Are we all feeling appropriately shitscared by the coverage of yesterday’s unemployment figures? Well cheer up – the reality could actually be a lot better than what was reported, but so saying, it could also equally be a lot worse.

As some of you may know (and some may not), the way the unemployment figures are calculated is actually via a poll that is run by the ABS called the Labor Force Survey. It’s a pretty complex type of poll that has a lot more to it than the type of political polling we’re used to – but a poll it is none the less (although I’m sure the ABS folks would prefer the word “Survey” – it has much more sophisticated connotations). For anyone interested, you can read all about the methodology of the survey and how it works over here.

As we all know, every poll and every survey has something in common – and something that not enough journalists and economic correspondents seem to appreciate with this data – they all contain sampling error.

Yes, as sad as some of you may find this, our official unemployment statistics from which a rather large number of important macroeconomic decisions are made, come with the dreaded Margin of Error [insert early Byzantine church music].

The actual Labor Force Survey results can be easily downloaded, and toward the end of the document – pages 28 and 29 to be exact – the ABS has gone to the trouble of providing the standard errors of not only the point estimates of all the unemployment metrics, but also the standard errors of the monthly change in those metrics. It’s quite nice of them to do that since the press doesn’t seem to pay any bloody attention to them whatsoever. But their incompetence aside, what these standard errors allow us to do is create a maximum margin or error for the unemployment figures using a 95% Confidence Interval – just as we do with the polling, and more particularly, Pollytrack.

So today we might look at the unemployment statistics through the prism of statistical reality – which makes a nice change from the guff that’s been oozing out of the MSM for the last 24 hours or so.

First up, the change in Full Time job numbers. The seasonally adjusted point estimate suggested that 43,900 full time jobs were lost between November and December of 08. We can be nearly 100% confident that the 43,900 figure that is getting so much attention isn’t actually true.

What we can say is that there is a 95% probability that the true change in full time job numbers was somewhere between a gain of 6300 full time jobs and a loss of 94100 jobs, for the margin or error attached to the 43900 full time job loss figure is a whopping 50200. We can run these MoE’s on the change in full time job numbers for Australia and each of the States. The black markers are the seasonally adjusted point estimates, the red lines are the Margin of Error.

Not only can we do this for the change in full time employment numbers, we can also do it for the total change in employment numbers as well as for the change in the unemployment rate.

The uncertainty contained within the sampling error of the survey (the error margins) provides a far more interesting and valuable picture than what has been thusfar painted around the usual traps, highlighting just how much we don’t actually know about yesterday’s unemployment figures.

As we know from polling, when in doubt the trend is your friend. Here we use locally weighted polynomial regressions to determine our polling trends (mostly because it works, but also because it makes us sound uber-nerdy), but there are other ways to create a trend – one being Henderson moving averages and their variants which is what the ABS uses. To show why the trend is more important than both the seasonally adjusted numbers and the original raw data, let’s have a squiz at how the various ABS metrics play out against each other from December 07 to December 08.

The original, raw data is extremely volatile – as we’d expect. The seasonally adjusted data is a bit tamer but still moves around with quite a bit of volatility while the trend is, well, the trend is just lovely.

On the trend figures, the unemployment rate remained steady at 4.4%, full time employment dropped by 11,200 nationally and total employment increased by 2000.

Far from this being a terrible result requiring widespread bouts of wrist slashing – in the broader scheme of things and considering the state of the international economy, it’s probably a remarkably good result. I say ‘probably’ because we must acknowledge the large uncertainty involved in the figures – the point estimates really aren’t the gospel they are too often made out to be.

What happens in the future is unknowable, things might tank, things might not – but what we should all be aware of is just how much uncertainty is actually contained in these figures.

(Visited 30 times, 1 visits today)