10 Ways Charts Can Lie, Cheat & Lead Astray
Charts; they’re like photographs: they never lie, they just show the data as it is right? – wrong. In the wrong hands they can be some of the most cunning and conniving creations ever conceived by man, they’re regularly put to work to lead the gullible astray and hide the all important information held within. What I’m going to show here isn’t necessarily to do with right or wrong per-se (although most areas are commonly held chart/visualisation errors) it’s about showing the same data in different ways to convey an entirely different meaning.
I first noticed this phenomenon from reading newspapers who along with Governments are the prime offenders when it comes to chart ‘mal-manipulation’, they use it to spin and twist the figures to suit their agenda and whilst not out and out altering the data they completely change the perception of that data through the use of charts. This is something we as Qlikview Developers have to be acutely aware of as it’s so easy to unwittingly convey additional meaning with our choice of chart detail; want to make annual sales data look flat; choose chart A, want the same data to seem erratic; choose chart B; it really can be that powerful as we’ll see.
A number of the examples shown here are taken from the wonderful MrExcel.com website (a Data Monkey’s dream; it taught me everything worthwhile I know about Excel) and specifically Bill Jelens post ‘Excel Chart Lies’ – nearly all of which are directly relevant to the Qlikview world. The original post can be viewed here: http://www.mrexcel.com/tip142.shtml
1. To Zero or Not to Zero
Here we’re looking at exactly the same basic dataset covering Visitor numbers per year, the chart on the left seems very placid and visitor numbers appear to be relatively steady year on year; not bad but not great, whilst the chart on the right looks much more positive; visitor numbers took a huge leap in 2010; cue extra investment and pay rises all round! The charts show identical data so what’s going on?
To accentuate the effect even further you can use the ‘Static Min/Max’ settings to zero in even more on the difference making the smaller value seem even smaller and the increase all the larger.
2. Flat or Erratic? – ‘I’ll take erratic please’
In a very similar way to the example above we can also use our chart detail choices to make data seem flat or erratic simply be altering the axis:
As before all we’re doing is switching between having ‘Forced 0’ applied (Top) and having it not applied (Bottom), again the choice completely alters the story that this chart tells and again the data isn’t changing just the viewers perception of it.
These first 2 examples are classic uses of chart mal-manipulation employed by media and Government to make things seem better or worse as the situation, story or agenda dictates.
Don’t get me started on Log Scales and how they can mess with peoples perceptions.
3. The Growing Pie Slice
It sounds like a cake lovers dream but in this instance it’s a dashboard nightmare and it’s our old friend the 3D Pie Chart at play again. Here we have 2 Pie Charts, the only difference; one is 3D and one is 2D; every segment is of equal size or at least it’s supposed to be because in the 3D version the Red slice looks larger (and is by pixel count) than the others; which is clearly wrong.
In my mind there’s no reason to ever (and I mean EVER) use 3D Pie Charts and certainly ‘The Users like them’ is not a valid reason. This has been covered in Stephen Fews ‘Save the Pies for Dessert’ which can be downloaded here: http://www.perceptualedge.com/articles/08-21-07.pdf
4. I Can Read You Like a Book…a Book in Another Language
Sticking with a 3D theme; it’s not just Pie Charts that cause confusion the same can be said of 3D Bar Charts. Take a look at the example below, without looking at the ‘Text on Axis’ actual values what would you say the number of Visitors was in 2007?
My initial reading would be about a 1,005 if I read from the top-front edge of the bar and about 992 if I use the top-back edge of the bar; so which is right; neither. What we are infact meant to do is follow the top-side edges back until they hit the rear ‘wall’ of the chart and then read off from there!? This isn’t a Qlikview problem as it’s the same with every 3D Bar Chart implementation I’ve come across. Due to the way the chart is rendered it appears that the 2007 bar never breaks through the ‘1,000’ Grid Line so at first glance the bar appears lower than it actually is.
5. Visitors are Down…We Can’t Have That!
This is a blatant ‘Don’t do this under any circumstances’ but I’ve seen it several times both intentionally used in the media and (probably) unintentionally used in Qlikview apps, it plays on our (Western) view that a time series progress’ from left to right. Look at the chart below; after a first quick glance you’d be forgiven for thinking that Visitor numbers are going up and up…
…you’d be wrong; look at the Dimension axis; it’s been reversed, Visitor numbers are infact going down year after year. This is achieved simply by reversing the sort order on the Dimension axis; easily done on purpose or by accident, whatever the reason we’ve completely changed the initial impression the chart gives.
6. Negative Visitor Numbers, Are You Sure?
This is one I commonly see in Qlikview dashboards; a Line Chart (or Combo Chart) showing negative values when negative values are infact impossible, or where they’ve never occurred.
From this chart it appears as though there were a negative number of visitors around 2006 which was never the case; the chart is telling lies. Due to the way the Smooth Line is drawn it can ‘peak or trough’ over or below the values around it creating at best a misleading impression and at worst in cases like the one above an illusion that the impossible has happened.
7. How Can I Hide my Departments Rising Costs?
This one is taken straight from Bill’s post as it shows the problem perfectly. On their own the Marketing costs are clearly going up and up; who would want to show that to their boss? So to get round this ‘problem’ all that has to be done is to bury the Marketing costs in a stack of other more positive data and there we have it; at first glance it even looks as though Marketing costs are falling!
This is all thanks to the R&D costs falling at a greater rate than the increase in the Marketing Department. Yet again; we aren’t changing the data; take the time to read the values and you’ll get exactly the same information out of both charts but on initial viewing one is bad and one is good.
8. Like a Good Politician; Tell a Different Story
This one is another classic employed to make a bad situation appear better and it’s again taken from Bill’s post. In the top chart it looks as though there’s been a crash in Population over the period (this drop is accentuated by using a non-zero Y-Axis) whilst in the bottom one things look to be getting decidedly better – how can that be? Simple; the 2 charts are driven by the same data but are showing 2 completely different things; the top one shows actuals and the bottom shows percentage change. As the rate of population decrease is lessening over time if we show this in a chart (Bottom) it looks positive thus radically changing the message taken from the data.
This technique is commonly trotted out for Unemployment and Inflation figures, the below is an example put out by the Obama Administration and appears to show that Employment is getting better…but it isn’t; the situation may be better in a sense (unemployment is increasing at a slower rate) but month after month (bar one) unemployment continues to rise. The chart and technique are perfectly valid as long as there’s no hidden agenda to deceive; show the same data on a simple line chart and it would be stoically downward; not something a politician wants to associate with, as with all these examples it’s the same data but a completely different message.
9. Even Scatter Charts Can Get in on the Act Too
Take a look at the first Scatter Chart showing yearly numbers of Male & Female visitors, everything looks fine, we can see for instance that once we get above around 50 men and 50 women the data points start to move away from the line of best fit; no issues here.
Let’s make some changes to the chart (not the data) and change the impression.
Firstly we’ll shorten one of the Axis – I’ve exaggerated this to make the effect more pronounced but I have seen Scatter Charts like this deployed in the wild:
It’s now far less apparent that the values over 50/50 are moving away from the line of best fit; they seem to be much more ‘normal’. This issue is documented by Stephen Kosslyn in his book ‘Graph Design for the Eye and Mind’: http://www.amazon.co.uk/Graph-Design-Mind-Stephen-Kosslyn/dp/0195311841
Next we’ll lengthen one of the axis.
This time we’ve created a mis-leading impression that the values along the X-Axis are greater than those along the Y-Axis; they aren’t they both run from 0-100. The vast majority of datapoints are very nearly equal yet looking at this chart you’d think that there were more Male visitors per year than Female.
As a rule I try to make Scatter and Grid chart Axis equal in scale – if they’re both ‘100’ of the same measure (eg Visitors) then they should be the same length, if one is double the other then the axis should technically be double the length to avoid creating a false impression.
10. Data You Don’t Like? – Just Ignore It
This final example is again taken from Bill’s blog and shows how selectively choosing the data you show can completely change the picture:
Again the data hasn’t changed at all but through a combination of a non-zero axis and a limited sub-set of the data the message is completely changed.
The aim of this post hasn’t been to give you the powers to lie and change the message a .qvw conveys or to be down on Qlikview; it’s no worse than other BI Tools, I’d hope instead that it’s demonstrated how easy it is to present data in wildly different ways and that we as Qlikview Developers need to be mindful of that fact. In most cases it’s not our call to choose whether to accentuate a value or to lessen it but we do have to be aware that the chart details we choose can convey just as much information – and in some ways more than the data itself.
One of Qlikview’s key strengths is how quick & easy it is to create a dashboard; you can have one working in minutes which is great, but conversely one of Qlikview’s greatest weaknesses (in the wrong hands) is that it’s so quick & easy to create a dashboard and then move on without thinking about what your dashboard is actually saying to the viewer. As dashboard developers we should always be taking the time to actually read and use our charts; what are they saying, what’s the conclusions that the user will draw and are they the right ones.
This ‘spend time reading your own charts’ applies just as much to general chart choice as it does chart mal-manipulation; the issues I raised in my post on the Qliktech Partner App Contest (https://qvdesign.wordpress.com/2012/03/09/qliktech-global-partner-mobile-app-contest-the-runners-riders-1-thoroughbred-a-few-also-rans-and-lots-of-old-nags/) would in the main be overcome by the Developers actually trying to use the charts they created – who can use a gauge chart without an annotated axis and come away thinking ‘I’m happy with that’…no-one, and if they do they’re in the wrong job; you’re giving us all a bad name.
So be careful and think about what you’re saying with your dashboard; your next pay rise may depend on it!
As always; all the best.