Australian employment fell 15,800 in July on a survey that ran one rotation group short, and the bureau prints the table showing who left and who joined

Twenty three paragraphs below the headline estimates, the Australian Bureau of Statistics prints a small table of who left its survey and who arrived to replace them. It is the most useful thing in the July release.

The headline first. On a seasonally adjusted basis, employment fell by 15,800 in July to 14,807,200, the employment to population ratio slipped 0.2 points to 63.9 percent, participation slipped 0.2 points to 66.9 percent, and monthly hours worked in all jobs fell by 12 million to 1,998 million. Unemployment rose by 4,200 people to 691,500 and the rate printed at 4.5 percent, against 4.4 percent in June. The release shows that rise as a monthly change of 0.0 points, because the movements are calculated from unrounded levels and are more accurate than the difference between two rounded figures.

The sample was one group short

In July the Labour Force Survey ran on seven rotation groups rather than the usual eight. Three groups left after responding in June and only two new ones entered. The bureau states the consequence plainly: with seven groups the sample is smaller and the standard errors on the estimates are bigger. One incoming group arrives in August, which restores the usual eight.

Then comes the table. Measured on original terms, the combined June outgoing groups had an unemployment rate of 4.1 percent, an employment to population ratio of 65.2 percent and participation of 67.9 percent. The combined July incoming groups had an unemployment rate of 4.6 percent, an employment to population ratio of 64.0 percent and participation of 67.1 percent. The July estimate as a whole sat between them at 4.4 percent, 64.6 percent and 67.6 percent.

So the households that stopped answering were, on the survey's own measurement, in better labour market shape than the households that started. That is not proof the composition drove the month. It is the bureau saying, in its own publication, where a reader should look before treating one month as a signal.

Two states were weighted differently

The June release adjusted the influence of about 2,700 respondents, roughly 6 percent of that month's sample, in one rotation group in New South Wales and Victoria. With July's data in hand the bureau reviewed both. The New South Wales adjustment was removed for June and not applied in July. The Victorian one was applied for July, following on from June.

The trend series tells a different story from the seasonally adjusted one this month, and the bureau names trend as the better measure of underlying behaviour because it strips month to month volatility. On trend, employment rose 29,300 in July to 14,824,200, and hours worked were flat at 2,004 million. The trend unemployment rate is the same 4.5 percent.

The survey covered the two weeks from Sunday 28 June to Saturday 11 July and was collected between 5 and 25 July. August figures are due on 24 September.