Presentation notes (click triangles to expand)

Best practices summary

Best practice #1: Differentiate data with color or shading rather than with shape.

Best practice #2: The rule of thumb is that for >4 series, you should not rely on color alone for differentiating series. Use direct labeling or some other technique instead.

Best practice #3: Use primarily neutral or natural colors for the main colors when presenting data. I think this avoids visual fatigue from trying to figure out which brightly colored thing I’m supposed to pay attention to.

Best practice #4: Don’t use a color scheme that has an implied meaning, unless that meaning matches the data.

Best practice #5: Even for encoding data, use shades of neutral colors rather than Excel defaults or making up a color scheme. It’s hard to make up a color scheme that is color blind safe but not too bright.

http://colorbrewer2.org is a good resource for getting color blind safe schemes with a variety of colors if you need one

Best practice #6: Remove unnecessary grid and axis lines from graphs. Use gray for remaining grid lines.

Best practice #7: Remove grid lines from tables, and use gray for headings.

Best practice #8: No one should notice the font you use. When I doubt, use Helvetica — while some think it is over-used, it is universally respected by typography designers and is very readable. Be careful with non-standard fonts for presentations.

Best practice #9:

For slides presented in:

Ambient light: light background, dark text

Dark room: dark background, light text

Avoid non-natural/bright colors for the background

Edward Tufte

He’s probably the most famous contemporary data presentation writer/presenter

He’s written some great books

There is a lot of good practical advice sprinkled through his books, but he is sometimes opaque about the reasoning behind his opinions.

Thinking about how we think

I’ve read some other books on data visualizations and presentation, and I really found it helpful to think about how our brains perceive visual information, and how this can be used to optimize the way I present data.

What I’m going to do in this presentation is talk briefly about how our brains work in a grossly oversimplified way that is probably inaccurate, but seems to me to be very helpful for improving how data are presented.

I’ll then go through some practical application of this for presenting data.

Avoiding tiger attacks

Our eyes are drawn to

Moving objects

Uncommon shapes

Bright colors

I assume this was an adaptation to help our ancestors avoid getting eaten by tigers.

This was one of the most helpful things I read about presenting data because there are a ton of implications from this for presenting data, in terms of

How to highlight important information, and

How to avoid distracting or overwhelming your audience

Pretty much everything else in this presentation references back to this slide.

Shapes vs. colors

Good data presentation often requires using visual techniques to distinguish elements of a graph or table.

We know that moving objects are annoying in presentations and impossible on paper

So we’re left with shapes and colors. Which is most effective?

However, we’re not as good at picking out shapes as we are at picking out differences in color.

The Functional Art has a great example of this that I’ve recreated here. I think it makes the previous point self-evident.

One practical application of this is with line graphs: it’s much easier to tell colored lines apart than lines versus different shaped markers.

Best practice #1: Differentiate data with color or shading rather than with shape.

However, there is a limit to this. We’ve all seen a graph with like 50 series in the legend, which is no good. The viewer has to keep referring back to the legend.

This has to do with limits in our working memory.

Best practice #2: The rule of thumb is that for >4 series, you should not rely on color alone for differentiating series. Use direct labeling or some other technique instead.

Choosing colors

I mentioned before that our eyes are drawn to bright colors, or in other words, colors that we don’t expect to see in nature.

Aside: before I started started learning about this, I would have probably made this slide like this (with much brighter colors).

But if you think about it, it makes sense: our eyes are drawn to the potentially poisonous tree frog or, say, flowers, but not the color of the sidewalk.

Tufte says that we should use bright, non-natural colors sparingly. Bright colors should be reserved for highlighting important data.

If you use bright colors for everything, it’s hard for the audience to know what to pay attention to. Bright colors have an implied meaning: pay attention to me!

Here’s an example of a weather graph based on a graph in one of Tufte’s books.

Here’s the same graph but with a bright color rather than a neutral color.

I find the 2nd graph much more fatiguing to look at.

But your graphics won’t be boring because you can still use bright colors to point out important data — especially in slides. (So if you don’t have any bright colors, you don’t have any important data.)

Choosing colors: when summarizing data

There’s a sub-class of the color scheme problem where the colors you use encode a summary of your data.

This already came up with the Seinfeld viewers graph. The color wasn’t helpful in that graph, but in some cases it is. For example, a bar chart with just 3 series.

Best practice #4: Don’t use a color scheme that has an implied meaning, unless that meaning matches the data. In Example 1, the colors imply to me that Site 1 is bad and Site 3 is good. This has nothing to do with the data though.

Ok, so how do you pick a good color scheme for this bar graph?

Example 2: I picked some colors that looked ok together to me. They are bright but not too bright.

The problem is that they are not color blind safe.

Why should you care? There’s a good chance someone looking at your data will be colorblind.

Example 3: Color blind safe colors are too bright

Example 4: Excel default colors are ugly, but they are color blind safe.

Example 5: Tufte uses gray in a lot of the graphs in his book. This lets you differentiate the series but is kind of boring.

Example 6: Shades of blue are “natural” colors, and for 3 series they should be different enough to allow you to differentiate on any screen. Four series might be pushing it.

They are color blind safe too.

And they work well with highlighting important data with a bright color.

http://colorbrewer2.org is a good resource for getting color blind safe schemes with a variety of colors if you need one

Eliminating visual noise

For the final graph in the last set of graphs, you might have noticed I did something different with the borders.

Here’s the default in my graphing software (lots of borders)

And again, here’s what I did

This goes back to the lion in the grass. Lots of extra dark lines make it harder to see the data.

Here’s another series of graphs that shows the whole spectrum from “aggressive grid” to Tufte’s anti-grid

Best practice #6: Remove unnecessary grid and axis lines from graphs. Use gray for remaining grid lines.

This is another thing that I was doing wrong before I started reading about this.

The same goes for tables. I’ve experimented with this and I actually like no grid lines for some tables. I also sometimes like using gray for headings to direct attention to the data.

This is another thing I was doing wrong before I started looking into this: I always did my tables with black horizontal lines, which now I think is wrong.

Best practice #7: Remove grid lines from tables, and use gray for headings.

More on tables

I also realized I was doing some other things wrong with tables.

I think the following are good guidelines for tables:

Words should be left-aligned

Numbers should be right-aligned and rounded so that the decimals align

Heading alignment should match the column (left for words, right for numbers)

You can do some fancy stuff in Excel to get p values to align properly, but you need:

A fixed-width font (like Courier New)

This formula: =TEXT(ROUND(A1, IF(A1<0.01, 3, 2)),"0.00?")

Nagging questions

There are two things I tried to find definitive answers on and could not

Fonts

Serif vs. sans serif? I couldn’t find anything particularly scientific about this. I think as long as the font is readable and not obnoxious, either is fine.

What is readable? Anything that looks normal and isn’t Comic Sans.

What is obnoxious? Depends on who you ask.

Note that if you use a non-standard font, you will need to either present from your own computer or from a PDF. If you bring a PPT file to a computer without your fancy font, your presentation will look bad.

I always present from a PDF for this reason and to avoid issues with compatibility between different version of PowerPoint.

Background colors for slides

Conventional wisdom for PowerPoint is to use either

White background, black text

Blue background, yellow text

This maximizes contrast

I think this was more of a problem in the past when projectors weren’t as powerful as they are now

My problem with the blue/yellow color combination is that the blue is too bright. Now that I know that my audience’s eyes will be drawn to bright colors, I want to use that as a tool in my slides, and I can’t if my background is the brightest color on the screen.

Of course, this all depends on the room and the projector. This is why you should test your slides in the same room with the same projector you’re going to present with.

You should also set your color scheme in your slide masters, so if you need to change it you can do it without having to adjust every single slide.

Best practice #9:

Ambient light: light background, dark text

Dark room: dark background, light text

Avoid non-natural/bright colors for the background

bit.ly/datatips

I’m working on a website on data presentation best practices, so these slides and my notes, as well as a bunch of other stuff are up on there.

If you want to look at the slides again or read more, this is a good place to start.