Communicating Statistical Results

The “90/10” rule

90% of the work you do will occupy only 10% of your presentation time/space.

• Students tend to make reports and dashboards using the timeline they experienced in the analysis

• Instead, focus on conclusions and takeaways

• Differentiate between communicating results and documenting a process (both are important!)

The “90/10” rule

10%:

• Data cleaning and wrangling details

• Assumption checking for models or statistical tests

• Combined results all in one visual

90%:

• Specific conclusions from tests - “we found evidence that…”

• Plots of model predictions and summary statistics

• Visuals highlighting individual takeaways - “California has higher predicted prices than the Midwest”

The “Newspaper Headline” approach

For each analysis, what is the headline a newspaper would use.

• NO: Statements of general accomplishment - “We fit a model to predict prices of fast food items.”

• NO: Model assessment - “Our model has 93% accuracy on a test set.”

• NO: Results in technical form - “The variable found to be most significant was location of store.”

• NO: Unspecific results - “Location of fast food store impacts price.”

• YES: Results of public interest in plain terms with specific knowledge: “Fast food prices are 50% higher in California than in the Midwest.”

YOUR TURN

Dashboards as “Q&A”

The goal of an interactive dashboard is to replace you, by giving the audience a way to ask and answer questions themselves.

• Anticipate follow up questions to the main “headlines”.

• Go from general to specific - e.g. “Here is a plot of the average price of burgers over time. Use this drop-down to filter by state.”

• Keep your dashboard sanitized against bad statistical choices - e.g., don’t allow users to select “Show logistic regression” for numeric data.

“Why should I care?”

Dashboards and presentations are for your target audience, not your peers.

• What is interesting to a data scientist might not be interesting to your target audience.

• Use personas:

  • Give your client a name
  • Why does this client want your analysis?
  • What is their specific stake in the data?
  • What level of technical content will they understand and appreciate?

• Stay focused on how the client will use the analysis to make decisions.

Example

My client is Nisha, a CEO of a small restaurant chain that competes with local fast food. She has a degree in business administration, so she understands basic statistics. She has hired me to analyze price trends in the food industry.

• Nisha is a CEO of a small business - she is probably interested in how to keep her restaurant solvent

• We can quote basic statistics, but we should not go deep into modeling math.

• She has asked specifically about prices, so our presentation should focus on that.

• She has asked about trends, so our “newspaper headline” should be about changes over time.

• We should state how our analysis will help her set her prices at her restaurants.

YOUR TURN

Activity ideas

Can the data answer this question?

• Give students a dataset and a list of questions

Is this conclusion supported?

• Give a visualization or test result, and a client quote - “This means I should sell more hamburgers!”

• Ask students if they think the result supports the conclusion.

“Client” feedback

• Invent a mock client for all student groups to target

• Have student groups swap projects and imagine themselves as the clients

Create your Project B Dashboard Assignment