I think to have more freedom on the data processing steps, the analysis should either

1. be split into a new package; or
2. be copied into this project and modified.

My preference is for 2, because I think it means that we can make separate decisions for the viz and the paper while the paper is under review, and reconcile differences down the track.

• 'Grouped' effects/outcomes should be computed as nested meta-analyses
• Need effect sizes for the exposure/outcome combinations at each level
• Need a method to identify the last time the data was updated

Have a look at flexdashboard

Have a look at shiny apps

Using this issue to store ideas for settings

• What to base the 'best' effect on (default: total N)
• Sort y by (default: effect size; other options: name)

Mostly on the x-axis.

E.g.,

There is weak evidence that internet use: cyberbullying victimization is associated with learning

is kind of a strange sentence. Same with all of the 'intervention' ones.

• The required environment variables (to get data from cloudstor)
• How to deploy to shinyapps
• How to modify the documentation website

The initial load of the app is very slow. Partly that's shinyapps, but there's probably some ways to make this a bit faster

Task One

• Convert to heatmap. Outcome on the y axis, exposure on the x axis
• Try with the subcategories as well
• Try with both effect size and effect count as value

Task Two

• Change to line for each of the subcategories
• Order by value, not name
• Use the size of the circle to indicate sample size or other forms of uncertainty
• For other uncertainty, use another heatmap on the side, or use a 'score' indicated on the background of the label
• Try using another heatmap instead of line plot

Task 5

• Try a spider/radar plot

• Fix the sketches with Kane's suggestions
• Build the interactive version in a couple of libraries

• Give each cell a 1 or 2 px border
• Add title with the current exposure/outcome level
• Filter for age
• Add a hint to click in for more info into the tool tip
• Hover over rest should explain what it does
• Add a previous level button
• Reset should only appear when needed (i.e., not at first level)
• Font sizes for axes need to be 1.5-2x larger
• reset location should be top left (@mnoetel in #21 (comment)_)
• Show effect size on hover (@cslonsdale in #21 (comment)_)
• Remove 'intervention' as a category

• Change the way that studies are selected (e.g., publication date, total n, k)
• Should show meta-analysis statistics ($I^2$, etc)
• What to base the 'best' effect on (default: total N)
• Sort y by (default: effect size; other options: name)

At the moment I've just used "RdBu" for a diverging colour scale, but it would be good to create a consistent theme across all of the plots

• Review level
• Effect size level
• Study level
• Study quality
• DOI/bibliographic information

Might need input from someone with more CSS experience