I liked Ed Balls’s Butch Cameron and the Flat Line Kid joke in his Labour Party conference speech yesterday. Not, it has to be said, a side splitter but funny enough to trigger a smile. However, a lot of economists think the joke will one day be on the shadow chancellor himself, who may eventually have to admit that the economy is doing a little better than official statistics at present suggest.
Last week the Office for National statistics (ONS) revised its estimates for GDP in the second quarter of the year. Output fell by 0.4%, a bit better than the previous estimate of minus 0.5% and a lot better than the 0.7% contraction the ONS originally reported. Many commentators reckon that without the distorting effect of the long June bank holiday to celebrate the Queen’s Diamond Jubilee the economy might have registered positive growth. Moreover, some go further in challenging the accuracy of the ONS numbers, which show contraction since the final quarter of 2011, thereby questioning whether there has been a double dip recession at all.
One reason for this is the strength of employment growth during this period. The number of people in work rose by 201,000 in the second quarter, while total unemployment fell by 46,000. Maybe, it’s argued, this isn’t a genuine puzzle, but simply the result of faulty GDP estimates. Yet while the growth figures get a lot of stick the employment and unemployment figures are taken at face value, so it was good to see last Saturday’s article on the Labour Force Survey (LFS) by Sam Fleming, economics editor of The Times.
The LFS provides estimates, published each month, of a variety of key labour market indicators drawn from a rolling quarterly survey of households. All those in the sampled households aged 16 or over are interviewed five times at 13 week intervals, initially face to face, and asked questions about their employment status, hours of work, pay etc. Households drop out of the sample after their fifth interview and are replaced by newly sampled households, so there is a quarterly turnover rate of 20% in the total sample.
The survey involves a variety of well-known statistical practices designed to ensure results can be obtained and published without too much delay. For example, if a household misses one of its interviews its responses from the previous interview are assumed to be unchanged (a process called imputation, though this is not extended if a further interview is missed). Similarly, typically, a third of responses are proxies, with answers for a temporarily absent household member given by an appropriate respondent, normally a spouse or parent of the absent person. While these proxy responses, which are much higher than average in the case of teenagers and ethnic minorities, amount in effect to informed guesses, an ONS study from the late 1990s suggests they provide fairly close approximations to the answers the absent respondent would have given except with regard to hours of work and earnings.
However, as The Times article notes, these normal statistical practices are being applied to a survey with a much lower response rate, and thus far fewer households, than it used to achieve. This is due in part to a methodological change in 2010 which saw households with respondents aged 75 or over removed from the sample after their first interview because it was found that their subsequent answers were very unlikely to change. But a long running downward trend is evident nonetheless, with the total sample having fallen from around 60,000 to around 45,000 between 1999 and 2010.
The significance of this lies in what it might mean for so-called ‘sampling variability’. This is the possibility that any given survey sample might give a somewhat different estimate of particular labour market indicators than other samples. This possibility gets bigger the smaller the sample size of the survey. Consequently, the ONS publishes substantial confidence intervals alongside its estimates of LFS indicators which show how much higher or lower the ‘true’ figure of an indicator might be than the survey estimate.
These confidence intervals don’t get much attention but in the second quarter of 2012 were as high as 159,000 for the quarterly LFS estimate of the level of total employment and 83,000 for total unemployment. The corresponding estimated quarterly rise in employment and fall in unemployment was 201,000 and 46,000. It is therefore quite possible that employment increased by no more than 40,000 in that quarter with unemployment flat.
I personally doubt that the puzzle of rising employment in a contracting economy can be entirely attributed to a ‘statistical illusion’, especially since the ONS’s alternative Workforce Jobs Survey also records net job creation of 93,000 in the second quarter of the year. But a closer look at the statistics might give pause for thought to those who accept the strength of the LFS figures without question.