under a defined collection plan. A survey might measure symptoms, unmet needs, attitudes, or experiences that never appear in a service record. Its usefulness depends on whom researchers invited, who responded, how questions were worded, and whether the resulting sample supports the intended population estimate.
Students learning these distinctions in a fast-paced public health bachelor's course format should check that the schedule allows time to examine a data dictionary and sampling plan - not just run calculations. Those documents establish what each source can actually measure.
How do administrative records compare with survey data in public health projects?
The comparison below concerns typical uses, not guarantees. Either source can be weak if its coverage or measurement does not fit the research question.
| Decision factor | Administrative records | Survey data |
| Primary purpose | Document operations, services, or required reporting | Collect responses to specified research or monitoring questions |
| People observed | Those captured by the contributing systems | Those selected and responding under the survey design |
| Common strength | Documented events and, sometimes, repeated observations | Self-reported experiences and measures selected for the question |
| Common limitation | Missing people or events outside the system | Nonresponse, recall error, and limits on subgroup precision |
| Typical access issue | Agency permissions and restrictions on identifiable records | Recruitment, consent, and restrictions tied to the survey protocol |
A useful distinction is coverage versus representativeness. A registry may cover nearly every recorded event within its jurisdiction yet say little about unrecorded events. A carefully designed survey may estimate a broader population despite observing fewer people. Neither a large record count nor a survey label establishes validity on its own.
For students considering analytical careers, comparing earnings across MPH career paths is separate from evaluating data quality: an occupation-wide pay figure cannot tell you whether a particular dataset supports a defensible estimate.
When should researchers use administrative records, surveys, or both?
Start with the estimand: the quantity the project aims to estimate. Specify for whom, what outcome, and over what period before requesting data. Suppose a local program wants to assess follow-up after referral. Records may show documented visits among enrolled clients; a survey may reveal why some clients did not seek care. Neither alone necessarily estimates follow-up among all residents referred elsewhere.
Choose records when the decision concerns documented service use, reporting timeliness, or changes among people reliably captured by a system. Choose a survey when the essential measure is perceived access, an unrecorded behavior, or an experience best reported by participants.
Consider both when the question requires a documented outcome and its context, provided access, cost, timing, and consent make integration feasible.
Before committing, identify the population each source excludes and whether that exclusion could change the decision. If only a modest methods course is needed to answer a bounded workplace question, a full degree - including a lower-cost online MPH program - may be a larger investment than the project requires.
Records can provide dates, service codes, and repeated observations without asking participants to recall every encounter. That makes them useful for tracking documented utilization or operational performance.
Availability is not the same as readiness, however: a researcher may need to reconcile duplicate patients, changing codes, transfers between systems, and delays between an event and its recorded appearance.
Records also reflect how an organization works. Billing incentives, eligibility rules, and documentation practices can change a recorded diagnosis or service count without an equivalent change in population health.
A denominator made up of enrolled clients answers a different question from one made up of all eligible residents. Record provenance, inclusion rules, and coding changes therefore belong in the project plan before analysis begins.
For prospective students, admissions convenience - such as an online public health master's program without recommendation letters - does not indicate whether supervised coursework will teach these record-audit skills. Evaluate the actual methods assignments and data-access opportunities instead.
Surveys let researchers ask about outcomes that administrative systems do not routinely capture, using wording and timing designed for the question. They can also ask respondents about barriers to services. Those advantages depend on a sound sampling frame and questions that respondents understand consistently.
A survey can mislead if people without reliable contact information are missed, certain groups respond less often, or respondents cannot accurately recall the requested period. Weighting may reduce some differences between respondents and the target population, but it cannot reliably repair an important unmeasured difference.
A short convenience survey can still inform a local service discussion; it should not be presented as a representative population estimate.
A student who needs stronger preparation in questionnaire design can ask whether a program offers introductory methods work or conditional entry into a public health master's program. Admission flexibility is distinct from being prepared to design and interpret a survey.
Assess quality against the intended decision, not against an abstract claim that a dataset is "clean." First, inspect definitions, collection periods, missingness, duplicate records, and changes in who or what could be observed. Then ask whether the available measure represents the concept of interest: a documented screening is not identical to receiving an appropriate screening, and a self-reported visit may not match a billing record.
Representativeness is question-specific. For records, compare the contributing systems and eligible population with the target population. For surveys, review selection probabilities, response patterns, weights, and whether estimates for small groups are stable enough to report.
For either source, examine whether missing observations differ on characteristics related to the outcome; describe uncertainty that those checks cannot resolve. Students evaluating online MPH accreditation and program quality should distinguish program recognition from evidence of methodological depth. Look for assignments that require a written assessment of source coverage and validity, not only software output.
Requirements depend on the information, its holder, the proposed use, and applicable law and institutional policy. HIPAA may govern protected health information held by a covered entity or business associate; it does not automatically govern every public health file.
An agency may impose a data-use agreement even for information described as deidentified, and a dataset without direct names may still pose reidentification risks.
Before accessing records or collecting survey responses, check who may authorize the use, whether consent or another permission is required, whether institutional review board review or a formal determination is needed, and how files will be stored, accessed, retained, and disclosed.
A class assignment is not automatically exempt from these checks. Community partners may also have legitimate expectations about interpretation and public release beyond the minimum legal requirements.
Restricted-data work can require secure computing and staff time. Doctoral researchers exploring scholarship options for epidemiology doctoral students should budget for those project requirements separately; funding or enrollment does not grant access to identifiable data.
Linkage connects observations believed to refer to the same person or event. With an approved shared identifier, matching may be straightforward; without one, researchers may use combinations of permitted fields and estimate how confidently records correspond. A match is still an inference, not proof, when identifiers are incomplete or inconsistent.
The main analytical risk is selective linkage. If respondents with unstable contact details are harder to match, the linked sample may exclude people whose experiences matter most to the project. Report which records were eligible, how many could be linked, which characteristics differed between linked and unlinked cases, and how uncertain matches were handled. Keep identifying fields and analytic extracts separate where the approved workflow requires it.
Before proposing a linked thesis or dissertation, confirm that the data holders permit linkage and that the project can absorb its time and security costs. Budget planning for a public health PhD should not assume a low tuition figure covers restricted-data infrastructure or linkage support.
Method follows the estimand and data structure. Administrative records may call for event rates with a defensible population denominator, models of repeated observations, or time-to-event analysis when follow-up and censoring are defined.
Probability surveys generally require analysis that accounts for their weights, strata, and clusters when estimating population quantities and uncertainty. Combining sources may additionally require attention to match uncertainty and selection into the linked sample.
Regression can adjust for measured differences between groups, but it does not turn an observational comparison into a causal effect. Check whether important confounders were measured, whether dates establish the intended sequence, and whether missingness or misclassification could explain the result.
Report uncertainty and sensitivity analyses where they address a plausible weakness rather than adding models for appearance.
A focused statistics or survey-methods course may be more useful than choosing among affordable shorter online public health degrees if the immediate goal is to analyze one project responsibly. Supervision matters most when the dataset has a complex design or restricted access.
How do public health degrees prepare students to work with these data sources?
Preparation varies by program and by the responsibility a student wants to take on. An MPH can provide a population-health foundation and applied epidemiology or evaluation work; a research-oriented MS may offer deeper quantitative training.
Doctoral study is more appropriate when the goal requires sustained independent research, not simply access to a larger dataset. None of these credentials substitutes for permission to use records, sound supervision, or experience explaining limitations to a public health audience.
Compare actual coursework and supervised projects: Will you define denominators, assess a survey design, inspect record provenance, write a reproducible analysis, and defend an interpretation?
Ask whether students can work with appropriately approved data and what happens if access is delayed. A public or synthetic dataset can support methods practice, while restricted projects require a separate access plan.
If cost affects the decision, online MPH university scholarship planning belongs alongside a comparison of methods depth, supervision, and time to completion. Choose the smallest credible training route that closes the skill gap for the work you intend to do.
Other Things You Should Know About Public Health
Yes. Clearly labeled synthetic data can help test a workflow and demonstrate coding skills. It cannot establish that an analysis will work on the real data or support claims about actual people.
Often, but check the data-use agreement and institutional policy first. Remove embedded credentials, identifiers, small-cell outputs, and details that could disclose protected information.
Follow the project's approval process. Typically, the data steward checks disclosure rules, while the project lead and relevant community partners review whether labels, context, and interpretation are appropriate.
Ask your supervisor about a revised schedule and an approved fallback, such as a documented analysis plan or work with public data. Do not substitute unapproved identifiable records to meet a deadline.
- Methodology | PLACES https://www.cdc.gov/places/methodology/index.html
- Assessing data linkage quality in cohort studies - PMC - NIH https://pmc.ncbi.nlm.nih.gov/articles/PMC7261400/
- A Guide for Using Administrative Data to Examine Long-Term Outcomes in Program Evaluation https://acf.gov/opre/report/guide-using-administrative-data-examine-long-term-outcomes-program-evaluation
- Summary of the HIPAA Security Rule | hhs.gov https://www.hhs.gov/hipaa/for-professionals/security/laws-regulations/index.html
- Data Sources for Health Care Quality Measures | Agency for Healthcare Research and Quality https://www.ahrq.gov/talkingquality/measures/understand/index.html
- Comparing the Accuracy of Health Record Data and Self-Reported Data - NCBI Bookshelf https://www.ncbi.nlm.nih.gov/books/NBK616906/
- Data Protection Law: An Overview | congress.gov https://www.congress.gov/crs-product/R45631
- What leads to measurement errors? Evidence from reports of program participation in three surveys - ScienceDirect https://www.sciencedirect.com/science/article/pii/S030440762300297X
- guide to evaluating linkage quality for the analysis of linked data | oup.com https://academic.oup.com/ije/article/46/5/1699/4107250
- 3 Using Multiple Data Sources to Enhance Data Equity | The National Academies Press https://www.nationalacademies.org/read/26804/chapter/5
