Following the 10 x 10 post series on tips for better charts, these are the 10 tips for line charts:
- Don’t use line markers unless you really need them to identify b&w printed charts;
- Don’t use a legend; directly label the series, instead;
- If you can’t easily see the pattern of each series you may have too many;
- In a time series, the spacing between markers in the x-axis should be proportional. For example, if you have data for years 1980, 1990, 2000 and 2008, the spacing between 2000 and 2008 should be smaller than between other dates; if you can’t do it with line charts use a scatter plot;
- If you are comparing two series like imports/exports or profit/expenses, chart the differences, not the actual series (or at least add a small chart with the differences, below the main chart;
- If you are comparing two time series with very different units of measurement, consider using a logarithmic scale;
- You don’t have to start the Y-axis scale at zero; break the scale if you need;
- If you are using different line styles you may be emphasizing some series more than the others; make sure that’s consistent with your users needs (emphasize what is important);
- Add a trend line (make sure the trend is plausible…);
- Don’t use line charts for categorical data; if you need a profile chart use a scatter plot and switch axis.
Feel free (and I would appreciate) to comment or add your own tips…
I know, I know, no one likes pie charts, but I can’t ignore them. A pie chart compares proportions but it is of limited use: either the data is too complex and a pie chart can’t handle it, or it is too simple and you should just use a table. So, the first tip should be:
- Do you really need a pie chart?
- Pie charts shouldn’t be compared (comparing market shares in two regions, for example);
- Don’t use the “exploded” option;
- Five is in general the maximum number of slices you can use in a pie chart, but two is better…;
- If there is no other meaningful order, order the slices from maximum to minimum;
- Put “other” in a gray slice;
- Don’t use a legend, just label the slices;
- Use a very small pie chart in a supporting role for a more complex chart;
- Use the appropriate color codes to identify groups of slices;
- Start the first slice at 0º (noon);
I am sure you can come up with some ideas to make a better pie. Please share your receipt in the comments.
This is the second of 10 posts where I’m listing tips for better charts. Please take a look at the first post where the project is discussed. These are my chart formatting tips:
- Use the right chart type for the data and the problem;
- Apply sound design principles;
- Use color strategically: mute axis and grid lines by graying them out; gray out some contextual data also; use soft colors; use saturated colors sparingly and with a clear purpose of emphasis;
- What the users see is not what you see in your monitor: if needed, test for other monitors and output formats (b&w print, color print, PDF, overhead projector);
- There is no rational justification to use pseudo-3D charts and other dubious effects (gradients, glow…), so never use them if you what to be rational;
- Use a clear font;
- Don’t emphasize everything (for obvious reasons);
- The y axis scale should start at zero; this is particularly important if you are using bar charts; make sure you have a good reason to break this rule;
- A chart is not a table: by labeling every single data point you make it harder for the user to search for trends or patterns; if you have to, place the labels where they can do no harm;
- Annotate: Add labels for the last, the lowest, the highest or any other relevant data point; add data or comments where appropriate;
- Bonus tips: Use smaller charts and never accept the Excel defaults;
What are your best chart formatting tips? Please share them in the comments!
This is the first in a series of 10 posts where I’ll suggest a (hopefully) coherent set of tips to improve our charts and, more important, to improve the way we make sense of the data. These are the planned posts:
- General charting;
- Formating;
- Column/Bar charts;
- Line charts;
- Scatterplots (XY charts);
- Pie charts;
- Other chart formats;
- Dynamic charts;
- Dashboards;
- Miscellaneous tips;
- Bonus post: online resources.
These will be very short tips, and I’m sure some of them will require further explanation. In time, all of them will be properly linked to a more detailed post (that will keep me off the streets for a while…). This is a kind of road map for the next few months and, when finished, a (partial) table of contents for the blog.
So, general tips on charting:
- A chart shows trends, patterns, outliers; if you already strive to make them apparent, you don’t need to read the next 99 tips…;
- Do you really need a chart? Sometimes the task and the data suggest another method of data analysis;
- Know your audience. If your audience is uncomfortable with some formats your message will be lost;
- Make sure you have enough data to create a pattern (two data points are not enough to create a trend line);
- Make sure you don’t have more than enough data: just because you have it, you don’t have to show it…; keep removing interesting data until only relevant data for your problem remains;
- A chart should be able to answer elementary, intermediate and global questions regarding the data;
- Don’t assume that the charts you see in the media are the ones you need to run your business;
- Learn how to lie with charts and, of course, avoid those lies;
- Let the reader see related charts simultaneously;
- Chart overload is as bad as information overload.
It’s your turn, now. What are your best tips in this category? Please share them in the comments (if you have specific tips please save them for the next posts). I believe that Tufte’s principles (avoid chart junk, maximize data/ink ratio, high data density…) are already implicit in some of these tips.
The relevance principle means that every variation should carry a meaning, derived from data variation, not from design variation. If it doesn’t, it can be confusing or misleading.
Suppose chart A displays population density by country. “Vary colors by point” is an option in Excel, but why should you use it? This is a design variation that confuses the reader and he’ll try to find a key for the color code (there is none, but he doesn’t know).
There may be some exceptions to this rule:
- When there is a strong association between color and entity, and that association is well known by the audience (for example, in sports);
- When there is a grouping variable (in chart B we are defining three groups);
- We want to emphasize a point (like in C, “my country”), although Kosslyn advises against it.
Another case of an spurious variation is discussed by Edward Tufte in The Visual Display of Quantitative Information:
The number of information-carrying (variable) dimensions depicted should not exceed the number of dimensions in the data.
What does this mean? Well, this is something that you see in the media all the time and is very easy to understand through a visual example. On the left you see growth of monitor sales (this is mock-up data). Regarding height, the proportions are correct, but since you want to maintains the aspect ratio, there is also a variation in width. But because the object is perceived as a whole, our perception tells us that the sales growth is much higher than in reality was: growth rate between 2003 and 2006 is 150%, but the image for 2006 is six times bigger!
These two examples are grouped under the same principle but, while the first one is only confusing, the second one can be dangerously misleading, and that is one of the mortal sins of information visualization.
You can’t write a novel just because you can type. You can’t create a chart just because you know how to do it in Excel. First, you have to know the job, then the tool. Research for best practices in your field. Read what some authors have to say about specific formats and options. Then, build a framework and let it guide your information display needs.
(And, of course, never use Excel default options…)
[Update: designers have a similar problem: software doesn't make design easy...]
How to create better charts? Search the web and you’ll find many specific advices, not always backed up by scientific evidence (can there be any?). Tufte’s advices are great for us, rational, positivist members of the human race, but what about those emotional poor fellows for whom a minimalistic chart is just a boring chart?
Can we remove personal aesthetics from the equation? Probably not, but we can minimize it. And if you have a set of basic principles that acts as a framework and guides you through the selection and design of charts you’ll end up getting a more efficient display than just relying on your preferences of the moment.
This six-part series focus on the generic design principles of simplicity, consistency, compatibility, congruence, relevance and conventionality (there will be a separate post for each one). These principles are defined by Michael Schiff in his master’s thesis, “Designing Graphic Presentations from First Principles” (you can get the PDF file here).
Let’s start with simplicity…
The simplicity principle states that a “simpler” chart will be easier to understand. By “simpler” it means fewer types of objects and properties used to encode information. This can be linked to Tufte’s minimalistic approach (the data/ink ratio, maximization of data density, no “chart junk”…) but extends beyond that, while having a more abstract nature.
What happens when you apply this principle to the well known Excel chart defaults? You can see it on the left: after an extreme makeover you can really see the data on the second chart. There is a quote by Michelangelo that expresses quite well what happened: “I saw the angel in the marble and carved until I set him free.“
Note that each of the formatting options has a different nature:
- The gray background is pure chart junk and must be removed;
- The decimal place on the Y-axis gives us an irrelevant illusion of precision that doesn’t make sense; you can also remove the percent sign;
- The gridlines are supporting actors that you can leave muted in the background;
- Removing the legend and replacing it with direct labeling of each series has a deep impact on the user experience: there is no need for the movement of the eyes between the data and the legend and you can free up your working memory;
As you can see, the simplicity principle alone can improve dramatically your chart message. Please remember: a chart is not a product, a chart is a delivery boy.
Suppose you are sharing a list of orders with some co-workers. One of them wants to see the higher sales orders [list]. Another one wants to know how much was exported to France [table]. The next one needs the average items per order [descriptive statistics]. You want to see the growth trend for several products [chart]. Only one of the tasks requires a chart. So, be sure to understand the nature of the task and then select the right tool.
A chart is just one of the available tools to communicate and help you and your audience to understand the data; sometimes using a chart is just plain wrong: if variation seems random or non-existent, what’s the point of displaying the data graphically (yes, I know, sometimes that’s what you want to show)? So, take a look at the available tools and ask yourself: Do I really need a chart?
