I wish people wouldn't take advantage of dodgy statistics

[andrewducker]

Look, I don't like The Conservatives any more than you do, but if you* publish a graph like this:

and expect me to not notice that the x-axis doesn't start at zero, then frankly you're just lying to people through misuse of statistics.

Thankfully, the lovely people at Full Fact put together a graph that shows this one in context:


I have no interest in winning through fraud. I would like everyone to have the truth, and make their own mind about it. Sure, put your own interpretation on things, but if you're trying to win through fraud then we are not on the same side at all.

*In this case Liberal Conspiracy, who I like to keep tabs on, but frequently get very annoyed by, largely due to their levels of ignorance and their dogmatic approach to things.

Comments

[ciphergoth.livejournal.com]

Straight out of Darrell Huff's How to Lie with Statistics.

[rhythmaning.livejournal.com]

I disagree completely. Rescaling the axes is perfectly ok - what they had done was absolutely clear.

Using stats to look at specific bits of data is fine, as long as it is transparent.

[andrewducker]

Rescaling axes is fine - if you point out that you've done it. If you say "Look how awful things are!" and present a falling curve, then the majority of people, not being trained to read graphs, will say "Gosh, things went down!" and not look any further at it.

As it is, what this graph shows is a fall under Labour, and a rise under the Coalition - but I doubt that's what the creator was trying to demonstrate.

[palmer1984.livejournal.com]

Writing on Lib Con doesn't necessarily mean being pro-Labour. Lib con writers tend to be anti-coalition, but they might support other parties (eg the greens).

[Also, off topic a bit, but I actually think the rise in employment was partly caused by Labour's response to the recession - the policies just didn't start to take effect until they were out of office. Coalition policies would not have had time to take effect by late 2010].

[andrewducker]

Oh yes - LibCon isn't exclusively Labour, and Sunny voted Lib-Dem at the last election (or at least intended to at one point).

And I tend to agree that it's not Coalition policy pulling unemployment up. I'm doubtful that the current cuts are needed nearly as hard as they're proposed (and I'm opposed to various of the areas they're in, obviously), or that they're having much of a positive effect.

[ninebelow.livejournal.com]

Why does anyone need to look further? The original graph is quite clear: the number of people in full time employment peaked in March 2008, dropped off precipitously and hasn't recovered since. There is nothing misleading about rescaling the axis.

[andrewducker]

It makes it look as if the fall was catastrophically large compared to the number of people in employment, when actually this is nothing like the case.

[ninebelow.livejournal.com]

It was catastrophically large. If you are saying that you think people will look at it and think that employment dropped by 80% then you obviously think people are stupid.

[andrewducker]

Having seen people's reaction to that graph, I think that people are easily confused by graphs which don't have zero at the bottom of the y-axis.

[fearmeforiampink]

Particularly as (as well as not more generally calling it out) it doesn't have the notation on the Y axis (think it's basically a zigzag shape) to be the part of the ellipsis in '0 … 21000, 21200, 21400'.

[wig.livejournal.com]

and expect me to not notice that the x-axis doesn't start at zero

Sorry to nitpick but I think you mean the y-axis.

And yes, I agree it is a misleading graph.

[andrewducker]

Gah. Yes!

[chuma.livejournal.com]

In fairness, the x-axis doesn't start at zero either ;)

[del-c.livejournal.com]

It's completely acceptable to have a line graph value scale not start at zero; you're misremembering something about bar graphs, which is why I've stopped saying "The value scale must start at zero" for bar graphs, and started saying "You must show the whole bar".

An insistence on showing the scale all the way to zero can just as easily be used to tell a false story ("Look! things have hardly changed really!"), when a "small" change is unacceptable in reality. Instead you should bring attention to the short period that the data in the original graph spanned, which was also corrected in the improvement. Now the graph says something a little more truthful ("Look. Things have changed by an amount not unreasonable in context.")

As Tufte says, "to clarify, add detail"

[andrewducker]

Aaah, yes. If you make it clear why you've chosen the layout you have, then it's absolutely fine to have any layout you like.

[del-c.livejournal.com]

Not in my expert opinion. There are many layouts that are never fine no matter how you try to justify them. I gave one example, the bar chart with bars cut off at the knees. Another is the 3-D pie chart.