The ONS podcast returns, this time looking at the importance of communicating uncertainty in statistics. Joining host Miles Fletcher to discuss is Sir Robert Chote, Chair of the UKSA; Dr Craig McLaren, of the ONS; and Professor Mairi Spowage, director of the Fraser of Allander Institute.

Transcript

MILES FLETCHER

Welcome back to Statistically Speaking, the official podcast of the UK’s Office for National Statistics. I'm Miles Fletcher and to kick off this brand new season we're going to venture boldly into the world of uncertainty. Now, it is of course the case that nearly all important statistics are in fact estimates. They may be based on huge datasets calculated with the most robust methodologies, but at the end of the day they are statistical judgments subject to some degree of uncertainty. So, how should statisticians best communicate that uncertainty while still maintaining trust in the statistics themselves? It's a hot topic right now and to help us understand it, we have another cast of key players. I'm joined by the chair of the UK Statistics Authority Sir Robert Chote, Dr. Craig McLaren, head of national accounts and GDP here at the ONS, and from Scotland by Professor Mairi Spowage, director of the renowned Fraser of Allander Institute at the University of Strathclyde. Welcome to you all.

Well, Sir Robert, somebody once famously said that decimal points in GDP is an economist’s way of showing they've got a sense of humour. And well, that's quite amusing - particularly if you're not an economist - there's an important truth in there isn't there? When we say GDP has gone up by 0.6%. We really mean that's our best estimate.

SIR ROBERT CHOTE
It is. I mean, I've come at this having been a consumer of economic statistics for 30 years in different ways. I started out as a journalist on the Independent and the Financial Times writing about the new numbers as they were published each day, and then I had 10 years using them as an economic and fiscal forecaster. So I come at this very much from the spirit of a consumer and am now obviously delighted to be working with producers as well. And you're always I think, conscious in those roles of the uncertainty that lies around particular economic estimates. Now, there are some numbers that are published, they are published once, and you are conscious that that's the number that stays there. But there is uncertainty about how accurately that is reflecting the real world position and that's naturally the case. You then have the world of in particular, the national accounts, which are numbers, where you have initial estimates that the producer returns to and updates as the information sets that you have available to draw your conclusions develops over time. And it's very important to remember on the national accounts that that's not a bug, that's a feature of the system. And what you're trying to do is to measure a very complicated set of transactions you're trying to do in three ways, measuring what the economy produces, measuring incomes, measuring expenditure. You do that in different ways with information that flows in at different times. So it's a complex task and necessarily the picture evolves. So I think from the perspective of a user, it's important to be aware of the uncertainty and it's important when you're presenting and publishing statistics to help people engage with that, because if you are making decisions based on statistics, if you're simply trying to gain an understanding of what's going on in the economy or society, generally speaking you shouldn't be betting the farm on the assumption that any particular number is, as you say, going to be right to decimal places. And the more that producers can do to help people engage with that in an informed and intelligent way, and therefore mean that decisions that people take on the basis of this more informed the better.

MF
So it needs to be near enough to be reliable, but at the same time we need to know about the uncertainty. So how near is the system at the moment as far as these important indicators are concerned to getting that right?

SRC
Well, I think there's an awful lot of effort that goes into ensuring that you are presenting on the basis of the information set that you have the best available estimates that you can, and I think there's an awful lot of effort that goes into thinking about quality, that thinks about quality assurance when these are put together, that thinks about the communication how they mesh in with the rest of the, for example, the economic picture that you have, so you can reasonably assure yourself that you're providing people with the best possible estimate that you can at any given moment. But at the same time, you want to try to guide people by saying, well, this is an estimate, there's no guarantee that this is going to exactly reflect the real world, the more that you can do to put some sort of numerical context around that the more the reliable basis you have for people who are using those numbers, and thinking about as I say, particularly in the case of those statistics that may be revised in future as you get more information. You can learn things, obviously from the direction, the size of revisions to numbers that have happened in the past, in order to give people a sense of how much confidence they should place in any given number produced at any given point in that cycle of evolution as the numbers get firmer over time.

MF
If you're looking to use the statistics to make some decision with your business or personal life, where do you look for the small print? Where do you look for the guidance on how reliable this number is going to be?

SRC
Well, there's plenty of guidance published in different ways. It depends, obviously on the specific statistics in question, but I think it's very important for producers to ensure that when people come for example to websites or to releases that have the headline numbers that are going to be reported, that it's reasonably straightforward to get to a discussion of where do these numbers come from? How are they calculated? What's the degree of uncertainty that lies around that arising from these things? And so not everybody is obviously going to have an appetite for the technical discussion there. But providing that in a reasonably accessible, reasonably findable way, is important and I think a key principle is that if you're upfront about explaining how numbers are generated, explaining about the uncertainty that lies around them in as quantified way as you can, that actually increases and enhances trust in the underlying production and communication process and in the numbers rather than undermining it. I think you have to give the consumers of these numbers by and large the credit for understanding that these things are only estimates and that if you're upfront about that, and you talk as intelligently and clearly as you can about the uncertainties - potential for revision, for example - then that enhances people's confidence. It doesn't undermine it.

MF
You mentioned there about enhancing trust and that's the crux of all this. At a time we're told of growing public mistrust in national institutions and so forth, isn't there a risk that the downside of talking more about uncertainty in statistics is the more aware people will become of it and the less those statistics are going to be trusted?

SRC
I think in general, if you are clear with people about how a number is calculated, the uncertainty that lies around it, the potential for revision, how things have evolved in the past - that’s not for everybody, but for most people - is likely to enhance their trust and crucially, their understanding of the numbers that you're presenting and the context that you're putting around those. So making that available - as I say, you have to recognise that different people will have different appetites for the technical detail around this - then there are different ways of presenting the uncertainty not only about, you know, outturn statistics, but in my old gig around forecasts of where things are going in the future and doing that and testing it out with your users as to what they find helpful and what they don't is a valuable thing to be doing.

MF
You've been the stats regulator for a little while now. Do you think policymakers, perhaps under pressure to achieve certain outcomes, put too much reliance on statistics when it suits them, in order to show progress against some policy objective? I mean, do the limitations of statistics sometimes go out of the window when it's convenient. What's your view of how well certainty is being treated by those in government and elsewhere?

SRC
Well, I think certainly in my time as a forecaster, you were constantly reminding users of forecasters and consumers of that, that again, they're based on the best available information set that you have at the time. You explain where the judgements have come from but in particular, if you're trying to set policy in order to achieve a target for a particular statistic at some point in the future, for example, a measure of the budget deficit, then having an understanding of the uncertainty, the nature of it, the potential size of it in that context, helps you avoid making promises that it's not really in your power to keep with the best will in the world, given those uncertainties. And sometimes that message is taken closer to heart than at other times.

MF
Time I think to bring in Craig now at this point, as head of national accounts and the team that produces GDP at the ONS to talk about uncertainty in the real world of statistical production. With this specific example, Craig, you're trying to produce a single number, one single number that sums up progress or lack of it in the economy as a whole. What do you do to make the users of the statistics and the wider public aware of the fact that you're producing in GDP one very broad estimate with a lot of uncertainty built in?

CRAIG MCLAREN
Thanks, Miles. I mean, firstly, the UK economy - incredibly complex isn't it? The last set of numbers, we've got 2.7 trillion pounds worth of value. So if you think about how we bring all of those numbers together, then absolutely what we're doing is providing the best estimate at the time and then we start to think about this trade off between timeliness and accuracy. So even when we bring all of those data sources together, we often balance between what can we understand at the point of time, and then equally as we get more information from our businesses and our data suppliers, we evolve our estimates to understand more about the complex nature of the UK economy. So where we do that and how we do that it's looking quite