The Internet of Truths

I spend a lot of time thinking about forms of measurement: what they are, how they can be implemented effectively in social programmes, and what they mean to people when they’re in use. There are a whole lot of implications attached to measurement and even simple forms can be controversial.

I’ve written before about the radical implications of measurement, but it’s an argument that bears repeating. The story of statistics is part of the great nineteenth century shift away from a predictable deterministic world in which actions determine reactions in inevitable and linear ways, towards a more chaotic universe in which prediction is at best a matter of probabilities. That makes the world more volatile, but fills it too with opportunity. You can’t act on something inevitable, but something only probable can be changed if you go about it in the right way. You may be born into poverty in Scotland, but you can end a multi-millionaire in Massachusetts. The American dream has its nebulous foundations at the least likely end of statistical probability.

This probabilistic world has to negotiate its relationship with the truth. Statistical models are never exactly true to life. They are representations. In the words of George Box, “all models are wrong, but some are useful”. The missing part of that axiom is the question: what they are useful for? We always need to consider the extent to which any statistical model usefully presents us with helpful truth values which make good predictions on which we can base our actions.

This is all rather learned and abstract and needs illustrating with a good straightforward example. BuzzFeed gave us a nice one a few weeks ago in the form of variable jean sizes. There was a US video, a UK article and many thousands of blogs and comments linked to the original pieces.

We’ve modelled the differences between the brands BuzzFeed tested and you can see what they look like in our infographic. There are indeed divergences between them. Not every brand has the same measurements reflected in its sizing. Where this becomes interesting – to me – is in the analysis. The BuzzFeed pieces didn’t bother with analysis, instead concentrating on what it felt like for their reporters to be confronted by elastic sizing systems (like totally belittling to their self-esteem). Other commentators, however, drew the conclusions that BuzzFeed only implied: “I honestly don’t understand why there isn’t some sort of industry-wide regulation when it comes to clothing size”, in the words of Lucia Peters on Bustle.com.

That statement is presumably based on the idea that there could be a real size 10 that is much truer to the shape of women. The world population of women, perhaps, divided into a predetermined number of sizes (based on achievable production targets and retail stock requirements?) and averaged to produce The Standard Sizes for industry-wide regulation.

Is that a workable idea? Sizes are at least a fairly easy place to start compelling mass standardization in that the measures themselves are already standardized in centimetres and inches and it is fairly clear which parts of the body to measure and in what way. This makes determining the true size 10 less of a challenge than, say, true social impact where the very idea of measurement itself is disputed.

In its purest form, as with all statistical models, the true size 10 would be based on accurate measurements of the whole population. In practice that would be impossible. There may even be some doubt about the population in question. In a globalised world is industry-regulated sizing national, international, or global? Different populations may sway the results. As will age. How old are the women you are measuring? Adolescent bodies are rather different from those of mature women but 14 year olds usually wear adult clothes. Where is the industry-regulated cutoff point for inclusion?

These are presumably not dissimilar to the kinds of questions that brands ask themselves when working out their sizing. A brand with a younger customer base may choose to adjust its sizing to reflect that. A brand with an older target market ditto. It will still make more sense, however, for them to use a common naming system because these will still, in all likelihood, be a useful shorthand for customers.

All of which is simply to say that sizing systems are probabilities, not accurate and unchanging representations of all real women. Is thinking of yourself as a size 8, or 10, or 12 a useful way of rifling a rack of clothes? Yes. Will you always be that size in all shops? No. But probably most of the time. If you want to test this out (and are a woman) you can do so here, on a handy site that runs your measurements against the sizing of different brands in the US and UK.

Has the Internet uncovered a flaw in our truth values? Not this time. Do we gain from insisting that all shops use the same measurements? It seems more likely that we lose by ruthlessly reducing to a narrow range of averages what are, in reality, a far more varied array of real bodies. After all, better to find a great fitting pair of jeans in one shop, than for all jeans in all shops not to fit in exactly the same way.

In this instance it just wouldn’t be useful to follow a single statistical model, however vast the dataset. It would take us much further from the truth.