My first study as a sociology graduate student collected 212 responses and produced nothing. The survey went out fast, came back fast, and died on my advisor's desk in four minutes. Leading questions. Two variables measured with one item. No pilot. An age field that let respondents type "old enough." The data was not wrong. It was unusable, which is worse, because unusable looks like progress until someone checks.

Nobody had taught me collection. My methods course covered theory and analysis, the cathedral and the interior decorating, and skipped the part where you pour the foundation. I relearned it the expensive way, and I now TA the course where students are about to make my mistakes at scale.

The reading I add to the syllabus, over the textbook's chapter, is this guide to collecting data for a research project. It is written for students running their first real collection, and it frontloads the decisions that ruined my survey: matching the method to the question, surveys measure breadth, interviews measure depth, and choosing wrong measures nothing, writing questions that do not smuggle the answer in, piloting on five people before launching on two hundred, and planning the recording format before the first datapoint arrives, so the spreadsheet does not become an archaeology site later.

The guide's sharpest section covers what to log while collecting: dates, conditions, refusals, anything that changed midstream. My 212 respondents answered two versions of question six, because I edited it after launch and told no one, including future me. The guide names that exact sin.

My students pilot now. They complain about piloting, run the pilot, and rewrite an average of three questions each. Every rewritten question is a dead dataset that never happened. As monuments go, absence is underrated.