Lies, Damn lies, and Statistics

The release of Australia's latest labour force survey was greeted by everybody with euphoria.  The headlines trumpeted a fall in the unemployment rate, and a huge surge in employment.  The digger dollar strengthened and the market rose. 

But all these reactions showed how little people -- including the oh-so-wise shakers and movers -- really understand about statistical processes.  The Ozzie population is around 22 million.  The sample from which these data are estimated is not much over 1200 people interviewed.  So from a small number the statistician has to estimate a much much larger number.  This would suggest fairly wide absolute errors in the data, even if in percentage terms they would be quite small.  And so it is.  Look at the table below:

Movements in seasonally adjusted series between October 2010 and November 2010

Monthly change

95% Confidence interval

Total Employment

54 600

-

to

109 200

Total Unemployment

-19 500

-50 500

to

11 500

Unemployment rate

-0.2 pts

-0.4 pts

to

0.0 pts

Participation rate

0.1 pts

-0.3 pts

to

0.5 pts

- nil or rounded to zero (including null cells)

Note that total employment is estimated to have risen by 54,600.  In the context of the US, the equivalent jump in employment would be close to 900,000.  That would indeed be a boom.  But the 95% confidence interval gives a range of between zero and 109,200.  We are 95% sure that employment didn't fall, nor that it rose more than 109,200.   That's a big gap, folks.  The markets assumed that the statistics were exact.  They're not.  They a blur, a smudge, a vision seen by a short-sighted person who isn't wearing his glasses.

Unemployment Rate

Now the statistical boffins in Canberra know this.  They would like to have a larger sample so that they could narrow the 95% confidence limits.  But the government is saving money, by keeping down the sample size.  Cretins.  Because the pollies don't understand stats either.  All the statisticians can do is to warn us to use the smoothed trend as the guide, not the seasonally adjusted data.   And note how the trend in unemployment has stopped falling and has started rising.  Gently, but nevertheless it's headed higher.  The RBA has been raising the cash rate.  Mortgage rates have risen even faster.  Loan demand has slumped.  Housing clearance rates (the number of houses offered for auction which are actually sold) have fallen sharply in the two main metropolises.  This indicator is an excellent guide to house prices. The Reject Shop's recent profit warning says that demand is weak.  The savings ratio has gone from below zero to 10%.

Total Employment

Now let's have a look at the level of employment.  Notice that it's above trend.  What normally happens when a time series deviates from its trend is that it snaps back to trend.  You can see how that happened in January 2010.  Employment rose very sharply, only to fall a little in the next month.  Remember: these are the putative centres of a blur.  They are not divinely inspired numbers accurate to the last decimal place.  But the trend is your best estimate to what's going on.

My conclusion  is that the economy is pausing, slowing.  Not to recession, not with China booming and our exports with it, but growth is definitely weakening.  And the economy outside the mining sector is very weak indeed.

Even if your interest in the Ozzie economy and its markets is minimal, the lessons about statistical variance and unreliability are the same in all economies.  And I wonder, when inevitably next month's data are weak, even if the trends are unchanged, will the digger dollar fall and the share market with it?