Thread by @VictimOfMaths

I posted this graph yesterday as part of the #30DayChartChallenge and a few people have asked some reasonable questions about why I chose to present this data in this way.

So I thought I'd write a thread to explain my thought process...

The first thing to say is that I don't think any dataset has some inherently 'right' type of graph. There are just different graph types which may be more or less effective at highlighting different aspects of the data.

The second is that there is no truly neutral presentation of data. Your choice of graph type, colour, scale, structure and annotations will all influence what message people take away or what patterns people see.

Some people (particularly boring academics like me) don't like this idea. We're conditioned to try and be neutral because otherwise it feels as though we are forcing our biases on the reader.

But I think this is wrong for two reasons:
i) this tends to lead to bad, ugly visualisations in most cases and what is the value of a graph nobody looks at or understands?
ii) people will always find their own messages in graphs, so why not help them find the 'right' one?

(other opinions are available)

What this means is that for almost* any graph, one of the first steps is to work out what the message I want to convey with that specific graph is.

Because what that message is will have a huge bearing on how you choose to present the data.

(* there are some exploratory, rather than explanatory visualisations where the aim is to let viewers explore the data and find their own patterns, but they are rarer and for another thread)

Sometimes people ask me for advice about making graphs and I spend about 90% of those conversations just saying 'what do you want the message to be?'.

A sign of this approach is to look at the title of the graph - is it neutral or directive.

A neutral title would be something like "Rates of alcohol, suicide and drug deaths in England and Scotland".

A directive one states the key message, e.g. "Scotland has a drug deaths problem".

If you look back over my posting history, you'll see that the vast majority (although not all) of the graphs I post have directive titles. I think there is something interesting in the data and I've made a graph to try and illustrate that thing.

So, back to this particular graph.

As the title suggests, I decided I wanted to tell the story of just how much higher the rate of drug-related deaths is in Scotland compared to England & Wales. Because I don't think many people appreciate quite how big the difference is.

I could have told that story just by using the overall population rates of death, but I've added in two additional bits of information - how death rates vary by age and how they compare to alcohol and suicide deaths.

I think the variation by age is useful as a lot of people wrongly assume that drug deaths mostly happen in people in their teens and twenties when it's actually highest in people in their 40s. The age distribution of alcohol and suicide deaths is also (IMO) interesting.

And the comparison of drugs with alcohol and suicide comes because these three causes are often looked at together - so called 'deaths of despair'. You might reasonably assume that they have the same underlying causes..

...and to some extent they do, but I wanted to highlight that the difference between England & Wales and Scotland is much bigger for drug deaths than for suicide or alcohol deaths.

After playing around with the data a bit, I decided I could add in these two bits of extra information without detracting from the key message. Essentially I think (hope) that the graph still conveys a simple message, but that there are also some other insights to be found in it.

Next, the choice of chart type. I chose a stacked bar chart for several reasons - a key one was aesthetics. I just think this type of plot looks visually appealing. Some people shy away from this, I think, but to me it's really important.

After all, you want people to look at your chart in the first place.

Of course, you have to be careful not to pick a chart type *solely* because it looks nice, even if the key message is obscured, but there is (IMO) nothing wrong with considering it in your chart choice.

I also think that the stacked bar neatly conveys 3 messages:
a) overall deaths of despair are much higher in Scotland
b) particularly in middle age
c) and that this is driven largely by higher rates of drug deaths

One big limitation of the stacked bar is that the ordering of the different series makes a big difference. It's much easier to see differences in the series at the bottom than those in the middle.

However, in this case there are only three series, I think the difference in drug deaths is so stark that it stands out however you order them *and* I've already chosen to focus on drug deaths - so it makes sense to put them on the bottom.

This is also behind the choice of colour - red against the two blue shades pops out much more. It's back to the idea of conveying a specific message - some colours are more eye catching than others, so it makes sense to harness that to help convey the message.

Another decision is the structure - why group the countries together, rather than the causes?

Normally, if the key comparison is between countries, I'd advocate for making that comparison as easy for the reader to see as possible, which this alternate structure does.

And this is a totally sensible way of presenting the data. But in this specific case I think the comparison still jumps out when you group by country *and* it's not so busy - 2 graphs instead of 3.

Finally, why a bar chart at all?

@statsgeekclare suggested a line chart like this.

I certainly don't think this is a bad option by any means, but I didn't chose it for 2 reasons:

1) The busyness argument - 2 graphs, rather than 3

2) The area under the lines (or the area of the bars) represents the number of deaths. Which is what really (sadly) matters. If that area is filled (as in the bars) it jumps out visually more than if it is not (as in the lines). I think.

I could talk more about some of the other design choices – I spent a while trying out versions where age was on the y-axis, or where the two countries were mirrored either side of a common axis, but they didn’t work as well, I don’t think. But this is already a long thread!

The only other thing I want to say is that there was a specific challenge here to try and create something in the style of The Economist. So some of the other, subtle design choices I made were informed by their principle that less is more

And some others by trying to replicate how their (excellent) design team have done things in the past (for example I was tempted to colour the words 'alcohol', 'suicide' and 'drugs' in the corresponding colours in the legend, but couldn't find any examples of them doing that).