RDP 2002-07: An Exploration of Marginal Attachment to the Australian Labour Market 3. Data and Definitions

3.1 The Survey of Employment and Unemployment Patterns (SEUP)

The SEUP is a longitudinal survey conducted by the ABS covering the period September 1994 to September 1997. Respondents were individuals aged 15 to 59, living in private dwellings in both rural and urban areas of Australia and were selected into the sample between May and July 1995. Data for the survey were collected in three waves of interviews.

The survey comprised three sub-samples. The largest sub-sample comprised 5,488 persons who were job seekers (the JS sub-sample). Members of this group were identified as being either unemployed, underemployed or marginally attached to the labour market. The underemployed included persons who were working part-time but who stated their desire to work more hours. Persons who were studying full-time and desired a part-time job were not included in this sample. A second smaller sub-sample, the Population Reference Group (PRG), was selected to be representative of the Australian population and comprised 2,311 persons. The third sub-sample was of people known to have participated in a labour market program. However, the data file released to the public excludes the labour market program information and this group cannot be separately identified.

At each interview, two levels of information were collected. First, data were collected about the respondent at the time of the interview. This information includes details on current demographic and social characteristics of the respondents such as gender, age, educational qualifications, family structures, place of residence, weekly income, hours of work and annual income for the previous financial year from various sources. In addition to current labour force status, information was also collected about selected characteristics of job(s) if a respondent was employed at the time of the interview. At each interview, respondents were also asked about the labour market experience of other family members such as their spouse and/or their parents.

The second type of data collected was episodal information relating to each respondent's experiences in the labour market over the previous 12 months or so. These data provide information on each spell of work, each period of looking for work and each period spent out of the labour market. Using these data it is possible to construct continuous labour market histories of employment, unemployment, marginal attachment and other time spent not in the labour force (other NILF) for the entire period of the survey (September 1994 to September 1997). Given the focus of this study is on the labour market dynamics of the marginally attached and the unemployed, most of the analysis in the rest of the paper is based upon measures of labour force status constructed using the episode data.[2]

Table 1 presents information on sample sizes for the JS and PRG samples at each of the three interviews (waves). As with all longitudinal surveys, there is sample attrition. Over the entire survey 19.3 per cent of the sample was lost, with 11.7 per cent being lost between waves 1 and 2 and 8.6 per cent being lost between waves 2 and 3. There are some differences in the rates of attrition between different groups. Young males have higher attrition rates than do young females or older respondents. Renters have higher attrition rates than do non-renters. There are also differences in rates of attrition between married respondents and those who are not married. See ABS (1995) for more information about the patterns of attrition in the survey.

Table 1: Sample Sizes at Each Interview
  Wave 1 Wave 2 Wave 3
PRG sample 2,311 2,120 1,983
JS sample 5,488 4,779 4,261
Total sample 7,799 6,899 6,244
Notes: There is some overlap between the PRG and JS sub-samples. In the first wave, 227 persons were members of both samples. Wave 1 covered the period 1 September 1994 to 31 August 1995, wave 2 the period 1 September 1995 to 31 August 1996 and wave 3 the period 1 September 1996 to 31 August 1997.

The period of time covered by the SEUP was one of considerable fluctuation in the rate of growth of employment. Between September 1994 and September 1995 employment growth was quite strong at 3.1 per cent. The rate of growth of employment for the rest of the survey period was much lower at 0.9 and 1.1 per cent in the year to the wave 2 and 3 interviews respectively. Given the prominence attached to the level of labour demand conditions in models of the determinants of labour force status, the variation in macroeconomic conditions should increase the power of the analysis.

Survey weights are available for each wave in the SEUP, and these can be used to generate the relevant population estimates. The survey weights are based on independently estimated distributions of various respondent characteristics in the general and the job seeker populations. They also account for sample design and attrition in the survey.

Since it is not possible to combine the JS and PRG weights in any simple manner, it is important to consider whether it is valid to combine the respective samples.[3] This depends upon the purposes for which the data are being used. The convention in economic studies is that unweighted data can be used for inferences about underlying behavioural patterns, but weighted estimates are required before the results can be interpreted as providing population estimates. In Sections 3 and 4, the JS and PRG samples are analysed separately using weighted data, and consequently the results can be interpreted as population estimates. In contrast, Section 5 combines the JS and PRG samples without weighting in order to analyse the factors underlying the dynamic behaviour of the marginally attached and the unemployed.

3.2 Definitions of Labour Force Status

Information on labour force status is available from the interview data or from the episode data. The analysis of the labour force dynamics presented in this paper is based upon labour force histories constructed using the episode data as they allow a complete labour force history to be constructed, and allow marginally attached workers to be identified.[4] The interview data are only used for the purposes of assessing the representativeness of the SEUP data.

The information about each episode is based upon retrospective questions at the time of each annual interview. While the questions asked are similar to the standard ABS questions, the fact that they are retrospective may introduce some distortions.[5] Using the episode data, employment episodes are defined as periods of time in which the respondent had one or more paid jobs. Unemployment episodes are defined as periods of time in which the respondent was actively searching for work and was not employed. This differs slightly from the standard ABS definition which also includes an ‘availability to start work’ test. It is not possible to include the availability criterion since the information was not collected for job search episodes.

Marginally attached episodes are defined as episodes in which the respondent is not actively searching for work but is willing and available to start work.[6] Using this definition, there were 449 respondents who were marginally attached at the approximate date of the wave 1 interview.[7] As outlined above, a subset of the marginally attached are discouraged workers, who can be identified as marginally attached respondents who give labour demand reasons for not actively searching for employment.[8] The number of discouraged workers at the approximate date of the wave 1 interview is 163, and falls to 106 if the main reason for not searching is related to labour demand. Although discouraged workers are an important subset of the marginally attached, the numbers of discouraged workers in the SEUP sample are too small to allow a reliable analysis, and the remainder of this paper focuses on the more general concept of marginal attachment.

3.3 Representativeness of the SEUP Data

Before proceeding to the analysis of the dynamics of labour force status it is important to establish the representativeness of the SEUP data and any biases which may be introduced by the use of the episode data. These may arise either from the retrospective nature of the data or subtle differences in the questions asked. This section compares the estimates of labour force status constructed using SEUP's episode data and interview data with estimates from the ABS's Labour Force Survey (LFS). As discussed, the PRG was selected to be representative of the general population whereas the JS sample is representative of job seekers. Given the very different nature of the population represented by the PRG and JS samples, the representativeness of the two samples are considered separately.

Table 2 shows the distribution of labour force status defined using standard ABS definitions at the wave 1 interviews. The first column presents the weighted estimates made using the JS sample and the second column the weighted estimates made using the PRG sample. At the time of the wave 1 interview, 74 per cent of the PRG sample were employed, 6.5 per cent were unemployed and 19.5 per cent were NILF. Breaking down the NILF category, the weighted estimates suggest that 0.6 per cent of the population are discouraged workers, 1 per cent wanted to work but were not actively looking for work because they were in education, and 0.9 per cent wanted to work but were unavailable to start work. These estimates are broadly consistent with those from the LFS (ABS 2000, 2001).

Table 2: Labour Force Status at First Interview: Standard ABS Definitions
Per cent
  JS sample PRG sample
Employed 36.8 74.0
Unemployed 45.0 6.5
NILF 18.2 19.5
Discouraged worker 1.8 0.6
Wanted to work but in education 3.0 1.0
Wanted to work but unavailable to start work 1.9 0.9
Sample size (unweighted) 5,488 2,311
Note: The numbers reported in the table are weighted estimates for the labour force status of the JS and PRG samples.

As expected, the distribution of labour force status in the JS sample is very different. The proportion of the weighted sample that is unemployed is 45 per cent, compared to 6.5 per cent of the PRG. This simply reflects the different populations. While the proportion of marginally attached groups is roughly three times higher in the JS sample, the distribution over the different types of marginal attachment is similar to estimates from the PRG. For example, the largest category in both cases is ‘Wanted to work but in education’ and there are roughly twice as many in this category as there are discouraged workers.

Table 2 provides information on the representativeness of the sample at a point in time. Given the focus of this study on the labour dynamics of the marginally attached and the unemployed, it is important to also consider whether the labour market dynamics of the SEUP sample are representative. One-month transition probabilities estimated from the SEUP can be benchmarked against the LFS gross flows data. The gross flows data do not separately identify the marginally attached and so for the purposes of benchmarking, the marginally attached identified in the SEUP data are recombined with other NILF. The labour force transition probabilities from the SEUP are estimated separately for the PRG and the JS samples. These one-month labour force transition probabilities are presented in Appendix A.

As already discussed, the samples were selected between May and June 1995, and exact interview dates are not available. We have chosen 1 June 1995, the midpoint of this period, as the starting date for calculating the transition probabilities on the basis that episodes current at this date are most likely to be representative of the JS and PRG samples. This also maximises the sample size.

Since the PRG sample is representative of the entire population, the lack of information on the interview date presents no difficulty. The patterns and size of labour force transitions estimated from the PRG sample are broadly similar to those obtained from the gross flows data. The major difference is that the estimated probability of remaining in unemployment after one month is higher in the SEUP sample than is estimated from the LFS gross flows data. This is mainly a result of the SEUP data having a lower transition rate from unemployment to NILF. This is consistent with the broader definition of unemployment used in the SEUP episode data.

The timing issue is more significant for the JS sample. At the chosen date, all episodes should either be of unemployment, underemployment or marginal attachment owing to the design of the JS sample. Another consequence of the sample design is that most episodes of employment and other NILF occurring at 1 June 1995 are not likely to be representative of those experienced by the general population. For this reason, the transitions from employment and other NILF for the JS sample are not presented in Appendix A, and these individuals are not included in the rest of the analysis. Unsurprisingly, the probability of unemployed individuals in the JS sample remaining unemployed after one month is also higher than that estimated from the LFS gross flows data, but there is not a statistically significant difference in this transition probability between the PRG and JS samples.


For more information on the SEUP, see ABS (1997, 1998). [2]

See ABS (1995) for further information about the weighting procedure. [3]

Although the original survey gathers the information to allow the official definition of marginal attachment to be derived, critical pieces of information are not included in the publicly released data set. [4]

If recall bias is significant, individuals may make systematic errors when asked about their past labour force experiences. On balance it appears that retrospective measures of labour force experience result in fewer spells of unemployment being reported than would be reported from contemporaneous surveys (Akerlof and Yellen 1985; Levine 1993). [5]

This excludes a small number of respondents who were actively looking for work but were not available to start work within four weeks and who are marginally attached according to the official definitions. It is not possible to separately identify this group using the public release data set. However, the numbers of respondents in the category will be small. In September 2000, this group made up only 2.2 per cent of the overall marginally attached workforce (ABS 2000). [6]

The interview date is not available in the public release data set. For all respondents, the wave 1 interview is assumed to be 1 September 1995. [7]

The labour demand reasons for not looking for work include: considered too young or too old by employers; lacks necessary schooling, training, skills or experience; difficulties with language or ethnic background; no jobs in locality or line of work; no jobs with suitable hours; and no jobs at all. [8]