Tuesday, November 27, 2018

beware of the semi-attached figure

What can you do if you need to convince someone of something, but you don’t have proper evidence?

One simple way is to demonstrate something else to be true and then just pretend it’s the same thing.

In statistics, this trick is known as the ‘semi-attached figure’.

Simply pick a couple of things that sound kind of the same – though they aren’t (this is the important point) – and make a comparison between them to validate your conclusion.

An everyday example would be the number of reports that contrast hours spent TV viewing with hours spent on the internet, as though those activities were the same thing.

One reputable market research firm recently tried to convince an audience I was in about the popularity of a particular on-demand Aussie TV channel.

‘Who is watching?’ they asked. ‘Well, 90% of viewers are aware of the service!”.

Sounds impressive, however, awareness of the existence of something is not the same as usage of the service.

It has long been a common tactic of persuasion to cite information that initially seems to uphold an assertion, but upon closer inspection is pretty much irrelevant to the actual claim.

This means stating one thing as a proof for something else.

For example, if some report claims says ‘85% of CEOs think that Blockchain will change the way their organisations do marketing by 2020”– what does that actually prove?

This implies that CEOs are some sort of authority on the application of Blockchain technology.

Or marketing.

There’s no shortage of reports showing the decline of advertising spends on printed news.

The implication is that advertisers should spend more on whatever the alternative is that’s being sold.

Of course, a decline in advertising spend does not necessarily mean a decline in readership.

When dealing with any ‘evidence’ of this nature, ask yourself how the evidence specifically proves the claim. Could there be alternative explanations that would make the claim false?

If the evidence isn’t necessarily relevant to the conclusion then you are probably dealing with a semi-attached figure.


(Note: For more fun with statistics I always recommend 'How To Lie With Statistics' by Darrel Huff, first published in 1954.)

blog comments powered by Disqus