April 2013

“…in British Gardens. AT least one in 20 homeowners now has an outdoor heated water spa.”
Percentage of GB households with private gardens?
“Government forecasts seen by The Daily Telegraph show the percentage of households with gardens is expected to have fallen from 91.8 per cent in 1995 to 90 per cent by next year.”

I’m not sure I believe that either. but the story in question, based on a survey of 2000 homeowners carried out by OnePoll for Lloyds TSB insurance found that between 4% and 10% of homeowners have a hot garden tub. I can’t find the original release, but I guess it said that the figures was 7% with a 3% margin of error.

“By 2008 the industry reckoned it was selling 12000 a year”. 7% of 90% (households with gardens) of 25 million (GB households) =
900,000. 900,000 divided by 1200 = 75 years. Or to achieve 900k, we have 20 years (guesstimate) at 12000, but a five year constant rate of increase such that last year 350,000 were sold (too boring to show the working but available to interested parties).

Crapstat rating: 2.
The guilty party: David Derbyshire in The Sunday Times 21-04-2013
Straw poll of one: I haven’t got a hot tub in my garden, but if I did I would spoil the view of about 20 neighboring households whenever I used it.


And here’s another one – and in this case the sample size given is 17 million.

“”Analysis by MoneySupermarket.com reveals that the type of car you drive is not just a matter of personal choice, but can also be influenced by your name. The UK’s number one comparison site analysed 17 million car insurance searches through its site and has come up with cars people are most likely to drive based on their name.“”
So, the usual confusion between ‘most’ – that means ‘the one with the highest count’ and ‘more’ – that means ‘higher than’, or in
this case ‘higher than we might expect’. To be fair to the PR company – down in the small print you’ll find:

“”For example, Jacob’s are listed as having the highest likelihood of all names to own a Volkswagen: –
According to MoneySupermarket quotes, 14.5% of Jacob’s own an Volkswagen in the UK, compared to 8.6% of the UK population owning an Volkswagen.
Jacob’s are therefore 69% more likely to own an Volkswagen than the average person, and are chosen for having the highest rate of Volkswagen ownership for any UK name.””

However later they show the 200 most popular names and their associated car brands. This must be differently calculated since in the Jacob example above there would be only 1 name associated with Volkswagon, but there are in fact 12. So probably this is done by taking the car brand for which the name has the highest relative index of ownership – that’s +69% in the Jacon example above and has the great advantage from the point of view of puffery that every name will have one.
And in addition because of random but meaningless variations, rare cars are more likely to have high indexes and of course that makes for better stories. I quote again:

“Surprisingly the analysis shows men with the name Justin are most likely to drive Porsches.”
No they aren’t – in either sense of the word – they aren’t more likely than any other name (except maybe relatively) and they aren’t more likely compared to any other car (AND if it was true it wouldn’t be surprising because we all know that Justin is a wanker’s name).

The Times had a story about this but it isn’t on their website version – maybe that’s because they had belatedly realised that they had basically been had.

In fact, there may even be something even more dishonest about the results. The top 200 names and 34 brands: each brand had 12 names associated with it except Peugeot (13), Ford (10), Citroen (10) and Vauxhall (8). As Ford and Vauxhall are the biggest brands it makes sense that they should have the fewest random outliers – but all the others at 12?
It looks to me like “give me the top 12 for each brand but only one per name” or something like that: and that would be very bad.

It’s been a bad couple of weeks for PR masquerading as statistical analysis. Maybe we’ve turned a corner yesterday with Kat Lay’s article in the Times yesterday about some rubbish from Littlewoods and perfect mothers. She wrote:

“Children want a mother who can bake them cakes, read them bedtime stories – and help them win £1 million on primetime TV, if the results of a new survey are to be believed. “

This idea that results of a survey might not be gospel is a bit of a breakthrough,  (I mean PR stunt surveys, not professional market research of course).  Later the article says:

“Researchers for Littlewoods.com concluded that the perfect mother would be 32 per cent McCall, 25 per cent Rowling, 25 per cent Berry, 13 per cent Klass, 3 per cent Beyoncé and 2 per cent Bussell.”

I guess that the analytical technique here is the little known ‘Add up all the responses and percentage them’ method.A

Combining the twin British characteristics of obsession with property prices and innumeracy, Zoopla has got had a lot of publicity with their analysis of the variations is the price of property by the letter of the alphabet with which the streetname begins.
The most expensive is U and least expensive is Z.

But hang on a minute, aren’t U and Z rather unusual letters, and since there is no causative reasons for initial street letters to differ in price, isn’t this just a way of finding out average house prices by a random sample? and the smaller the sample the less acccurate the results?

I can’t find the number of streets in GB by letter, but I can for London. 5 letters have particularly low counts: I,J,U,Y,Z. Three of these are in fact amongst the two highest or the two lowest results.