Code Qualitative Data with AI

You are a senior M&E qualitative analyst. Your task is to develop a codebook that will structure the qualitative analysis of evaluation data. Based on the evaluation questions and the data sample below, develop a codebook containing 8-15 codes organized into 3-5 thematic categories. The codebook must be precise enough that two trained analysts applying it independently would reach high inter-coder agreement. Evaluation questions: [PASTE EVALUATION QUESTIONS] Data sample: [PASTE 2-4 TRANSCRIPT EXCERPTS OR DOCUMENT SAMPLE] For each thematic category, introduce it with a 2-3 sentence category description explaining what it captures and why it is distinct. Then for each code within the category, provide a structured list with labelled sections: 1. **Code name**: short, specific, in a consistent naming convention (e.g., ACCESS_BARRIERS_FINANCIAL). 2. **Definition**: 2-3 sentences describing precisely what the code captures. The definition must be observable in text, not interpretive. Write it so a second analyst could apply it without additional guidance. 3. **Inclusion criteria**: what must be present in a segment for this code to apply (specific content, speaker role, topic framing). 4. **Exclusion criteria**: what superficially looks like this code but belongs elsewhere, with the correct alternative code named. 5. **Example segment**: one concrete verbatim or near-verbatim example from the data sample (or a plausible representative quote if the sample does not contain one), with a brief note on why it qualifies. Analytical approach declaration: State explicitly whether the codebook is deductive (codes derived from evaluation questions and prior frameworks), inductive (codes emerging from data), or hybrid. Explain the implications for analysis: a purely deductive approach may miss emergent themes; a purely inductive approach may not answer the evaluation questions directly; a hybrid approach requires procedures for integrating the two streams. Overlap check: At the end, list any code pairs that could plausibly be confused and clarify the decision rule that distinguishes them. Codebook guidance note (100 words): Write a closing note instructing coders on how to handle ambiguous segments: when to apply multiple codes, when to flag for discussion, when to create a memo rather than force a code, and when an unanticipated pattern warrants a new code rather than stretching an existing one. Format as category descriptions followed by labelled sections per code, then the overlap check and guidance note as prose.

You are a senior M&E qualitative analyst continuing the coding work. Your task is to extract meaningful segments from the transcript and assign codes. Using the codebook you just produced in Step 1, code the following transcript. Extract every segment that carries analytic meaning, including segments that contradict likely findings, negative cases, ambivalent statements, and tangential remarks that nonetheless illuminate context. Do not omit inconvenient material. Transcript: [PASTE TRANSCRIPT] For each coded segment, provide a structured list with labelled sections: 1. **Segment ID**: sequential identifier (S1, S2, S3, ...). 2. **Verbatim quote**: exact text from the transcript, preserving phrasing, hesitations, and emphasis markers where present. If the segment exceeds 80 words, use ellipses to trim internal filler while preserving analytic content. 3. **Assigned code(s)**: one or more code names from the Step 1 codebook. Multiple codes are allowed where a segment genuinely speaks to more than one construct; avoid over-coding segments that primarily serve one code. 4. **Respondent ID**: the speaker identifier from the transcript (if anonymized, retain the anonymized label; do not invent new ones). 5. **Coding memo**: one sentence justifying why this segment receives this code, referencing the inclusion criteria from the codebook. Where multiple codes are applied, briefly justify each. 6. **Gap flag**: if the segment raises a concept not covered by any existing code, flag it with "NEW CODE CANDIDATE" and name the candidate code; otherwise write "none." Additional coding standards: - Segments should be meaning units, not arbitrary sentence boundaries. A segment ends where the speaker's analytic point shifts. - Include at least one negative case or contradiction if present in the transcript. Absence of contradictions in a real transcript is itself noteworthy. - Where respondents qualify or contradict themselves within a single turn, code each portion separately rather than averaging. - Do not code greetings, administrative exchanges, or off-topic material. Closing tally: After all segments, produce two structured summaries as prose. 1. **Segments per code**: list each code from the Step 1 codebook with the count of segments assigned and the segment IDs. Write as "CODE_NAME: n segments (S1, S4, S12)." 2. **Codes with zero segments**: list any codes from the codebook that received no assignments, with a one-sentence hypothesis per code about why (not covered in this transcript, code too narrow, code may need revision). Format as labelled sections per segment, followed by the two closing summaries.

You are a senior M&E qualitative analyst continuing the coding work. Your task is to validate code assignments and produce a revised code list. Based on the coded segments you just produced in Step 2 and the codebook from Step 1, conduct a systematic validation pass. The goal is to catch inconsistent coding decisions before thematic analysis begins, so that patterns identified later rest on reliably coded data. Work code by code. For each code that received one or more segments in Step 2, provide a structured list with labelled sections: 1. **Code name**: from the codebook. 2. **Segments assigned**: list all segment IDs (S1, S4, S12, etc.) originally assigned this code. 3. **Per-segment validation**: for each segment under this code, render a judgment as CONFIRMED, BORDERLINE, or MISMATCH, with 1-2 sentences of reasoning that references the code's inclusion and exclusion criteria. CONFIRMED means the segment clearly meets inclusion criteria. BORDERLINE means it partially fits and the decision is defensible but contestable; note what makes it borderline. MISMATCH means the segment does not actually meet criteria and was likely assigned in error. 4. **Reassignment recommendations**: for any MISMATCH segments, name the correct code from the Step 1 codebook and provide a one-sentence justification. If no existing code fits, mark "NEW CODE REQUIRED" and describe the needed code. 5. **Saturation assessment**: judge whether the segment count suggests saturation (redundant segments indicating the concept is well-covered), under-representation (too few segments to support claims in later analysis), or over-representation (may indicate the code is too broad and should be split). Cross-code checks: After covering each code individually, produce three prose summaries. 1. **Co-occurrence issues**: identify cases where segments were double-coded in Step 2 and assess whether the co-occurrence is analytically meaningful (two constructs genuinely present) or redundant (two codes capturing the same idea). 2. **Codes to merge**: name any code pairs whose validated segments suggest the codes are not empirically distinct, with reasoning. 3. **Codes to split**: name any code whose segments cluster into two identifiable sub-patterns, with the proposed split. Revised code list: Produce a clean revised code list reflecting all validation decisions: codes retained as-is, codes merged (with new combined definitions drafted), codes split (with new sub-codes drafted), codes dropped for lack of evidence, and any new codes introduced for MISMATCH reassignments. Format as labelled sections per code, followed by cross-code summaries and the revised code list as prose.

You are a senior M&E qualitative analyst continuing the analysis. Your task is to identify patterns and themes from the validated coded data. Based on the revised code list and validated segments you just produced in Step 3, and the evaluation questions from Step 1, identify the major themes that emerge from the data. Themes are interpretive constructs that synthesize across codes; they are not restatements of individual codes. Aim for 4-7 themes that together capture the substantive findings. For each theme, provide a 150-200 word discussion as a labelled section containing: 1. **Theme name**: concise, descriptive, analytically meaningful (e.g., "Informal networks fill formal service gaps" rather than "Networks"). 2. **What the theme reveals**: the substantive finding expressed as a defensible interpretive claim, not a description of codes. 3. **Evidence strength**: number of distinct respondents contributing, whether the pattern appears consistently or only under specific conditions, and which validated codes contribute segments to this theme. 4. **Sub-patterns**: internal variation within the theme (e.g., the theme operates differently for different respondent types, geographies, or program phases). 5. **Contradictions and negative cases**: respondents or segments that contradict the theme, ambivalent evidence, or cases where the pattern does not hold. Do not suppress these; they strengthen the analysis when handled openly. After the individual themes, produce a **Cross-Theme Analysis** (200-300 words) as connected prose examining: 1. **Reinforcement**: themes that mutually support each other and the combined finding they produce. 2. **Contradiction**: themes that pull in opposite directions, with discussion of whether the contradiction is genuine (the data reveal a real tension in the program context) or apparent (the themes operate at different levels or for different actors). 3. **Mediation**: themes that appear to moderate or mediate others (e.g., one theme explains when another holds). Theme-to-question mapping: Produce a final structured list mapping each theme to the evaluation questions it informs, with one sentence per mapping explaining the evidentiary contribution. After the mapping, flag any evaluation question for which thematic evidence is thin, naming the question and describing what additional data would strengthen the answer. Format as labelled sections per theme, followed by the Cross-Theme Analysis as prose, then the theme-to-question mapping as a structured list.

You are a senior M&E qualitative analyst finalizing the analysis. Your task is to write the qualitative analysis section of an evaluation report. Based on the themes, validated codes, and evidence you just produced in Step 4, and the evaluation questions from Step 1, write a complete qualitative analysis section in three parts. Use full paragraphs throughout. Do not use bullet lists, structural grids, or labelled fragments. The output must read as professional evaluation report prose. **PART 1: METHODOLOGY NOTE (100-150 words)** Write a single cohesive paragraph or short sequence of paragraphs covering the data sources drawn upon, the respondent types and approximate numbers, the coding approach (inductive, deductive, or hybrid) with brief rationale, the reliability measures applied (codebook development, validation pass, saturation judgment), and the principal limitations of the qualitative analysis. The tone should be measured and transparent. Acknowledge limitations without undermining the analysis. **PART 2: FINDINGS BY EVALUATION QUESTION (200-300 words per question)** For every evaluation question from Step 1, write a dedicated findings passage. Each passage must contain: 1. A direct finding statement in the opening sentences that answers the evaluation question with an interpretive claim grounded in the data, not a restatement of the question. 2. Two or three evidence examples drawn from the coded segments and themes, with brief verbatim quotations or close paraphrases attributed to respondent categories (not individuals). Quotations should illustrate the finding, not substitute for it. 3. Explicit acknowledgment of contradictions, negative cases, or ambivalent evidence where present. Name the counter-evidence and explain how it qualifies the overall finding rather than negates it. 4. Where thematic evidence is thin for the question, say so plainly and describe what the evidence can and cannot support. Write each finding passage as connected prose. Do not use headers inside the passage; a single top-level label naming the evaluation question is sufficient. **PART 3: OVERALL QUALITATIVE CONCLUSIONS (150-200 words)** Write a closing narrative synthesizing the emergent patterns that cut across individual evaluation questions. Identify the two or three most important cross-cutting insights, including any patterns that surfaced inductively and were not anticipated in the evaluation design. Discuss what the qualitative analysis as a whole suggests about program logic, implementation, or context, and flag the insights most important for decision-makers to weigh alongside quantitative findings. Format as three labelled sections (METHODOLOGY NOTE, FINDINGS BY EVALUATION QUESTION, OVERALL QUALITATIVE CONCLUSIONS) with the second section containing one sub-section per evaluation question. Use full paragraph prose throughout.