A mixed latent class Markov approach for estimating labour market mobility with multiple indicators and retrospective interrogation
Section 3. The data

• Previous
• Next

The Continuous Labour Force Survey (CLFS), conducted by ISTAT (Italian Institute of Statistics), is the main and official source of statistical documentation on the Italian labour market. The CLFS has been conducted since 1969 and has been modified many times. In 2004, major updating was carried out, mainly dictated by the requirement to adapt the survey to new EU (European Union) standards. The principal changes involved interviews distributed throughout the years of the study, new criteria to classify respondents’ status in the labour market, computer-assisted data collection techniques, and dependent interviewing. Every year the survey collects information on about 280,000 households, for a total of about 700,000 individuals. The reference population consists of all household members officially resident in Italy.

The Italian CLFS sampling design has two stages: 1) municipalities were denominated as primary sampling units (PSUs) with stratification, and households as final sampling units (FSUs) with rotation. PSUs were stratified according to demographic size. Large municipalities, with population over a given threshold (also called self-representative municipalities), were always included in the sample; smaller municipalities (not self-representative) were grouped in strata, so that one municipality in each stratum was selected with probability proportional to its population; 2) households were randomly selected from the population registers in all municipalities drawn at stage 1.

The survey was quarterly with a 2-2-2 rotating design. Householders were interviewed in two consecutive quarters. After a two-quarter break, they were interviewed again, twice in the corresponding two quarters of the following year. As a result, each household was included in four waves of the survey over a period of 15 months. This rotation system meant that half of the sample remained unchanged in two consecutive quarters and in quarters one year apart, and 25% of the sample remained unchanged over three quarters.

All the following statistical analyses are made on the so-called longitudinal population. The CLFS is not designed as a proper panel: the initial population changes during the observation period due to demographic events and migrations. Although ISTAT has proposed a procedure to calculate longitudinal weights (Boschetto, Discenza, Lucarelli, Rosati and Fiori 2009), they are not available to researchers, so that we could not take into account the complex sample design. However, we consider that it was reasonable to assume that respondents belonging to the same households were independent.

Information on labour force condition in one reference quarter was collected three times: (i) each respondent was classified as employed, unemployed or not in the labour force according to the definition of the ILO on the basis of answers given to a selected group of questions; (ii) in a subsequent section of the questionnaire, all respondents were asked to classify themselves in the labour market, in order to collect the “self-perceived” condition; (iii) after one year, a retrospective question asked about respondents’ state in the labour market one year before the first interview.

According to the ILO definition, respondents were classified as employed in the reference quarter if, aged 15 years or over, during the reference week they performed some kind of work, for at least one hour, for pay, profit or family gain, or were not at work but had a job or business from which they were temporarily absent because of illness, holidays, industrial dispute, or education and training. Respondents were classified as unemployed if, aged from 15 to 74, they were: (a) without work during the reference week; (b) currently available for work in the two weeks following the reference week; (c) actively seeking work, i.e., had taken specific steps, in the four-week period ending with the reference week, to seek work or who did not seek work but who had found a job to be started later, within a period of up to three months (International Labour Organization (ILO) 2008).

Current self-perception and the retrospective question classified respondents in eight categories: employed; unemployed looking for new employment; unemployed looking for first employment; fulfilling domestic tasks; student; retired; disabled for work; other.

Table 3.1 shows the rotating design of the survey for two consecutive calendar years. Letters identify rotation groups: four rotation groups were interviewed in each quarter. With reference to one calendar year, information on labour market condition came from nine rotation groups. However, the rotation design generates a specific pattern of missing data. For example, for units of rotation group A who are interviewed for the fourth time in the first quarter of year 1, only the ILO (I) indicator and self-perception (S) of labour market condition in the first quarter of year 1 are available. For units in rotation group F, who were first interviewed in the first quarter of year 1, we only have information on labour force state based on the ILO definition, self-perception and the retrospective question (R) for the first and second quarters of year 1.

We examined data collected from 2005 to 2010. (Excluded from these analyses are data collected in 2004, the first year of implementation of the new labour force survey, because the data may not be totally reliable; with reference to 2010, here we use only information collected with the retrospective question and referring to labour condition in 2009.) Table 3.2 lists labour market composition in the first quarter from pooled data over the five-year period. The ILO indicator clearly counts a lower percentage of unemployed and a higher percentage of persons not in the labour force than the other two indicators. The two measures based on self-perception give a higher unemployment rate because ILO applies a very strict definition of unemployment. To be classified as unemployed, respondents between the ages of 15 and 74 must not be in employment at the moment of the interview but would accept suitable jobs in the next two weeks if the opportunity arose, and had actively looked for ways of obtaining jobs in the preceding two weeks. ILO provides these guidelines in order to facilitate comparisons of labour market performance over time and across countries (ILO 2008). However, this framework was set up when the prevailing type of employment was full-time and under permanent contract; since then, the employment situation has changed to one of more flexibility, with more part-time and fixed-term types of work, especially for those about to enter the labour market.

Other studies in the literature show that the distinction between labour market states is not always clear-cut: people may not know official definitions or perceive their labour condition as different from that arising from standard criteria (see, for example, Clark and Summer 1979; Flinn and Heckman 1983; Gonul 1992). In most cases, it is difficult to distinguish between unemployment and not in the labour force: the most critical condition seems to be that of actively seeking a job, since respondents may perceive themselves as unemployed even when they are not actively looking for a job. Inconsistencies may consequently arise between information collected in surveys and effective behaviour. Another explanation of the differences between the ILO and the self-perceived classifications is that respondents with temporary jobs in terms of hours of work per week may not classify themselves as employed.

Table 3.3 lists inconsistencies, i.e., different labour conditions observed for the same respondent with two indicators, among the three indicators for the period in question. Data over quarters and years were pooled for reasons of space. The number of inconsistencies is clearly higher for the state of unemployment than for the other two states, and most of the misclassifications tend to refer to people out of the labour force rather than in employment, as many previous studies show (see, for example, Poterba and Summers 1986). Comparing the labour condition according to the ILO definition with that reported according to answers to the retrospective question generated the highest number of inconsistencies. Examining consistencies over quarters and years for couples of the three indicators (not reported here for reasons of space) we note that consistency tends to increase slightly over time, perhaps because all the actors involved in the survey process - interviewers, respondents, etc. - learn how to collect and supply good-quality information while participating in the survey. Although we did not observe seasonal effects in the number of inconsistencies, the number of inconsistencies indicated non-negligible measurement error in the data, which means that one of the two indicators, or both, were reported incorrectly.

However, the inconsistencies emerging from Tables 3.2 and 3.3 may also occur because all three indicators are exposed to measurement error. Previous studies have investigated the causes of labour condition misperception, finding that it is influenced by social, demographic, economic and institutional factors (e.g., Richiardi 2002). Inconsistencies between the two self-perceptions (actual and retrospective) may mainly be due to memory decay (Bound, Brown and Mathiowetz 2001). Lastly, the higher consistency between the self-perception indicators suggests the possibility of correlated measurement errors.

Table 3.4 lists observed quarterly transition probabilities among the three labour force conditions from the first to the second quarter of the years from 2005 to 2009 with the three indicators. The ILO indicator describes a much more dynamic labour market, especially for unemployed respondents, than that described by the self-perceived and retrospective indicators. This difference is another piece of evidence revealing measurement error in the data. From the existing literature, we know that even small degrees of classification error may lead to severe bias in the estimation of transition probabilities (Hagenaars 1994; Pavlopoulos, Muffles and Vermunt 2012). If errors are uncorrelated over time, we can expect to observe a more dynamic labour market than the true one, and the opposite if error correlation over time also exists.

Table 3.5 compares observed gross flows, as an example from the first to the second quarter of 2005, by gender and age. The three age intervals were obtained by dividing the samples into three groups, with equal dimensions (i.e., 33^rd and 66^th percentiles). In detail, for the year 2005, in age 1 we find respondents aged between 16 and 36; in age 2 they are between 36 and 55, and in age 3 between 56 and 75. The evidence is that women are more dynamic, especially with regard to unemployment, than men. When leaving unemployment, women tend to leave the labour market more often than to become employed. There are also some important differences in observed gross flows across ages. The older respondents were more stable when out of the labour market and had higher probabilities of moving out of the labour market than of becoming unemployed after being employed. Younger respondents have lower probabilities than those in the second age-group of leaving unemployment and the condition of not being in the labour market by finding jobs. This evidence suggests that gender and age should be included as covariates in our model, to estimate corrected gross flows in the labour market.

• Previous
• Article main page
• Next