Decision Tree Exercise

A participant-led group exercise. The goal is to work through scenarios and surface the questions that matter when choosing how to handle reporting delays and other data challenges. The output will help inform the interactive decision tree.

For reference during the exercise, see the guide’s challenges and modelling options pages.

How it works (~20 minutes)

Please split yourselves into up to four small groups, sized to the room. Please aim to mix STLT practitioners with modellers, and to be with people you do not already work with closely.

Step 1: Sketch in pairs (~10 minutes)

Pair up within your group. Pick a scenario closest to your setting, or bring your own. On paper, sketch a decision tree that captures the choices you would make and why.

Sketching trees is a skill in itself, so do not worry about being neat. Boxes for choices, lines for branches, short labels. If a step does not fit a clean yes/no, write it in as a note next to the branch.

Some questions to consider as you sketch:

Are delays stable or changing?
Do you have enough historical data to estimate delay patterns?
Do you need uncertainty estimates or just a point correction?
How much time and technical capacity do you have?

Step 2: Compare in your group (~10 minutes)

Bring your pair sketches together at the flipchart or board. Where do your trees agree? Where do they diverge, and why? Try to draw a shared tree, or mark the points where the group does not agree.

Please nominate one person in the group to take brief notes on the discussion.

Things that are hard to put on a tree

Trees are good for binary choices but awkward for some real decisions. If any of these come up, capture them as notes next to the branch rather than forcing them into the structure:

Trade-offs that are not binary (e.g. complexity vs. accuracy, speed vs. precision).
Multiple valid paths where more than one approach would work.
Constraints that change through a season (capacity, leadership attention, data volume).
Stakeholder or communication needs that cut across the technical choice.
Dead ends where no current tool fits well.

Example scenarios

These are starting points; use your own data and challenges if you prefer.

Weekly syndromic surveillance with stable delays: weekly ED visit counts for respiratory illness, typically “complete” within 6 weeks, fairly stable delay pattern, need to report prior week’s trends midweek.
Daily case counts during an emerging outbreak: daily counts by symptom onset, delays getting longer as labs come under strain, leadership wants daily updates.
Site drop-in/drop-out in sentinel surveillance: hospitalisations across 12 sentinel sites, two reporting intermittently, one new site, need a jurisdiction-wide estimate.
Low counts with long delays: rare condition with 5-10 cases per week, delays up to 8 weeks, 18 months of historical data, weekly bulletin needs a current case count and trend statement that does not over- or under-state recent activity.

What we will collect

To feed into the interactive decision tree, we would like to take away:

Your paper sketches from the pair stage (we will collect them at the end if you are willing to share).
Photos of each group’s flipchart.
The note-taker’s summary of the group discussion.

After the session we will write up this material and circulate it for your feedback before it informs the guidance.