Pies fail to deliver

The Wall Street Journal reported that the Ritz-Carlton brand of hotels has been hit worse in the slump than other brands in the Marriott family, and has recently launched a loyalty program as a result after holding out for a long time.

The following serving of pie charts shows the occupancy rates in the past three years. In the second layer of charts, I removed the data from the chart in order to show why this chart is not self-sufficient. Without the data printed directly on the chart, it is difficult to read the individual occupancy rates; and it is even harder to figure out that the decline was worse on the Ritz-Carlton brand.

A line chart brings out the message clearly and directly.

Comments

[dan l]

Oh that's bad. Using pies in a series like that always seems like it could work in the very early conception stage. After about 3 or 4 times executing it, I've learned that it just doesn't go.

[Mike]

Great point! I don't know why pie charts are as popular as they are when they only serve a very specific purpose. They get abused the most often. As one comment I would say that the line chart may be better served to begin at zero in stead of 60%. From my experience most people, at first look, might assume the line goes down to almost zero in 2009.

[Rhett]

The author of this post is correct: the line chart brings out the message more clearly. but, the pie chart series look better- it looks more robust and comprehensive.
! LOoK aT aLl thE DIfFERenT SHaPes OF GreEn! InTeLLigencE!

Sadly, it is not about what it is. It is not about what information it portrays. It is about how it looks.

Someone should be held accountable for this riff-raffy dross!

[Facebook]

I'm actually not a huge fan of the line chart either. With the axis set from 60 thru 74% the graph makes it look like there was this plumet from nearly full to nearly empty when there was only about a 10% change in occupancy.

[dan l]

----
Sadly, it is not about what it is. It is not about what information it portrays. It is about how it looks.
----

Thank you, Michael Bay.

[Rob Easterling]

I, too, do not think the line chart with the suppressed origin is a good way to convey the occupancy information. For one thing, the connected lines make it look like you could interpolate, say to July 2008 occupancy rates. Also, the wispy lines and small-font labels don't have much visual impact. I'd do a bar chart and with the vertical axis ranging from 0 to 75% or 100%.

[Kaiser]

Facebook, Rob, Mike: for a line chart, we do not always have to start the axis at zero. It depends on whether we want to draw the reader's attention to the relative values or the absolute values. For this data, the issue is the change from year to year, how much it has declined so the previous year's number provides the benchmark. A drop of 10% from a base value of 72% is very significant, as the chart shows. Conversely, if one starts the chart at zero, it sends the impression that the drop is not important but it is the wrong perception because the occupancy rate is just never going to dip below say 50%... the block of empty space that would be produced at the bottom of the chart is meaningless.

I'll end up writing a separate post on this because it's something that is brought up constantly.

What would really improve this chart is if a longer time horizon is included.

[derek]

I think the line chart can be improved by much bigger numbers on the vertical scale, and fewer of them. Because all most people see at a glance is the lines, the Cleveland-style "box of containment", and many small blobs of number on the left, it appears that the bottom of the box is a base line.

Reduce the labels on the left to 60, 65, 70 and 75%, show them big and legible, and it becomes much less easy to misunderstand the line chart (as Kaiser says, it's completely legitimate to have a line chart not start at the origin).

Another good step would be to abandon the box that ties the vertical and horizontal scales together, in favour of a Tufte-style pair of floating scales, but that is hard work to do in default Excel. An alternative would be to scrap the horizontal scale line completely, and just let the year labels stand alone. That's easy to do in Excel.

[derek]

My attempt here.

[Facebook]

"for a line chart, we do not always have to start the axis at zero"

Sure. But, as you point out, starting it elsewhere has meaning.

"it sends the impression that the drop is not important but it is the wrong perception because the occupancy rate is just never going to dip below say 50%"

That's exactly what I was taking exception to. During this recession there indeed have been hotels that went *way* below 50%. For example, in Dubai there were hotels that went down to 30%. The US average was just a hair above 50%. So for this dataset, a graph implying that you just can't go any lower seems to say that these hotels were exceptionally hurt when they were actually doing better than many.

[Cat]

I prefer the line chart but to the viewer it seems to drop to near zero, I have put together another way to look at the information, without including the data in between and all the figures, after all do we care what happened last year?
Why not just track the changes from 2007 - 2009 and use a bar chart.

here