What to settle before you collect data

Organizations collect data because a decision feels safer with numbers or with quotes. The files then sit unused, because the question was never specific enough to analyze. The cheapest time to fix that is before the questionnaire, the interview guide, or the export from an existing system. Collection is a cost. Analysis is the point.

Start from the decision, then the question

Write the decision and the date it will be made. Then write the question the data must answer for that decision. “How are we doing?” is not a question a spreadsheet can answer. “Which of these three service points recorded a waiting time last quarter, and which did not?” can be answered if the records exist.

If you cannot name the decision, do not collect yet. A dataset gathered for “future use” almost never fits the future question, because the categories were chosen for a different reason. It is better to keep the records you already have in good order than to launch a new instrument with no reader.

Check what you already store

Ask which system, spreadsheet, or notebook already holds the fact you need. Many collection exercises repeat a record that finance, a programme officer, or a previous study already keeps. The first product of a data assignment is often a map of those sources, not a new form.

When you use existing files, budget time to clean them. Names differ. Dates are missing. Categories changed halfway through the year. Cleaning is part of analysis, not a delay before the real work. A finding built on an unexamined export will not survive the first question in the meeting.

Choose a source you can actually reach

A survey requires a defined group and a way to reach that group. Interviews require people who will give the time, and a guide tied to the question. Observation requires access. If the source is unreachable, change the question or change the method. Do not write a sample size you cannot meet and then quietly interview whoever answers.

For qualitative material, decide how notes will be written and who may see them. For quantitative material, decide the definitions before the form goes out. “A client” or “a completed case” must mean the same thing in every row. Changing the definition after collection makes the rows impossible to add together.

Design the analysis before the last response arrives

Sketch the table or the themes you expect to report. If you cannot sketch them, the instrument is asking the wrong things. The sketch is not a prediction of the finding. It is a check that each question on the form has a place in the note.

Report the limits with the finding. A partial export is not a census. Ten interviews are not every community. Deepnex Consulting’s data collection and analysis support covers this sequence: the question, the sources, the instrument when one is needed, the cleaning, and a note that stays tied to the decision. The assignment does not claim that the numbers represent more than the people or records that were actually included.

Keep a short record of decisions made while cleaning: which rows were dropped, which categories were merged, and which dates were unusable. That record belongs with the analysis, not in a separate archive nobody opens. When someone asks why a total changed between the export and the note, the answer should be a sentence in the method, not a reconstruction. Settling this before collection, and writing it down during cleaning, is what makes the file usable the next time the same question is asked.