Quantitative Findings Writer

Project Integration

project.yaml

# From project.yaml
type: quantitative  # This skill is for quantitative projects
paths:
  drafts: drafts/sections/
  tables: output/tables/
  figures: output/figures/

progress.yaml

status:
  results_draft: done
artifacts:
  results_section: drafts/sections/results-section.md

Connection to Other Skills

File Management

1. Stage and commit current state:

   git add [files] && git commit -m "quant-findings-writer: Phase N complete"

2. Then proceed with modifications.

-v2

-final

-working

Workflow

Phase 1: Orient

1. Method type: secondary-survey-analysis, administrative-data, or content-analysis
2. Key results: tables, model output, or thematic findings to present
3. Theoretical predictions: hypotheses or expectations the results address
4. Target length: typical is 12-25 paragraphs (2,000-5,000 words)

Phase 2: Select Cluster

• Survey data + model progression → Progressive Model Builder
• Admin data + quasi-experimental design → Causal Inference Specialist
• Admin data + inequality decomposition → Decomposition Analyst
• Any method + explicit hypotheses → Hypothesis Tester
• Any method + group comparisons as central question → Subgroup Comparator
• Content analysis + thematic coding → Thematic Explorer
• Panel/longitudinal + change over time → Temporal Tracker

clusters/{cluster-name}.md

Phase 3: Build the Arc

1. Map each major finding/table to a MOVE from the standardized vocabulary
2. Sequence moves following the cluster's canonical arc
3. Allocate paragraphs using the cluster's paragraph budget
4. Identify the opening and closing moves

Phase 4: Draft

techniques/techniques.md

• Table reference (58% of corpus): "Table 2 presents results from..."
• Sample description (20%): "Before turning to multivariate models, I describe..."
• Hypothesis restatement (14%): "Recall that H1 predicted..."
• Methodological setup (5%): "To estimate the causal effect, I use..."

• Lead with the finding, not the method
• Translate every key coefficient into substantive terms (85% of corpus does this)
• Use attenuation tracking when adding controls: "the coefficient falls from .34 to .21"
• Connect to theory at moderate density: ~1 theory reference per 3-4 paragraphs for most clusters
• Report null findings transparently (45% of corpus does this)

• Robustness cascade (18%): "Results are robust to..."
• Strongest finding (18%): save the most important result for the end
• Subgroup analysis (17%): end with heterogeneity
• Supplemental reference (14%): "Additional specifications in Appendix Table A3..."
• Summary (11%): brief recap of all findings

• Median section length: ~18 paragraphs, 3 tables/figures referenced
• 75% use hybrid table strategy: tables anchor the narrative but prose interprets
• 55% link results to theory heavily; 40% moderately; only 5% minimally
• Distinguish statistical from practical significance when warranted

Phase 5: Calibrate

• Does the arc match the canonical sequence?
• Is the paragraph budget balanced?
• Are tables referenced with interpretive guidance, not just pointed at?
• Is theory linking at the right density for the cluster?
• Are robustness checks present if the cluster expects them?
• Are null findings acknowledged rather than buried?

Reference Files

• Cluster guides (read the one matching the selected cluster):

  • clusters/progressive-model-builder.md

  • clusters/hypothesis-tester.md

  • clusters/decomposition-analyst.md

  • clusters/subgroup-comparator.md

  • clusters/temporal-tracker.md

  • clusters/thematic-explorer.md

  • clusters/causal-inference-specialist.md

• techniques/techniques.md

  — 20 writing techniques with descriptions and frequency data

• references/corpus-statistics.md

  — summary statistics from the 83-article analysis corpus