From gap to protocol
8 min read
Lesson objectives
- Turn a practical problem into a researchable question and documented evidence-search plan
Please sign in to use all the tools
Once you sign in, the reading tools are yours: larger or smaller text, typeface, line spacing, focus mode, highlighting and bold, and notes saved to your account and gathered in My study.
After searching and screening, do not write “No studies exist” because a bounded search found none. Describe what was examined: sources, date, eligibility criteria and records reaching full-text assessment. Frame a tractable gap: available studies may examine another sector, measure satisfaction rather than departure, or fail to separate voluntary resignation from contract completion. The gap may concern transferability, not a total lack of publications. Link it to the decision: what would change if the question were answered? If neither a decision nor understanding would change, reconsider the value of collecting new data.
A protocol records decisions before execution. Start with purpose, question, analysis unit, population, inclusion and exclusion rules. Specify design and procedures for recruitment or record extraction. Define each outcome, exposure and explanatory variable by source, unit and time window. “Employee departure” can mean voluntary resignation or any contract ending; the choice changes the numerator and interpretation. Make a short codebook with variable name, definition, permitted values, missing-value rule, source and reviewer. Writing a definition does not make a dataset accurate. Pilot it on synthetic records and, if access is permitted, a small sample of actual records.
State in advance how indicators will be calculated or text analyzed, including handling missing and duplicate entries. “Use appropriate analysis” leaves room to choose a method after seeing results. Equally, do not freeze a detail that cannot be known before inspecting the record structure. Write conditional decision rules: if missingness exceeds a justified threshold, perform a sensitivity analysis or reassess feasibility. Record the threshold and why it matters. Seek statistical review for complex designs instead of selecting a test from a menu. For qualitative work, describe participant selection, interview documentation, analytic memos and treatment of contradictory accounts without imposing a numerical hypothesis.
Set authority and ethics boundaries before the timetable. Who may see person identifiers? Who may authorize contact? Does the plan require local ethics or institutional review? A specialist cannot infer an exemption because records are “internal” or seem anonymous to them. Pause collection or refer it to the responsible body if authorization is unclear. Specify access, security and retention under institutional policy; do not invent a universal legal retention period. Name the person responsible for each decision in the protocol itself.
Add a change log. Keep a protocol version and date, describe what changed and why, and record whether the change preceded or followed inspection of outcomes. Revisions can be justified, but their timing affects how evidence is interpreted. Preserve the earlier version rather than rewriting history. When reporting, compare what was planned, done and found, then explain deviations and their possible effects. That trace lets another researcher assess a conclusion even when they disagree with it.
In a protocol, distinguish fixed decisions from conditional ones. Set population and primary outcome before viewing results. A fallback for a failed data source can be specified prospectively: if exit reason is unavailable for a consequential share of records, pause resignation estimation and consult the data owner or version a revised question. Such flexibility cannot be used to switch goals after an inconvenient result, nor must a protocol be rigid in the face of a foreseeable obstacle. Name who may activate the condition.
Add a short knowledge-risk table: possible claim, what might make it misleading, a signal revealing the issue and a response. For “the policy reduced resignations,” concurrent pay changes are a threat, an HR decision log is a possible signal, and a narrower statement or better comparison is the response. This differs from an information-security risk list, although both matter. It keeps the principal reservation from being buried late in the report.
Run a readiness review before collection. Is the instrument understandable in both working languages? Does the codebook match the question? Is the outcome source available and authorized? Is there a deviation and incident channel? Is follow-up timing feasible? Give each answer, evidence and decision owner. A heading in a file is not completion. Walk one synthetic case from eligibility through a proposed result table and find where the route breaks. Repair that point before expanding collection, retaining reason and version.
Separate when an idea arose, when the protocol was fixed and when outcomes were inspected. A plan written after collection can be useful, but must not be called prespecified. Preregistration can make decisions transparent where appropriate; it does not certify execution or question quality. Correct a discovered error through a visible version rather than knowingly following a defective rule for appearances. Show which analyses were designed before results and which emerged afterward.
Lesson summary
From gap to protocol: write objective, operational definitions, decision and change log.