How to show OUTLIERS in charts (and how not to)

 
 

Have you ever tried to visualize a set of values only to find that an outlier in the data (i.e., a value that's much larger than the other values) was messing up the chart by making the other values too small to see clearly?

 
 

Now, it's important to note that this chart may or may not be problematic. For example, if the "job" of this chart is to show that Russia is much larger than other European countries, this chart does that job perfectly and doesn't need to be "fixed." If, however, the job of this chart is to say something about how the smaller values compare with one another, then, obviously, this chart wouldn't cut it.

Chart creators typically use one of four methods to deal with outliers, one of which I think is the most effective by far.

BTW, if you want a lot (like, a lot a lot) more tips like these, I'll be delivering my Practical Charts course live online, beginning on June 2nd, 2025! Info/registration: https://www.practicalreporting.com/june-2025-online-workshop Hope to see you (online) there!

The first method is to simply exclude the outlier(s) from the chart, but this isn't ideal:

 
 

The second method is to "break" the quantitative scale, but this isn't a great solution either:

 
 

The third method is to switch to a logarithmic quantitative scale but, again, this isn't ideal:

 
 

Most audiences struggle to read values on logarithmic scales accurately (assuming they know how to read them at all). Logarithmic scales can be useful for certain types of analysis, but generally not for “everyday” charts.

The fourth method, and the one I use myself in virtually all situations, is to use an inset chart:

 
 

When designing inset charts, subtle design details help readers understand how the “zoomed-in” chart relates to “big-picture” chart:

 
 

Note that inset charts aren't limited to bar charts, and can be used to handle outliers in almost any chart type:

 
 

If you're designing charts that will be dynamically generated in a live dashboard/reporting application (Power BI, Tableau, etc.), unfortunately, these applications generally don’t support subtle design details like those shown above. Inset charts can still be used, but the relationship between the two charts will be less visually obvious without the boxes, drop shadows, etc.:

 
 

As I mentioned earlier, if you want more tips like these and to become a true dataviz pro, I'll be delivering my Practical Charts course live online, beginning on June 2nd! Info/registration: https://www.practicalreporting.com/june-2025-online-workshop Hope to see you (online) there!

"Economic Chaos" Sale! 25% off courses and books until May 20th!

 
 

I get it. With the unpredictability that's been injected into the global economy, many of us are hesitant to make any kind of new investments, including new upskilling investments.

That's why I've decided to make it easier to take my courses and get my books by discounting prices by 25% for seven days, from now until May 20th:

My courses and books don't go on sale often, so take advantage of it!

 

New video! Should you be using AI to create charts instead of Excel/Tableau/etc.?

 
 

Generative AIs like ​ChatGPT​, ​Claude​, and ​Gemini​ have been steadily improving their chart-making abilities during the past year or so and now claim to be top-notch chart creation tools. Is this just AI hype, or is it time to switch from Excel/Tableau/etc.?

❓ Is it faster and/or easier to create charts in AI?

❓ Can AI handle the kinds of charts that you need to create? How good are the results?

❓ How much human help does AI need? Can it produce well-designed charts without human intervention?

❓ Are AI-generated charts reliable and accurate?

Having recently put various AIs through their chart-making paces, I can say that the answer to all of these questions is a resounding “it depends.” Sometimes it makes sense to use AI (although there are important caveats to be aware of). Sometimes, however, a tool like Excel or Tableau will be the better choice. When should you use which tool? What are the caveats around using AI? Find out in my latest video (see links below).

One thing that became clear in my testing was that, if you want top-notch charts, you still need top-notch dataviz skillz—even if you're using AI. How can you acquire those skills quickly? Well one pretty good way would be to take my ​Practical Charts course​, which I’ll be delivering in a live online public workshop in June! Info/registration: ​https://www.practicalreporting.com/june-2025-online-workshop​

Watch the "Should you use AI to create charts?" video on...

Why it wouldn’t make sense to adapt Practical Charts to Power BI’s limited visuals

My Practical Charts course is tool agnostic, that is, it doesn’t assume that any particular dataviz software product (Excel, Tableau, Google Sheets, etc.) is being used. There are three reasons for this:

  1. The guidelines for making design choices like choosing a chart type or choosing a color palette are the same, regardless of which dataviz software product is being used.

  2. All the chart types and techniques in Practical Charts are pretty standard and can be implemented in just about any modern dataviz software product.

  3. If a workshop participant is unsure of how to create anything that I show in the course, high-quality tutorials for just about every major dataviz software product are just a Google search or AI prompt away.

You might have noticed that, in those last two reasons, I referred to just about any major data visualization software product. The “just about” qualifier is there because of one unfortunate exception:

Power BI.

In this post I explain why I don’t think it makes sense to change my Practical Charts course due to Power BI’s limitations.

Read more

No, everyone DOESN’T know how to read a scatterplot

When I’m chatting with other chart creators, it sometimes feels like there are two different groups that live in two completely different worlds:

The first world is populated by those who create charts for relatively data-savvy audiences. In this world, chart types like scatterplots and histograms are “basic” chart types that everyone knows how to read.

The second world is populated by those who regularly create charts for “non-data” audiences who often struggle with anything other than simple bar, line, or pie charts.

Read more

Should you avoid using “advanced” chart types? (+ Black Friday Sale!)

I’ve seen the following scenario play out many times in the organizations with which I work:

A chart creator decides to “get creative” by using a histogram, connected scatterplot, ribbon chart or some other chart type that they know to be unfamiliar to the audience. They could have used a simpler, more familiar chart type to say the same things about the data, but they wanted to “challenge the audience,” or “teach them new chart-reading skills.”

The chart then goes over like a lead balloon, however. The audience misreads the chart, skips reading it altogether, or gets annoyed with the chart creator, who then feels bitter, believing their audience to be intellectually lazy or just dumb.

Read more

“I don’t want red on my dashboards. It looks too negative.”

I first heard this objection from a client a number of years ago and it took me so off-guard that I just stared at them and then mumbled something about getting back to them on that. It just seemed like such a bizarre thing to say…

Since that time, I’ve heard this objection at least half a dozen times and have had a chance to formulate a couple of responses that usually convince dashboard users to rethink their “no red” policy.

Read more

Dashboards should only show the “most important” metrics… right?

I regularly hear complaints from dashboard creators that go something like this:

  • "My users consider dozens of metrics (or more) to be ‘KPIs’ and they want me to put them all on the dashboard."

  • "My users don’t understand that, among all the metrics that we could show on the dashboard, only three to five of them are truly important. The rest are just noise that distract from the truly important numbers, and so don’t belong on the dashboard."

Think about that for a second, though. Does it seem plausible that someone could run an entire team, department, or organization based on just five numbers?

Read more