Presenting Graphs

I’m a sucker for nuggets of technology history, especially understanding how core parts of company’s cultures developed. The Amazon Weekly Business Review is one piece of culture I’ve come across recently (from Cedric Chin’s articles). So when I came across a post from Eugene Wei about the very first analytics packets at Amazon, I was hooked. You should read the article and do everything Eugene suggests when designing your graphs, but I realized the article doesn’t talk about giving presentations with graphs.

The difference with a presentation is that you can build complex graphs with your audience. This allows you to stuff more data on a page than you could with a static graph (not that you always should) in a way that is comprehensible to your audience. I learnt this technique from my research mentor a long time ago, and it’s served me well for every technical presentation I’ve had since - here he is using it in one of his talks. I used to think that this was fairly obvious, but time after time I’ve seen people throw graphs onto slides with little thought for how an audience will consume those graphs.

As a very simple example, let’s look at this chart from “Our World in Data” comparing literacy rates among age groups across countries (that’s already a mouthful). The simple thing would be to dump the chart into a single slide:

This data was basically incomprehensible to me until I broke it down over many minutes, so either I suffer from terrible visual reasoning skills or this is not a graph that’s suitable for rapid consumption in a presentation. If you’re also talking the whole time you show this graphic, your betting on your audience either being extremely familiar with this data already, or only diverting all their energy to understanding the graph instead of hearing what you have to say about it. Instead, we can split out our graphs into many transitions:

Now you understand these things in order:

• The graph is comparing literacy rates of older vs. younger populations.
• We’re comparing data across 6 continents.
• Europe generally has a high literacy rate across both young and old people.
• Some Asian countries like China have high literacy rates across old and young people. Others like India/ Bhutan/ Bangladesh have much higher literacy rates among young people suggesting more recent progress in education systems.
• Africa is extremely varied in it’s distribution, but generally seems to have literacy levels lower than other parts of the world. It’s also the first time we see the bottom left quadrant populated.

Even with this split I’m not fully satisfied with how complicated this graph feels. For a real presentation about this data I would have done the following:

• Introduced the x and y axis separately.
• Introduced the population size circles separately.
• Introduce the legend entry for each continent one at a time.
• Split the Asian cohort into the two or more slides based on elderly literacy rate and country size.
• Split the Africa cohort into a few slides based on where on the graph they show up.

This might seem like overkill, but I haven’t ever noticed breakdowns like this going too far in talks. For every tiny atomic change you introduce incrementally, you can spend less time establishing simple correlations once all your data is on the screen. It lets you focus on bigger revelations, or even let your audience come to your conclusions before you! An added advantage is that you no longer need slide notes - see how easy it is to think of something to say with each slide. You’re not only building the graph up for your audience, your building it up for yourself.

Making these animations is not usually hard. Ideally you can script out your changes logically if you’re generating your graph programatically. My preferred way is to export graphs as SVGs and just paste them on 10 slides in the same spot. Then incrementally remove various parts of the graph until you have a logical progression. No fancy animations necessary, and it’s easy to flip back and forth through your slides if you ever want to go back. If you only have an image of a graph (less ideal), you can just put rectangles over various parts of the graph to hide and reveal it slowly.