Primary vs Secondary Data: Sources, Applications, and Selection

Quick Answer: Primary data are collected firsthand by the investigator for a specific research or program purpose; secondary data are pre-existing records or datasets originally collected by someone else for another purpose. The CPH exam expects you to distinguish the two, know the major U.S. secondary sources (BRFSS, NHANES, NVSS, Census, SEER), and choose the right source for a given analytic question.

Definitions and the Core Distinction

Primary data are generated when a public health team designs and executes its own collection — a community health survey, key informant interviews, environmental sampling, biomonitoring, or a new registry. The investigator controls the variables, the sampling frame, the instrument, and the timing. The cost and time burden are high, but the data fit the question exactly.

Secondary data are records or datasets that already exist before the current question is posed. They were collected by another organization, often for administrative, clinical, or surveillance purposes — vital statistics death certificates, hospital discharge data, the U.S. Census, or BRFSS. The investigator cannot change what was asked or how, but gains speed, scale, and geographic breadth at low cost.

The CPH distinction is purpose-driven, not format-driven. A state health department running BRFSS is producing primary data for its own surveillance objective; a university researcher downloading the same BRFSS file is using secondary data. The same dataset is primary or secondary depending on the relationship of the user to the collection.

Comparison at a Glance

Major U.S. Secondary Data Sources to Know

The exam repeatedly tests recognition of the flagship population health data systems:

• BRFSS (Behavioral Risk Factor Surveillance System): state-based telephone survey of adults on health risk behaviors, chronic conditions, and preventive services; the largest continuously running telephone health survey in the world.
• NHANES (National Health and Nutrition Examination Survey): CDC program combining interviews and physical examinations with laboratory tests in a mobile examination center; the source of national prevalence of obesity, hypertension, and nutritional biomarkers.
• NVSS (National Vital Statistics System): CDC/NCHS compilation of birth and death certificates; the foundation for mortality rates, life expectancy, and cause-of-death analyses.
• U.S. Census / American Community Survey: demographic and socioeconomic denominator data for nearly every public health rate.
• SEER (Surveillance, Epidemiology, and End Results) Program: NCI cancer incidence and survival registry covering roughly half the U.S. population.
• WONDER (Wide-ranging Online Data for Epidemiologic Research): CDC portal that serves NVSS, census, and other public-use data with query tools.
• National Notifiable Diseases Surveillance System (NNDSS): case reports of reportable conditions transmitted by states to CDC.
• Healthcare Cost and Utilization Project (HCUP): AHRQ family of hospital inpatient discharge databases.

Choosing the Right Source

Match the question to the source's strengths. To estimate diabetes prevalence by state, BRFSS self-report suffices; to measure undiagnosed diabetes including those unaware they have it, NHANES laboratory values are required because they capture HbA1c in examined persons. For county-level death rates, NVSS via WONDER. For cancer survival, SEER. For social determinants, the American Community Survey. For infectious disease counts, NNDSS. For hospitalization burden, HCUP. Each source was designed for a specific inference; using it for another inference introduces bias the analyst must name and, where possible, quantify. The strongest analytic designs often link two secondary sources — for example, NVSS death counts as the numerator and ACS population estimates as the denominator — to produce a standardized rate.

Worked Scenario

A county wants to quantify food insecurity among older adults. The team could run a new primary survey using the validated USDA Food Security Module for precise local data, or pull the BRFSS optional food insecurity module if their state fielded it. The primary route yields exact local estimates but costs staff time and risks low response; the secondary route is faster but may lack county-level detail because BRFSS is designed for state-level inference. A mixed approach — secondary BRFSS for state context, primary survey for local granularity — is often the defensible answer on the exam.

Exam Traps

• Treating "survey data" as automatically primary. A survey you field is primary; the same instrument fielded by CDC is secondary to you.
• Confusing surveillance with secondary. Surveillance systems (NNDSS, syndromic surveillance) are secondary to an outside researcher but primary to the agency running them.
• Forgetting that secondary data cannot add variables you need. If the dataset lacks the social determinant you care about, no cleaning recovers it.
• Assuming secondary data are always cheaper to analyze. Complex restricted files (NHANES restricted branches, SEER-Medicare) require application, fees, and secure data enclaves.
• Assuming denominators come free. Census denominators must match the geographic and temporal granularity of your numerator; county-level ACS five-year estimates differ from one-year estimates.