2.1 Scope and definition

2.2 Labour Accounting Systems (LAS)

2.3 Defining labour market dynamics

2.4 LMD questionnaire

2.5 Analytical units

2.6 OECD work

3.1 Breakdown of database

3.2 Support documents

3.3 Country coverage

3.4 General note on Section 3

4.1 Policy uses

4.2 Response to questionnaire request for information on policy uses

4.3 Literature sources

4.4 Institutional responsibility

4.5 Accessibility

4.6 International comparability

4.7 Data presentation

4.8 Developing countries

5.1 Study findings

5.2 Possible ILO action

5.3 Final considerations

Foreword

In 1996 the Committee on Employment Policies of the 83^rd session of the International Labour Conference requested the International Labour Office to develop an expanded range of indicators of labour market performance, and to assist member States in improving the collection of labour market information and widening the range of labour market indicators.

Much of the response to this request is being taken forward in a project known as Key Indicators of the Labour Market (KILM). The KILM project entails developing, maintaining and disseminating a database of up-to-date and relevant indicators for as wide a range of countries as possible. Guidelines and modules for measuring, collecting and analysing the selected labour market indicators are being prepared. The work is being carried out in close collaboration with the ILO Bureau of Statistics and the field structure, in order to achieve consistency and wide coverage in the data.

An initial list of 18 indicators has been chosen for data collection. These indicators will become known as the ILO Key Labour Market Indicators and will cover underemployment, educational attainment, poverty and income distribution, wages, labour costs and productivity. A new ILO publication (ILO Key Indicators of the Labour Market, to be published later in 1999) will present the KILM database, along with succinct descriptions of the situation in labour markets around the world.

The work described above is known as KILM Activity 1 and by its very nature can only concern itself with the more traditional and well-established statistical indicators. Activity 2 of the project will develop and promote new indicators, capable of revealing new insights in the labour market and the changing nature of employment. Bearing in mind that the 18 Activity 1 indicators are almost entirely 'static' or 'snapshot' measurements at a point in time, it was thought appropriate to begin Activity 2 with a study of the sources and analyses currently used to measure labour movement and flows, in other words labour market dynamics (LMD).

This paper presents the results of the study, which are based partly on an examination of literature on LMD, assembled by various means, and partly on questionnaire responses from national statistical offices. A shorter account of the study will be presented in one of the chapters of ILO Key Indicators of the Labour Market, but the ILO is publishing this near-complete version of the consultant's report in advance, as a response to the fast-growing interest in the measurement of labour market dynamics.

Gek-Boo Ng
Chief
Employment and Labour Market Policies Branch

1. Introduction

Increasingly, analysts and commentators realize that the most familiar and frequently used labour statistics - stock measurements of the number and composition of the employed and unemployed and derived net flows - provide only a limited contribution to understanding the workings of labour markets. This realization is not confined to labour specialists. There is a growing appreciation of the pivotal role of labour markets in understanding economic and social developments generally and a consequential widening of interest in labour dynamics.

This is because a labour market is a system in continual movement. Counts of employment, unemployment and people outside the labour force give only still pictures at a point in time; this is inadequate even when the comparatively rapid snapshots of a monthly series are available. Much of what is going on is not revealed. To shed light on the factors that underlie net changes in stocks, figures are wanted of gross as well as net flows between the count dates or periods; also the labour market history of people and other analytical units over several years.

This paper describes the ways in which the need for data describing labour market dynamics (LMD) are being met; the Terms of Reference are reproduced at Annex l. The study is based partly on an examination of documents compiled in literature searches by ILO Headquarters, within which OECD work - much of it published in Employment Outlook - is given a deservedly high profile; and partly on responses from 40 countries to a questionnaire (see Annex 2). The report also uses material assembled by the author when organizing a one-day session on LMD, held as part of the Conference of European Statisticians (CES) in June 1996, including the organizer's discussion paper prepared for that occasion. The author's attendance at meetings of the Groupe de Paris on Labour and Compensation over the past two years and the July 1998 KILM Workshop have also benefited this study. Collectively, the material gathered from these several sources falls short of being fully comprehensive, but should be generally representative of recent practices and, thereby, should bring out the main issues that need to be addressed in taking the topic further forward.

The layout of this paper is as follows: Section 2 deals with the coverage and other conceptual issues relating to LMD. Section 3 describes the database of references created for this study and provides some analyses derived from it. Section 4 discusses some key aspects that have emerged during the study. Section 5 draws some conclusions and offers some pointers for the future.

2. The scope and meaning of 'labour market dynamics'

2.1 Scope and definition

There appears to be no definition of the full scope of LMD in common use; in fact there have been scarcely any attempts to define the subject. This probably reflects the fact that, for most labour statisticians, it is not a recognized topic that can or should be separated from other labour statistics. In drawing a boundary round the subject it must be acknowledged, notwithstanding the introductory remarks made above, that the conventional stock figures are not devoid of information on dynamics. Time series of stock data give net changes which tell us whether or not inflows exceed outflows. More pertinently, a single observation of an unemployment stock analysed by short- and long-term unemployment is, in effect, an LMD statistic (and is included in the 18 Activity 1 KILM indicators as No.10). The same can be said about an employment stock analysed by length of job tenure.

2.2 Labour Accounting Systems

Labour Accounting Systems (LAS) are intended to be a comprehensive framework for looking at labour markets and therefore should encompass LMD. They have been described by Eivind Hoffmann as a means for 'the description and analysis of the state and dynamics of the labour market and its interaction with the rest of the economy (author's underlining) ....[including].....studies of 'gross' changes ('flows') in the number of jobs and persons and their activities'_. The author provides a description of LMD data in an LAS context as follows:

'...User[s] ...will focus on various changes, such as

(c) the net changes between reference periods in the number of persons in each status category;

(d) the total number of changes occurring in the reference period;

(e) the total number of persons who experience at least one change within the reference period; and/or

(f) the number of persons who have changed status from one period (or one reference date) to the next.

The numbers for (c)-(f) are only equal for short reference periods - periods which are too short for a post or a person to experience more than one change. We must expect that in practice an LAS will mostly be concerned with (c) and (f) type changes for reasons relating to the availability of data ...

... In discussions about LAS a fair amount has been said about 'flows' in the sense of gross changes from one reference date to the next, for example item (e) ... above. Making sure that all possible forms of such changes have been identified and estimated, given the periodicity and reference periods, is one of the 'accounting relationships' necessary within the LAS ...

On ... issues ... such as a core set of change (or 'flow') tables ... it would seem useful to develop international guidelines.'

LAS have scarcely developed beyond the theoretical stage and the few applications in individual countries have so far concentrated largely on stock data. However, this study has shown that Austria, Germany and Switzerland do have LMD data (gross flows) in an LAS framework.

2.3 Defining labour market dynamics

2.3.1 Some examples
To convey the idea of data describing LMD to other people, one has to resort to near- tautologies such as 'statistics that directly measure movement in labour markets'. More specific definitions tend to be rather lengthy; for example, a definition presented at the 1996 CES session was:

Measurement over time of changes in the activity status of individuals (employed, unemployed and inactive) and of changes in jobs of employed persons either in terms of persons who have experienced changes; or the duration of completed spells in status or job situations.

This definition excludes the dynamics of income from employment, and the demography and dynamics of workplaces. The latter is of interest in itself - describing aspects of entrepreneurial job-creating and job-destroying activity - and also as a means of explaining some changes in jobs and activity status.

2.3.2 Sources and methods approach
Another approach - perhaps particularly suitable for an audience of statisticians - is to get them to think in terms of the sources and methods that are used to generate data describing LMD. The classification used for the 1996 CES paper was as follows:

1. Surveys of individuals and households

1.1 Recall questions

1.2 Rotation sample designs

1.3 Longitudinal surveys

2. Establishment surveys

3. Registers and administrative records: persons

4. Registers and administrative records: establishments

5. Combination and other methods

This categorization was utilized to demonstrate that there is no ideal single source of LMD data - all of them had relative advantages and disadvantages; these are reproduced from the CES paper at Annex 3.

2.4 LMD questionnaire

2.4.1 Classification of LMD analyses
This 'sources and methods' approach was elaborated in the ILO questionnaire used to collect and code information on current practices for this report, but was also supplemented by a classification of types of LMD analyses. These are reproduced for convenience below and are also shown in the questionnaire included in Annex 2 (Questions 3 and 4 respectively):

Sources and methods for data describing LMD

Explanatory notes for most of these 'sources and method' codes were appended to the questionnaire; explanatory notes of 'types of analyses' mainly took the form of illustrations of LMD statistics (see pages 6-8 of the questionnaire included in Annex 2). By these means it is hoped that respondents to the questionnaire shared our understanding of the scope of the topic.

2.4.2 Design of questionnaire
The general approach to the design of the questionnaire was governed by two factors:

• to explain and clarify the meaning of data describing labour market dynamics and how they differ from the more familiar stock data;
• to achieve a satisfactory response rate.

This approach meant that the questionnaire could not be too lengthy nor complicated and thus imposed constraints on the classification systems used for 'sources and methods' and 'type of analysis' questions.

2.4.3 Classification of sources and methods
Regarding the classification systems used for 'sources and methods', although we have used this familiar term, it is more a classification of 'sources' than 'methods'. Even for 'sources' there is room for further elaboration. But for 'methods' there is much highly relevant information which was not gathered, for example on:

• the types of survey questions asked (or recorded in the cases where administrative sources were used) e.g. the date past labour market 'events' happened as opposed to establishing what was the labour market status a fixed period ago a year or quarter;
• whether 'dependent interviewing' techniques are employed or some other device to reduce error;
• whether specific weighting systems are used to compensate for the special problems of non-response and sample attrition which affect longitudinal surveys;
• whether the reporting unit for workplaces is the establishment or the enterprise.

If we had attempted to collect all this information it would have resulted in a questionnaire of great complexity which only a few enthusiasts would have completed. However, respondents were asked to supply references for their 'sources and methods' material and this gives us the means to probe further, possibly at some future stage.

2.4.4 Classification of types of analysis
Regarding the classification systems used for 'types of analysis', the approach was to use reasonably familiar research headings topics that respondents would easily recognize. A problem with this approach is that the ideal properties of classification systems of exhaustiveness and mutual exclusion are largely absent. There is work to be done in future to classify the types of analyses rigorously in this way. As a first step towards developing such a typology, the 'type of analysis' used in the questionnaire is mapped against the graphical expression of the familiar 3 X 3 transitions matrix as shown below. This shows the six flows between the three basic labour market states of employment (E), unemployment (U) and 'not in the labour force' (N).

Figure 1

This exercise brings out the following points:

• To achieve reasonable correspondence, a seventh flow - within E - is necessary. This shows that much of the interest in LMD analysis is about flows within employment - for example between jobs; full-time and part-time employment; industries; occupations; or between employee status and self-employment.
• LMD data are of two kinds:

(i) short-term gross flows indicating a single change of status; the focus here is on the status and the number of individuals entering and leaving it. Examples are categories A and G(i) above which - besides being statistics in their own right - are also used as supporting explanatory statistics for net changes in stock data.

• longer-term work history data, tracking successive changes of status - which often involves more than one spell in a status U, E or N. The focus here is the individual and what happens to him or her over an extended period. Examples are starred in the above diagram and they comprise the majority of the categories.

• The inability of categories C and D to fit into this model shows that the analytical unit of interest is not necessarily the individual person.
• It follows from (a), (b) and (c) above that LMD analyses go well beyond what can be shown in the above gross flow diagram (or its tabular equivalent, the 3 X 3 matrix).

2.5 Analytical units

It is worth commenting further on (c) in the previous paragraph. Although the usual analytical unit in LMD is the individual person, a full understanding also requires data on other analytical units:

• jobs (including vacancies) - their creation, duration and destruction;
• workplaces, enterprises, establishments - their demography (births, deaths and life cycle);
• events - for example, the beginning and end of labour market states: a hiring and a separation - defining a period of employment; or a separation and a hiring - defining a period of unemployment. This is described as 'episodal' data in the Australian Survey of Employment and Unemployment Patterns and is needed for the work history studies mentioned above;
• the household or family this is increasingly the focus of 'static' labour market studies and will surely apply equally to LMD studies as they develop.

2.6 OECD work

Moving from matters of general scope to particular elements of LMD studies, the investigations for this study show that the most sustained attempts to conceptualize or define specific elements of LMD - and (by implication) harmonize concepts and definitions internationally - have emanated from the OECD - either published in Employment Outlook (EO) or associated with the OECD in some other way. A digest of their work over the past ten years or so is presented in Annex 4. Particularly useful features of OECD work are the meticulous recording of national sources of LMD data (a valuable input to this study) and the frankness with which the formidable difficulties of international comparisons in this field are discussed.

The most frequently visited subject in Annex 4 is labour turnover - decomposed into hirings and separations, the latter further decomposed into quits (voluntary separations) and layoffs (involuntary separations). A distinction is made between labour turnover and job turnover, and there is an associated interest in job, enterprise and employer tenure, which appear to be alternative names for the same concept. All their work is valuable to the furtherance of LMD studies, but to the author's mind their special contribution is to draw attention to the need for more and better enterprise-based data, and the contribution this would make to understanding the process of job creation and loss. There is also interest in job-losers and displaced workers and unemployment turnover, i.e. flows in and out of unemployment relative to the stock of unemployed people.

3. Database of references

3.1 Composition of database

This section presents analyses from the database of references, which has entries for:

• each piece of literature identified during the study (see Introduction for their origin);
• all the questionnaires dispatched - whether or not a response has been received. In the case of those received, there is a separate entry on the database for each source/method separately identified in the completed questionnaire(s);
• every time a country's data has been used in an OECD study.

Each entry is grouped by country and coded, where possible, according to the type of:

• source and method;
• analytical topic.

3.2 Support documents

Three working documents were created to support the project:

• a schedule to record completed questionnaires received by the author via the ILO;
• a 'catalogue' of literature identified and received;

• the database of references, which is an Excel 4.0 spreadsheet recording the details mentioned in para 3.1 above.

Annex 5 explains the analysis base for each of the tables in this section and their relationship to each other.

3.3 Country coverage

Table 3.l ranks the 80 countries on the database by the total number of useful entries (column d) and each case shows the make-up of those entries as follows:

Column a. Country-specific literature entries identified, received, examined and coded;

Column b. Questionnaire entries received and coded;

Column c. OECD studies received, examined, and coded (see also table at para. 34 in Annex 4);

The table also shows, for the record:

Column e. Country-specific literature entries identified but not coded because they were not received;

Column f. Questionnaire dispatched but not received (also includes 'nil returns' reported indicated by #);

Column g. Total entries on the database.

[N.B. OECD studies (and others by international organizations) identified but not received are excluded from this and other tables.]

It can be seen from column d of Table 3.2 that for 17 countries no information was obtained (Côte d'Ivoire and those below), leaving 63 countries for which information has been used to produce the remaining tables in this section.

For each of these 63 countries, Table 3.1 shows the origin of the information on the database. For example, for Germany, entries were obtained from all three - literature references, questionnaires and OECD studies. In contrast, the only information available for Greece and Portugal was from OECD studies and the only information about Tunisia was from the questionnaire they completed. Mainly due to the effect of column c, the upper part of the table is dominated by OECD countries. Pages 26 and 27 in Section 4 discuss the coverage of developing countries. Overall, OECD studies contributed 38 per cent of the useful row entries, literature 34 per cent and questionnaires 28 per cent.

3.3.1 Sources and methods by country
All 63 countries identified in Table 3.l for which we have useful information are shown in Table 3.2, indicating the number of times a particular source and method occurs in the database. The column headings used for the table closely follow those used in the questionnaire - see paragraph 2.4 above; the meaning of the letter codes is shown at the end of the table. There is some duplication between the database references for a particular country - either because different research projects identified in the literature have used the same source, or because official statistical agencies have reported the same source in their questionnaire response as the literature or an OECD study. No attempt has been made to remove the duplication for two reasons: (a) in nearly all cases the derived analyses were different and had to be preserved to show complete analyses in Tables 3.3 and 3.4 to 3.10; (b) retaining all references is necessary to indicate the relative popularity and use of each source. The final row, beneath the total, shows the number of countries for which at least one column entry was recorded.

Table 3.1 Country coverage of database entries

Table 3.2 shows the contrast between countries such as Australia, some EU members and those in North America - for which LMD studies using all or nearly all of the sources are available; and others such as many developing and transition countries where only one or two sources are available. Looking at the column totals:

• the first four columns A, B, C and C1 nearly all relate to Labour Force Surveys of some kind or other and are clearly an important source of LMD data. The database records 31 countries using the longitudinal element of Labour Force Surveys (i.e. column C) to generate LMD data;
• longitudinal surveys of persons (columns D and E) are comparatively rare - recorded respectively for six and ten countries - and these are found almost exclusively in a few developed countries, reflecting the considerable resources needed to launch and maintain them;
• workplace surveys (columns F and G) are used more extensively than one might have expected for LMD analyses (see Tables 3.4(f) and 3.4(g) below), but note that their numbers in this analysis are exaggerated by the difficulty in distinguishing between F and G in some literature references - in which case both were recorded.
• administrative records or registers for persons (column H) and workplaces (column I) are equally represented in the totals, but the former are rather more evenly spread among countries (representing the considerable use of unemployment registers - also seen in Table 3.4(h) below) whereas the latter were used relatively intensively in countries such as Germany, Italy and the United States.

Table 3.2 Number of references to each source and method type, by country

3.3.2 Type of analysis by country
All the 63 countries identified in Table 3.l for which we have useful information are shown in Table 3.3, indicating the number of times a particular type of analysis occurs in the database. The column headings used follow those used in the questionnaire and the meaning of the letter codes is shown at the end of the table. The numerical basis of analysis in this table is much greater than in Table 3.2 - and there are fewer gaps - reflecting the fact that, many times, more than one type of analysis was reported for a given source (see Annex 8 for further explanation). The final row, beneath the total, shows the number of countries for which at least one column entry was recorded.

There were 141 instances of measures of 'Gross flows measuring transitions between labour market 'states' (column A) from 39 countries. The most frequently reported type of analysis was 'labour and job turnover rates' (column B ) - 45 countries. Other analyses frequently recorded were:

C. The creation and termination of jobs - 31 countries;

E. Job tenure - 32 countries;

G. Frequency and length of periods of unemployment - 36 countries.

Table 3.3 Number of references to each analysis type, by country