What is baseline data?
Baseline data describes the starting condition of a person, group, place or process before a change is introduced. It gives later measurements a reference point. A baseline might be a skills score before training, average service waiting time before a process change, or household income before a livelihood program.
The baseline is not necessarily one number. It may include distributions, interview findings and differences between groups. Its value depends on a clear definition of what was measured, who was included and when the measurement happened. A baseline does not, by itself, tell you what would have happened without the intervention.
Choose the comparison before collecting the baseline
| Question | Appropriate design | What to retain |
|---|---|---|
| How did each participant change? | Measure the same people over time. | A stable identifier and consistent measures. |
| How did conditions in the service population change? | Repeated representative samples may be appropriate. | Comparable sampling, population definitions and timing. |
| Did a process improve? | Measure a consistent process or location over multiple periods. | Volumes, seasonality, operating conditions and definitions. |
An anonymous repeated survey can describe population trends if the samples and questions are comparable. It cannot show an individual trajectory. Conversely, matching the same people does not make a poorly designed question valid or eliminate selection bias. Choose the design for the question rather than treating an identifier as a substitute for sound measurement.
A practical baseline collection plan
- Define the outcome. Write the change you expect and the unit: a person, household, company, site or transaction.
- Choose a measure. Specify a score, percentage, count, rate or qualitative description that reflects that outcome.
- Set the window. Collect before exposure to the intervention where possible. Record the actual date, not only the reporting quarter.
- Define who is eligible. Keep a list or sampling rule so the denominator can be reconstructed.
- Pilot the questions. Check understanding, language, accessibility and whether people can answer accurately.
- Plan follow-up. Agree when to repeat the measure, how to contact people where appropriate, and how to handle missing observations.
Collect only information needed for the evaluation or service purpose. For sensitive information, explain its use and restrict access. Keep the questionnaire version, sampling method and any collection problems with the data.
How to calculate a baseline
There is no single baseline formula: the calculation depends on the indicator. These examples use illustrative data.
| Measure | Example calculation | Interpretation |
|---|---|---|
| Mean score | Total score 720 ÷ 12 valid assessments = 60 | Average score among the 12 assessed people. |
| Percentage meeting a criterion | 18 meeting the criterion ÷ 60 assessed × 100 = 30% | Share of assessed people meeting a defined threshold. |
| Rate | 15 incidents ÷ 3,000 service hours × 1,000 = 5 | Five incidents per 1,000 service hours. |
| Median | Ordered waiting times: 2, 3, 4, 8, 18 days; median = 4 | The middle waiting time, less influenced by the longest wait. |
Report the number of valid observations and missing values. If 60 of 80 eligible people completed the assessment, the response coverage is 75%; it is not correct to treat the remaining 20 as having failed the criterion. For a sample estimate, consider uncertainty and the sampling design before generalizing.
Calculate change with the same definitions
Suppose the share meeting a skills threshold rises from 30% to 45%. The difference is 15 percentage points. Relative to the baseline, the increase is 50%: (45 − 30) ÷ 30 × 100. These are different statements. When the baseline is zero, this relative percentage-change formula is undefined; report the absolute change.
For paired data, calculate change among people observed at both times and report how many were matched. Also show coverage of the original group. If people with the greatest barriers are less likely to answer follow-up, the matched group may give an overly favorable picture.
For repeated samples, compare population estimates only after checking whether sample composition and collection methods changed. A change in the mix of respondents can move an average even when no individual improves.
What if the baseline is missing?
First look for relevant earlier administrative records, assessments or routine measures. Check whether they describe the same population, outcome and time period. Label any reconstructed baseline and document its limitations.
A retrospective question—asking people now to recall their earlier condition—can be useful in some settings, but recall and changed understanding can affect the answer. Do not present it as equivalent to a measurement collected before the program. If no credible baseline exists, describe current conditions and establish a starting point for future measurement rather than inventing a before figure.
Keep qualitative evidence beside the numbers
A starting interview can explain barriers that a score misses: limited equipment, irregular shifts or uncertainty about a process. At follow-up, ask what changed and what made that change easier or harder. Apply a consistent coding approach while allowing unexpected findings.
Where linking is appropriate, keep responses, documents and observations attached to the relevant record. AI can help organize text, but a reviewer should check themes against the source, especially uncommon or contradictory responses. Collection-time validation reduces errors; it does not remove the need for quality checks.
Use the baseline to make a decision
Before follow-up, agree what finding would change the plan. For example, low starting skill may call for prerequisite support; long waiting times at one site may call for a process review. Use the baseline to understand starting conditions, not to blame participants or to choose only easy-to-serve groups.
For collection design, continue to pre and post surveys. For the broader evaluation purpose, see CDC’s evaluation framework.
Frequently asked questions
Is baseline data always collected before a program?
Ideally it describes conditions before the intervention. If you reconstruct it later or use an existing earlier source, state that clearly and assess comparability.
Does baseline data need to be quantitative?
No. Interviews, observations and document evidence can describe starting conditions. Use a systematic approach and retain the collection context.
Can baseline and endline use different people?
Yes, for some population-level questions. Comparable repeated samples can describe group trends; they cannot establish individual change.
Does improvement from baseline prove the program worked?
No. Other changes, measurement effects and differences in respondents may explain part of the result. The evaluation design determines what causal claims are justified.
