The bars seem to show decile ranges, until you count the segments and realize there are only nine per bar. It turns out that each bar connects ten values, the average incomes within each decile. This strikes me as a little odd: I think if you had to leave off the extremes, perhaps taking the 5th, 15th, 25th, . . . and 95th percentile values might be a bit more natural. But maybe it’s a matter of taste.

The data originated with the folks who bring us the OECD Factblog, by the way. The data is available in an Excel workbook at Growing Unequal? Income Distribution and Poverty in OECD Countries.

Data in hand, I decided to slice it and dice it. My first impression was to look at cumulative distributions. I assumed that the value for each decile was the value for the middle of that decile, so I used 5% for the first decile, 15% for the second, etc. There were a lot of lines cluttered together, so I made myself an Interactive Multiple Line Chart with a listbox to highlight one country at a time. I have highlighted the US in each of these charts.

I did not include the protocol for creating this interactive chart, but if I get more than two comments asking for it, I’ll be glad to oblige.

Using linear values on the horizontal scale tends to compress the lower values together. I plotted the log of the income data to spread out the data. On a log scale, an equal distance on the chart relates to an equal percentage difference of the values.

The rightmost gray line is Luxembourg. It is steeper than the line for the US, meaning there is less disparity between the lowest earners and the highest. Most of the other curves have a similar slope (similar percentage disparity between low and high). The US and a few others are less steep, meaning they have a greater difference between the poorest and richest earners.

I converted the percentile values on the vertical axis to their Z-scores. This straightens out most of the curvature at the ends of these S-shaped curves.

I took this one step further, and computed linear regression coefficients for each country’s distribution using Z-scores as the independent variable and log income as the independent variable. The computed intercept provides a rough approximation of the median income, while the computed slope is related to the disparity between highest and lowest income.

The median income is shown below left and the income disparity below right. If you believe that higher income and less income disparity is good, then it’s better to be located toward the top of each of these charts. Luxembourg is close to the top of both charts; Mexico, Turkey, and Poland are near the bottom of both; and thee US is mixed, near the top of median income and near the bottom of income disparity.

I like XY charts, so I plotted Income Disparity against Median Income. The data shows a negative trend in the bottom right corner is Luxembourg, while in the top right are Mexico, Turkey, and Poland. The US is the furthest above a diagonal, meaning its combination of high median income and high income disparity is unmatched. To reduce clutter, I removed the labels from the data points in the middle of the chart.

While this analysis is interesting, none of these charts gives a really good overview of all of the data at once. The cumulative distributions give a sense of the data, but it is not realistic to try to label all countries on a single chart. Ironically I drifted back to the stacked bar representation that The Economist showed. With less distracting background, of course.

In Kaiser’s post, Chris Jackson mentioned his paper in The American Statistician, Displaying Uncertainty With Shading, which essentially applies a gradient to a bar that relates to thee density of data at each point along the bar. I decided to apply this approach, but without a gradient that fades too severely at its ends (too reminiscent of the conditional formatting data bars in Excel 2007). I applied his approach by applying lighter shades to the outer deciles and darker shades to the central ones. This worked reasonably well, considering that I made no effort to develop a color palette for this purpose.

Here is my own rendition of The Economist’s chart.

This has the same problem with the lower deciles being compressed. But having just plotted log income in my cumulative distribution plots, I decided it would be worthwhile to apply logs to this stacked bar chart.

This is a more effective display of income disparity than the linear version put forth by The Economist, but it assumes that the audience understands Logarithmic Axis Scales. The median income is clearly identified by the lateral position of the bars, and the disparity of income is directly related to the total width of the bars.

Federal Spending on Education and Training

The first compares two years of Federal education spending data using side by side pie/donut charts, which commits two sins: using one pie/donut charts, and using more than one of these charts. Tony recommends using a table instead, while I’ve created a line chart instead. Tony’s table and my chart contain all of the data from the Business Week charts, with additional display of percentage changes.

The Original Business Week Donut Charts

Tony’s Tabulated Data

The Peltier Tech Line Chart

Tony’s table is certainly the most compact display of the data. The Business Week dual donuts are eye-catching, perhaps, but the line chart shows the trends best of all.

Percentage of Children Living in Poverty

The second chart is a simple bar chart of poverty rates by state, ranked by the rate of high school graduation in each state. Tony likes this chart, and I don’t think it’s too bad either, except that it’s divided into three sections because of its aspect ratio.

Before merely duplicating the chart above, I thought the individual graduation and poverty rates deserved some investigation. I think a heat map on a geographical state map would be interesting.

Here is my rendition of the Business Week chart, differing only by not splitting mine into three panels. It’s pretty tall, but it tabulates the data clearly. I say “tabulates” because it seems to show a trend between graduation and poverty rates, it does not allow an easy determination whether there is a correlation between these rates.

Being the geek that I am (did I say I used to be a scientist?) I decided to plot the two variables against each other. As expected, there’s a general inverse relationship. As poverty goes down, high school graduation goes up. I plotted it both ways, because there’s only a correlation, and I’m sure there are conflicting models for which variable causes the other.

I picked the latter XY chart, because I think that economic stress leads to higher drop out rates, and replaced the markers with labels showing the state abbreviations. I enlarged the chart significantly to reduce overlap of state abbreviations.

This chart is more cluttered that I would like because of the overlapping state labels, but I think it’s better to show correlations between measurable quanitites, rather than to simply list them in a table, or in a chart that is essentially like a table. What do you think?

JPG is for photographic images, and for art with continuously varying color gradients. PNG is for line art, screen shots, text; in short, for most anything computer-related. Even the ancient GIF format is usually better than JPG.

To the rescue comes Louis Brandy. In My First and Last Webcomic Louis cleverly shows the difference between the two image formats, in xkcd style.

Thanks to Karl Peterson for the link.

Nathan of Flowing Data first showed me the electoral cartograms, and also led me to the following approach.

Andy Woodruff has written about an approach Axis Maps has developed which does not distort the geographical geometry of the map, in A new kind of election map and also in I hate your favorite election map in his personal blog. This visualization approach keeps the map undistorted, but uses shades of red and blue to shade counties according to their population.

This is an interesting approach. I have had some difficulties interpreting this map (here is a larger version). First, the preponderance of black distracted me. I had the feeling of floating in space above disconnected objects.

Second, when I see different shades of color in a map, my first inclination, ingrained by staring at thousands of maps, is to interpret the different shadings as a measure of some sort of density. Population density, perhaps, or comparative strength of the vote in that region for one candidate or the other.Using color for total population of a county is integrating population density over the square miles of the county.

Another issue is comparing brightness of different regions. If two regions are abutting, and surrounded by a uniform third color, it is easy to compare their shades. If however the two regions are located at a distance and surrounded by different swatches of colors (say, Detroit and the region NE of Los Angeles in the map above), then I challenge anyone to compare their shades accurately. In Practical Rules for Using Color in Charts Stephen Few illustrates the difficulty of comparing colors of regions that are surrounded by different shades.

In general, I’m not opposed to this visualization approach. I think with a little development it may actually be practical. What we should keep in mind while comparing all manner of visualization techniques is what Andy said to me, when I described my difficulties with these charts, both area-distorted and color intensity-distorted, is that:

1. It’s a matter of taste and other personal factors.
2. In the end, neither map is really going to provide observations more qualitative that “there’s more blue in this area than that area.”

I could not have said it better.

Update:

As I ruminate on these charts, I have thought of a couple enhancements that may help make them more instantly comprehensible: Use a white background instead of black, and draw in the state boundaries, especially if the map shows separate shadings down to the county level. Perhaps Andy may humor me and toss out a couple samples?

The states won by Republican candidate John McCain seems to dominate the regular map, even though Democratic candidate and president-elect Barack Obama controlled the Electoral College results.

But the map above is based on area, not on the amount of voters. The Northeast and California have relatively high population densities, while many of the rectangular states in the middle of the country are less densely populated. Mark creates “cartograms” which distort readjust the areas of geographical units to reflect other quantities besides land mass area (”distort” is such an ugly word, normally reserved here for 3D charts).

The cartogram below adjusts the areas of the states so each is proportional to the population of that state. Notice how most of the red states in the middle of the map have shrunk, particuarly in the northern plains. The Northeastern stats and to a lesser extent California have expanded, and the colors on the map match the election results. My boyhood state of Rhode Island is larger than Wyoming (even DC is larger than Wyoming), the Dakotas, and Montana. If Alaska were on this map, it would also be dwarfed by Rhode Island.

Pretty cool. The shapes of the redrawn geographical units are still recognizable, and you come away with a better sense for distribution of votes. On his web site, Mark shows more views of the results of this and previous elections, as well as cartograms based on other measures.

Panel Chart Data

The data used for the chart is shown below. This first table contains data to be plotted in two XY chart series. The lines stretch across category labels at X values 1, 2, 3, 4, and 5. I arbitrarily selected ±0.33, which produces the pairs {0.67, 1.33}, {1.67, 2.33}, etc. In fact, these values are not hard coded. Outside of the range shown here I have the X values for the category labels {1, 2, 3, 4, 5} and the value 0.33. Formulas produce the X values; if I want a change, I can merely change the vaue from 0.33 to something different, and all of the X values update. Blank rows in the data will produce gaps in the otherwise continuous series. For example, a line is plotted between 0.67 and 1.33, a gap is present between 1.33 and 1.67, a line is drawn between 1.67 and 2.33, and so forth.

The data for my dummy category axis labels is shown below. I will make a simple column chart with this series. Excel will place the C1 to C5 labels at X=1 to 5. The Y axis range is blank, so the column chart will have bars with zero height. (I could have used zero values as well.)

Building the Panel Chart

Start by selecting the axis data and creating a column chart. I temporarily inserted values of 1 so it’s obvious what is going on. I’ve also already changed the axis and plot area lines to gray.

Clean up the chart. Since we’re concerned with relative trends, and because these are tiny little charts, I have removed the Y axis labels. I also removed the title and the X axis tick marks, and moved the legend to the top of the chart.

Now I’ve removed the “1″ values in the axis data range.

Add the brand 1 and brand 2 data. Copy the range, select the chart, and use Paste Special to add the data as new series, with categories in the first column (but not replacing the existing categories) and series names in the first row. These series are added as more column series.

Select one of the added series, and change it to an XY type (right click on it, choose Chart Type or Change Chart Type, and select the type from the dialog). Excel helpfully (NOT!) adds secondary axes, because nobody could possibly want XY and column series on the same axis. Except us.

In Excel 2003 or earlier, immediately select the other added series, and press the F4 function key to repeat the previous action. In Excel 2007, the F4 key has lost most of its former glory, so you’ll have to use the right-click approach to change it to an XY type.

Select one of the XY series (’brand1′ below) and format it so it is on the primary axis.

Select the other XY series, and in Excel 2003 or prior, click F4 again. Excel 2007 users, you know the drill.

Well, that’s lines up the way we want. Let’s tidy up, shall we? First, select the legend entry for the ‘axis’ series (the text label, not the colored rectangle), and press delete.

Add vertical (X-axis) major gridlines (Chart menu > Chart Options > Gridlines tab in Excel 2003, one of the contextual Chart Tools tabs in 2007). Format them the same gray as the border.

Finally, use formatting to make the brand data stick out a bit. I used brighter colors and a thicker line.

Not too tricky. We needed a little hint about data arrangement, and one dummy series to handle the X axis labels.

In Visualization Challenge - How to show market share changes? our colleague Chandoo presented the following data and challenged his readers to display it meaningfully. The data shows market share for two brands and five competing retailers, at two times. The challenge was to show the trends over time. I took on this challenge in Challenge - Show Market Share Changes. While working on this challenge, I realized that the data made for a good example of the power of pivot tables for fast and detailed analysis.

Pivot Table Data

I’ve rearranged Chandoo’s sample data into the form required for an effective pivot table. There are no blank rows or columns, the first row has headers (field names), and there is one row (record) per observation.

Pivot Tables and Charts

To create a pivot table, select all of the data or a single cell within the data (Excel will try to use the contiguous range surrounding the selected cell), and from the Data menu or the Insert tab in 2007, choose Pivot Table. To include a pivot chart, you can choose the appropriate option in the dialogs, or you can select the pivot table and insert a chart, or (and this can be dangerous) you can select an existing chart, edit the source data, click in the Data Range entry box, and click in the pivot table. This last operation is dangerous, because it converts the chart into a pivot chart, and pivot charts cannot be reverted to regular charts. Of course, it is possible to create Regular Charts from Pivot Tables.

The pivot chart is joined to the pivot table at the hip. Any changes to the pivot table are reflected in the pivot chart, and vice versa. For clarity, I have hidden the pivot buttons from these pivot charts. They are redundant since the pivot table and chart are so close together, and they consume valuable chart real estate.

This first arrangement shows each competitor as a different color series. This is fine, but there is too much spacial separation between the two times to allow a good assessment of the trends.

This is easy to fix with a pivot table. You can click and drag the gray field buttons around to change the arrangement instantly. So I dragged the time field inside the product field. The times are closer, but the trends are still not so easy to discern.

Okay, maybe the competitors shouldn’t rank their own series. I dragged the competitor field down to the rows area and the product field up to the columns area. Hmm, the times are again too far apart.

Click, drag. Okay, the times are pretty close now, but the comparison between brands is given more emphasis.

So if we swap the time and product fields, we put the data for the two corresponding time periods next to each other. This view compares the competitors for each brand…

… while this view compares the two brands for each competitor.

The Ultimate Chart

This analysis gave me just enough insight to know how I should build the chart I answered Chandoo’s challenge with. It was not a pivot chart, nor was it based directly on a pivot table, but it utilized the insights from the analysis above. I created a panel chart with each competitor in its own vertical panel, and with lines connecting the data.

I’ll show the particular data arrangement and protocol I used to create this chart in an upcoming post. Stay tuned.

More About Pivot Tables

For more about pivot tables, check out these web pages, mostly from my site:

• Pivot Tables and Pivot Charts (on the PTS web site)
  Originally contributed by fellow Excel MVP Debra Dalgleish.
  Check out Debra’s Excel Tips page and Pivot Table and Pivot Chart FAQs.
      - Introduction to Pivot Tables in Excel
      - Working with Pivot Charts in Excel
      - Pivot Table Programming
      - Pivot Table and Pivot Chart Links
• Pivot Tables (related posts on this blog)
• Pivot Tables, Pivot Charts, and Real Charts (TechTrax Article)
• Books on Excel Pivot Tables
      - Excel Pivot Tables Recipe Book: A Problem-Solution Approach by Debra Dalgleish
      - Beginning PivotTables in Excel 2007: From Novice to Professional by Debra Dalgleish
      - Pivot Table Data Crunching by Bill Jelen and Michael Alexander
      - Pivot Table Data Crunching for Microsoft Office Excel 2007 by Bill Jelen and Michael Alexander

Chandoo came up with a pair of stacked charts, with the stacked columns staggered laterally.

Perhaps the staggering and the connecting lines make it a bit easier to discern the edges of the different colored bars than in an unstaggered stacked column chart. But in general the charts have some deficiencies (these are my versions of Chandoo’s charts, by the way). First, the stacking of bars make it more difficult to judge the magnitude of the segments between top and bottom, and having two charts forces the eye to move too much to see all the data. The legends take up a large amount of space, though of course duplicate legends are overkill.

Here is Chandoo’s sample data, if anyone wants to play along.

I approached this challenge by first investigating some different chart types. The stacked column chart is less cluttered looking than the staggered charts above, but they don’t really improve on the spatial difference between the two brands, and the large space needed by the legends (we could delete one of the legends, of course).

How about a Clustered-Stacked Column Chart? Below is the data arrangement and resulting chart. I’ve colored the data range using the scheme that Excel uses to highlight data when a chart series is selected: the blue range contains Y values, the purple  contains category labels, and green denotes series names. The two columns of category label data produces the two-layer labels (see Chart with a Dual Category Axis). To preserve the lateral spacing, the purple cells with the hatching pattern contain spaces. The rest of the cells in this table are blank.

The separation of brands and the excess real estate required for the legend(s) are both improved, and it sure shows off a clever data arrangement. But the values and trends are no easier to compare than in the other chart types above.

Since we’re interested in trends, a line chart may help with comparisons. Labeling the series directly is less obtrusive than the legends, and more effective at identifying the data. The charts still look somewhat cluttered, and the slopes are all rather too shallow, flattening out the trends.

Perhaps clustering the data instead of stacking it can provide a better perspective (see Stacked vs. Clustered). The trends are visible even though the charts have bars instead of lines. The legends are less obtrusive; that’s not an intrinsic property of clustered charts, but rather a consequence of having fewer series. The separation of the two brands still makes comparisons difficult.

Maybe we can combine the good features of the line and clustered column charts. Enter the panel chart. The two charts below show the values and trends clearly, without a lot of clutter.

We still have the separation of the brands in the two panel charts above, but the openness of the panel chart allows us to display both brands in the same chart. This is pretty effective, perhaps the best of all the types I’ve reviewed.

Does anyone have any better ideas? Are there better approaches that I’ve neglected?

In the next few days, I plan to write a set of instructions for building this chart. In addition, I thought this data would be a good vehicle to show a little exercise in data exploration, using pivot tables and pivot charts. I’ll comment on this post when I’ve added each of these pages, so if you subscribe below (comments are always welcome, but not needed to subscribe) you’ll receive a notification when I’ve updated.

Update 16 November 2008

I’ve written a couple of follow-up posts. In Explore Your Data With Pivot Tables I used this data to show how pivot tables can be used to investigate the data quickly and easily. Then in a second follow-up I showed How to Build a Simple Panel Chart.

Mike has written Excel 2007 Dashboards and Reports For Dummies, which is among the most comprehensive Dummies books I’ve come across. The book is geared towards Excel 2007, but its principles and techniques are valid for any version of Excel. Mike takes his readers through dashboard principles, data preparation and chart creation, advanced data techniques including pivot tables and dynamic ranges, automation of dashboards using VBA, and designing interactive dashboards.

Like all Dummies books, Excel 2007 Dashboards and Reports For Dummies has a “Part of Tens” chapter, including Ten Chart Design Principles, which Mike will write about in this post. While Mike has written these principles, they are in agreement with principles that other graphing and visualization experts would have presented. Without further ado, Here’s Mike:

Ten Chart Design Principles by Mike Alexander

Excel makes charting so simple, it’s often tempting to accept the charts it creates no matter how bad the default colors or settings are. But I’m here to implore you to turn away from the glitzy lure of the default settings. You can easily avoid charting fiascos by following a few basic design principles:

1. Avoid Fancy Formatting
2. Skip the Unnecessary Chart Junk
3. Format Large numbers where possible
4. Use Data Tables instead of Data Labels
5. Make Effective Use of Chart Titles
6. Sort your data before Charting
7. Limit the use of Pie charts
8. Don’t be afraid to parse data into separate charts
9. Maintain Appropriate Aspect Ratios
10. Don’t be afraid to not use a chart!

1. Avoid Fancy Formatting

• Don’t apply background colors to the Chart or Plot Area.  Colors in general should be reserved for key data points in your chart.
• Don’t use 3D charts or 3D effects.  No one is going to give you an Oscar for special effects./p>
• Avoid applying fancy effects such as gradients, pattern fills, shadows, glow, soft edges, and other formatting.  Focus on the data and not shiny happy graphics.
• Don’t try to enhance your charts with clip art or pictures.  Not only do they do nothing to further data presentation, they often just look tacky.

2. Skip the Unnecessary Chart Junk

• Remove Gridlines
• Remove Borders
• Skip the Trend Lines
• Avoid Data Label Overload
• Don’t Show a Legend if you don’t have to
• Remove Axes that Don’t Provide Value

3. Format Large numbers where possible

When plotting very large numbers on a chart, you should consider formatting the values so that they are truncated for easy reading.

For instance, in this chart, I’ve formatted the values to be displayed as 10M and 17M instead of the hard-to-read 10,475,000 and 16,906,714.

4. Use Data Tables instead of Data Labels

A data table allows you to see the data values for each plotted data point, without overcrowding the chart itself. Although data tables increase the space your charts take up on your dashboard, they respond well to formatting and can be made to meld nicely into your charts. Data tables come in particularly handy if your clients are constantly asking to see the detailed information behind your charts.

5. Make Effective Use of Chart Titles

You can use chart titles to add an extra layer of information, presenting analysis derived from the data presented in the chart.

6. Sort Your Data Before Charting

Unless there is an obvious natural order such as age or time, it’s generally good practice to sort your data when charting. By sorting, I mean sort the source data that feeds your chart in Ascending or Descending order by data value.

7. Limit the use of Pie charts

• Pie Charts typically take up more space than their cousins the line and bar charts.
• Pie charts can’t clearly represent more than two or three data categories.
• Bar Charts are an ideal alternative to Pie Charts.

8. Don’t be afraid to parse data into separate charts

A single chart can lose its effectiveness if you try to plot too much data into it. Step back and try to boil down what exactly the chart needs to do. What is the ultimate purpose of the chart?

9. Maintain Appropriate Aspect Ratios

A skewed aspect ratio can distort your charts, exaggerating the trend in charts that are too tall, and flattening the trend in charts that are too wide.

Generally speaking, the most appropriate aspect ratio for a chart is one where the width of the chart is about twice as long as the height is tall.

10. Don’t be afraid to not use a chart!

You typically use a chart when there is some benefit to visually seeing, trends, relationships, or comparisons.

Ask yourself if there is a benefit to seeing your data in chart form. If the data is relayed better in a table, then that’s how it should be presented.

Jon’s comments on Mike’s top ten.

You didn’t think I could let this go, did you? As I noted in the introduction, I concur with all of Mike’s points, but there are a couple of things that would make Mike’s example charts even better.

2 and 3. The suggestions are spot-on, but the charts can be further clarified. Reduce confusion in the chart by maintaining a proportional axis scale (XY not Line chart), so the gaps between 2000 and 2003 and between 2003 and 2004 don’t look like the five year gaps earlier in the chart. This makes the upswing in the last few years even more striking.

6. This is a good practice. However, you should sort by category if it makes sense. The example in item 7 should be sorted in order of increasing age. In this kind of plot, it also is important that the bins (i.e., age ranges of each category) be of equal width. If this is age of drivers, then the first age range should be something like 16 to 24, which becomes by far the largest value in the chart.

7. Jorge Camoes showed the best collection of pie charts in The Best Pie Charts Come From Germany.

8. This is a nice example of the use of what Edward Tufte calls small multiples. Small multiples is part of the basis of panel charts.

9. Adjust the aspect ratio of the chart and its X and Y axis scales so the “average” slope is around 45°. This is described by Kelly O’Day in Banking to 45° to Enhance Visualization.

10. This was discussed by Tony Rose last week in Best Method for Illustrating a [Single] Data Point.

Among Mike’s other books are Pivot Table Data Crunching for Microsoft Office Excel 2007 (with Bill Jelen), Microsoft Excel and Access Integration: With Microsoft Office 2007, Microsoft Access 2007 Data Analysis, and Excel 2007 VBA Programmer’s Reference (with John Green, Stephen Bullen, and Rob Bovey).

This linked-picture trick is related to Excel’s Camera tool, which means two things. First, it’s pretty cool. Second, using it more than a few times will consume resources and slow down the workbook’s response. If you are linking to one picture of many, this is the way to do it, but if you merely want to show different set of data in your chart, it’s more than you need. In this post I will show you an easy and resource-friendly way to achieve the same effects in your dynamic report.

The Report Layout

I’ve set up my report as shown below. Cell F2 contains the selected quantity to be plotted, and I’ve named the cell “WhichChart”. The original data is in B17:E25, and a second data range linked to the first is in B28:E36. The borders around parts of the range B28:E36 show the chart source data range of the working chart, which I will discuss shortly.

One or both of the data ranges can be moved to a different place on this sheet, or onto another sheet, so that the focus is on the chart itself.

Data Selection

To make the selection of a chart more flexible, I changed the data validation list source of cell F2 (”WhichChart”) so it refers to the range C17:E17 rather than a hard-coded list (see the dialog below). This way, if I change one of the headers, I don’t need to update the data validation list.

I generally prefer using a dropdown control or a listbox. A dropdown always shows the drop down arrow icon, so the user knows he can choose something even when the cell is not selected. A listbox shows all options without having to go to all the trouble of dropping down a list. But data validation is pretty easy to use, so it’s really a matter of choice.

The Dynamic Chart

I then constructed a chart that contained all of Chandoo’s data and the formatting from Chandoo’s three charts. I used B18:B25 for the category (X-axis) labels, C17:E17 for the series names, and C18:E25 for the series Y data. To make the chart title dynamic, I selected the title, typed = in the formula bar, then selected the range C17:E17, and pressed Enter. The chart title then displayed the formula =Report!$C$17:$E$17. When multiple cells are selected like this, Excel concatenates the cell values, inserting a space between non-blank items.

The Dynamic Data

The second data range is linked to the original data, with a twist. It compares the column header to the selection in F2 and shows or hides the column as appropriate. I selected the range C28:E36 with C28 the active cell, entered this formula, then pressed Ctrl+Enter to enter the formula into the entire selected range:

=IF(C$17=WhichChart,C17,"")
 

This hides the headers and Y data for the non-selected column of data. I changed the source data of the chart so that B29:B36 (purple outline) is used for the category (X axis) labels, C28:E28 (green outine) for the series names, and C29:E36 (blue outline) for the series Y data. I also changed the title link formula to =Report!$C$28:$E$28. The outline colors correspond to the colors that Excel highlights these ranges when the chart’s plot area or chart area is selected. When Sales is selected, the data range looks like this (with only the Sales column displayed):

and the chart looks like this:

Final Touches

This isn’t quite perfect yet, but it’s easy to finish. First, apply a custom number format of 0;;; to the data labels in the chart, so the zero values (corresponding to the blank cells) are not shown. Second, format any one of the series in the chart so that it has an overlap of 100, so that all columns overlap, and there is no horizontal space wasted for columns with no data. Here is the finished chart:

Download this dynamic chart (in a zipped workbook).