Prepare Coded Discourse Data for ENA
Turn ordered qualitative records into an auditable ENA table with complete unit, conversation, raw-evidence, and binary code fields.
O corpo do tutorial revisto está atualmente disponível em inglês.
Fontes do método
Tutorial completo
ENA normally starts from rows of coded interaction or other ordered records rather than from a finished network. The table must preserve enough structure for the software to decide which rows belong to each network and which rows can contribute to a shared relational context.
A clean file also keeps the qualitative evidence close. Identifiers and code columns make computation possible; speaker, line order, and raw text make interpretation and audit possible. Removing the raw evidence may produce a valid-looking network that is difficult to explain responsibly.
Caso didático
A synthetic design-review transcript
Each row represents one ordered turn from a fictional team discussing how to improve a feedback activity. The dataset includes four teams per condition and five codes: Goal, Evidence, Strategy, Tradeoff, and Revision. Several rows contain more than one code so connections can be inspected directly as well as across a moving window.
Dados de prática
Passo 1
Download and preserve the source file
Save the teaching CSV without opening and resaving it in software that may change delimiters, encodings, or identifiers. Record the filename and date. For a real project, also retain a read-only source export and a scripted or logged transformation path to the analysis file.
Ponto de verificaçãoYou can distinguish the untouched source, the analysis-ready file, and any later model export.Passo 2
Identify metadata columns
team_id identifies the network unit. condition supports grouping. discussion_round is a readable round label, while conversation_id combines team_id and discussion_round to create a boundary that is unique across the full dataset. line_number preserves order. speaker and utterance retain context. These fields should not be selected as codes. Every row needs a complete unit and conversation value, even when a spreadsheet visually repeats the same group.
Ponto de verificaçãoAll 48 rows have team_id, condition, discussion_round, conversation_id, line_number, speaker, and utterance values, and no conversation_id belongs to more than one team.Passo 3
Inspect the code matrix
The five code columns contain 1 when the coded idea is present in a row and 0 when it is absent. Multiple 1 values are allowed when the coding scheme supports multiple ideas in one segment. Do not replace missing judgments with 0 unless absence has actually been established by the coding process.
Ponto de verificaçãoEvery code cell is an explicit 0 or 1, and the difference between absent, missing, and not-applicable is documented.Passo 4
Verify order and boundaries
Sort by team_id, conversation_id, and line_number, then look for duplicates and gaps. Confirm that each conversation_id maps to one team and one discussion round. Conversation boundaries prevent moving windows from leaking across unrelated teams or events. Ordering is essential because the window uses nearby rows as its operational definition of context.
Ponto de verificaçãoEach team has one uniquely keyed six-line conversation numbered 1 through 6 with no duplicate row identifiers or cross-team conversation keys.Passo 5
Audit the codebook against excerpts
Choose at least one positive and one negative example for every code. Ask whether a second trained coder could apply the definition, whether two codes overlap by design or by ambiguity, and whether segmentation hides a meaningful connection. A binary table does not remove the need for a defensible qualitative coding process.
Ponto de verificaçãoEach code has a definition, inclusion rule, exclusion rule, and linked example outside the numeric matrix.Passo 6
Create a data dictionary and validation record
Record every column name, role, type, allowed values, and missing-data rule. Note that this dataset is synthetic and tiny. In a real analysis, include consent and de-identification decisions, coding procedures, reliability or consensus processes where appropriate, transformations, exclusions, and the software-ready file hash.
Ponto de verificaçãoAnother analyst can explain every column and reproduce the validation checks without guessing.