Presentations

SURE 2024

Department of Statistics & Data Science
Carnegie Mellon University

Timeline

EDA project presentation

8-min presentation on Tuesday, June 18 during lecture time
- aim for 7 min + 1 min for Q&A
- make sure you do not exceed 8 min (practice and time yourself)
- I’ll cut you off if you exceed 8 min
Slides are due Monday, June 17 at 5pm ET

Capstone project presentation (a preview)

2 presentation checkpoints (both during lab time)
1 final presentation on the final day (July 26) (+ other deliverables)
Details will be provided next week

Presentation tips

Begin with background and motivation

Every presentation has a story:

What is the motivation? Why should people care about your work?
You want to build up what your work is trying to address

Example: Ron’s nflWAR talk at NESSIS 2017:

Do NOT begin with: “We’re introducing [project topic]” (WAR for NFL)
Instead begin with current state of NFL analytics and need for better, reproducible player level-metrics

Do NOT include an outline slide!

Your presentation should flow naturally

Or you can begin with a “creative” intro…

Describing the data

You want to provide a general overview of your dataset:

What are your observations? i.e., what does each row of your dataset represent?

What are the relevant variables/features? i.e., what are the columns of interest?
- Be careful though with many variables - avoid just listing everything!
- Simplify by describing groups of variables together

Use examples - makes your data explicit and concrete for the audience

But do NOT print out raw R console output or screenshots of output!
Use text or a clean formatted table (via e.g., knitr, DT, gt, etc.)

Introducing and describing methods

Prior to presenting results, you want to clearly state any transformations and methods used in the analysis
Your presentation should provide the general steps for someone to replicate your work
- e.g., Used complete-linkage hierarchical clustering with [INSERT VARIABLES], determined \(k\) number of clusters by [INSERT REASON]
- e.g., Modeled [INSERT RESPONSE VARIABLE] as a function of [INSERT EXPLANATORY VARIABLES]

For more complicated methods, you’ll want to provide a brief review of the methodology
If introducing new methodology: walk through the steps clearly

Always justify your choice of methodology
- Why you used a flexible tree-based model over linear regression

Presenting results

Use the assertion-evidence model
Assertion: title of the slide should be the key takeaway in brief sentence form
- Indicates the point of the visualization or whatever means used to display the results
Evidence: the body of the slide containing the results
- Display of the results in some format that is simple to explain and understand

Limit the amount of text in your Evidence portion - brief statements with important context
Treat the Assertion as the title of your Evidence
- Plot titles are then redundant and not necessary with an effective assertion

Assertion

Evidence

(e.g., plots, animations, tables, etc.)

MLB strikeout rates have been increasing throughout history

(Explain the aes of your graph - what is each axis, color, shape, etc referring to? And what is the unit scale?)

Discussion (and ending a presentation)

Conclude with a recap of the main points of your work

Then point out limitations and indicate future directions

Either end with the Discussion slide (or Acknowledgements but this is sometimes placed at the beginning)
Never end a presentation with lone Thank you slide!
- Want the audience to focus on the final points in your Discussion slide

Include back-up Appendix slides with additional info, ready for questions

Slides for References should not be displayed during your talk
- Their purpose is just for sharing with others
- Alternative option: include references directly on slides either in text or via footnotes ¹

Additional tips and reminders

Use pauses effectively to highlight points and explain steps

Showing all of your text at once can overwhelm your audience

But

don’t

ridiculous

Remember: memory overload is real!

Do NOT introduce too much notation at once
Repetitive language and usage of words are useful and reminders for the audience
- Use consistent language and terminology throughout the talk

Know your audience!

Design tips

Less ink, less ink, less ink
Your plot should be big enough (font size, line width, point size, etc.)
Reformat variable names (don’t show the original names in the data)
If a table can be represented by a figure, turn it into a figure (e.g., comparison of model evaluation metrics)

More general tips

Your slides should support what you say, not create interference between speech and vision brain areas – your slides serves as a summary of what you said so someone who may have been distracted can catch up quickly.

Do not create your talk at the last minute (it will be obvious if you do). Designing good slides takes time. Practicing a talk takes time.

Advices from Andrew Patton:

You don’t need 17 decimal points for anything ever.

Make sure your figures and charts, etc. all display the correct size. If I have to zoom in 27 times to read the legends that’s not going to help your case.