New webinar on 3rd March 2026 | 9:00 am PT
In this webinar, discover how Sopact Sense revolutionizes data collection and analysis.
Secondary data offers speed and scale primary collection can't match. Learn when to use external sources, where to find them, and how to validate quality.
A program officer at a community foundation opens a spreadsheet she downloaded from her state's labor department. The dataset shows employment outcomes by county, by year, by industry — clean, well-documented, publicly available, and free. She plans to use it to evaluate whether the workforce programs her foundation funds are producing outcomes above or below the regional baseline. Three weeks later, her evaluation team tells her the comparison isn't defensible. The state counts anyone working 20+ hours as "employed"; the foundation's grantees count only full-time, benefits-eligible placements. The state data includes workers under 24; the foundation's cohorts don't. Two datasets, two different definitions of the same word — and the mismatch was invisible in the column headers.
This is The Provenance Gap: the contextual information about how data was collected, defined, and bounded that does not travel with the data when it changes hands. Every dataset you reuse carries invisible assumptions you cannot see until they break your analysis. This article explains what secondary data is, how its types and sources differ, where examples show up in nonprofit evaluation work, and how to tell when secondary data is the right tool — versus when it is masking the need for primary collection.
Last updated: April 2026
Secondary data is information that was collected, compiled, or recorded by someone other than the researcher using it, for a purpose other than the current research question. Government statistics, academic databases, administrative records from other organizations, published survey datasets, and your own organization's historical records all qualify. Unlike primary data — which you design and collect yourself — secondary data arrives with a fixed definition, a fixed sampling frame, and a fixed collection moment that you cannot change.
The practical consequence: you can only decide whether the data as it exists fits the question you are trying to answer. You cannot redefine "employed" to mean full-time. You cannot add a demographic field that was never collected. You cannot re-interview respondents from three years ago to ask the question you actually care about now. Tools like SurveyMonkey and Qualtrics export clean files that look identical whether the underlying questions were answered by your stakeholders last month or by a different population three years ago — the format hides the provenance.
For a side-by-side methodological breakdown between the two data types, see the primary vs secondary data comparison page. For the analytical workflow once you have secondary data in hand, see analyzing secondary data.
Secondary data falls into two primary categories along the most important axis — who collected it. Internal secondary data comes from your own organization's historical records: past intake forms, case management files, donor CRM exports, program activity logs, previous evaluation surveys. External secondary data comes from outside sources: government statistics, academic research datasets, industry reports, and other organizations' published findings. The internal/external distinction matters because The Provenance Gap presents differently in each case — internal data is archaeologically hard to reconstruct; external data is evaluatively hard to fit.
A second axis separates quantitative secondary data (census tabulations, administrative counts, survey aggregations, financial records) from qualitative secondary data (interview transcripts, policy documents, case studies, archived meeting minutes). The analytical tools differ dramatically — thematic coding versus statistical regression — but the provenance risk is the same. Both require auditing before they enter an analysis.
A third practical axis separates published secondary data (peer-reviewed datasets, government reports, industry publications) from unpublished secondary data (internal records, gray literature, administrative files shared on request). Published sources typically have better methodological documentation; unpublished sources often have better contextual fit to a specific local question.
The five main sources of secondary data each carry a different Provenance Gap signature:
Government and public agencies — census bureaus, labor departments, health ministries, court records, tax authorities. Methodologically rigorous, statistically representative, typically free. The main gap is temporal: publication lag of 18–36 months means the data describing 2024 program outcomes often does not arrive until 2026.
Academic and research institutions — published datasets, longitudinal cohort studies (Framingham Heart, Add Health, PSID), dissertation archives. Well-documented with codebooks and technical reports. The main gap is population fit: an academic sample drawn for a national study rarely matches the specific population your program serves.
Industry and trade bodies — sector associations, industry benchmarking studies, trade publication surveys. Current and contextually relevant to a specific field. The main gap is methodological transparency: sampling methods and response rates are frequently undisclosed.
Internal organizational records — your own past intake data, donor CRM exports, program activity logs, historical impact reports. Contextually aligned to your work. The main gap is definitional drift across time — the intake form from 2021 measured different variables than the 2024 form, and the 2024 form's "mandatory" fields were "optional" when the 2022 cohort filled them in.
Commercial data providers — market research firms (Nielsen, Gartner, Forrester), syndicated consumer panels, third-party behavioral datasets. Current and often granular. The main gap is cost and transparency: syndicated panels rarely disclose sampling methodology in full and single reports run from a few hundred to tens of thousands of dollars.
The internal/external split is the most consequential distinction in secondary data strategy because it determines what kind of audit work you need to do before using the dataset.
Internal secondary data looks safer because your organization collected it. But the program director who designed the 2021 intake form left two years ago. The definition of "program completion" shifted when the grant scope changed. The field you want to analyze was added to the form mid-cohort. Internal data requires archaeological work — rebuilding the collection context from whoever is still around, whatever documentation survived, and any version history the system retained. Most nonprofit organizations discover, mid-analysis, that their own historical records have The Provenance Gap baked into them just as severely as any external source.
External secondary data is collected by a party outside your organization entirely. The methodological transparency is often better than internal records — peer-reviewed datasets come with codebooks; your 2021 intake form does not. But the definitional alignment with your specific question is almost never exact. External data requires evaluative work — assessing fit between the data as designed and the question you are trying to answer, and documenting every mismatch explicitly so downstream readers understand the limits.
The practical rule: internal secondary data requires archaeology; external secondary data requires evaluation. Both require audit before analysis. Neither is safer by default.
Concrete examples from nonprofit and foundation evaluation work:
Each of these examples has legitimate uses. None of them can substitute for primary data collection when the question requires causal attribution to your program's specific intervention.
Before any secondary dataset enters an analysis, answer six questions in writing:
Most organizations skip step 6. That is where The Provenance Gap does its damage — not because secondary data was used, but because the primary data that should have accompanied it was never designed.
Secondary data belongs as the outer ring around primary evidence, not as the evidence itself.
Correct use: "Our program's 12-month employment retention rate is 74%; the regional baseline from BLS data is 58%. Our cohort outperformed the regional baseline by 16 points." The secondary data provides comparison context. The primary data provides the attributable outcome.
Incorrect use: "We reviewed state employment data and saw improvement in counties where we operate; therefore our program contributed to the improvement." No causal attribution is possible without primary measurement of your actual participants. Secondary aggregate data cannot tell you what happened to the individuals your program served.
Once you have audited your secondary sources and identified the gap, design primary collection specifically for the variables secondary data cannot provide. For most nonprofit programs, three things:
This is where platforms like Sopact Sense fit — as the origin for primary data designed specifically for the evidence secondary data cannot provide. Persistent stakeholder IDs assigned at first contact mean the same participant at intake, at exit, and at 12-month follow-up is the same row forever. No ID reconciliation, no "which Sarah is this?" mid-analysis, no lost linkage when a staff member leaves.
Three misuses show up repeatedly in nonprofit evaluation:
Substituting secondary for primary because it is cheaper. Budget pressure pushes evaluators to skip primary collection. The result is an evaluation with no attributable outcome measurement, only context. Funders increasingly recognize this pattern and ask directly for participant-level evidence.
Ignoring the publication lag. Government data is often 18–36 months old by the time it is usable. If your program ran in 2024 and the most recent comparable BLS data is from 2022, your comparison is already misaligned on time. Name the lag explicitly in reporting so readers understand what is being compared.
Treating internal records as ground truth. Your own past data carries the same Provenance Gap risks as external data — sometimes worse, because internal users assume internal data is authoritative. Run the same six-question audit on your 2022 intake records that you would run on BLS.
Done well, secondary data is a force multiplier for impact measurement strategy. Done poorly, it is a defensibility risk that only becomes visible when a funder asks a pointed question.
Secondary data is information originally collected by someone other than the current researcher, for a purpose other than the current research question. Common examples include census statistics, academic study data, industry reports, and an organization's own historical records. Unlike primary data, secondary data has a fixed definition and sampling frame that you cannot change.
The secondary data definition is data that was collected, compiled, or recorded by a party other than the current researcher, for a purpose other than the current use. The defining characteristic is that all collection decisions — what was measured, who was included, when, and how — were made by someone else and cannot be modified by the person now reusing the data.
Secondary data falls into two primary types: internal (your organization's historical records) and external (government, academic, industry, or commercial sources). A secondary classification separates quantitative secondary data (numerical aggregations, administrative counts, survey tabulations) from qualitative secondary data (transcripts, policy documents, case studies, archived meeting minutes).
The main sources of secondary data are government and public agencies, academic and research institutions, industry and trade bodies, internal organizational records, and commercial data providers. Each source carries a different Provenance Gap risk — government data has long publication lags, commercial data lacks methodological transparency, internal records suffer from definitional drift across time.
External secondary data is data collected by a party outside your organization and made available for reuse. Examples include census tabulations, government labor statistics, academic longitudinal studies like Framingham Heart or Add Health, industry benchmarking reports, and commercial market research panels. External data typically has better methodological documentation than internal records but rarely matches your exact population frame.
Internal secondary data is historical data your own organization collected for a previous purpose, now being reused for a new question. Examples include past intake forms, donor CRM records, program activity logs, and prior evaluation surveys. Internal data feels safer because your organization owned the collection, but it typically carries definitional drift as forms, staff, and program scopes changed over years.
Examples of secondary data include US Census ACS estimates, BLS county-level employment data, NCES education statistics, CDC BRFSS health survey data, Candid 990 nonprofit filings, academic longitudinal studies like Framingham or Add Health, and your own organization's past intake records. Each has legitimate uses, and each carries its own Provenance Gap that must be audited before use.
Secondary data collection is the systematic process of sourcing, auditing, and preparing existing data for reuse. Unlike primary data collection — which involves designing instruments and collecting new responses — secondary data collection focuses on provenance auditing, source evaluation, documentation review, and assessing fit between the data as it exists and the current research question.
Primary data is collected by the researcher specifically for the current research question; secondary data is reused from a prior collection effort by someone else. Primary data has definitional control and temporal alignment; secondary data has cost, scale, and availability advantages but inherits all the collection decisions of whoever originally compiled it. See the dedicated primary vs secondary data comparison for a full breakdown.
The Provenance Gap is the contextual information about how data was collected, defined, and bounded that does not travel with the data when it changes hands. It is the invisible tax on every reused dataset — definitions drift, sampling frames do not align, collection moments do not match your question. The Provenance Gap is why secondary data should provide context, not carry attribution.
Advantages of secondary data: lower cost, faster availability, larger samples than most organizations can collect themselves, and baseline context for comparison. Disadvantages: fixed definitions you cannot change, publication lag that misaligns time periods, unknown or poorly documented sampling decisions, and inability to attribute outcomes to your specific program intervention without paired primary measurement.
Nonprofits should use secondary data for regional baselines, environmental scanning, comparative context, and literature reviews — any use where aggregate context complements primary evidence. Nonprofits should not rely on secondary data alone for participant-level attribution, longitudinal outcome tracking with their actual cohort, or qualitative narratives of change from the people they directly serve.
Government and most academic secondary data is free or low-cost. Commercial datasets from providers like Nielsen, Gartner, or syndicated consumer panels range from a few hundred to tens of thousands of dollars per report. The real cost of secondary data, however, is the time spent on provenance audit and the risk of building analysis on data that does not actually fit the question being asked.
