UK Labour Market Statistics: Combining Continuous Survey Data into Monthly Reports

David Caplan Office for National Statistics, Labour Market Division B3/11, 1 Drummond Gate London, SW1V 2QQ, [email protected]

Marta Haworth Office for National Statistics, Labour Market Division B3/06, 1 Drummond Gate London, SW1V 2QQ, United Kingdom [email protected]

David Steel University of Wollongong, School of Mathematics and Applied Statistics New South Wales 2522, [email protected]

1. Introduction

The United Kingdom labour Force Survey (LFS) produces statistics on employment, and inactivity using internationally agreed concepts and definitions developed by the International Labour Organisation (ILO). The survey has been continuous since 1992 but, until April 1998, its main use was to meet requirements. In the UK, it supplemented other sources of labour market statistics, notably business surveys and administrative records.

In 1998 the UK Office for National Statistics (ONS) undertook a major review and user consultation about official labour market statistics. This has resulted in a fundamental change. In response to user demand for a more rounded and coherent picture of the labour market, a new presentation was introduced. This gives much greater prominence to the LFS as the only source which can present coherent information on all of these aspects. New LFS data are now published monthly.

2. The Design of the UK Labour Force Survey

The UK LFS is a continuous survey designed to survey around 60,000 households each quarter and produce publishable UK level estimates quarterly. A sample rotation scheme is used in which four fifths of the households selected in a particular week are included in the survey 13 weeks later. Altogether, a household is surveyed 5 times at 13 week intervals. This design was introduced in 1992 and also conforms to the requirements of the 1998 European Council Regulation governing EU labour force survey yielding quarterly results.

3. Producing Monthly Reports from the UK Labour Force Survey

There is a very strong user demand for timely estimates of short term changes particularly in employment and unemployment and sub-categories of economic inactivity. The design of the LFS permits production of monthly estimates, but the volatility of these estimates is very high as they are based on a smaller sample and there is no sample in common between two consecutive months. The design of the sample means that estimates can be produced for any period of 13 consecutive weeks using an unclustered sample of 60,000 households. These estimates have the same sampling standard errors as the quarterly estimates and the estimates of change from the previous 13 weeks also have the same SEs as the estimates of change between consecutive quarters. Hence, it is possible to produce each month an estimate of the average unemployment in the three months up to and including the latest reference month.

Within the constraint of the current sample, the UK is addressing this by using the continuous survey as a source of monthly series constructed as a three month moving average. Even then, the information on employment and unemployment is subject to significant lags. Moreover, sampling variability for estimates of changes in levels for unemployment but also for employment and inactivity is large, particularly changes a rega\rds changes, even when sampling variability is an even more serious issue at regional and small area levels. In order to overcome the sampling variability problems, the ONS has set up new methodological work. This covers trend estimation and construction of model based estimates for small areas. This paper focuses on the UK work on trend estimation. The objective is to improve the assessment of short-term changes and to establish the properties and reliability of trend estimates as compared with the changes observed in direct survey estimates.

4. Survey Estimates and Trend Analysis

The rolling average approach is a way of addressing the issue that whist the current survey can give information about each month, the estimates of change between consecutive months are very volatile.

The monthly release of latest three-months averages is not the same as releasing monthly estimates. Let Yt be the population value of the characteristic of interest, for example total unemployment for month t. Each month is defined as the 4 or 5 weeks that approximate it. Let Xt denote the value that would be obtained if the series of values Yt were seasonally adjusted. When data for month t have been collected the estimate of average monthly change that is highlighted is ~ ~ (y, t-1 - y, t-4)/3 which is unbiasedly estimating [ (Yt-Yt-3) +(Yt-1-Yt-4)+(Yt-2-Yt-5)]/3. An ~ ~ alternative is to use the change in the overlapping 3 month average, y, t-1 - y, t-2, which unbiasedly estimates (Yt-Yt-3)/3 as an indication of current direction of the series. The corresponding changes in the seasonally adjusted series can also be examined. Analysis of the overlapping change will detect changes earlier, but has higher standard errors than the estimate obtained from the non-overlapping 3 month average.

An additional method analysing trends to calculate the smoothed trend cycle estimate produced by X-11 (i) using monthly estimates from the LFS, (ii) the monthly published series of rolling averages. These estimates may assist in interpreting the underlying trends and changes. The usefulness and reliability, in particular the expected degree of revision, of these estimates are being evaluated.

5. Conclusions

Within the UK, there is a growing demand for timely and accurate information about the labour market. The approach of combining estimates into monthly reports, which has been described here, has allowed for the timely and regular production of meaningful statistics. These provide users of labour market data with a frequently updated indication of latest trends in the labour market.