Statistical Communication

The Thanksgiving Table: explaining one study to someone you love

Mine Doğucu

Why statistical communication?

Statistics doesn’t speak for itself

A result only helps if the people who need it understand it.

  • a patient deciding on a treatment
  • a manager deciding on a budget
  • a voter reading a poll
  • your family, deciding what to believe

Communicating a result is part of doing statistics, not an extra step at the end.

Same number, two stories

In 1995, the UK Committee on Safety of Medicines warned that some newer birth control pills doubled the chance of dangerous blood clots.

As a ratio

“The chance doubles.” (100% increase)

As counts of people

From about 1 in 7,000 women to about 2 in 7,000.

Many women stopped taking the pill. The following year there were an estimated 13,000 additional abortions in England and Wales.

Gigerenzer, G., Gaissmaier, W., Kurz-Milcke, E., Schwartz, L. M., & Woloshin, S. (2007). Helping doctors and patients make sense of health statistics. Psychological Science in the Public Interest, 8(2), 53–96.

Facts alone rarely change minds

It’s tempting to think people disagree with science because they’re missing facts, so we just need to give them more facts.

But people also ask:

  • Do you understand my worry?
  • Do you respect me?
  • Are you honest about what you don’t know?

If the answer is no, the facts don’t get through.

Good statistical communication

Four qualities of good statistical communication
Quality What it sounds like
Accurate Says what the study found, and no more.
Clear Plain words, one key message, counts of people.
Honest about uncertainty “Here’s what this study can’t tell us.”
Respectful Starts with the other person’s view.

The Thanksgiving Table

Thursday

Someone at the table will ask you about science.

Today’s goal

You will not win Thanksgiving. Nobody does.

The goal: leave the table with your relative

  • understanding one study a little better, and
  • still wanting to talk to you.

A relative who moves a little is a success.

What you’ll practice

Statistics

  • who was sampled
  • experiment or observational
  • p-values and intervals
  • what generalizes
  • how evidence builds up

Communication

  • plain language
  • counts of people
  • one key message
  • checking for understanding

Respect

  • listen first
  • start from their view
  • find a shared value
  • admit what you don’t know

How it works

The roles

The three roles
Role Your job
Explainer Explain one real study to your relative. Use your study card.
Relative Play a family member with a real reason for their view. Use your secret relative card.
Cousin Sam (groups of 3 or 4) On the fence and a bit lost. Asks “What does that mean?”

The conversation: five courses

  1. Appetizer: listen first. Ask, reflect back, name a shared value.
  2. Who and how. Who was in the sample? Experiment or observational?
  3. What they found and how sure. One count of people. The p-value or interval in plain words. One thing it is not.
  4. Who it applies to. What generalizes? One real limitation.
  5. Dessert: one brick, door open. One study is one brick. Summarize. “What’s still bugging you?”

Course 1: listen first

You may not explain anything until you have done all three.

Ask

“What worries you about it?”

“Where did you hear that?”

Reflect back

“So it sounds like you’re worried that…”

Name a shared value

“I want kids to be safe too.”

“I don’t like being lied to either.”

Course 3: explaining a p-value

Say it like this

“Suppose [the vaccine] did nothing at all. How surprising would results like these be, just by chance? The p-value measures that surprise. A small p-value means: this would be pretty weird if nothing were going on.”

Two analogies: coin flips (9 heads out of 10 makes you suspicious of the coin) and the courtroom (“not guilty” is not “innocent”).

Course 3: what a p-value is not

Say at least one of these:

  1. It is not the chance that the result is true, or that the treatment works.
  2. “Not significant” ≠ “proven: no effect.” It means results like these would not be unusual if there were no effect.
  3. “Significant” ≠ big or important. A tiny effect can be significant in a big study.

Course 3: explaining an interval

Say it like this

“Here’s the range of effects that fit the data reasonably well.”

Then ask two questions out loud:

  • Does the range include “no difference”? (For a slope or a difference, that’s 0. For a ratio, it’s 1.)
  • Does the range rule out a big effect? Sometimes this is the more convincing part.

Course 3: use counts of people

Hard to picture

“There were 95% fewer cases.”

“The sample proportion was 0.0088.”

Easy to picture

“Out of every 1,000 people who got the fake shot, about 9 got sick. Among those who got the vaccine, fewer than 1.”

Counts of people are easier to understand than percentages, ratios, or probabilities.

You already know the statistics

From STAT 100

  • population, sample, sampling bias
  • observational study or experiment
  • random assignment, confounding variables
  • p-values, statistical and practical significance
  • confidence and credible intervals
  • holding other variables constant

New today

Two cards compare groups with a ratio:

  • a ratio of 1 means no difference
  • below 1: less common in the first group
  • above 1: more common

Your card explains it.

What the relatives are told

Soften when you…

  • feel heard
  • hear honest uncertainty or a real limitation
  • hear an example from your own life

Dig in when you hear…

  • unexplained jargon
  • “just trust the science”
  • a condescending tone
  • “that’s been debunked” (no explanation)
  • a textbook definition, recited

Move only as much as a real person would.

The studies

The four study cards
Card Study Design
A MMR vaccine and autism (Denmark) Observational
B Pfizer–BioNTech COVID-19 vaccine Randomized experiment
C Mediterranean diet (Spain) Randomized, with problems
D Quitting Facebook (US adults) Randomized experiment

In your group, pair A with D or B with C.

Groups

2 people

Round 1: One member explains, the other is the relative.

Next round: swap.

3 people

Add Cousin Sam.

Rotate roles each round.

4 people

Tag-team siblings: two explainers together, plus a relative and Cousin Sam.

One person alone? Join a pair as Cousin Sam. Five? Split into 2 + 3.

The full rotation tables are on your guide.

On Zoom

  • Rename yourself to your character when you play a relative (“Uncle Ray”). Rename back afterward.
  • Explainers and relatives: cameras on. Faces are part of listening.
  • Nods are hard to see on video. Check for understanding more often.
  • Need a pause to look at your notes? Say “Pass the gravy.”
  • Breakout rooms are not recorded.

Open your page

In your group: step by step

1. Assign roles and pick study cards (A with D, or B with C). Links are on Open your page.

Steps 2 to 5 for each role
Step Explainer Relative Cousin Sam
2. Read your study card your secret relative card the Cousin card
3. Take notes key message, one count of people, one limitation who you are, what you care about words you wouldn’t use at home
4. Talk listen first, then explain ask your p-value question ask “What does that mean?”
5. Feedback listen go first: what moved you? add what you noticed

6. Switch roles when I call it, and start again at step 2.

Round 1

Pass the (vegan) turkey. Start with course 1.

Feedback

Relative goes first

  1. Did they ask about my view before explaining?
  2. Did I understand the p-value or interval explanation?
  3. Did they name something the study can’t tell us?
  4. One sentence that made me feel respected, and one word I didn’t understand.

Round 2

Switch roles. Same group.

Back to Open your page for your new card.

Another round

Feedback, switch roles, go again.

Step out of the role

“I’m not Uncle Ray. I’m ___.”

On Zoom: change your name back.

Debrief

What happened at your table?

Discussion

  1. Relatives: what made you soften? What made you dig in?
  2. Which sentence worked best for explaining a p-value or interval?
  3. With Card D: how did it feel to tell a parent the evidence was weaker than they hoped?
  4. Did admitting a limitation make you look weaker or stronger?
  5. Did talking over video change how you showed respect or checked understanding?

Briefing

What we practiced

Statistics ideas and plain-language versions
Idea Say it like this
Sample “Who was in the study? How were they found?”
Random assignment “A coin flip decided who got what.”
Sample proportion “Out of every 1,000 people…”
p-value “If it did nothing, how surprising is this?”
Interval “The range of answers that fit the data.”
Generalization “It can’t tell us about…”

One brick

A single study is one brick.

We trust an idea when many studies, by different people, with different methods, point the same way.

And when science gets something wrong, the fix often happens in public, like PREDIMED.

Your statistical communication checklist

  1. Listen first. Ask, reflect back, find a shared value.
  2. One key message.
  3. Counts of people, not only percentages or ratios.
  4. Say what the p-value or interval means and doesn’t mean.
  5. Name a limitation before anyone asks.
  6. Leave the door open. “What’s still bugging you?”

Exit ticket

Text your relative

Write a 3-sentence text message to your relative that

  1. summarizes the study,
  2. gives one number in plain language, and
  3. names one thing the study can’t tell us.

Submit at PollEv.com/minedogucu.

Happy Thanksgiving

Listen first. One brick. Leave the door open.