Why Collecting Less Can Improve a Study
A data collection form can easily grow beyond what an analysis requires. A project may begin with a clear question, then acquire extra fields because they are easy to ask for, might be useful someday, or appeared on an earlier form. Each extra field takes time to collect and manage. It may also reveal something private, make a person easier to identify, or create a new opportunity for error or unauthorized access.
Collecting only necessary data does not mean collecting as little as possible regardless of consequences. Omitting a variable needed to define the study group or measure its outcome can make the analysis invalid. The goal is to collect the minimum data that are adequate for the stated purpose, including justified requirements of the study design and responsible analysis.
The key word is stated. A project cannot reliably decide what is necessary until it says what population it concerns, what question it will answer, and what analysis it plans. For example, “learn about health” is too broad to guide a collection form. “Estimate the proportion of patients at this clinic who missed a scheduled appointment during April” gives a more useful starting point: the group, outcome, and time period are specified.
A Practical Test for Each Field
For each proposed variable, ask what job it does. Common legitimate jobs include defining eligibility, measuring the outcome or exposure in the question, supporting a planned comparison, or linking records when the analysis genuinely requires that link. A field that does none of these jobs is a strong candidate for removal. A field that has a job should still be collected at no greater level of detail than needed.
Name the target group, the question, and the planned summary or comparison. A vague purpose makes it difficult to judge necessity.
Identify the outcome, any exposure or comparison group, and information needed to establish eligibility or carry out the design.
Ask what concrete part of the stated analysis would fail or become materially less valid if the field were omitted.
Consider a broader category, a yes-or-no indicator, an on-the-spot eligibility check, or a temporary code instead of detailed personal information.
Do not collect information merely because it might support an unspecified future question.
This test is a reasoned comparison, not a claim that every field has a numerical risk score. If removing a variable leaves the study’s stated output and validity unchanged, then its analytical contribution is not established. If keeping it adds privacy exposure, collection burden, or the chance of data-entry mistakes, collecting it has a cost without a demonstrated benefit for that purpose. Conversely, if removing it prevents the study from identifying eligible records or measuring its outcome, it is not redundant. This logic supports a field-by-field decision rather than a blanket rule to collect either everything or almost nothing.
Worked Decisions in Different Study Settings
Worked Example: Estimating a Commute-Time Average
A hypothetical city project will estimate the average one-way commute time, in minutes, among adults who work within the city. The team proposes asking for commute time, work location, age, date of birth, home street address, employer name, and name.
Start with the target group and outcome. The outcome is one-way commute time in minutes. The target group has two eligibility conditions: participants must be adults and must work within the city. Commute time is necessary because it is the quantity to be summarized.
Check eligibility fields separately. Work location can help determine whether the work-within-city condition is met, but it does not establish that a respondent is an adult. Adult eligibility must be established separately—for example, by screening participants and recording only a minimal adult-eligibility indicator if the analysis needs an eligibility check in the dataset. Alternatively, if recruitment screening ensures that only eligible adults enter the analysis and no later record-level verification is needed, the team may avoid retaining an age field. The team should not treat a city work location as evidence of adult status.
Compare the remaining fields with the purpose. Exact date of birth is not required if adult status can be checked without retaining it. A home street address is more detailed than needed to establish that the respondent works within the city; it should not be collected for this purpose. Employer name is also unnecessary unless the project has a separate stated need involving employers. A participant’s name does not help calculate the average commute time and would make the records more directly identifiable.
A suitably limited collection could therefore contain commute time and, if needed for record-level eligibility checks, minimal indicators for adult eligibility and city-work eligibility. The study should also decide whether any linking code is necessary; it should not add direct identifiers by default. The proposed fields are judged by their role in this particular estimate, not by whether they might be interesting in a different project.
Worked Example: Comparing Two Appointment Reminders
In a hypothetical randomized study, a clinic compares its usual reminder with a new reminder to see whether the new approach changes the proportion of scheduled appointments that are missed. Each appointment is assigned one reminder approach, and the clinic records whether it was missed.
Identify the analysis. The comparison needs the assigned reminder group and the missed-or-attended outcome. It also needs enough information to pair each outcome with its assignment. Those fields support the stated comparison directly.
Consider whether a record link is needed. If the study receives appointment and assignment information from separate systems, a temporary study code may be needed to match the two records correctly. The code’s purpose is linkage, not identification for its own sake. If one collection process records assignment and outcome together, a separate link field may not be needed.
Review tempting additions. The clinic might propose collecting each patient’s name, full date of birth, home address, and all prior diagnoses. None of these is automatically required to compare the two reminder groups on missed appointments. If the design has a prespecified, justified need to account for a particular factor, the team should identify that need specifically and collect an adequate measure of that factor—not an open-ended medical history. If no such need is part of the stated analysis, the extra fields should be left out.
The appropriate collection is not necessarily a single universal list: the method of record linkage and the prespecified design determine whether a code or additional measure is needed. But “it could be useful later” is not, by itself, a sufficient reason to collect names, addresses, or broad sensitive histories for this comparison.
Worked Example: Measuring a Student Survey Outcome
A hypothetical university survey asks whether students used a particular campus service during the current term. The intended result is the proportion of responding students who report using it. The draft form also requests student ID, exact home address, date of birth, and a detailed list of health conditions.
Keep the outcome measure. A response indicating whether the student used the service is necessary to calculate the requested proportion. If the target population is currently enrolled students, the study also needs a sound way to determine that respondents are in that group. That check might occur during recruitment rather than by retaining student IDs in the response file.
Ask whether each proposed identifier is needed. If the survey is anonymous and does not need to prevent repeat responses or link records, student ID is not necessary for the stated estimate. Exact address and date of birth add detail without measuring service use. A detailed health-condition list is sensitive and does not contribute to this particular outcome unless a separate, specified analysis justifies collecting particular health information.
Consider the limits of the answer. The resulting estimate describes the respondents represented by the survey’s design; removing identifiers does not itself guarantee that the respondents represent every enrolled student. If a later project needs to study differences by a defined characteristic, it should state that purpose and collect an appropriately limited measure then. The present form should not gather broad personal details on the assumption that an unspecified future question will arise.
Necessary, Useful, and Merely Available
A field can be interesting without being necessary. “Could we explore this later?” is weaker than “Which part of the stated analysis requires it?” A planned subgroup comparison, for example, may justify collecting a particular grouping variable. An imagined possibility of many future subgroup comparisons does not automatically justify collecting a large set of sensitive details.
Likewise, a variable should not be labeled unnecessary just because it is not the outcome. Some fields are needed to determine who belongs in the target group, to distinguish the groups being compared, or to carry out the study design. The standard is not “collect only the outcome.” It is “collect only what has a defensible role in this purpose, at an appropriate level of detail.”
Identifiers deserve special attention because they often add little to the statistical question while increasing the consequences of a data exposure. Sometimes a code is needed to join records or prevent counting the same unit twice. In that case, use the least identifying approach that works, and avoid keeping direct identifiers alongside analytical data unless there is a specific reason. This decision concerns whether information needs to be collected and retained for the analysis; it does not replace secure handling.
Common Mistakes
- Collecting everything for possible future use. A vague future possibility does not establish necessity. Define a future purpose before adding fields for it.
- Confusing a screening question with a retained variable. Information may be checked to decide eligibility without being stored in detailed form in the analysis data.
- Using one eligibility field to stand for another. A location field does not establish age, enrollment, or any other separate eligibility condition. Each condition must be checked on its own.
- Calling every non-outcome field unnecessary. A comparison group, a design variable, or an eligibility indicator may be essential even though it is not the measured outcome.
- Collecting more detail than the decision requires. If a broad category is adequate, exact dates, addresses, or other fine-grained details may add risk without improving the planned analysis.
Check Your Understanding
For each situation, explain which information is necessary for the stated purpose and what could be omitted.
- A survey estimates the proportion of adults who use a public library. What must be established about eligibility, and why would a library-card number not automatically be necessary?
- A study compares two assigned reminder messages on the proportion of missed appointments. Name the core variables needed for that comparison and give one circumstance in which a study code could be justified.
- A project estimates average travel time among employees whose workplace is in a particular town. Why does workplace location alone not establish that respondents are adults if adulthood is also an eligibility condition?
- A survey asks whether students used one campus service and proposes collecting a detailed health history. What question should the team ask before retaining that history?
- Explain why “this variable might be useful someday” is not, by itself, a sufficient justification for collecting sensitive information.