ETC5523: Communicating with Data

Workshop 5: Turning statistical output into prose

Author

Michael Lydeamore

Published

1 May 2001

🎯 Objectives

By the end of this workshop, you will be able to:

  • select results that matter for a stated purpose;
  • translate statistical output into accurate prose; and
  • distinguish estimates, uncertainty, association and causation.
Scenario and fictional results

A company examined whether completion of an optional training program was associated with analysts’ scores on an internal data-communication task. Scores range from 0 to 100. The model adjusts for each analyst’s years of experience. These data were created for this exercise.

Term Estimate 95% confidence interval p-value
Training completed (yes vs no) 4.8 2.1 to 7.5 <0.001
Years of experience 0.6 0.2 to 1.0 0.004

Additional information: 240 analysts were included; 96 completed the training. Participation was optional rather than randomly assigned.

1. Quickwrite: what deserves a sentence? (5 minutes)

Imagine you have space for only one sentence about a statistical analysis. Write what you would need to know before deciding which number deserves that sentence.

2. Read for meaning (5 minutes)

Without drafting report prose yet, annotate the output:

  1. What does 4.8 compare?
  2. What is the unit of the outcome?
  3. What does the confidence interval add?
  4. What can the analysis not establish?

3. Write the takeaway (7 minutes)

Write one sentence for a technically competent, time-poor manager. Include the comparison and its practical size. Do not begin with the p-value.

4. Build a results paragraph (10 minutes)

Expand the takeaway into a paragraph of 80–120 words. Use this sequence if helpful:

  1. main result;
  2. numerical evidence and uncertainty;
  3. relevant adjustment or context; and
  4. limitation on interpretation.

Do not describe every table cell. Do not claim that training caused the difference.

5. Write a self-contained caption (7 minutes)

Write a caption for the table. It should help a reader understand the table without searching the surrounding prose. Include:

  • the takeaway;
  • what the estimates represent;
  • the sample size; and
  • any abbreviation or essential model detail.

6. Evidence check (8 minutes)

Exchange the paragraph and caption with a partner. For every factual claim, the reader marks:

  • S – supported directly by the supplied information;
  • I – a reasonable interpretation, but not directly observed; or
  • U – unsupported or too strong.

Discuss disagreements, then revise the strongest unsupported claim.

7. Writer’s note (3 minutes)

Complete: “The difference between reporting a number and communicating a result is…”