Science Communication

Science Communication

Translating technical data science findings for non-technical audiences. Covers audience analysis, narrative frameworks (Pyramid Principle, SCQA, AIDA), plain-language translation, executive summaries, policy briefs, causal language guidance, hedging and uncertainty communication, and accessibility standards. Complements data-scientist visualization references — handles what story to tell and to whom, not how to build charts. Use when presenting findings to stakeholders, writing executive summaries or policy briefs, communicating statistical results to non-statisticians, or reviewing a draft deliverable for clarity and audience fit.

Guidance for translating rigorous data science work into clear, compelling communication for non-technical audiences. This skill is additive to data-scientist's visualization references — it covers what story the chart tells and to whom, not how to build the chart.

Boundary with data-scientist: The data-scientist skill handles chart construction, encoding, color palettes, and export standards. This skill handles audience adaptation, narrative structure, plain-language translation, deliverable formatting, and communication quality review.

How to Use This Skill

Reference File Structure

Reading Order

1. Writing a report or brief? Start with audience-analysis.md, then narrative-frameworks.md, then deliverable-templates.md 2. Translating technical findings? Read plain-language.md first 3. Reviewing a draft? Go straight to communication-review.md 4. Concerned about equity or accessibility? Read accessibility-equity.md

Quick Decision Trees

"Who am I writing for?"

Identifying your audience?
├─ Academic researchers or peer reviewers
│   └─ ./references/audience-analysis.md (Academic section)
├─ Policymakers or legislative staff
│   └─ ./references/audience-analysis.md (Policy section)
├─ Executives or board members
│   └─ ./references/audience-analysis.md (Executive section)
├─ General public or community members
│   └─ ./references/audience-analysis.md (Public section)
├─ Journalists or media
│   └─ ./references/audience-analysis.md (Media section)
└─ Mixed or unclear audience
    └─ ./references/audience-analysis.md (Assessment checklist)

"How should I structure this?"

Choosing a narrative structure?
├─ Need to deliver a recommendation quickly
│   └─ Pyramid Principle → ./references/narrative-frameworks.md
├─ Framing a problem that needs solving
│   └─ SCQA → ./references/narrative-frameworks.md
├─ Walking through a discovery journey
│   └─ Three-Act Data Story → ./references/narrative-frameworks.md
├─ Translating findings into action
│   └─ "So What?" Framework → ./references/narrative-frameworks.md
├─ Presenting data with narrative and visuals together
│   └─ Data-Narrative-Visual Triad → ./references/narrative-frameworks.md
├─ Persuading stakeholders to act
│   └─ AIDA → ./references/narrative-frameworks.md
└─ Not sure which to use
    └─ Selection guide table → ./references/narrative-frameworks.md

"How do I say this in plain language?"

Translating technical language?
├─ Statistical jargon (p-value, confidence interval, etc.)
│   └─ Jargon translation table → ./references/plain-language.md
├─ Expressing how certain you are
│   └─ Hedging language scale → ./references/plain-language.md
├─ Using calibrated uncertainty terms
│   └─ IPCC uncertainty framework → ./references/plain-language.md
├─ Describing causal vs correlational findings
│   └─ Causal language guide → ./references/plain-language.md
├─ General readability improvement
│   └─ Reading level guidance → ./references/plain-language.md
└─ Replacing formal/bureaucratic words
    └─ Word replacement list → ./references/plain-language.md

"What format should this take?"

Choosing a deliverable format?
├─ One-page summary for decision makers
│   └─ Executive summary → ./references/deliverable-templates.md
├─ Informing policy decisions
│   └─ Policy brief → ./references/deliverable-templates.md
├─ Presenting to a room
│   └─ Stakeholder presentation → ./references/deliverable-templates.md
├─ Talking to a journalist
│   └─ Media talking points → ./references/deliverable-templates.md
└─ Full research report
    └─ (Use REPORT_TEMPLATE.md from agent_reference/)

"Is this ready to share?"

Reviewing before finalization?
├─ Comprehensive quality check
│   └─ 10-point checklist → ./references/communication-review.md
├─ Statistical interpretation errors
│   └─ Seven deadly sins → ./references/communication-review.md
├─ Common communication pitfalls
│   └─ Pitfall catalog → ./references/communication-review.md
├─ Accessibility compliance
│   └─ WCAG checklist → ./references/accessibility-equity.md
└─ Equity and framing review
    └─ Do No Harm checklist → ./references/accessibility-equity.md

The Communication Gap

Research findings fail to reach their audience not because they lack rigor, but because they lack translation. The core challenge is bridging two worlds:

The AAAS communication framework captures this as Goal → Audience → Message: define your purpose first, identify who can act on it, then craft language for that specific reader. This inverts the typical scientific structure where content comes first and audience comes last.

Core Principles

1. Audience determines everything. The same finding requires different structure, vocabulary, detail level, and uncertainty communication for each audience type. Always identify your audience before drafting.

2. Lead with the "so what." In scientific communication, conclusions come last. In stakeholder communication, the bottom line comes first. The reader's time and attention are scarce — earn continued reading with immediate relevance.

3. Translate, don't simplify. Plain language is not dumbing down. It is precise communication calibrated to the reader's vocabulary. A well-translated finding preserves its meaning while making it accessible.

4. Show uncertainty honestly. Hedging is not weakness — it is intellectual honesty. But hedge appropriately: too much hedging obscures the finding; too little overstates the evidence. Match your language to your inferential capacity.

5. Every deliverable gets reviewed. Use the 10-point communication checklist before sharing anything. Communication errors — jargon leakage, causal overclaiming, missing the "so what" — are as consequential as analytical errors.

6. Equity is not optional. How you frame data about people reflects values. People-first language, equity-aware color choices, inclusive framing, and accessible design are baseline requirements, not enhancements.

Topic Index

Citation Responsibility

When reporting standards or communication frameworks from this skill shape how findings are presented in a DAAF report (causal language guidelines, uncertainty frameworks, equity-sensitive visualization principles), the report-writer includes the relevant citation in the report's Reporting Standards references section.

Each citation must include a brief rationale explaining why it is included.

For the master citation index and inclusion thresholds, consult agent_reference/CITATION_REFERENCE.md.

Accessibility and Equity in Data Communication

Standards and guidance for ensuring that data communication is accessible to all audiences and equitable in how it frames findings about people. This reference covers two intertwined concerns: accessibility (can everyone perceive and understand the communication?) and equity (does the communication respect and accurately represent the people in the data?).

This file focuses on communication-level accessibility and equity framing. For chart construction standards (color palettes, encoding hierarchy, export specifications), see the data-scientist visualization references. For audience-specific adaptation, see audience-analysis.md.

WCAG Compliance for Data Communication

Web Content Accessibility Guidelines (WCAG) 2.1 establish the baseline standards for accessible data presentation. Even for non-web deliverables (PDFs, presentations, printed reports), WCAG principles provide a useful framework.

Contrast Requirements

Practical tips:

• Adjacent bars in a bar chart each need 3:1 contrast against the background; separating bars with white space simplifies compliance
• Pie chart segments can use a contrasting border between adjacent slices
• Light gray gridlines on white backgrounds often fail contrast requirements — check with a contrast checker tool

Non-Text Content (WCAG 1.1.1)

Every chart or data visualization must have a text alternative. The approach depends on complexity:

Writing effective alt text for charts:

• State what the chart shows AND what the key finding is
• Not: "Bar chart of graduation rates" (says the type but not the finding)
• Better: "Bar chart showing graduation rates by selectivity tier; highly selective institutions graduate 89% vs. 42% at open-access institutions"
• For published documents, the UK Government Analysis Function recommends marking the chart image as decorative (empty alt) and placing the meaningful description in adjacent body text — this serves both screen reader users and sighted readers

Color-Alone Encoding Prohibition (WCAG 1.4.1)

Color must never be the sole means of conveying information. Every use of color must be paired with at least one redundant encoding:

The golden rule: Never rely on color alone. Always pair color with shape, pattern, line style, direct label, or position.

Multi-Format Presentation Strategy

The evidence-based approach is to present data through multiple concurrent formats:

1. Visual chart — for at-a-glance pattern recognition 2. Prose summary — immediately adjacent, stating the headline finding in plain language 3. Data table — for precise values, screen readers, and independent verification 4. Accessible download — Excel/ODS file for further analysis (when applicable)

This strategy simultaneously satisfies WCAG requirements, serves diverse user needs, and aligns with the IES recommendation to "lead with the headline" in both visual and text form.

Screen Reader Considerations

When data tables accompany charts:

• Use semantic HTML: <table>, <th> with scope attributes, <caption> elements
• Ensure header-to-cell relationships are programmatically determinable
• For downloadable tables, provide Excel or ODS files with clear link text stating file type and content
• SVG format for chart images retains clarity at high zoom levels and can embed accessible text

Accessible Chart Type Selection

Based on the UK Government Analysis Function checklist:

---

People-First Language

Default to language that foregrounds personhood, then describes characteristics. This is not political correctness — it is precision. "Students experiencing poverty" conveys something different from "poor students": the former describes a circumstance; the latter defines an identity.

General Principles

Nuance: Identity-First Language

People-first language is the default, but some communities prefer identity-first language. Many autistic adults prefer "autistic person" over "person with autism." Many Deaf community members prefer "Deaf" as an identity marker. The consistent principle across all groups: talk with the communities you are studying to understand their preferred terminology, rather than imposing a universal rule.

Evolving Terminology

Language evolves continuously. Some notes on current usage in research contexts:

• Hispanic/Latino/Latina/Latinx/Latine: A 2020 Pew survey found only 3% of Hispanic/Latino adults use "Latinx." "Latine" is an emerging alternative. When survey data use one term, you may use updated language in your communication if explicitly noted. When uncertain, use the term the community uses most widely in your context.
• BIPOC: Useful as shorthand but can obscure meaningful differences between the groups it aggregates. Prefer specific group names when the analysis supports disaggregation.
• Gender language: Many major surveys still offer only "male/female." When such groups are absent from data, call it out explicitly: "This survey did not offer nonbinary response options."

---

Urban Institute Do No Harm Principles

The Urban Institute's "Do No Harm Guide: Applying Equity Awareness in Data Visualization" (Schwabish & Feng, 2021) provides a comprehensive framework for equity-aware data communication. Its central diagnostic question is: "If I were one of the data points on this visualization, would I feel offended?"

Core Principles

1. Put people first. Data reflect real lives. As Jacob Harris wrote: "If your data is about people, make it extremely clear who they are or were."

2. Critically examine the data before visualizing. Applying equity thinking to a visualization will not fix inherently biased underlying data. Ask: How were these data generated? Who is included and excluded? Whose voices are missing?

3. Name forces of oppression directly. Titles and annotations are prime real estate for naming racism, sexism, and structural discrimination — don't bury these forces in body text or leave them for the reader to infer.

4. Frame around systems, not deficits. "Black students are suspended at 3x the rate" locates the problem in Black students. "Schools suspend Black students at 3x the rate" locates the problem in the system. The second framing is almost always more accurate.

5. Question default comparison groups. Major federal surveys list "white" as the first response option and code it as "1." This causes analysts to default to white as the comparison baseline — not by active choice, but by data structure. Consider: if your study focuses on a particular community, present that group first. Sort by the variable of interest, not by dataset order.

6. Be thoughtful about aggregation. Combining racial or ethnic subgroups into broad categories can mask critical variation. Aggregated "Asian or Pacific Islander" poverty rates range from 4.5% to 27.8% across detailed subgroups. When you must aggregate, note what is hidden.

7. Acknowledge absent groups. When data exist for a group but sample sizes are too small, say so explicitly rather than silently omitting the group. Transparency about absence is better than invisible exclusion.

Stereotype-Aware Color Choices

Ordering Data Purposefully

The order in which demographic groups appear in charts and tables communicates values:

• Do not default to dataset order. Raw data ordering inherits historical biases (white first, male first).
• Consider your study's focus. If your analysis centers a particular community, present that group first.
• Sort by the variable of interest when there is a natural ordering (e.g., by magnitude of the result).
• Alphabetical order is a neutral default when no other ordering is justified.
• Interactive selection (when the medium allows it) lets readers choose their own comparison group, avoiding any static hierarchy.

Titles as Equity Instruments

Titles are among the first things readers scan. They present an opportunity to frame the narrative:

The systems framing is almost always more accurate — outcomes are produced by systems, not by the characteristics of affected people.

Do No Harm Checklist

Before finalizing any deliverable containing demographic data:

• [ ] Has the team communicated with the people/communities represented in the data?
• [ ] Does the team understand the data's sourcing, inclusion/exclusion criteria, and who benefits or is harmed?
• [ ] Have words, phrases, and labels been carefully reviewed for people-first language?
• [ ] Have colors been chosen to avoid stereotypes and meet accessibility standards?
• [ ] Has the ordering of groups in charts and tables been considered deliberately (not defaulting to dataset order)?
• [ ] Have alternatives to "Other" been considered? ("Other" literally others individuals)
• [ ] Have all icons and images been reviewed with an equity lens?
• [ ] Would alternative chart types better serve the data and the communities represented?
• [ ] Does the framing locate problems in systems rather than in affected communities?
• [ ] Does the final product meet the needs of its audience, including members of the communities studied?

---

Colorblind-Safe Communication Design

Approximately 8% of males and 0.5% of females have some form of color vision deficiency. The data-scientist visualization references cover color palette selection and redundant encoding in detail. This section covers the communication-level considerations that go beyond chart construction.

Communication-Level Principles

1. Never describe findings using color alone. "The red line shows..." fails if the reader cannot distinguish red. Use: "The dashed line (labeled 'District A') shows..."

2. Reference chart elements by label or position, not color. In narrative text accompanying a chart: "The top bar (high-poverty districts) shows..." not "The dark blue bar shows..."

3. Test all deliverables in grayscale. Print or display in black and white. If any information is lost, the design relies too heavily on hue.

4. Provide data tables alongside every chart. Tables are inherently colorblind-accessible — the information is in the numbers, not the colors.

5. In presentations, describe visual elements verbally. "You can see in the chart that the solid line — representing literacy programs — rises steeply while the dashed line — class-size reduction — remains flat." This also benefits audience members with visual impairments.

---

Comprehensive Accessibility Review Checklist

Apply before releasing any deliverable:

Content Accessibility

• [ ] All charts have text alternatives (alt text or adjacent prose descriptions)
• [ ] Data tables accompany complex visualizations
• [ ] Text contrast meets 4.5:1 minimum (standard text) or 3:1 (large text)
• [ ] No information is conveyed by color alone — redundant encoding is present
• [ ] Chart elements are referenced by label/position in narrative text, not by color
• [ ] Font is sans-serif, minimum 12pt (print) or 11px (screen)
• [ ] Reading level is appropriate for the target audience

Structural Accessibility

• [ ] Document has proper heading hierarchy (H1 → H2 → H3, no skipped levels)
• [ ] Tables use proper header markup
• [ ] Links have descriptive text ("View the full dataset" not "click here")
• [ ] Document is navigable by keyboard (for digital deliverables)

Equity Review

• [ ] People-first language used throughout
• [ ] Demographic groups named respectfully and specifically
• [ ] Default comparison group considered deliberately
• [ ] Aggregation does not mask meaningful subgroup variation
• [ ] Absent groups acknowledged rather than silently excluded
• [ ] Framing locates problems in systems, not in affected communities
• [ ] Color choices avoid demographic stereotypes
• [ ] Group ordering reflects the analysis purpose, not dataset defaults

References and Further Reading

Urban Institute. (2021). "Do No Harm Guide: Applying Equity Awareness in Data Visualization." https://www.urban.org/research/publication/do-no-harm-guide-applying-equity-awareness-data-visualization

Urban Institute. "Data Visualization Style Guide." https://urbaninstitute.github.io/graphics-styleguide/

WCAG 2.1. W3C Recommendation. https://www.w3.org/TR/WCAG21/

UK Government Analysis Function. "Accessible Charts: A Checklist of the Basics." https://analysisfunction.civilservice.gov.uk/policy-store/charts-a-checklist/

Elavsky, F. "Chartability." https://chartability.fizz.studio/

Schwabish, J. (2021). Better Data Visualizations: A Guide for Scholars, Researchers, and Wonks. Columbia University Press.

Tufte, E.R. (1983). The Visual Display of Quantitative Information. Graphics Press.

D'Ignazio, C. and Klein, L.F. (2020). Data Feminism. MIT Press. https://datafeminism.io/

Audience Analysis

How to adapt your communication strategy based on who is reading. Every dimension of your deliverable — structure, vocabulary, detail level, uncertainty communication, visuals, and length — changes based on the audience.

The AAAS communication framework puts it simply: Goal → Audience → Message. Define your purpose, identify who can act on it, then craft language for that specific reader. Never write first and identify the audience second.

The Five Audience Types

1. Academic Researchers

Peers and reviewers who share your methodological vocabulary and expect full technical transparency.

What makes academic communication distinct: Comprehensiveness is valued over conciseness. The reader is assessing your methodology as much as your findings. Every analytical choice needs justification.

---

2. Policymakers and Legislative Staff

Decision makers working under time constraints who need evidence translated into actionable implications. They often have staff who brief them, so your materials may be read by intermediaries with varying technical backgrounds.

What makes policy communication distinct: Policymakers are buying decision-relevant evidence, not evaluating your methodology. They need to know: What did you find? How confident should I be? What does it mean for my decision? The IES framework emphasizes: "summarize, don't extrapolate" — present what the evidence shows without overstepping into prescription.

Preparing for policymaker engagement (AAAS):

• Understand where in the policy cycle your evidence is relevant (agenda setting, option identification, implementation, evaluation)
• Define your "ask" before the meeting — what action are you requesting?
• Be prepared for conversations of only a few minutes
• Identify overlapping interests between your work and their agenda

---

3. Executives and Board Members

Senior leaders making resource allocation decisions. They want strategic implications, not analytical details. Their question is always: "What should we do differently?"

What makes executive communication distinct: Executives optimize for decision speed, not comprehension depth. They will not read beyond the first page unless you have earned their interest. The Pyramid Principle (see narrative-frameworks.md) was designed for this audience: lead with the answer, then provide supporting logic only as deep as needed.

---

4. General Public and Community Members

Diverse audiences with varying educational backgrounds, lived experiences, and reasons for engaging with your findings. Includes community stakeholders, parents, local advocacy groups, and interested citizens.

What makes public communication distinct: The public audience is the most heterogeneous — they have no shared technical vocabulary and no obligation to keep reading. You must earn their attention with immediate relevance and maintain it with clarity. NIH's research shows that clear, simple communication benefits all readers, including experts — simplicity is a feature, not a concession.

Critical framing consideration: People may see themselves in your data. If your findings are about communities, schools, or demographic groups, readers may be members of those groups. Communicate with empathy and respect. See accessibility-equity.md for equity-aware framing guidance.

---

5. Journalists and Media

Intermediaries who translate your work for their own audiences. Journalists work to tight deadlines, write in their own style, and aim to inform rather than educate. You cannot control the final product — you can only provide the best raw material.

What makes media communication distinct: You are providing raw material, not a finished product. Journalists will select, condense, and reframe. Your job is to make the accurate version of the story the easiest one to tell. The AAAS recommends building ongoing relationships with journalists (not just responding reactively) and connecting with institutional communications staff who can facilitate.

Pre-interview preparation (AAAS checklist):

• Who is the journalist's audience and what do they value?
• What is your three-point core message?
• Why should their specific audience care?
• What are the most common misinterpretations you want to prevent?

---

Tone and Register by Audience

The same content changes not just in vocabulary and structure but in tone — the emotional register that signals respect, competence, and appropriate relationship with the reader.

The key rule: Your tone should signal respect for the audience's intelligence while meeting them at their vocabulary level. A warm tone for the public is not condescending; a formal tone for academics is not cold. Match the register to the relationship, not just the content.

What Changes Across Audience Levels

This comparison table highlights the dimensions that shift most dramatically across audiences:

Audience Assessment Checklist

Before drafting any deliverable, answer these questions:

• [ ] Who is the primary audience? (Name the specific type, not "stakeholders")
• [ ] What is their existing knowledge level? (Can they interpret a regression table? A bar chart? A percentage?)
• [ ] What decision or action does this communication support? (If none, reconsider whether the deliverable is needed)
• [ ] How much time will they spend with this? (This determines your length and structure)
• [ ] What is their vocabulary? (Use their words, not yours)
• [ ] What do they already believe about this topic? (Align your framing with or explicitly address their priors)
• [ ] Will they see themselves in this data? (If yes, review equity framing — see accessibility-equity.md)
• [ ] Is there a secondary audience? (A policy brief may be read by both the policymaker and their staff, who have different knowledge levels)
• [ ] What is the delivery format? (Print, screen, presentation, conversation — each has different constraints)
• [ ] What is the single most important takeaway? (If you can't state it in one sentence, you're not ready to write)

Mixed Audiences

When your audience is heterogeneous (e.g., a presentation to a room with both technical and non-technical attendees), layer your communication:

1. Lead at the lowest common denominator — state findings in plain language first 2. Provide depth on demand — "For those interested in the methodology..." or appendices for technical readers 3. Use the "newspaper test" — would the headline version of your finding be accurate? Start there, then add nuance 4. Label technical content explicitly — "Technical note:" or "For researchers:" signals that non-technical readers can skip ahead

References and Further Reading

AAAS. "Communication Toolkit." https://www.aaas.org/resources/communication-toolkit

Dykes, B. (2019). Effective Data Storytelling: How to Drive Change with Data, Narrative, and Visuals. Wiley.

IES. "Six Strategies for Effectively Communicating Research Findings to Decision Makers." https://ies.ed.gov/blogs/research/post/six-strategies-for-effectively-communicating-research-findings-to-decision-makers

Nussbaumer Knaflic, C. (2015). Storytelling with Data: A Data Visualization Guide for Business Professionals. Wiley.

NIH. "Clear & Simple: Developing Effective Print Materials for Low-Literate Readers." https://www.nih.gov/institutes-nih/nih-office-director/office-communications-public-liaison/clear-communication/clear-simple

Communication Review

A systematic review framework for checking the quality of any data communication deliverable before finalization. Apply the 10-point checklist to every deliverable, then scan for the common pitfalls and statistical misinterpretation sins cataloged below.

10-Point Communication Review Checklist

Apply to every deliverable — executive summary, policy brief, presentation, report, or talking points — before sharing with any audience.

The Checklist

How to Use the Checklist

1. Complete drafting first. Don't apply the checklist during writing — it interrupts flow. Write the full draft, then review. 2. Check each item independently. Don't let a strong "so what?" compensate for inappropriate causal language. 3. Mark failures explicitly. If any item fails, fix it before sharing. No item is optional. 4. Have someone else apply it if possible. Self-review is limited by the curse of knowledge (see pitfalls below).

---

Seven Deadly Sins of Statistical Misinterpretation

From Louis & Chapman (2017), adapted for data communication review. Each sin represents a way that statistical findings can be misrepresented or misunderstood in communication.

Sin 1: Treating Small Differences as Meaningful

The error: Reporting a difference as noteworthy when it falls within the margin of error or random variation. A 1-point polling lead within a ±3-point margin tells you nothing.

In communication: "Group A scored higher than Group B" when the difference is 0.3 points on a 100-point scale.

Prevention: Always contextualize differences with their uncertainty. Ask: "Is this difference larger than the noise?" If you cannot report a confidence interval, at minimum note whether the difference is "meaningful" vs. "within the range of normal variation."

Sin 2: Equating Statistical Significance with Practical Significance

The error: Conflating "statistically significant" (unlikely due to chance) with "meaningful" or "important." A statistically significant effect can be trivially small.

In communication: "We found a significant relationship between X and Y" without reporting effect size or practical magnitude.

Prevention: Always report the size of the effect, not just its existence. "The program improved test scores by 2 percentile points (statistically significant)" lets the reader judge whether 2 points matters for their context. For non-technical audiences, drop the significance language entirely and just state the effect size with appropriate hedging.

Sin 3: Ignoring the Tails

The error: Focusing only on group averages when the distribution matters. Two groups with identical averages can have dramatically different distributions — and the extremes are often the most policy-relevant.

In communication: "The average student performs at grade level" when 25% of students are two or more years below grade level.

Prevention: When averages tell an incomplete story, show the distribution. Report percentiles, ranges, or proportions above/below thresholds. "The average masks considerable variation: while most students are at or above grade level, 1 in 4 are two or more years behind."

Sin 4: Trusting Spurious Correlations

The error: Reporting a correlation found through extensive data mining as if it were a meaningful relationship. With enough variables, spurious correlations are guaranteed.

In communication: Presenting a data-mined association without noting that it was discovered post hoc or that many comparisons were tested.

Prevention: Distinguish pre-registered hypotheses from exploratory findings. If a result was discovered during analysis (not predicted beforehand), label it as "exploratory" and note that it requires replication. For non-technical audiences: "We found an interesting pattern, but it needs to be confirmed with additional data."

Sin 5: Getting Causation Backwards

The error: Assuming the direction of causation without considering reverse causality. Does unemployment cause depression, or does depression increase unemployment risk?

In communication: "Factor X leads to outcome Y" when the arrow could run in either direction.

Prevention: For every association, explicitly consider: could the outcome be causing the exposure? If you can't rule it out, use bidirectionally neutral language: "X and Y are associated" rather than "X leads to Y." See plain-language.md for causal language guidance.

Sin 6: Forgetting Confounders

The error: Presenting an association between two variables without acknowledging that a third variable (a confounder) might explain both. Eating more vegetables "correlates with" better health — but income drives both vegetable consumption and health outcomes.

In communication: "People who do X have better outcomes" without noting that people who do X may differ systematically from people who don't.

Prevention: State what you controlled for and what you couldn't control for. "After accounting for income, education, and age, the association between X and Y remained — though we cannot rule out other unmeasured factors."

Sin 7: Misleading with Visualization

The error: Using visual design choices (truncated axes, distorted scales, cherry-picked time windows) to exaggerate or minimize findings.

In communication: A bar chart with a y-axis starting at 50 instead of 0, making a 3-point difference look dramatic.

Prevention: Follow the integrity checklist in data-scientist visualization references. Bar charts start at zero. Axes are labeled. Scales are consistent across comparison panels. The visual impression matches the actual magnitude of the finding.

---

Additional Communication Pitfalls

Beyond statistical misinterpretation, these pitfalls are specific to the act of communicating findings to non-technical audiences.

Jargon Leakage

The problem: Technical terms slip into non-technical communication unnoticed by the writer, who is so familiar with the terms that they no longer register as jargon.

Examples: "We controlled for confounders" / "The regression coefficient was..." / "After propensity score matching..." in a policy brief.

Prevention: Read the deliverable aloud to someone outside your field. Every term they stumble on is jargon that needs replacement. Consult the jargon translation table in plain-language.md.

Curse of Knowledge

The problem: Once you understand something, you cannot easily imagine not understanding it. You unconsciously assume the reader shares your background, your context, and your mental model.

Examples: Skipping explanation of a concept that feels "obvious" to you but is unfamiliar to the reader. Using acronyms without definition. Presenting a chart that requires domain knowledge to interpret.

Prevention: Write for the least knowledgeable member of your target audience. Define acronyms on first use. Ask: "Would I understand this if I hadn't done this analysis?" Have someone outside the project review the draft.

Over-Caveating

The problem: Listing every limitation until the reader concludes the findings are worthless. Researchers are trained to enumerate limitations exhaustively; non-technical audiences interpret extensive caveats as "this study is unreliable."

Examples: Three paragraphs of limitations after two paragraphs of findings. "However, it should be noted that... Furthermore, it is important to caveat that... Additionally, one must consider that..."

Prevention: Apply the IES "too important to exclude" filter: include only the limitations that could meaningfully change the interpretation. Frame limitations as open questions: "We don't yet know whether..." rather than "A limitation of this study is..."

Under-Caveating

The problem: The opposite of over-caveating — omitting important limitations to make findings appear more definitive than they are. This is an integrity issue.

Examples: Presenting observational findings as if they were causal. Omitting that results only apply to a specific subpopulation. Not mentioning that a key variable was unavailable.

Prevention: Include the limitations that would change a reader's interpretation if they knew about them. The test: "Would an informed critic point out something I haven't mentioned?" If yes, mention it.

Schrödinger's Causal Inference

The problem: Using cautious associational language in findings but implying causation in recommendations. Per Haber et al. (2022), 44.5% of published papers with action recommendations implied stronger causality in recommendations than in findings.

Examples: Findings: "X is associated with Y." Conclusion: "Policymakers should adopt X to improve Y." The causal claim exists and doesn't exist depending on which section you read.

Prevention: The strength of language in recommendations must not exceed the strength of language in findings. If you found an "association," your recommendation cannot assume a causal mechanism. See plain-language.md for the full causal language framework.

Numerical Overload

The problem: Presenting too many numbers too quickly, overwhelming the reader's working memory. Research shows most people can hold 3-4 items in working memory simultaneously.

Examples: "The coefficient was 0.34 (SE = 0.08, p < 0.001, 95% CI [0.18, 0.50]), indicating that a one-standard-deviation increase in X is associated with a 0.34-standard-deviation increase in Y, which corresponds to roughly 12 percentile points (N = 4,523)."

Prevention: Present one number at a time. Lead with the most intuitive number ("about 12 percentile points"). Put technical details in footnotes or appendices. For non-technical audiences, round aggressively: "about 12 points" not "11.7 percentile points."

Missing the "So What"

The problem: Reporting findings without explaining their significance. The reader is left with data but no meaning.

Examples: "Students in District A scored 14 points higher than students in District B." (So what? Is that a lot? What should someone do about it?)

Prevention: Apply the "So What?" framework (see narrative-frameworks.md) to every finding. Finding → Implication → Action. "Students in District A scored 14 points higher — roughly one full grade level. This suggests that District A's literacy program model may be worth replicating."

Confusing Verbal Probability

The problem: Using probability words that mean different things to different people. "Likely" means ~70% to most laypeople but 66-100% in the IPCC framework. "Rare" could mean 1% or 10% depending on context.

Examples: "Adverse effects are unlikely" — does this mean 5%? 20%? 33%?

Prevention: Pair verbal probability with concrete context: "Adverse effects are uncommon — in our study, fewer than 3% of participants experienced any side effect." See the IPCC calibrated uncertainty framework in plain-language.md for a systematic approach.

Deficit Model Thinking

The problem: Assuming that the audience's failure to adopt your conclusion stems from their lack of information, rather than from legitimate differences in values, priorities, or interpretation. "If they just understood the data, they'd agree with us."

Examples: Responding to audience skepticism by presenting more data instead of engaging with their concerns. Framing public disagreement as "science illiteracy."

Prevention: Recognize that evidence is one input to decisions, not the only input. People can understand your findings and still reach different conclusions based on different values. Communicate findings clearly and let the audience draw their own conclusions. Engage with concerns rather than repeating the data louder.

Equity-Blind Communication

The problem: Presenting data about people without considering how the framing, language, and defaults affect those people. Using demographic groups as decoration rather than treating them as audiences who will encounter the work.

Examples: Using "minorities" when you mean "communities of color." Defaulting to white/male as the comparison group. Framing achievement gaps as deficits of one group rather than as systemic inequities.

Prevention: See accessibility-equity.md for comprehensive guidance on people-first language, equity-aware framing, and the Urban Institute Do No Harm principles.

---

Quick Pre-Release Review Protocol

Before sharing any deliverable, spend 10 minutes on this rapid review:

1. Read only the first paragraph and the last paragraph. Do they tell a coherent story? Is the bottom line clear in both places? 2. Scan every verb connecting a cause to an effect. Is the causal strength appropriate? (Checklist item 4) 3. Count the caveats. Are there too many (>3 in a brief) or too few (0 in a brief)? 4. Read chart titles in isolation. Do they state findings? Could someone understand the story from titles alone? 5. Search for jargon. Ctrl+F for technical terms from the jargon translation table. Replace any that survived. 6. Read the recommendations. Do they exceed the inferential capacity of the findings? 7. Check the "so what." After the last finding, ask yourself: "Would the reader know what to do next?"

References and Further Reading

Haber, N.A. et al. (2022). "Causal and Associational Language in Observational Health Research." American Journal of Epidemiology, 191(12), 2084-2097.

Louis, W. and Chapman, C. (2017). "The Seven Deadly Sins of Statistical Misinterpretation, and How to Avoid Them." The Conversation. https://theconversation.com/the-seven-deadly-sins-of-statistical-misinterpretation-and-how-to-avoid-them-74306

IES. "Six Strategies for Effectively Communicating Research Findings to Decision Makers." https://ies.ed.gov/blogs/research/post/six-strategies-for-effectively-communicating-research-findings-to-decision-makers

Cairo, A. (2016). The Truthful Art: Data, Charts, and Maps for Communication. New Riders.

Tufte, E.R. (1983). The Visual Display of Quantitative Information. Graphics Press.

Deliverable Templates

Structural outlines, rules, and example skeletons for four core communication deliverable types. Each template is designed for a specific audience and purpose — choose based on who will read it and what decision it supports.

For audience-specific strategy, see audience-analysis.md. For narrative structure choices, see narrative-frameworks.md.

1. Executive Summary

Audience: Executives, board members, senior decision makers Purpose: Deliver the bottom line and its implications in the shortest possible format Framework: Pyramid Principle (answer first, then supporting logic)

Structural Outline

Rules

1. Maximum 10% of full document length. A 20-page report gets a 2-page executive summary. A 5-page brief gets a half-page summary. 2. Self-contained. The reader should never need to consult the full report to understand the executive summary. Every claim must be interpretable without additional context. 3. No new information. Everything in the executive summary must appear in the full document. The summary distills — it does not introduce. 4. One page is ideal. If it goes to two pages, it's not executive enough. Cut ruthlessly. 5. Declarative titles on any embedded charts. "Rural Districts Lost Teachers Fastest" not "Teacher Attrition by District Type." 6. No methodology in the body. The methodology note is a courtesy sentence, not a methods section. "We analyzed enrollment data for 13,000 districts over 8 years" is sufficient.

Example Skeleton

Bottom line: Early literacy interventions should replace class-size reduction as the district's primary reading improvement strategy, based on evidence showing 2-3x larger effects at 40% lower cost per student.

>

Key findings:

- Schools piloting structured literacy programs improved reading proficiency by 14 percentage points; schools with class-size reduction improved by 5 points

- Cost per student served is $1,200 for literacy programs vs. $2,000 for class-size reduction

- Effects are largest for students who entered below grade level — the highest-need population benefits most

>

Based on: Analysis of 200 district reading programs (2019-2025), controlling for student demographics, baseline performance, and prior funding levels.

>

Implications:

- Current $14M class-size investment is producing below-benchmark returns

- Redirecting $8M to literacy programs could serve 2x more students at higher effectiveness

- Implementation timeline is 6-12 months vs. 2-3 years for further class-size changes

>

Caveat: These findings are based primarily on elementary schools; middle school effects are less certain.

>

Next steps:

- Approve $2M pilot in 6 schools for 2026-27 (Board action needed by April)

- Commission independent evaluation design by June

- Report interim results to the Board in January 2027

---

2. Policy Brief

Audience: Policymakers, legislative staff, agency officials Purpose: Inform a policy decision with evidence without prescribing a specific policy choice Framework: SCQA (establish the problem, then present the evidence)

Structural Outline

Rules

1. 2-4 pages maximum. Policy briefs that exceed 4 pages lose their audience. If you need more space, write a report and attach a 1-page summary. 2. No jargon in the body. Every technical term must be translated. See plain-language.md. 3. Present alternatives, not prescriptions. The researcher's role is to present evidence and illuminate tradeoffs. The policymaker's role is to decide. "Based on our analysis, three approaches merit consideration..." not "We recommend..." 4. Lead each paragraph with the finding in plain language. Follow IES Strategy 2: the first sentence of each evidence paragraph states the insight without numbers. Numbers come in the supporting sentences. 5. Include only the most essential caveats — those "too important to exclude" (IES Strategy 4). Over-caveating undermines confidence without serving the reader. 6. Every figure must be self-interpreting. Declarative title, clear labels, source note. A policymaker should understand the chart without reading surrounding text. 7. Summarize, don't extrapolate. Per IES Strategy 6: present what the evidence shows and its implications, not explicit policy recommendations.

Example Skeleton

Title: Early Literacy Programs Outperform Class-Size Reduction for Reading Improvement

>

Key Takeaways:

- Structured literacy programs improved reading scores 2-3 times more than class-size reduction

- Literacy programs cost 40% less per student served

- Benefits were largest for students entering below grade level

- Evidence is based on 200 districts over 6 years; results were consistent across diverse contexts

>

Context: Since 2005, states have invested heavily in reducing class sizes as a strategy for improving student outcomes. [Situation.] However, recent evidence suggests that how instructional time is used may matter more than how many students share the classroom. [Complication.]

>

What the evidence shows:

Schools using structured literacy programs saw substantially larger reading gains than those relying on class-size reduction alone. [Headline — no numbers.] Specifically, literacy program schools improved proficiency rates by 14 percentage points compared to 5 points for class-size reduction schools, after accounting for student demographics and baseline performance. [Supporting evidence.]

>

The benefits were not uniform across all students. [Headline.] Students who entered below grade level showed the largest gains — 18 percentage points in literacy programs versus 4 points with smaller classes. [Supporting evidence.]

>

Policy alternatives:

- Option A: Full reallocation — redirect class-size funds to literacy programs...

- Option B: Phased transition — pilot literacy programs alongside existing class-size commitments...

- Option C: Targeted deployment — maintain class-size investments in early grades, redirect in upper elementary...

>

What we don't know: The evidence does not yet tell us whether these effects persist beyond elementary school, nor whether the same programs work equally well in rural and urban settings.

---

3. Stakeholder Presentation

Audience: Mixed groups (board members, managers, practitioners, community members) Purpose: Build shared understanding and move toward a decision or next steps Framework: Three-Act Data Story or AIDA, depending on whether the goal is understanding or buy-in

Structural Outline

Rules

1. One message per slide. If you need two charts, use two slides. 2. Slide titles are declarative sentences that state the finding. "Rural Districts Lost Teachers Fastest" — not "Teacher Attrition Data." The title should tell the story even if the audience doesn't look at the chart. 3. Maximum 3-5 bullets per slide, maximum 6-8 words per bullet. Slides are visual aids, not documents. If you're reading your slides, you've written a document. 4. Presenter notes, not slide text. The detail goes in your spoken remarks, not on the slide. The audience should be listening to you, not reading ahead of you. 5. Design for the back row. Minimum 24pt for body text, 32pt for titles. If they can't read it from the back of the room, cut it. 6. Consistent visual language. Same color means the same thing across all slides. Same chart style throughout. See data-scientist visualization references for chart construction standards. 7. Build in pauses for questions. After key findings slides, pause. "Does this match your experience?" invites engagement and surfaces misunderstanding early. 8. End with a specific ask or clear next step. Never end with "Questions?" without first stating what you need from the audience.

Example Skeleton

Slide 1 (Hook): "Our highest-need students are falling further behind — but three schools found a way to reverse the trend." [Photo or simple chart showing the gap]

>

Slide 2 (Context): The district invested $14M in class-size reduction since 2019. Reading scores have remained flat. [Simple trend line]

>

Slide 3 (Finding 1): "Literacy programs produced 3x the reading gains of class-size reduction." [Paired bar chart, two bars]

>

Slide 4 (Finding 2): "The biggest gains went to students who needed them most." [Chart showing effect by baseline level]

>

Slide 5 (Finding 3): "Literacy programs cost 40% less per student." [Cost comparison chart]

>

Slide 6 (What it means): Redirecting resources could serve twice as many students at higher effectiveness.

>

Slide 7 (Recommendation): Proposed pilot: 6 schools, $2M, starting fall 2026. Board approval needed by April.

>

Slide 8 (Discussion): "What would you need to see from the pilot to support district-wide expansion?"

---

4. Media Talking Points

Audience: Journalists and through them, the general public Purpose: Equip journalists to tell an accurate story about your findings Framework: SCQA (compressed) — the hook and the answer

Structural Outline

Rules

1. Prepare for 1-2 quoted sentences. Even from a 30-minute interview, most stories use 1-2 direct quotes. Make those quotes count. Practice saying them out loud until they're natural. 2. No jargon. Not even "statistically significant." Journalists will either skip the term or misdefine it. 3. Provide the headline. If you don't give the journalist a clear headline, they'll write their own — and it may not be the most accurate version of your finding. 4. Prepare three tiers of depth. (1) The headline in one sentence. (2) Supporting context in 2-3 sentences. (3) Full methodology on request. Let the journalist choose their depth. 5. Proactively state limitations. If you don't provide the caveat, it likely won't appear. But frame it constructively: "This study focused on elementary schools — we're now extending to middle schools" rather than "A limitation of this study is..." 6. Analogies travel further than numbers. "That's the equivalent of losing one full year of learning" is more quotable than "0.8 standard deviations below the mean." 7. Never say "no comment." If you can't answer, explain why and offer to follow up.

Example Skeleton

Headline finding: "Schools that switched from smaller classes to structured reading programs saw three times the improvement in student reading scores."

>

Why it matters: "Millions of dollars are being spent on a strategy that produces smaller results than a cheaper alternative — and the students who need the most help are benefiting the least from the current approach."

>

Supporting context: "We studied 200 school districts over six years. Districts that adopted early literacy programs saw reading proficiency rise by 14 percentage points. Districts that focused on reducing class sizes saw only a 5-point improvement. The difference was even larger for students who started behind."

>

Prepared analogy: "The gap between the two approaches is like the difference between walking and driving — both get you there, but one is dramatically faster, and the students with the longest journey benefit the most from the faster route."

>

Proactive caveat: "This evidence is strongest for elementary schools. We're now investigating whether the same pattern holds in middle schools."

>

Quotable hedged statement: "Our analysis provides strong evidence that how instructional time is used matters more than class size for reading improvement — and that the students most in need benefit the most from structured literacy approaches."

---

How to Distill: Turning Long Reports into Short Deliverables

One of the most common tasks: "I have a 15-page report and need a 1-page executive summary." The deliverable templates above describe the output format. This section describes the process of distillation.

The Distillation Process

1. Read the report's conclusion first. The bottom line is usually buried at the end. Extract it — this becomes the opening sentence of your short deliverable.

2. Identify the 3 most important findings. Not the 3 most interesting, not the 3 you worked hardest on — the 3 that most directly answer the original research question or inform the audience's decision. If you can't choose 3, ask: "If the reader remembers only one finding, which one?"

3. For each finding, extract: the claim, the evidence, and the implication. Strip away methodological detail, robustness checks, and supporting analysis. These belong in the full report, not the summary.

4. Translate vocabulary. Run every sentence through the audience-appropriate vocabulary filter (see plain-language.md). Replace jargon. Shorten sentences. Use active voice.

5. Cut methodology to one sentence. "We analyzed enrollment data from 13,000 districts over 8 years" is usually sufficient. If more is needed, it goes in a brief appendix or "methodology note" box.

6. Apply the caveat filter. Include only caveats that could change the reader's interpretation of the bottom line. "More research is needed" is never a caveat worth including. "These results only apply to elementary schools" might be.

7. Check: does the short version faithfully represent the long version? The distilled deliverable must not say something the full report does not support. Simplification is acceptable; misrepresentation is not.

What to Cut vs. What to Keep

Managing Nuance Loss

Distillation inevitably loses nuance. Manage this honestly:

• Acknowledge what is lost. "This summary covers the headline findings. For methodology, robustness checks, and subgroup analyses, see the full report."
• Don't overstate in the short version. If the full report says "the evidence suggests," the summary should not say "the evidence proves."
• Provide a path to depth. Always link or reference the full report for readers who want more.

Cross-Deliverable Consistency

When a single analysis produces multiple deliverable types (e.g., a full report, an executive summary, a policy brief, and a presentation), maintain consistency:

• [ ] The same finding is stated the same way across all deliverables (adapted for audience, not contradicted)
• [ ] Numbers are identical everywhere they appear
• [ ] Charts use the same color coding and style across formats
• [ ] Caveats included in shorter deliverables are a subset of those in longer ones (never introduce new caveats in the summary)
• [ ] The "bottom line" is consistent: if the report says A, the executive summary cannot say B

References and Further Reading

IES. "Six Strategies for Effectively Communicating Research Findings to Decision Makers." https://ies.ed.gov/blogs/research/post/six-strategies-for-effectively-communicating-research-findings-to-decision-makers

Minto, B. (2010). The Pyramid Principle: Logic in Writing and Thinking. 3rd Edition. Pearson.

Nussbaumer Knaflic, C. (2015). Storytelling with Data: A Data Visualization Guide for Business Professionals. Wiley.

Schwabish, J. (2021). Better Data Visualizations: A Guide for Scholars, Researchers, and Wonks. Columbia University Press.

AAAS. "Communication Toolkit." https://www.aaas.org/resources/communication-toolkit

Narrative Frameworks

Six proven structures for organizing data-driven communication. Each framework serves a different purpose — the right choice depends on your audience, your goal, and the nature of your finding.

The key insight behind all of these: data alone does not persuade. As Brent Dykes argues, data, narrative, and visuals form a triad where all three must work together. Data provides the evidence, narrative provides the meaning, and visuals provide the intuition.

Framework Selection Guide

Use this table to choose the right framework before you start writing:

Decision Tree

What is your primary goal?
├─ Deliver a recommendation or answer
│   ├─ Audience is time-constrained → Pyramid Principle
│   └─ Need to build the case first → SCQA
├─ Tell the story of a discovery
│   ├─ Findings were surprising or counterintuitive → Three-Act Data Story
│   └─ Findings confirm expectations but need action → "So What?" Framework
├─ Create a comprehensive deliverable
│   └─ Data-Narrative-Visual Triad (use as a completeness check)
└─ Persuade someone to commit resources
    └─ AIDA

---

1. Pyramid Principle (Minto)

Origin: Barbara Minto at McKinsey & Company. The gold standard for business and consulting communication.

Core idea: Start with the answer. Then provide supporting arguments. Then provide evidence for each argument. The reader should be able to stop at any level and have a complete (if less detailed) understanding.

Structure

Level 1: ANSWER (one sentence)
  ├─ Level 2: Supporting Argument 1
  │   ├─ Level 3: Evidence A
  │   └─ Level 3: Evidence B
  ├─ Level 2: Supporting Argument 2
  │   ├─ Level 3: Evidence C
  │   └─ Level 3: Evidence D
  └─ Level 2: Supporting Argument 3
      ├─ Level 3: Evidence E
      └─ Level 3: Evidence F

Rules

1. Start with the answer. The very first sentence of the deliverable states the conclusion or recommendation. Not the background, not the methodology, not the context — the answer. 2. Group supporting arguments logically. Three arguments is ideal (human working memory). Each argument must independently support the answer. 3. Arguments before evidence. A busy reader reads Level 1 and Level 2. Only interested readers drill into Level 3. 4. MECE at each level. Arguments should be Mutually Exclusive and Collectively Exhaustive — no gaps, no overlaps. 5. Use vertical and horizontal logic. Read top-to-bottom (each level supports the one above) and left-to-right (arguments at the same level are parallel in structure).

Example Skeleton

Answer: School districts should prioritize early literacy interventions over class-size reduction to improve reading outcomes.

>

Argument 1: Early literacy programs show 2-3x larger effect sizes than class-size reduction in controlled studies.

- Evidence: Meta-analysis of 47 RCTs...

- Evidence: Our district-level analysis confirms...

>

Argument 2: Early literacy interventions are more cost-effective per student served.

- Evidence: Cost-per-student comparison...

- Evidence: Scalability analysis...

>

Argument 3: Implementation is faster and less politically complex.

- Evidence: Timeline comparison...

- Evidence: Stakeholder survey results...

When to Use

• Executive summaries and recommendations
• Board presentations
• Any situation where the audience's first question is "What should we do?"
• When you have limited time or space

When NOT to Use

• When the audience needs to be convinced the problem exists before hearing the answer
• When the finding is politically sensitive and needs careful framing
• When the journey of discovery is itself the story

---

2. SCQA (Situation-Complication-Question-Answer)

Origin: Also from Barbara Minto, as a setup technique for the Pyramid Principle. Widely used in consulting and policy communication.

Core idea: Before giving the answer, establish why the audience should care by framing a problem. SCQA creates the cognitive "itch" that the answer then "scratches."

Structure

Rules

1. Situation must be uncontroversial. The audience should nod along — "yes, that's the world I know." No new information here. 2. Complication creates tension. It disrupts the comfortable situation. This is where the energy comes from. 3. Question must feel inevitable. If the complication is well-crafted, the audience is already asking the question before you state it. 4. Answer resolves the tension. It should feel satisfying — the puzzle pieces fit together.

When to Use

• Policy briefs where you need to motivate the reader before presenting findings
• Funding proposals where you need to establish the need
• Any situation where the audience doesn't yet recognize the problem
• When your findings challenge the conventional wisdom

When NOT to Use

• When the audience already knows the problem and just wants the answer (use Pyramid)
• When the situation is too complex for a brief setup

---

3. Three-Act Data Story

Origin: Classical dramatic structure (Aristotle's Poetics) applied to data communication. Championed by Brent Dykes and Cole Nussbaumer Knaflic.

Core idea: Treat your analysis as a narrative with setup, conflict, and resolution. This structure works because human brains are wired for stories — narrative activates more brain regions than data alone.

Structure

Rules

1. Act 1 must be brief. Don't spend half the presentation on background. Set the stage quickly — the audience came for Act 2. 2. Act 2 is the star. The data insight is the turning point. Make it vivid, specific, and surprising. This is where your best visualization goes. 3. Act 3 must be actionable. Don't end with "more research is needed" unless that is genuinely the most important takeaway. End with what the audience can do. 4. One story per deliverable. If you have three insights, tell three stories — don't weave them into one confused narrative.

Example Skeleton

Act 1 (Setup): Between 2015 and 2020, our district invested $14M in class-size reduction, bringing the average from 28 to 22 students per class. The expectation was measurably improved reading scores.

>

Act 2 (Confrontation): Reading proficiency remained flat — and in 12 of our 18 schools, it actually declined. Meanwhile, three schools piloting early literacy programs with a fraction of the budget saw double-digit gains.

>

Act 3 (Resolution): The evidence suggests that how instructional time is used matters more than class size. We recommend redirecting $8M toward structured literacy programs while maintaining current class sizes.

When to Use

• Presentations to mixed audiences
• When the finding is surprising or counterintuitive
• When you need to take the audience on a journey to appreciate the conclusion
• Conference talks and public lectures

When NOT to Use

• When the audience is impatient and wants the answer immediately (use Pyramid)
• Written deliverables for busy executives (the "story" wastes their time)

---

4. "So What?" Framework

Origin: A practitioner framework common in consulting, journalism, and science communication. Each finding must answer "so what?" before it earns inclusion.

Core idea: Every finding in your deliverable must carry three layers: the finding itself, its implication, and the action it suggests. If a finding can't answer "so what?" at all three levels, it doesn't belong.

Structure

For each finding:

Finding → Implication → Action
  "What did we observe?"
    → "Why does this matter?"
      → "What should someone do about it?"

Example

Rules

1. Apply the "so what?" test to every finding. If a finding is interesting but has no implication or action, it belongs in an appendix, not the main text. 2. The implication is where you add value. Anyone can report what the data shows. Explaining what it means is the analyst's contribution. 3. Actions should be proportional to evidence. Strong evidence supports specific recommendations; weaker evidence supports "areas for further investigation." Match the action to the inferential capacity (see plain-language.md on causal language). 4. Layer the full chain for your most important findings. Minor findings may stop at the implication level.

When to Use

• As an editing filter for any deliverable (apply to every paragraph — does it answer "so what?")
• Reports that combine multiple findings
• When your audience is action-oriented
• As a completeness check: if you can't state the implication, you may not understand your finding well enough

When NOT to Use

• This is more of a universal editing tool than a standalone structure. It works within other frameworks rather than replacing them.

---

5. Data-Narrative-Visual Triad (Dykes)

Origin: Brent Dykes, Effective Data Storytelling (2019).

Core idea: Effective data communication requires all three elements — data, narrative, and visuals — working together. Missing any one creates a deficiency:

Rules

1. Data provides the evidence. It must be rigorous, well-sourced, and correctly analyzed. (This is covered by data-scientist skill — not duplicated here.) 2. Narrative provides the meaning. It explains why the data matters, what the audience should take away, and what should happen next. Use one of the other frameworks in this file to structure the narrative. 3. Visuals provide the intuition. A well-designed chart makes the pattern feel obvious. (Chart construction is covered by data-scientist visualization references — not duplicated here.) 4. All three must tell the same story. If the chart shows one thing, the narrative says another, and the data supports a third, the communication fails.

When to Use

• As a completeness check for any deliverable: Do I have the data? Have I explained what it means? Have I shown it visually?
• When reviewing someone else's work: identify which element is missing
• When planning a report or presentation: ensure all three are planned from the start

When NOT to Use

• This is a meta-framework, not a structure. It tells you what must be present, not how to organize it. Pair it with another framework for structure.

---

6. AIDA (Attention-Interest-Desire-Action)

Origin: Marketing and persuasion framework, attributed to E. St. Elmo Lewis (1898). Adapted for data communication in contexts requiring stakeholder buy-in.

Core idea: Move the audience through four psychological stages — from noticing, to caring, to wanting, to committing.

Structure

Rules

1. Attention must be earned in the first sentence. If you lose them here, nothing else matters. 2. Interest requires narrative, not more data. Data captured attention; now explain what it means with a story or analogy. 3. Desire requires evidence of feasibility. "Here's what happened when District X tried this approach." Comparable examples are more persuasive than theoretical arguments. 4. Action must be concrete. "Consider our findings" is not an action. "Approve the pilot budget" is.

When to Use

• Budget presentations and funding requests
• Change management communications
• When the audience needs to be persuaded, not just informed
• When inaction is the default and you need to overcome it

Example Skeleton

Attention: "For every dollar we spend on class-size reduction, we get 15 cents of reading improvement. For every dollar we spend on structured literacy, we get 42 cents."

>

Interest: Over the past six years, we tracked 200 districts that took different approaches to improving reading. The pattern was consistent across rural, suburban, and urban settings: the type of instructional investment mattered far more than class size. Three schools in our own district piloted literacy programs and saw double-digit gains — while our $14M class-size investment produced flat results.

>

Desire: District X, which is demographically similar to ours, switched to structured literacy three years ago. Their proficiency rates have climbed from 52% to 71%. Their per-student cost dropped by 40%. And their largest gains came from the students who were furthest behind — exactly the population we're most concerned about.

>

Action: We are requesting Board approval for a $2M pilot across 6 schools starting fall 2026, with an independent evaluation reporting interim results in January 2027. The pilot is designed to produce evidence sufficient for a district-wide decision by spring 2027.

When NOT to Use

• Academic publications (AIDA is persuasive; academic communication should be neutral)
• When the audience is already motivated and just needs information
• When the action is unclear or premature — don't force an ask that isn't ready

---

Combining Frameworks

These frameworks are not mutually exclusive. Common combinations:

Matching Framework to Deliverable Type

References and Further Reading

Dykes, B. (2019). Effective Data Storytelling: How to Drive Change with Data, Narrative, and Visuals. Wiley. https://www.effectivedatastorytelling.com/

Minto, B. (2010). The Pyramid Principle: Logic in Writing and Thinking. 3rd Edition. Pearson.

Nussbaumer Knaflic, C. (2015). Storytelling with Data: A Data Visualization Guide for Business Professionals. Wiley. https://www.storytellingwithdata.com/

Alan Alda Center for Communicating Science. https://www.aldacenter.org/

AAAS. "Communication Toolkit." https://www.aaas.org/resources/communication-toolkit

Plain-Language Translation

How to communicate statistical and analytical findings in language your audience can understand. Plain language is not dumbing down — it is precise communication calibrated to the reader's vocabulary. As the NIH states: "People with average reading skills also appreciate messages conveyed simply and clearly."

This reference covers four areas: translating statistical jargon, calibrating reading level, expressing uncertainty through hedging, and choosing appropriate causal language.

Statistical Jargon Translation Table

When writing for non-technical audiences, replace statistical terminology with accessible equivalents. The goal is to preserve meaning while removing the barrier of unfamiliar vocabulary.

Reading Level Guidance

The NIH recommends writing for the actual audience, not for a universal grade level. Match complexity to the reader:

Practical Reading Level Tips

Based on NIH and federal plain language guidelines:

1. Sentence length: Average 15-20 words per sentence. One idea per sentence. 2. Paragraph length: 2-3 sentences maximum. One topic per paragraph. 3. Word choice: Pick the familiar word over the unusual one. "Use" not "utilize." "Before" not "prior to." "About" not "pertaining to." 4. Active voice: "We found that X causes Y" not "It was found that Y is caused by X." Active voice clarifies who did what. 5. Concrete over abstract: "1 in 5 students" not "20% of the student population." Concrete examples stick; abstract percentages slide off. 6. Define once, use consistently: If you must introduce a technical term, define it clearly the first time, then use it consistently. Don't alternate between synonyms.

Common Word Replacements

From federal plain language guidelines (PlainLanguage.gov), selected for research communication:

Hedging Language Scale

Hedging communicates your confidence level to the reader. The challenge: too little hedging overstates your evidence; too much hedging buries the finding. Match your hedge to your inferential strength.

Five-Level Hedging Scale

Hedging Anti-Patterns

IPCC Calibrated Uncertainty Language

The Intergovernmental Panel on Climate Change developed a formal framework for expressing uncertainty that has become a model for science communication across disciplines. It uses two independent scales.

Likelihood Scale (Probability of Outcome)

These terms express quantified probability. In formal usage, they appear in italics to signal calibrated meaning.

Key design features:

• The scale is asymmetric — terms overlap (e.g., "very likely" is a subset of "likely")
• "About as likely as not" (33-66%) is the pivot point between affirmative and negative likelihood
• The everyday meaning of "likely" (~70-75% to most people) roughly aligns with the calibrated meaning (66-100%), but "unlikely" is often misread as near-zero rather than the actual 0-33%

Confidence Scale (Epistemic Quality)

Confidence reflects the author team's judgment about the validity of a finding, based on evidence quality (limited/medium/robust) and degree of agreement (low/medium/high):

Using Both Scales Together

Confidence certifies the epistemic quality of the assessment; likelihood communicates the probability. They can be stacked: "It is very likely (high confidence) that X is occurring."

Decision rule: Use likelihood terms only when confidence is high or very high. Expressing a precise probability for a finding with low confidence creates false precision.

Adapting for Non-Climate Research

While the IPCC framework was designed for climate science, its two-dimensional approach (probability + confidence) is valuable for any research communication that needs to express uncertainty clearly:

• For policy audiences: Use the likelihood terms directly (they're intuitive) and translate confidence as "strength of evidence"
• For general audiences: Map to everyday language: "very likely" → "the evidence strongly points to..."; "about as likely as not" → "the evidence is mixed"
• For academic audiences: Use the framework as a supplement to standard statistical reporting

Causal vs. Correlational Language

Choosing the right linking language is one of the most consequential decisions in research communication. The wrong word can either overstate your evidence (implying causation from correlation) or understate it (hedging a well-identified causal effect into meaninglessness).

The Core Distinction

Linking Language Spectrum

Adapted from Haber et al. (2022), who systematically evaluated how readers interpret different linking phrases. Words are arranged from strongest to weakest causal implication:

The "Schrödinger's Causal Inference" Problem

Haber et al. (2022) identified a pervasive pattern in published research: papers use hedged associational language in their results ("X was associated with Y") but then imply causal sufficiency in their conclusions and recommendations ("clinicians should consider X to reduce Y"). The causal claim simultaneously exists and doesn't exist depending on which paragraph you read.

Prevention: Ensure that the strength of language in your recommendations does not exceed the strength of language in your findings. If you found an "association," your recommendation should say "further investigation is warranted" — not "policymakers should adopt X."

Practical Causal Language Rules

1. Default to associational language unless your study design explicitly supports causal claims. "Is associated with" is almost always safe.

2. State the study design when using causal language. "Our difference-in-differences analysis shows that the policy reduced dropout rates by 4 percentage points" is honest because it names the design. "The policy reduced dropout rates" without design context invites misinterpretation.

3. Separate findings from implications. Use causal language in findings only if the design warrants it. In the implications section, you can discuss causal possibilities with appropriate hedging: "If this association is causal, it would suggest..."

4. Watch for direction-of-causation assumptions. "Poverty affects test scores" assumes a direction. "Poverty is associated with lower test scores" is neutral about the direction. Only claim directionality if your design supports it.

5. Avoid "weasel causality." Words like "impact," "influence," and "affect" are verbs that imply the exposure did something to the outcome — they are causal in ordinary English even if the writer thinks they're being cautious. If your design is observational, these words overclaim.

6. Never let a word list substitute for clear thinking. Per Haber et al., policing individual words fails because context, sentence structure, and rhetorical positioning all contribute to how readers interpret causal claims. The solution is structural transparency about your design and estimand, not vocabulary restrictions.

Number Formatting Standards

How you present numbers affects comprehension as much as how you present words. Match formatting to the audience.

General Rules

Framing Effects to Watch

The same number can feel very different depending on how it is framed:

Rule of thumb: Present both absolute and relative comparisons when the finding is important. Lead with whichever framing is most appropriate for your audience and purpose, but always include the other for context.

Domain-Specific Jargon

The jargon translation table above covers statistical and methodological terms. However, every research domain has its own specialized vocabulary (e.g., "Title I," "FRPL eligibility," "AYP" in education; "incidence," "comorbidity," "morbidity" in public health).

Guidance for domain-specific translation: When working with a specific data domain, consult the relevant domain skill (e.g., education-data-context, education-data-source-ccd) for the authoritative definitions and caveats of domain-specific terms. Then apply the same plain-language translation principles from this file: replace with familiar equivalents, define on first use, and provide concrete examples.

The jargon translation table in this file is deliberately domain-agnostic — it handles the statistical and analytical language that appears in every research project. Domain-specific translation is handled by domain skills to avoid duplication and ensure accuracy.

References and Further Reading

Haber, N.A. et al. (2022). "Causal and Associational Language in Observational Health Research." American Journal of Epidemiology, 191(12), 2084-2097. https://doi.org/10.1093/aje/kwac137

NIH. "Plain Language at NIH." https://www.nih.gov/institutes-nih/nih-office-director/office-communications-public-liaison/clear-communication/plain-language-nih

NIH. "Clear & Simple: Developing Effective Print Materials for Low-Literate Readers." https://www.nih.gov/institutes-nih/nih-office-director/office-communications-public-liaison/clear-communication/clear-simple

Federal Plain Language Guidelines. https://www.plainlanguage.gov/guidelines/

IPCC. "Guidance Note for Lead Authors of the IPCC Fifth Assessment Report on Consistent Treatment of Uncertainties." https://www.ipcc.ch/site/assets/uploads/2017/08/AR5_Uncertainty_Guidance_Note.pdf

Mastrandrea, M.D. et al. (2011). "The IPCC AR5 Guidance Note on Consistent Treatment of Uncertainties: A Common Approach Across the Working Groups." Climatic Change, 108, 675-691.