Before the questions, a codebook has an introduction.
What the documents are, what the data is for, and anything else true of the whole corpus.
#Survey Party Likes and Dislikes Codebook
Input documents
Open-ended survey responses, four feedback columns per respondent and a case ID.
Purpose for extraction
Clean and align the responses, flag the difficult ones, and pull structured lists of topics.
A note on “Not reported”
A sentinel meaning the respondent gave no usable text. It is never text to be cleaned or categorised.
## is_valid_response
Did this respondent give any substantive feedback?
· · ·
a field you invented, and every reader still sees it
When an item does not cover the case, this is what the reader has left.
@karlrohe
Two of those fields are required. What the documents are, and what the extraction is for. The reader sees the whole block on every document, before every item.
It carries three things. The decision the data feeds. The kinds of documents coming in. Which fields matter most for that decision. On Data Mint the assistant drafts this in your own words and asks you to approve it before it proposes a single column.
The decision the data feeds is what sets how careful the reader should be. Screening abstracts for a systematic review and building a rough map of a literature are different jobs against the same documents. One wants a reader that flags anything borderline. The other does not. Nothing in an item says which you are doing.
The preamble says what kind of thing the reader is holding. “What was the main intervention?” means one thing against a corpus of randomised trials and another against a corpus of clinic notes. The item does not carry that. The preamble does.
It is also where you settle something once instead of item by item. The card shows a real one. That codebook uses Not reported as a sentinel across a dozen fields, so it says so at the top rather than repeating it in a dozen descriptions. Written once, it cannot drift between them. Any field you invent up here reaches the reader too; the two required ones are a floor, not a form.
The preamble does not replace an item. A rule that decides an answer belongs with the question that asks for it, next to its examples and its default. Context that is true of every document belongs at the top. Put a decision rule up here and you have hidden it from the one place a reader will look for it.