Bad Graphs

[jimbob]

I thought we had a topic for this, but spotted by a friend on Twitter, who described the scale s as "novel"

[TAFKAsoveda]

Erm, wow

[Little waster]

The longer I look at it the more wrongness I see.

I clocked the variable Y scale straightaway, it took a few seconds to also twig the X-axis was non-standard and embarrassingly long before I noticed the weird shenanigans of setting the intercept at 2%.

[bagpuss]

Someone's copied it and somehow just f.cked up all of the axis labels barring the 8% and 4% at the top. Correct chart can be seen here.

https://www.bbc.com/news/business-55072447



EDIT: add "and 4%"

[bagpuss]

Acksherly, the BBC one has the y axis correct too so now I'm wondering if the original was wrong and BBC just has a corrected version, while that newspaper was fed the original cocked up one.

[El Pollo Diablo]

So the scale was wrong in the OP as well (the bars should've been every 4 points, rather than every two), which is amazing. Good effort. I'd say it's easier to do it right than to do it this badly.

[jaap]

So the scale was wrong in the OP as well (the bars should've been every 4 points, rather than every two), which is amazing. Good effort. I'd say it's easier to do it right than to do it this badly.

It seems to me that the only thing wrong with the bad graph was the numbers along the vertical scale. Those numbers were so catastrophically bad that it looked like several errors (the baseline of the bars not being at zero, irregular step sizes along the vertical axis), and invalidated the whole graph of course. But I don't see anything else wrong - the 2006 on the x axis seems a bit weird, but it is located at the correct tickmark (the original graph in the BBC article has 2005 at the larger tickmark but no bar above it, the data starts at 2006).

[Little waster]

So the scale was wrong in the OP as well (the bars should've been every 4 points, rather than every two), which is amazing. Good effort. I'd say it's easier to do it right than to do it this badly.

It seems to me that the only thing wrong with the bad graph was the numbers along the vertical scale. Those numbers were so catastrophically bad that it looked like several errors (the baseline of the bars not being at zero, irregular step sizes along the vertical axis), and invalidated the whole graph of course. But I don't see anything else wrong - the 2006 on the x axis seems a bit weird, but it is located at the correct tickmark (the original graph in the BBC article has 2005 at the larger tickmark but no bar above it, the data starts at 2006).

The x-axis is definitely the "jaywalking" part of that "murder, arson and jaywalking" graph but nevertheless it still isn't good practice.

The implicit assumption is that the sections between major tickmarks represent equivalent 5-year chunks, having one randomly represent 4 years can distort the interpretation of the data.

It's not a major issue here, especially given the other f.ck-ups, but still not good.

[Martin Y]

Yebbut in defence of 2006, it doesn't have a major tickmark like the 5-year dates do.

I do note that it's rather hard to spot the difference between the faint and bold ticks yet on the other hand the bold ones don't actually serve any particular purpose I can see. (edit) On reflection, they make it plain which tick the printed years refer to but, like 2006, 2025 doesn't have one. Because reasons.

[Millennie Al]

It seems to me that the only thing wrong with the bad graph was the numbers along the vertical scale.

I'd take issue with two other aspects. Firstly, it's about GDP which is a fairly bogus statistic in the first place - if I buy a coffee at a cafe that counts as part of the nations productivity, but if I make the same coffee at home it doesn't. And secondly it is a sequence of growth percentages presented as if that's meaningful. Specifically, a sequence of 5% growth, -5% growth looks like each of the 5% and -5% mean the same thing, but they are percentages of different things. 5% growth followed by -5% growth is not 0% but -0.25%. Rather than showing percentage growth, the graph show show actual GDP. That would make it easy to see when a recovery from a dip actually ends.

[El Pollo Diablo]

I had a related argument recently with an innumerate misogynist, who was claiming that a 20% rise in the number of women working for my employer (which is 80% 84% male) meant a 20% drop in the number of men.

[jimbob]

I had a related argument recently with an innumerate misogynist, who was claiming that a 20% rise in the number of women working for my employer (which is 80% male) meant a 20% drop in the number of men.

percentages vs percentage points - the classic one.

Did you manage to get through with any concrete examples? I have before now explained similar by talking about the case of shopping costing £100 of which one item costs £20.... and what happens if that rises by 20%.

[El Pollo Diablo]

Well unfortunately he'd claimed it was "basic maths", at which point I told him he failed the basic maths test. I did then point out the difference between percentage increase and percentage points, but probably he'll have stopped listening after that.

[Martin Y]

I think we* bamboozle ourselves for the sake of brevity using phrases like "... increases by 20%" allowing ourselves to avoid the laborious rigour of explaining 20% of what. All too often we're not thinking about the of what bit either.

* Not us, obviously. We're far too clever for that. You know. Other people.

[snoozeofreason]

I don't imagine it's often that one would get to say this, but here is a bad graph eloquently corrected by a Tory MP on twitter.

tweet_smaller.png (120.41 KiB) Viewed 5610 times

tweet2.png (52.88 KiB) Viewed 5610 times

I appears that the "Statistics Guy" who produced the original had massaged the figures downwards by assuming that the population of the UK had increased by 6.8% in a year and then "correcting" for that fictional increase.

While everything should be questioned (that's what science is), some of the papers need to apply the same standards to covid-denial content from random people with cartoon avatars that they do to the real scientists and clinicians on SAGE.

[jimbob]

I don't imagine it's often that one would get to say this, but here is a bad graph eloquently corrected by a Tory MP on twitter.
tweet_smaller.png
tweet2.png
I appears that the "Statistics Guy" who produced the original had massaged the figures downwards by assuming that the population of the UK had increased by 6.8% in a year and then "correcting" for that fictional increase.

While everything should be questioned (that's what science is), some of the papers need to apply the same standards to covid-denial content from random people with cartoon avatars that they do to the real scientists and clinicians on SAGE.

From the Pandemic Arena :

The back-story to this:

The Daily Mail used a graph by a "Twitterer" called "The Statistics Guy" @Jon_statistics

On one of his conspiratorial threads, 2 weeks ago, I posted this:
https://twitter.com/gordonrlove1/status ... 22657?s=19

"[Excess deaths in England &Wales are] over 10% up on the standard seasonal thing" (I was quoting the ONS figures for week 43)

He told me that my figures were a rookie mistake, as I had not
accounted for population-growth

He then "corrected" me, using figures adjusted in a way that, to this day, I cannot work out.

I then published a population-normalised version, & pointed out how it looked as you'd expect such a thing to look

I waited a day, pointed out more explicitly how his looked quite different, and asked him to explain his version.... and he blocked me

Then the other day, someone pointed out that his week 44 version of the same graph was being used to discredit the lockdown

Over the past 5 years, the population has grown, up to 2.7%
He has upped the historic data by 6.98%

Coincidentally, that makes the current year and his version of the historic data look the same

He then manages to graph it so the current actual very slightly larger number, looks very slightly smaller than his bizarrely wrong adjusted historic figure looks

Gordon has been very productive in this.

Also Gordon shared this:

https://twitter.com/GregoryDavisHNH/sta ... 7730279427

It's also worth noting that his previous account was banned by Twitter, which means he's not allowed to operate this new account.

with this in the thread below:

[Stephanie]

I thought we had a topic for this, but spotted by a friend on Twitter, who described the scale s as "novel"

there was a topic for this :( viewtopic.php?f=8&t=1191

[jimbob]

I thought we had a topic for this, but spotted by a friend on Twitter, who described the scale s as "novel"

there was a topic for this viewtopic.php?f=8&t=1191

That was the one... I knew I'd seen it.

Didn't think to search for "terrible" though.

Maybe this could a topic be for ones that are appropriate for weighty matters and the other for ones that are wrong in nerdier ways

[shpalman]

What the f.ck is this? Labelling the x-axis every two days while talking about weeks, and the y-axis seems to be every 6, except for one interval which is only five. f.ck's sake.

[bagpuss]

What the f.ck is this? Labelling the x-axis every two days while talking about weeks, and the y-axis seems to be every 6, except for one interval which is only five. f.ck's sake.

The x axis is fine by itself - labels are every 2 days but there are points every day. There just isn't enough space to label every point. And I'm not that bothered about the fact that they're comparing with a week ago as you can click on the graph and then select to view just the last week or month or whatever - so the week on week figure and the graph are just 2 headline things that you can delve further into. No idea what the blithering wotsit is going on on the y axis though (it is every 96/5, not every 6/5, though).

[discovolante]

I've split a post about good graphs, rather than bad graphs, to here (friendly reminder that starting new threads is allowed!)

[Allo V Psycho]

Via Popehat, 'surge in crime' on CNN. The points on the abcissa are not evenly spaced either.

[Sciolus]

The assertion that there is a "drop in crime" is also iffy. June 2020 is clearly an anomaly, for tolerably obvious reasons, and the other three numbers are the same, within margin of error (and credit to the graph-maker for including that).

[Gfamily]

Is there a surge in violent crime? Who knows.

This chart is just the number of people (in the poll) who expressed that as an opinion.

[Little waster]

Is there a surge in violent crime? Who knows.

This chart is just the number of people (in the poll) who expressed that as an opinion.

There’s a Dara clip for that.

ETA also even if the numbers were legit, I wonder if we all were to think really hard we might be able to come up with a reason why the period around “June 2020” might not be particularly representative of ... well ... absolutely f.cking anything?