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

3.3.3 Analyses by source
Tables 3.4(a) to 3.4(j) show, in turn, the most common types of analyses derived from each source, ranked in order of popularity. (N.B. To avoid a long list tailing off with small numbers, only the analyses accounting for three-quarters of the total number of entries are shown in each table).

The most 'general purpose' sources are:

A. Ad hoc survey of persons or households: recall questions - seven types of analysis;

C. Regular survey of individuals or households: rotational sample design - six types of analysis;

D. Longitudinal survey of individuals or households: cohort design - six types of analysis;

E. Longitudinal survey of individuals or households: panel design - six types of analysis;

H. Registers or administrative records: persons - six types of analysis;

and the most specialized is:

I. Registers or administrative records: workplace establishments - two types of analysis.

Table 3.4(k), which is derived from Tables 3.4(a) to 3.4(j), gives a summary overview and ranks the number of times a particular type of analysis occurs in those tables. Two-thirds of the occurrences consist of:

A. gross flows measuring transitions between labour market 'states';

B. labour and job turnover rates;

E. job tenure;

G. frequency and length of periods of unemployment;

H. aspects of labour market flexibility and mobility;

which broadly confirms the findings of Table 3.3 on the relative popularity of the different types of analyses - reported in the preceding paragraph.

Table 3.4(a) Ad hoc survey of persons or households: recall questions

Ranking of most common types of analyses [total no. of entries: 120]

Code		No. of entries
G	Frequency and length of periods of unemployment	21
H	Aspects of labour market flexibility and mobility	15
A	Gross flows measuring transitions between labour market 'states'	14
B	Labour and job turnover rates	11
E	Job tenure	10
J	Absences to have or raise children and re-entry to the labour market	9
L	The move from work into retirement	9

Table 3.4(b) Regular survey of persons or households with a different sample each time:

recall questions

Ranking of most common types of analyses [total no. of entries: 168]

Code		No. of entries
B	Labour and job turnover rates	41
E	Job tenure	37
F	Job security	27
G	Frequency and length of periods of unemployment	15
A	Gross flows measuring transitions between labour market 'states'	11

Table 3.4(c) Regular survey of individuals or households: rotational sample design

Ranking of most common types of analyses [total no. of entries: 172]

Code		No. of entries
A	Gross flows measuring transitions between labour market 'states'	40
G	Frequency and length of periods of unemployment	27
B	Labour and job turnover rates	20
H	Aspects of labour market flexibility and mobility	20
E	Job tenure	13
F	Job security	9

Table 3.4(d) Longitudinal survey of individuals or households: cohort design

Ranking of most common types of analyses [total no. of entries: 42]

Code		No. of entries
K	The transition from full-time education into the labour force	6
E	Job tenure	5
I	The profile of individuals' earnings over a period of time	5
M	'Other'	5
B	Labour and job turnover rates	4
G	Frequency and length of periods of unemployment	4

Table 3.4(e) Longitudinal survey of individuals or households: panel design

Ranking of most common types of analyses [total no. of entries: 60]

Code		No. of entries
E	Job tenure	11
G	Frequency and length of periods of unemployment	8
A	Gross flows measuring transitions between labour market 'states'	7
B	Labour and job turnover rates	7
H	Aspects of labour market flexibility and mobility	6
I	The profile of individuals' earnings over a period of time	5

Table 3.4(f) Survey of workplace establishments: ad hoc

Ranking of most common types of analyses [total no. of entries: 83]

Code		No. of entries
B	Labour and job turnover rates	24
D	Births and deaths of firms and their life cycle	14
C	The creation and termination of jobs	12
H	Aspects of labour market flexibility and mobility	11

Table 3.4(g) Survey of workplace establishments at regular intervals

Ranking of most common types of analyses [total no. of entries: 133]

Code		No. of entries
C	The creation and termination of jobs	32
B	Labour and job turnover rates	31
D	Births and deaths of firms and their life cycle	24
H	Aspects of labour market flexibility and mobility	14

Table 3.4(h) Registers or administrative records: persons

Ranking of most common types of analyses [total no. of entries: 114]

Code		No. of entries
A	Gross flows measuring transitions between labour market 'states'	20
G	Frequency and length of periods of unemployment	18
B	Labour and job turnover rates	14
H	Aspects of labour market flexibility and mobility	14
I	The profile of individuals' earnings over a period of time	13
E	Job tenure	6

Table 3.4(i) Registers or administrative records: workplace establishments

ranking of most common types of analyses [total no. of entries: 87]

Code		No. of entries
C	The creation and termination of jobs	39
D	Births and deaths of firms and their life cycle	28

Table 3.4(j) Combinations of sources, or other methods

Ranking of most common types of analyses [total no. of entries: 63]

Code		No. of entries
A	Gross flows measuring transitions between labour market 'states'	28
C	The creation and termination of jobs	6
M	'Other'	6
H	Aspects of labour market flexibility and mobility	4

Table 3.4(k) Ranking of frequency of types of analysis, by source and method

(derived from Tables 3.4(a) to 3.4(j))

Type of analysis code

Source and method codes

	A	B	C	D	E	F	G	H	I	J	Total
B	x	x	x	x	x	x	x	x			8
H	x		x		x	x	x	x		x	7
A	x	x	x		x			x		x	6
E	x	x	x	x	x			x			6
G	x	x	x	x	x			x			6
C						x	x		x	x	4
D						x	x		x		3
I				x	x			x			3
F		x	x								2
M				x						x	2
J	x										1
K				x							1
L	x										1

Total

7

5

6

6

6

4

4

6

2

4

50

3.4 General note on Section 3

Question 2 of the questionnaire (see Annex 2) asked countries to supply information on LMD statistics under development as well as those currently or recently in operation. Where these were reported they have been included indistinguishably in all the analyses in this section. For the record, the following sources and methods under development for the following countries were included in the analyses:

A B C G H I

Ad hoc survey of persons or households: recall questions Survey at regular intervals of persons or households, with a different sample each time : recall questions Survey at regular intervals of persons or households: rotational sample design Survey of workplace establishments at regular intervals: Registers or administrative records: persons Registers or administrative records: workplace establishments

Kenya South Africa South Africa Netherlands Slovak Republic South Africa Norway Norway

N.B. Similar information for 'types of analysis' under development can be obtained from the database if required.

4. Important aspects

This section discusses in turn some important aspects of LMD data that have emerged during the project.

4.1 Policy uses

Generally one would expect the main demand for flow data to come from those concerned with formulating and monitoring labour market policies. This is because labour market policies are usually designed to stimulate flows in certain groups of the population. An examination of flow data helps to decide whether a policy is worthwhile and, if it is implemented, assists in tracking its effectiveness.

Policies to create employment usually take the form of trying to move people out of unemployment and into employment. But they can sometimes include stimulating or facilitating movements, within the employed population, from outdated and declining industries and occupations to those expected to expand in future. Other programmes, such as vocational youth training, are aimed at guiding young people into employment when first entering the labour force, rather than unemployment. Also, throughout the 1980s and 1990s, policies in an increasing number of countries have attempted to encourage the enterprise culture and in that regard there is interest in transitions into and out of self-employment, and activity in the 'small firm' sector. All these policies should require good quality LMD data.

4.2 Response to questionnaire request for information on policy uses

Does this study confirm or alter the above propositions? Respondents to the questionnaire were given an opportunity to provide information of this kind in Question 6 (see questionnaire at Annex 2). Unfortunately, the quality of the answers to this question was on the whole disappointing. Of the 40 countries returning a questionnaire, eight did not answer this question. Of the remainder, 22 provided answers that were not very enlightening, for various reasons, examples of which are set out below.

4.2.1 Less informative replies
Over-generalized replies

These answers could have equally applied to stock data or labour statistics of any kind.

The most succinct was:

'labour market policy'

and other typical replies were:

'To study the registered unemployed and their composition which include sex, unemployment duration, etc.'

'description of labour administration and steering this system (also as general labour market indicators)'

'To get detailed information about the mobility in the labour market and economic activity of the population'

'to quantify the employment situation in order to ease the decision making process of the government, the business sector and trade unions'

'la poursuite de l'évolution mensuel du chômage ... le taux mensuel du chômage au niveau national et régional'

'to provide current estimates on employment, unemployment and labour force ... on a continuous basis'

Circular replies

These replies were tantamount to saying that LMD data are used to monitor LMD indicators. Examples are:

'The main use of dynamic statistics are ... the measurement of entry, exit and rotation rates'

'To develop information about Socio-Economic Dynamics ... the relationships within and between the dimensions education, labour market, income ...'

Self-evident replies Replies stated that LMD data are used for statistical publications, research, conferences or inputs into other statistics or statistical systems. Examples are:

' for information at the Annual Conference of the Association of Labour Economists'

'the most important academic institutions use the ... data for several postgraduate theses and specialized research. They are also used in the Labour Secretariat for government planning and for the evaluation of the employment programmes'

'A number of indicators from the taxpayer register are used in compiling the balance of labour resources'; 'Used as a basis for drawing a sample of establishments when conducting sample labour force surveys'

'Disposer d'une information sur la situation de l'emploi à la veille de la préparation du IX^ème Plan du Développement économic et social 1997-2001'

4.2.2 More informative replies
Replies from the remaining ten countries were more focused and detailed and therefore rather more helpful. They were also generally longer, which prevents their reproduction here. The countries concerned were:

Australia

Austria

Brazil

Canada

Germany

Hong Kong (China)

Norway (enclosure A)

Poland

Sweden

United States and it is recommended that each reply to Question 6 be examined to appreciate the extent and variety of policy uses. Those from Australia, Canada, Germany and the United States were particularly carefully prepared. (The same applied to the response from France but it arrived too late to be used in this report's analyses.)

4.3 Literature sources

Regarding the literature sources, many were of academic origin and often did not mention a policy use. These might be described as examples of 'basic' as opposed to 'applied' research. Their stated purpose is to test economic theories or to investigate previously unexplored aspects of the labour market (although their results may well have some influence on policy eventually). This was sometimes the case for statements by statistical agencies: according to its 1994 report, the purpose of Statistics Canada's Survey of Labour and Income Dynamics (SLID) is

'to measure movement - to map out patterns of labour market activity and changes in income, and to record the events that trigger these changes'.

However, some pieces of literature were more targeted. The publicity booklet for the Australian Survey of Employment and Unemployment Patterns [SEUP] says:

'the survey is designed to provide information on the labour market and, more specifically, to assist in the assessment of the impact of labour market assistance initiatives in alleviating the extent of joblessness' and ' ... affords a unique opportunity to answer a number of questions concerning ... how social security payments affect individuals' labour supply behaviour'.

Some 14 policy questions expected to be answered by SEUP have been identified and several 'project outlines' using results from the Longitudinal Surveys of Australian Youth give considerable detail of the policy context.

An information leaflet for the British 'Lifetime Labour Market Database' (LLMDB) says:

'Specific ways in which the LLMDB will be used to improve policy development will include: Dynamic simulation;[e.g.] to massively enhance our ability to reliably model the future effects of pensions policy. Understanding Processes: by following a large number of individuals through their working lives, and examining the patterns of transition between the different work types and benefits, and finally into retirement, we shall gain an unrivaled insight into the factors which affect peoples ability to participate in the labour market'.

A random selection of other policy issues mentioned in the literature follows:

'... the focus has shifted ... to the rise in long-term employment and, more generally, the increase in unemployment duration ... Longitudinal data are valuable in this area not only because of the information on the beginning and end of spells but also because they allow researchers to control for unobserved individual-specific effects' (Jones and Ridell).

'... the authors use ... the Survey of Income and Program Participation to investigate whether employer-provided health insurance reduced worker mobility' (Buchmueller and Valletta).

'The analysis exploits the short-run capability of the Current Population Survey ... [which] allows some examination of the following important questions; is the experience of extensive unemployment in one year associated with extensive unemployment in the following year? How important are repeat spells of unemployment? ...' (Bowers).

'One of the major challenges of the transition ... will be to formulate effective labour market policies and programs, as well as an adequate social safety net ... This paper attempts to summarize existing hard evidence concerning the pattern of job mobility and changes in returns to education, experience, and gender ...The analysis of worker displacement and the effect of unemployment compensation on duration of unemployment is also presented. Data are drawn from unusually rich administrative data sets that include personal characteristics, work history and earnings ...' (Vodopivec).

'The key questions taken up here are:

How good an indicator is job turnover of adjustment in employment?

What flows of workers are generated by job turnover differences across OECD countries and what, if any, implications does this have for understanding unemployment?

What are the effects of policies on job turnover and labour turnover, and with what implications, if any, for unemployment?' (OECD, 1996).

'The purpose of this paper is to quantify some of the pressures being placed on the Canadian labor force by change ... Three questions are posed. Are intersectoral shifts in the relative size of industries important? To what extent are jobs reallocated because some firms expand and others contract? How does job change at the industry and firm level translate into worker separations?' (Baldwin and Gorecki).

4.4 Institutional responsibility

The author's prior perception was that much LMD analysis is carried out by academics and published in the academic literature; or that it is done for, and kept within, official policy circles. And that only few official statistical agencies analyse and publish the data themselves. This is in spite of the fact that, due to the cost and elaborate nature of LMD data-gathering processes, official statistical agencies almost always have the production and processing responsibilities. Because of the way the study was carried out, it has shed little light on this matter but has not discredited the above perception.

The questionnaire was sent to official statistical agencies and they were encouraged to distribute it to other departments, institutions and organizations in their countries which compiled or analysed LMD data (see covering letter at Annex 2). In fact, this invitation was scarcely taken up and virtually all questionnaires returned related to the activities of the national statistical agency or some other government department. However, some of the replies to Question 6 did indicate that research institutions and academics had access to the processed LMD data for the purposes of analysis and research. The database references originating from the literature were dominated by work by academics and research institutions, almost always using data produced by national statistical agencies. It was interesting to read the following remarks in the questionnaires from two Nordic countries:

'So far Statistics Sweden has not published or analysed data from the Swedish Labour Force Surveys in order to describe labour market dynamics ... if everything works out all right there will be a publication in 1999'.

'Description of 'dynamic aspects' of the labour markets have not gained much attention from the public. This (is) probably due to the fact that 'flow' statistics are rather difficult for users to interpret. Also changes in the structure of the flows seem to be quite minor. (The) same purposes can be fulfilled by frequent enough cross-sectional statistics. Dynamic statistics would require more interpretation by official statisticians' (Finland).

4.5 Accessibility

It follows from the above remarks that one of the reasons for LMD having such a low profile among users of labour market statistics, especially those outside official circles, is its lack of accessibility. This is due to a number of interrelated factors:

• LMD data are fragmented at present, non compatible and not thought of as a single, cohesive, topic; there is no overall conceptual framework;
• in disseminating labour market data, official agencies give stock data far more prominence;
• the academic studies using LMD data, and those published by international organizations, tend to appear in relatively obscure journals, are not well publicized and therefore not widely read.

In responding to the questionnaire, some countries volunteered information on availability of data, for example:

'Flows between employment, unemployment and not in the labour force, numbers and transition responsibilities available on request' (New Zealand).

Although first developed some 20 to 30 years ago, gross flows are not published by the statistical agencies in North America; they have only been made available to researchers and academics and some of their work has been published. In January 1997, in a paper to (the forerunner of) the Groupe de Paris, Statistics Canada said:

'As early as the 1960s, attempts were made to exploit the potential of the LFS to produce true flow data using the sample overlap ... However, the algebraic sums of the flows have been inconsistent with the net change in the corresponding stocks, leading to serious (but unprovable) response error biases. To date these gross flow estimates have never been published although a number of them have been available on request.'

However, through attending meetings of the Groupe de Paris, the author detects a renewed determination to tackle the technical problems in several countries.

In other cases the considerable resources needed to exploit LMD sources are not available, as is seen in the following remark in the questionnaire from Brazil regarding their 'integrated system of pensions':

'These registers allow the monthly follow up of a large part of the labour force and the enterprises. The possibilities of (this) database are almost infinite but they require a processing capacity which is too large for us.'

4.6 International comparability

As has already been mentioned in paragraph 2.6, over the past ten to 15 years the OECD has put considerable effort into international comparisons of trends of certain LMD data. Although not attempting to collect and publish these data on a regular basis, nor seeking to impose or propose any international definitions or standards, valuable preparatory work to this end has been done in their work in Employment Outlook and in other publications - which has been exploited in this report. And, although OECD has found scarcely any international harmonization, the organization nevertheless appears to have found the disparate country sources of some use in its economic analyses.

As far as the author has been able to establish, the best hope for standardization of LMD data, in the short term, lies with Eurostat influencing the 15 EU Member States (and other countries applying to be members). This is most likely to happen in the following developments:

European Community Household Panel (EHCP)

This is a longitudinal survey of persons with a considerable degree of harmonization across the Community. Part of the questionnaire is devoted to economic activity (labour force status). As with any longitudinal survey, a long time elapses before useful results begin to emerge. The author was informed by Eurostat that ECHP wave 2 (for all 13 wave 2 participating countries) would become available by the end of September 1998. Moreover, in early 1999, the second ECHP quality report will be submitted to Eurostat's Statistical Programme Committee. To prepare for this, a questionnaire was recently sent to all 14 ECHP National Data Collection Units. (One section of this deals with the use of ECHP data that is made by them and other researchers in their country.) Replies were expected in September 1998.

European Labour Force Survey (ELFS)

There are already common definitions of stock data across countries, in line with ICLS decisions; common coding of responses; and Eurostat partly processes and tabulates the results in parallel with national statistical agencies. Current efforts are devoted towards harmonizing the sample design - in particular moving more countries to quarterly frequency and introducing a standard rotating sample design. This should lead, in time, to the production of gross flow and maybe other LMD data on a standardized basis. There is still other harmonization work to do on the sequence of questions and several technical features of data collection and the weighting of results. Nevertheless, some work on LMD has been done using the ELFS.

Labour Market Accounts

The production of LMD data will, in time, also be stimulated by the accounting systems which are being developed by some EU countries with the encouragement of Eurostat - also involving Switzerland from outside the EU.

European Enterprise Panel Project

In 1995 Eurostat was commissioned to begin this project to study the effects of the single market on enterprises.

Guidelines of the European Council on Employment

The following types of statistical indicators will be required for this:

• the frequency and length of unemployment spells;
• the probabilities of unemployed young persons and adults finding employment, with or without being offered employability measures;
• the flows between employment, unemployment and inactivity for men and women;
• the duration of searching for employment.

4.7 Data presentation

Undoubtedly, one of the barriers to achieving greater knowledge and appreciation of LMD data among users is the difficulty of presenting the data in an easily digestible manner. The illustration on the following page demonstrates how complex the data can become. By the addition of just one more labour market state to the simple model presented under section 2.4.4 above (in this case separating self-employment (S-E) within employment), the number of gross flows generated is doubled.

Figure 2

With the addition of more labour market states the number of flows generated increases rapidly, as is shown in the following table:

No of labour market 'states' = N	No of flows = N(N-1)
3	6
4	12
5	20
6	30

Presentation is further complicated by the need to provide separate analyses for sub-populations likely to be of policy interest:

• males and females;
• young people and older workers;
• regions;

In longer-term studies there is also need to account for:

• births (the labour market equivalent is an individual reaching an age when it is legal or customary to enter the workforce) and deaths;
• immigration and emigration;

all of which generate additional extra flows that have an effect in the dynamics of the labour market and need to be taken into account. As pointed out earlier, much LMD data comes from work history data; this also has presentational problems. In its report on this subject to the 16^th ICLS, the ILO Bureau of Statistics presented an illustrative 'retrospective' typology of labour market experiences - to be more precise, a typology for the 'pattern of activities during year T for persons observed at the end of T'. This consists of 19 detailed categories grouped into seven main categories. The report observes:

' ... (this) is complicated in comparison with the basic distinction between E, O and N and also with a typology parsimonious enough to serve as a basis for newspaper headlines and cabinet briefs. However, relative to the total number of situations and 'careers' of analytical and descriptive interest over a reference of one year the typology represented by these 19 categories is quite simplistic'.

In all the literature received for this project, the author was hoping to see some imaginative examples of good practice in the presentation of LMD data. There may be some in the official statistical publications mentioned in answers to Question 5 of the questionnaire - most of which the author has not seen. But in the literature actually received - much of it academic or semi-academic in nature - there were scarcely any graphic or tabular presentations which effectively communicated the message of the data to the layman, and those which were received were only concerned with short-term flow data. No examples of graphical presentations of longer-term longitudinal data were found.

In the future development of LMD data, there is a danger that efforts will be concentrated on aspects of data collection and processing - definitions, sample design, weighting systems, international harmonization, etc. While all this is essential, it is equally important not to overlook the means by which the resulting data is to be communicated to busy decision-makers, overcoming the challenge of its novelty and complexity. An especially imaginative approach will be needed, recognizing the dynamic nature of the data - including perhaps taking advantage of the increasing possibilities of computer-generated animated graphics.

4.8 Developing countries

The Terms of Reference (see Annex 1) made a special point of mentioning that developing countries should be covered by this study. Accordingly, a selection of developing countries were approached in the questionnaire survey and they were particularly sought in the literature search. Using the categorization of countries used for KILM Activity l, the following tables are of interest:

Table 4.1 Response of countries to questionnaire

Grouping	No. of countries approached	No. of countries responding	Percentage responding
Developed (Industrialized)	18	11	61
Transition	15	10*	67
Developing	32	19**	59
Total	65	40	62
Derived from Annex 1 * inc. one nil return ** inc. two nil returns

It is heartening to see that the response rate for developing countries was similar to that for other countries.

Table 4.2 (extracted from Table 3.1) looks at the database references to developing countries picked up in the literature search, as well as the questionnaire responses. Of the 63 countries with useful references on the database, 27 were developing countries; however, they accounted for only 65 of the 483 corresponding row entries on the database.

4.2 Coverage of database entries: developing countries

	Literature coded	Questionnaires coded	OECD studies	Total a+b+c
	a	b	c	d
Argentina	2	3		5
Brazil	2	3		5
Hong Kong (China)		5		5
Mexico*	3	3		6
South African Rep.	1	3		4
Paraguay		4		4
India	3			3
China		3		3
Ethiopia		3		3
Nicaragua		3		3
Tunisia		3		3
Chile	1	1		2
Colombia	1	1		2
Egypt		2		2
Malaysia		2		2
Turkey*	1	1		2
Israel	1			1
Kenya		1		1
Korea, Rep. of*			1	1
Peru	1			1
Philippines	1			1
Venezuela	1			1
Indonesia	1			1
Morocco	1			1
Pakistan	1			1
Sri Lanka		1		1
Tanzania	1			1
Total entries * OECD countries	22	42	1	65

The sources and methods and types of analyses available for these 27 countries can be seen by inspecting Tables 3.2 and 3.3 respectively. Especially for the entries originating from questionnaires, further probing would be necessary to determine the quality of the data and, indeed, to verify that they really are LMD data.

5. Conclusions

5.1 Study findings

To the author's mind this study has broadly confirmed that, in spite of their obvious relevance to labour market policy issues for many years, LMD data are in a relatively primitive state. Some (interrelated) reasons for this are:

• LMD is not a topic recognized by many labour statisticians, separable within labour statistics generally, especially those in national statistical agencies;
• insofar as it is recognized, it is given low priority as a 'fringe' concern;
• there is no conceptual framework relating the strands of LMD in part reflecting that there is no all-embracing framework for labour statistics as a whole;
• there are scarcely any international standards relevant to LMD;
• LMD data are expensive to collect and process;
• the limitations of some LMD data require especially careful explanation e.g. the difference between current job or unemployment tenure and completed spells;
• there are severe difficulties in reconciling short-term flow data with stock data which, if they were more widely known among users, might unreasonably cast doubt on the quality of the stock data;
• long-term longitudinal data tends to be delivered late with the risk that they are no longer policy-relevant;
• there are severe data presentation problems, with complex tables and diagrams;
• the dissemination of LMD data is fragmented and unrelated.

5.2 Possible ILO action

The author's view is that, initially, the ILO can do little more than publicize the subject, educating its audience on the nature of the data, at the same time drawing attention to the dilemma of a body of statistics that are apparently in great demand yet are in short supply. It is certainly too early to start collecting data for purely statistical publications such as the ILO Statistical Yearbook. There is more of a case for assembling them first in a future KILM publication, perhaps on a illustrative basis, as part of the publicity and education process. This could be a follow-up to the envisaged introductory article. Even this modest step will be pioneering work; at present, no individual country seems to bring together all types of LMD data in this way. As far as the author is aware, the nearest any country comes to this is the UK, which publishes a guide to longitudinal data and this has a wider remit than just labour market data. LMD data comprise a variety of outputs the complicated results of longitudinal surveys, gross flows, probabilities of change in status, durations and for different analytical units. While we should never forget that this complexity reflects the complexity of the labour market (in fact LMD data simplify and summarize the underlying realities), a set of summary measures will have to be developed to give LMD more customer appeal. Judging from this study, the indicators that would attract most interest and would likely be most widely available are:

• unemployment inflows and outflows/unemployment turnover;
• hirings and separations/job or labour turnover;
• unemployment duration/probability of leaving unemployment;
• frequency and length of unemployment spells/work history over long periods;
• job tenure/probability of becoming unemployed;
• duration of job search.

It will aid understanding if these statistics are presented as part of a system describing LMD using a development of the presentation on page 5 rather than as isolated and unrelated indicators. Some examples of real data of gross flows between labour market states can be included where countries are prepared to see them published.

This study has shown (see Table 3.3) that only about 30 to 40 countries would currently be capable of supplying indicators of the above types. This figure may be under 30, if the number of countries for which data are being assembled for KILM 10 (Long-term unemployment a quasi-LMD statistic) is any guide.Only three countries on the KILM 10 list are developing countries, although this study (see Table 4.2) has suggested a few more possibilities. However, at this stage there is no need to put great effort into maximizing the number of countries for which data are shown. It is better to restrict the exercise to volunteer countries that are ready, willing and able to take part. Another point is that statistical progress has to begin with data confrontation: there should no qualms about including data which does not conform with international standards if they are the only LMD data available, e.g. unemployment inflows and outflows into registered unemployment.

5.3 Final considerations

One final point to consider. Probably even more than stock data, meaningful interpretation and international comparison of LMD indicators require an in-depth knowledge of the institutions, labour market conditions and demography of individual countries also the current stage of the economic cycle. This is not an argument for never publishing them in a purely statistical publication but it is another reason for a measured programme of preparatory work, which hopefully will stimulate an exchange of information and ideas between national statistical offices and indirectly improve exploitation of the microdata sets that are potentially a rich source of LMD data. To an extent, this is already happening through the Groupe de Paris on Labour and Compensation and the proposed actions through KILM should help to propel this work forward.

Annexes

Annex 1. Terms of reference

"The External Collaborator will prepare a report on the state-of-the-art of labour market dynamics measures, covering both developed and developing countries. This will be accomplished by analysing the responses to a survey questionnaire sent to 60 countries. The report will also be informed by analysis of documents and reports obtained in a literature search done at ILO Headquarters, with such reports submitted to the External Collaborator, as well as documents independently available to the External Collaborator.

The report will discuss the worldwide availability of data and studies ongoing or completed on labour market dynamics, the methodologies of the various approaches to the measurement and the policy uses of such data. Labour market dynamics includes but is not confined to studies of labour force flows, labour turnover, job creation and job destruction, and labour market transitions."

[Extract from Contract No. 3164, signed 3 April 1998]

Background note

This study is part of Activity 2 of the ILO's Key Indicators of the Labour Market (KILM) project.

Annex 2. Survey questionnaire

Questionnaires seeking information on countries' practices regarding data describing labour market dynamics were sent to 65 ILO member States early in 1998, under cover of a letter from the Director of the Employment and Training Department on behalf of the Director-General. There was no prior pilot stage. The questionnaire was prepared in three languages French, Spanish and English. The English language version of the covering letter and the questionnaire are reproduced in this annex.

Countries were asked to respond by end-March. Reminders to non-respondents were sent out between 27 March and 20 April. By the beginning of August, 40 countries had responded with completed questionnaires (of which three countries registered a 'nil return'). Responding countries are listed below; their cooperation in the survey is much appreciated.

Argentina

Australia

Austria

Brazil

Canada

Chile

China

Colombia

Croatia

Czech Republic

Egypt

Ethiopia

Finland

Germany

Hong Kong (China)

Hungary*

Kenya

Latvia

Lithuania

Mauritius*

Malaysia

Mexico

Netherlands

New Zealand

Nicaragua

Norway

Paraguay

Poland

Romania

Russian Federation

Slovak Republic

Slovenia

South Africa

Sri Lanka

Sweden

Switzerland

Tunisia

Turkey

Uruguay*

USA *nil return (i.e. answering 'no' to Questions 1 and 2).

The following 25 countries were approached but did not respond:

Algeria

Belgium #

Bulgaria

Central African Rep.

Côte d'Ivoire

Denmark

Djibouti

France #

Georgia

India

Ireland

Israel

Italy

Japan

Korea, Republic of

Macedonia

Mauritania

Papua New Guinea

Peru

Philippines

Spain

Ukraine

Uzbekistan

Venezuela #

Zimbabwe # Questionnaires for Belgium, France and Venezuela were received but unfortunately were too late to be included in the analyses in this report.

6463

EMP 62-2-02

CS/rf

Data describing labour market dynamics: An international survey

Dear Sir/Madam,

The most well-known labour statistics are measurements of 'stocks' - for example, the number of people in employment or the number unemployed at a point in time. Other familiar statistics are derived from these:

• the relationship between two stocks at the same point in time: " the unemployment rate - i.e. unemployment as a percentage of the workforce - was 6.3% in June";
• the difference between two successive stock measurements - the net change: "employment rose by 30,000 over the last month"

A major limitation of these stock figures is that they portray only a static picture of a dynamic system in continual movement. Labour market analysts have long demanded statistics of labour market dynamics that record the volume and speed of labour market movements and thereby reveal the factors that determine net changes in stocks. Yet in spite of this long-standing demand, statistics of this kind are relatively unknown and under-developed.

I am writing to seek your help in an international survey of data describing labour market dynamics. Our intention is to publish the findings to improve awareness of these data and their sources and methods - thereby transferring knowledge between countries and stimulating their further development. Eventually it may be possible to draw up internationally standard recommendations on methods, definitions and forms of presentation, based on best practice. For the present, however, our work is limited to discovering the extent, characteristics and use of these statistics among ILO member States.

Accordingly, I would be grateful if you could arrange for the attached questionnaire to be completed. It is possible that statistics of labour market dynamics are compiled or analysed by another government department, by a research institute or some other organization. Or perhaps this work is done in more than one organization in your country. Whatever the situation, I would be pleased if you could include all activities in your return, irrespective of the organization carrying it out. Alternatively, please make a photocopy of the questionnaire

- 2 -

for other organizations to complete and return separately. Please let us know the name and postal address of the other organization and the name, position and telephone number of the person responsible in each case.

Besides surveys or studies entirely or mainly devoted to labour market topics, please also include in this questionnaire those of wider social or economic scope within which labour market topics are one of the constituents.

If you wish, please qualify or expand on your answers by annotating the questionnaire, or use Question 7 to do so. The questionnaires will be processed manually and therefore written comments can easily be taken into account in the analysis. If you have any major problems with completing the questionnaire please contact:

either: Constance Sorrentino Employment and Labour Market Policies Branch (Room 8-47) International Labour Office 4, route des Morillons CH-1211 Geneva 22 Switzerland Tel (direct): 41-22-7996463 Fax: 41-22-7997678 E-mail: sorrentinoc@ilo.org

or * Peter Stibbard, ILO Consultant 62 Allington Drive Tonbridge Kent TN10 4HH United Kingdom Tel: +44 1732 358747 Fax: +44 1732 361064 or +44 1732 367231 E-mail: 106405.1433@compuserve.com [* responses available from 1 March onwards]

It would be very helpful to have your reply by 6 March 1998. Please send it to Ms. Constance Sorrentino at the address noted above.

Even if your answers to Questions 1 and 2 are 'no' please send in your return to prevent us troubling you with reminders.

Thank you.

Yours faithfully,

For the Director-General:

Werner Sengenberger

Director

Employment and Training Department

DATA DESCRIBING LABOUR MARKET DYNAMICS

AN INTERNATIONAL SURVEY

---

COUNTRY:

Please provide contact details below of a person who will be able to assist in any queries about the completed questionnaire and any other questionnaires returned by organizations in your country.

Name __________________________________

Position __________________________________

Postal address __________________________________

__________________________________

__________________________________

Telephone no. __________________________________

Fax no. __________________________________

E-mail address __________________________________

LABOUR MARKET DYNAMICS

Before completing the questionnaire please carefully study the explanatory notes

Mark correct answers like this

X

and leave incorrect answers blank

		Yes	No
1	Have statistics describing labour market dynamics been
	in your country during the last ten years
2	Are statistics describing labour market dynamics being
	the intention of publication within the next 5 years?
Sources and methods
3	If the answer to Q1 and/or Q2 is 'yes' please indicate whether or
	each of the following sources and methods are used:	Yes	No
	A. Ad hoc survey of persons or households: recall questions
	B. Survey at regular intervals of persons or households, with a
	different sample each time : recall questions
	C. Survey at regular intervals of persons or households:
	D. Longitudinal survey of individuals or households: cohort
	E. Longitudinal survey of individuals or households: panel
	F. Survey of workplace establishments: ad hoc
	G. Survey of workplace establishments at regular intervals:
	H. Registers or administrative records: persons
	I. Registers or administrative records: workplace
	J. Other methods or combination of above sources (please
	details)

LABOUR MARKET DYNAMICS

Type of analysis

4. For each of the 'yes' answers in Question 3 please indicate, in the second column, the type of analyses derived from these statistical studies. Use the letter codes provided at the foot of the second column. If there is more than one 'yes' answer to Question 3 please enter the appropriate letter A to J in the first column to distinguish them, together with the survey or project title. Space is provided for only three entries; if you have more please continue on a separate piece of paper using the same format.

Q3 reference letter and survey/project title	Types of analyses: please use letter codes at foot of this column and provide comments and explanations if you wish.
	Type of analysis:- A. gross flows between labour market 'states'; B. labour and job turnover rates; C. the creation and termination of jobs; D. births and deaths of firms and their life cycle; E. job tenure; F. job security; G. frequency and length of periods of unemployment; H. aspects of labour market flexibility and mobility; I. the profile of individuals' earnings over a period of time; J. absences to have or raise children and re-entry to the labour market; K. the transition from full-time education into the labour force; L. the move from work into retirement; M. other - please specify.

LABOUR MARKET DYNAMICS

Publication

5. For each of the 'yes' answers in Question 3 that relate to published work please give the full publication source(s) below, indicating with an X the types of information provided, using the boxes A, B and/or C. If there is more than one 'yes' answer to Question 3 please enter the appropriate letter A to J in the first column to distinguish them, together with the survey or project title. Space is provided for only three entries; if you have more, or if you have insufficient space to provide full bibliographical details, please continue on a separate piece of paper using the same format. It would be helpful if reproductions of references (or key extracts from them) could accompany your completed return. Alternatively, if references are accessible on the Internet, please give Internet address.

Q3 reference letter and survey/project title	Publication reference: also indicate type(s) of information provided in each case using the following code: A. latest data B. interpretative commentary C. description of sources and methods
	A
	B
	C
	A
	B
	C
	A
	B
	C

LABOUR MARKET DYNAMICS

Uses of statistics

6. Briefly describe below, for each of the preceding entries in answer to Q4 and Q5, the main objectives of the work and the actual or potential policy applications. Continue on a separate piece of paper if necessary.

7. If you would like to expand or comment on any of the answers given above, please use the space below, referring to question numbers as appropriate.

Thank you for your co-operation.

Explanatory notes

Q1 and Q2: Data describing labour market dynamics: definition and coverage. Before deciding how to answer Questions 1 and 2, please check through the statements using statistics of labour dynamics provided on page 8; also the different sources and methods used for compiling data on labour market dynamics which are listed in Question 3. In addition you may find the following remarks helpful: Data on dynamics are measurements over time in the activity status of individuals or changes in jobs of employed persons, either in terms of the number of persons who have experienced changes, or the duration of completed spells in status or jobs. The dynamics of income from employment and the life-cycle of workplaces are also relevant. Data on dynamics depends on collecting information about the same person (or some other analytical unit such as a household, job or workplace establishment) relating to two or more points in time; they usually reflect experiences at the level of the individual person. An analogy often used to illustrate the difference between stock figures and data on dynamics is to liken the former to 'snapshots' of a labour market in continual movement - whereas the latter are 'videos' recording those movements.

Q3A and Q3B: Recall questions. Flow data can be derived from retrospective questions in surveys, for example by asking about labour market status at some past date, say a year previously, and about prior status. Also, retrospective questions can be asked about how long ago certain events happened (e.g. when the respondent's present job began). These can be asked in single ( i.e. ad hoc or 'one off') surveys, or in surveys conducted at regular intervals such as Labour Force Surveys.

Q3C: Rotational Sample Designs. Many Labour Force Surveys include a rotational scheme in their sample design. The sampling units are partially replaced according to a prescribed pattern, so that a reducing proportion of the sample remains in successive survey periods. Although increasing the precision of monthly or quarterly estimates of changes of levels (stocks) is the prime reason for these designs, flow data can be produced as a by-product. By linking the labour force status at these different points in time for individual respondents, statistics of gross flows between each status, in both directions, can be compiled.

Q3D and Q3E: Longitudinal survey of persons or households. These are surveys deliberately designed to monitor the behavior and experiences of individuals or households over quite long periods of time. They can take the form of cohort surveys, in which a sample of individuals of a particular age is selected at a point in time and then re-surveyed at intervals; or panel surveys, in which a sample representative of the population is selected and then re-surveyed at intervals.

Q3F and Q3G: Survey of workplace establishments. Labour turnover statistics can be derived by collecting aggregate information on the number of employees at the end of the reference period who were not employed at the beginning of the period; or by asking direct questions on the number of employees joining and leaving. There are also surveys of establishments which seek information on individual employees, enabling analyses which link firms' characteristics with employee characteristics. Establishment surveys are usually at regular intervals and involve a panel of establishments but ad hoc surveys collecting historical information are sometimes used.

Q3H: Registers or administrative records: persons: Gross flow statistics can be derived from administrative registers which record successions of events pertaining to individual persons (e.g.hirings by and separations from employers, or registration with employment services) and their dates; or which match records of individual persons at different points in time. They can be used to show how long individuals were claiming unemployment benefit, and how many spells of unemployment they experienced. Such registers are often associated with taxation or social security schemes. Sometimes the information about the same individual persons and/or establishments in more than one register can be combined through the use of common identifiers.

- 6 -

Q3I: Registers or administrative records: workplace establishments. Registers of establishments can generate statistics used to study certain aspects of labour market dynamics. The traditional focus on net change in employment totals derived from establishment surveys hides most of the dynamics of employment. Irrespective of whether net employment is rising, falling or constant, large numbers of jobs are being created and terminated. Data are needed where the jobs are being created (in existing or new establishments?) and where they are being terminated (in existing or closing establishments?). Survey data on labour turnover can supply a partial answer (see Q3F and Q3G) but register analysis, preferably linked with surveys, is potentially a better source for these data, as it provides 'business demography' data on establishment openings and closings. Thus it is possible to link data on labour turnover and establishment turnover.

Q3J : Other methods and combination of sources. Administrative records, supplemented by surveys, can provide useful data on the operations and outcomes of government training programmes. Individual 'starters' and 'leavers' records can provide overall numbers of flows and stocks while on the programme, also demographic details on participants and operational information on actual duration of training, qualifications achieved and immediate destination on leaving (if known). 'Follow-up' surveys can then be conducted after an interval of six or nine months to seek information on participants current activity. These provide information on the progression into employment.

Q4. Type of analysis. To correctly identify the analyses derived the different sources' please enter the Q3 letter code in the first column - also the survey/project title or name; this will be particularly helpful in distinguishing two or more sources of the same type (for example, there may be more than one longitudinal survey with labour market content in a country). Pre-coded 'types of analysis' are listed at the foot of the second column to reduce the amount the text you have to provide; however, add further information if you wish. Space is provided for only three entries; if you have more please continue on a separate piece of paper.

Q5. Publication. To reduce the amount of information you have to provide please direct us to published material, indicating the type of information provided, using one or more of the letter codes A, B & C. Where there has been several pieces of work derived from a particular source, or more than one instance of a particular source, please give priority to recent publications and to articles which provide good descriptions of sources and methods . Enter full bibliographical details and, where possible, provide photocopies or offprints of articles and papers. References to publications in any language are acceptable but those in Spanish, French, [Russian] or English are preferred. Space is provided for only three entries; if you have more please continue on a separate piece of paper.

Q6. Uses of data. Please briefly describe the main objectives, uses and policy applications of the analyses. Space is provided for only three entries; if you have more please continue on a separate piece of paper.

The following statements make use of data describing labour market dynamics [Q4 letter codes for types of analysis are provided in each case for guidance]:

'The net increase in employment in 1995 of 200,000 was the outcome of gross flows between the three basic labour market 'states' of employment {E}, unemployment {U}and economic inactivity outside the labour force {O}, as follows:-

500,000 U to E ; 400,000 E to U ; 700,000 O to E ; 600,000 E to O' [A]

'Men are more likely than women to make the transition from full-time to part-time employment and less likely to make the converse transition from full-time to part-time' [A]

'12% of people in employment have been in their present job for less than one year' [B]

'20% of the unemployed have been unemployed for more than a year' [B].

'last month 350,000 people were added to the official count of the unemployed and 320,000 left the count - a turnover of 15%' [B]

'6% of employed people changed their occupation during last year compared with 8% in the previous year' [B, H]

'In the past year the creation of new businesses and the growth of existing ones created new jobs that amounted to 14% of total employment; whereas the contraction and closing of businesses simultaneously was responsible for the termination of jobs amounting to 13% of total employment'[C, D]

'Two thirds of women are continually employed by the same employer for more than two years' [E]

'Median job tenure is 3.5 years and 11 years at the 90th percentile' [E]

'The jobs taken by the previously unemployed are usually temporary and these individuals subsequently either upgrade or leave those jobs' [H]

'About one third of unemployment spells last less than 3 months' [G]

'Between 1990 and 1995 about 10 million people experienced at least one period of unemployment' [F,G]

'The number of jobs a person can expect to have in a lifetime has risen sharply over the last 20 years from 7 to 11' [F,H]

'individuals that escape from low paid jobs are more likely to re-enter low-paid jobs in subsequent years than those with no previous experience of low pay' [I]

'The percentage of young people in gainful employment in November of the same year as they completed their secondary education has decreased dramatically in the 1990s' [K]

Labour turnover is rising markedly for older workers ... primarily through increased exits from the employed labour force to early retirement' [L]

Annex 3. Sources of data: their respective advantages and disadvantages

Surveys of individuals and households: recall questions
Advantages

• Convenient; relatively inexpensive;
• Produce quick results.

Disadvantages

• Errors due to respondents' memory lapses, made worse by proxy responses;
• Less scope to employ detailed questions to determine previous status than present status; thus a risk of inconsistent definitions which would affect accuracy of resulting flow data.
• Analyses based on questions about status at some past date exclude additional changes of status since then.

Surveys of individuals and households: rotation sample designs
Advantages

• Absence of recall errors;
• Flow data based on same definitions as corresponding stock data.

Disadvantages

• Low response rates and high attrition rates affect quality of data;
• Sampling error of flow data greater than corresponding stock data;
• Difficulty of establish links between successive interviews when respondents change their address can cause bias in results;
• Ability to provide data on particular flow periods (month, quarter, year) depends on suitability of sample design;
• Conditioning effect of repeated interviews;
• Generally not suitable for deriving long-term flow data;
• Necessary to wait until after termination of reference period for flow data before results are available;
• Spurious flows may be generated if people respond differently to face-to-face and telephone interviews.

Surveys of individuals and households: longitudinal surveys
Advantages

• Absence of recall errors;
• Flow data based on same definitions as corresponding stock data.

Disadvantages

• Expensive to operate and require a long-term budget commitment;
• Attrition affects quality of data;
• Rarely suitable for providing regular and timely data for national statistical infrastructure.
• Problems of tracking households over time.

Establishment surveys
Advantages

• Employment statistics derived from establishment surveys tend to be more accurately coded to industry than those from household surveys;
• Analyses by size of establishment or enterprise possible.

Disadvantages

• Requires special record-keeping by respondent establishments which may affect speed and level of response;
• Coverage of employees may be affected by accuracy and up-to-dateness of register used to select sample or panel of enterprises;
• Very limited information available on the socio-economic characteristics of employees joining, leaving or retained in the establishment without increasing compliance burden significantly.

Registers and administrative records: persons
Advantages

• Less subject to measurement errors;
• Register analysis usually less expensive than surveys unless register especially constructed for statistical use only;
• Non-response not a problem (although incomplete or out of date records can be a difficulty);
• Comprehensive coverage facilitates analysis of small geographical areas and other sub-populations.

Disadvantages

• Concepts, definitions and accuracy coding usually mainly determined by administrative rather than statistical requirements;
• Vulnerable to changes in administrative rules and practices which could exaggerate true flows;
• Use of registers, especially in combination, raises confidentiality issues.
• Problems of tracking analytical units (i.e. households) over time.

Registers and administrative records: establishments
Advantages

• As for 'person' registers.

Disadvantages

• As for 'person' registers;

• Problems of tracking analytical units i.e. enterprises over time;
• Very limited information on the socio-economic characteristics of employees unless linkage established with person registers.

Annex 4. OECD work

1. This digest of OECD work over the past ten years or so is generally but not entirely presented in chronological order this does entail some repetition but it also captures the development of thought.

2. Employment Outlook Chapter 2 (EO90 Ch.2) discussed Displacement and Job Loss and the approach 'imposed by the availability of relatively comparable data' was to identify and analyse two 'apparently similar' groups of workers:

• job-losers; i.e. those who permanently lost a stable or temporary job in the last few years and are currently unemployed (U) or out of the labour force(N)
• displaced workers; i.e. those who permanently lost a stable job in the last few years and are currently unemployed , out of the labour force or re-employed.

The former are analysed as a stock, based on replies to labour force survey questions on the reason for being U or N, the last job held, etc. and at the time could be analysed for 15 OECD countries. Analyses of displaced workers, on the other hand, rely on flow analysis of what happens to these workers after displacement. These studies are based on longitudinal or retrospective follow-up surveys of widely varying scope in a few countries in the 1980s. OECD said '... undeniably the displaced workers ... are the most interesting group (for) ... analysing the ... effects of structural change' but the lack of data at the national level made it necessary to concentrate on stock analyses of job-losers instead. The flow analysis of displaced workers in the Chapter was restricted to:

• matched files for labour force surveys Australia, Spain and USA only;
• national surveys of displaced workers USA and Canada only;
• examples of case studies of redundancies from a single enterprise;
• examples of studies of groups of workers made redundant.

_____

3. A paper prepared for December 1992 meeting of the OECD Working Party on Employment and Unemployment Statistics looked at labour turnover data and (their part in) labour market analysis (and made proposals for collecting these data from quarterly labour force surveys).

4. At the beginning of Part 1, entitled 'Turnover - a conceptual and definitional overview', it is pointed out that, to an employer, the term 'turnover' usually means the loss of employees through quits to other firms or activities, often requiring the hiring of replacements. But labour market analysts use the term in a wider sense. Attention might be focused on 'job turnover' or the loss of jobs in certain firms or sectors and the creation of new ones. However, the term is 'more commonly used to refer to aggregate measures of gross change in employment for establishments; it includes employment inflows (hirings or accessions) as well as outflows (separations)'.

Hirings or accessions occur to fill new jobs, or jobs directly or indirectly freed up by a quit, a dismissal, a retirement or some other reason. Hirings can fill either temporary or permanent jobs. Hiring data can give a picture of the jobs available over a period but only a general one in that unfilled vacancies are excluded and some 'hirings' may be left out of data collected from firms because the jobs may not be publicly advertised or made generally available to all prospective candidates.

Separations consist of two types of transition, quits and layoffs, which should be distinguished:

Quits (worker-initiated separations): these can occur for job-related reasons, because workers seek better pay, opportunities or working conditions. Or for personal reasons pregnancy, a return to full-time education, illness or retirement. Generally, jobs vacated as a result of quits are eventually filled, except where firms take advantage of them in order to downsize or when workers quit in anticipation of a layoff.

Layoffs (employer-initiated separations which may be temporary or permanent): layoffs usually result from a decline in business activity, firm closures, technological change, redundancies caused by mergers, cost-cutting measures or other reasons related to the economic situation of the firm. However, they can also occur for reasons related to the individual, e.g. dismissal for dishonesty or incompetence.

5. Different denominators can be used to calculate hiring rates and separation rates over a particular period. Those used in analysis have included:

• number of employees at the beginning of the period;
• number of employees in a reference week during the period;
• annual average number of employees;
• number of person-jobs (one person-job = one job held by one person) over the period.

Data on hiring rates and separation rates can mislead if it is not accompanied by other information, including the age structure of the workforce, the state of the business cycle, the size of firms, occupational structure and the prevalence of temporary jobs.

6. Part 1 of the paper continues with a lengthy discussion of the uses of turnover data and concludes with a classification and critique of sources of turnover data, the key points of which are reproduced in the table below:

7. Part 2 of the paper includes some observations on the availability and comparability of turnover data in OECD countries. The main points are: Labour Force Surveys have a rich potential which is rarely realised in practice because the right questions are not asked or, if asked, are not standardized. Where they exist, Work History surveys provide good data on labour mobility. In countries with establishment data, coverage is often limited to manufacturing industries with little additional information collected; only two countries had administrative data. Sources that are limited to larger establishments understate turnover because it is usually higher in small establishments and turnover also varies considerably between industrial sectors.

8. EO93 Ch.4 focused on enterprise tenure derived from household survey questions about how long employees have been with their present employer. It is pointed out that this is a measure of 'spells in progress', not of completed spells and goes on to say 'it is only with exacting datasets that all completed spells of employment could possibly be captured'.

Enterprise tenure is said to complement measures of labour turnover when individuals leave or are dismissed by their employer.

Annex 4.A gives sources of enterprise tenure data from 13 countries for 11 of them published or unpublished data from labour force surveys or other household surveys, from employer responses mainly in the case of Japan and from an administrative source in the case of Finland.

9. The analytical unit in the above studies was the individual worker and the data came mainly from surveys of persons. EO94 Ch.3 studied the related topic of job gains and job losses in firms, and here the main focus of study was the job and the firm. Developing some earlier work in EO 1987, this chapter develops a conceptual framework for studying job turnoverand labour turnover.

The chapter defines:

• jobs as filled employment positions;
• job turnover is based on comparisons of the stock of employment at a workplace at two points in time - only net job changes are counted. True job turnover is therefore underestimated and moreover changes in unfilled vacancies are not included;
• labour turnover (distinct from job turnover) as the movement of individuals in and out of jobs.

A graphic (reproduced on the next page) explains the relationship between these concepts and includes the concept of 'excess job turnover'.

10. Annex 3.A of this issue of EO provides much detail on the sources, definitions and methods of data collection for the data presented in the chapter. Data from ten countries are used; there are plenty of warnings of pitfalls of international comparisons in this subject, which include differing:

• employment coverage (e.g. treatment of the self-employed);
• industry coverage;
• industry classification;
• type of analytical unit establishment or enterprise/firm.

As an example of the problems of definition, interpretation and comparison in this field, page 108 is devoted to the difficulties in the measuring job turnover in a USA database.

11. In November 1994, a meeting of experts was convened by OECD on Job Creation and Loss: Analysis, Policy and Data Development. This meeting was a result of the March 1994 G-7 Jobs Summit asking the OECD to 'push forward its work on the dynamics of job creation and job loss, focusing both on improving data comparability and deepening ... analytical understanding ...' A number of papers were commissioned (and subsequently published) and the brief introductory paper by Bowers and Grey said that the work 'was partly driven by the recognition that there is a lack of internationally comparable data about the processes through which workers find, lose and upgrade their jobs, as well as little information on establishments' hiring, firing and adjustment strategies'. Three of the papers addressed questions about concepts, definitions, sources, methods and the main points from these are presented below.

12. A paper by Blanchflower examines research questions arising from the uses of establishment-based data. By way of introduction, the paper says the policy-maker needs to know what kinds of policy helps or hinders job creation and admits that theoretical analysis on job creation and destruction is at a 'very rudimentary stage' and the empirical evidence does not take one very far because of the considerable number of problems in comparing data between countries.

13. Blanchflower makes a plea for the increasing availability of micro-data at the level of households or individuals over the last 20 years ( he gives examples from Germany, UK, USA) to be matched by data collected from workplaces to be available at the level of the firm or the establishment. A reason for the scarcity of these data is worries over confidentiality and he gives examples how this is being overcome in Canada, UK and USA. He also says that data are becoming available that match information on employees and their employers; and match establishments on the owning firm. This is followed by a discussion of the relative merits for various purposes of using the establishment or the firm as the unit of observation.

14. He produces a formidable list of 17 definitional problems that arise in using establishment-based data for empirical work, especially in making international comparisons, all of which would need attention in any attempt to harmonize. Summarizing, these are:

1. Representativeness of sampling frame/sample; stratification, size cut-off.

2. Up-to-dateness of sampling frame.

3. Definition of establishments initially and then through time when premises or ownership changes

(counted as death and birth?).

4. Sectors covered treatment of services, public administration, agriculture.

5. Industry definition when a variety of products are produced.

6. Treatment of part-time workers.

7. Measurement of labour inputs by hours.

8. Treatment of the self-employed and family workers.

9. Subcontracted work.

10. Job characteristics that indicate 'good' or 'bad' quality.

11. Structure of the labour force over the economic cycle; labour hoarding.

12. Treatment of temporary layoffs.

13. Definition of 'large' or 'small' firm measured by turnover, assets as well as employment.

14. Definition of denominator in calculating job creation rate.

15. Suitability of a standard measure of 'small'.

16. Information on other factors of production besides labour.

17. Position of respondent in organization.

15. After working through a long list of information needs required 'to open up the workings of the black box that is the firm' he proposes, agreeing with Hamermesh, that an ideal data series should have the following properties:

(a) a monthly survey based on a random sample of establishments, defunct workplaces replaced with appropriate substitutes, with information collected on:

• employment desegregated into several skill and demographic categories, including sex, race, broad occupation and age;

• hours, both paid for and worked for above categories;
• flows of workers into and out of establishments for above categories;
• payroll and other costs for above categories;
• output;
• inventory and capital expenditure.

(b) A sample of workers from the establishments should be followed while they are employed at the establishments, with substitutes when they leave. Data to be collected on demographic characteristics, self-reported earnings and hours, including those associated with second jobs held. Corresponding data on other household members is also desirable.

• Complement the establishment data at (1) with detailed interviews at intervals with owners or managers to examine the decision- making process and the factors influencing it.

Blanchflower concludes 'The only way to properly inform ... policy is to have high-quality, matched panel data across countries on individuals and their employers that are comparable. Such data files need to contain a lot more detail than has been available in most existing establishments data sets, which are usually based on administrative records ... [Also] we need to observe more directly what is going on by interviewing interested parties'.

16. The second paper, by Grey, begins by repeating the definitions of labour turnover and job turnover, and the graphic, reproduced in this Annex as Chart 3.1, and quotes Hamermesh as follows 'While they are extremely interesting in their own right, data on gross flows tell us nothing about the magnitude of the wage or output elasticities of employment changes through the births or deaths of establishments or growth or contraction in existing establishments'. There are references to panel studies of enterprises or establishments in several countries (p.41) and also the recent development in a few countries of longitudinal information linking establishments and workers.

17. The third paper, also by Grey, is a comparative overview of sources, definitions and methods of data collection on job creation and loss also begins by repeating the concepts of EO94 Ch. 3 on labour turnover and job turnover and then discusses the type of information needed to understand job turnover and makes the following points:

• the data can come from a number of sources but there must be some degree of longitudinality;
• although (by design) they carry limited information on individual establishments and enterprises, business registers do have longitudinal information, and surveys can be associated with them or administrative sources linked to them;
• for several reasons the establishment is the preferred unit of analysis but there are many definitional problems, some of which are being tackled by Eurostat;
• however, enterprise level data are also desirable;
• definitions of opening and continuing establishments are difficult to operate;
• the data needed from each workplace includes:

turnover/sales

output/value added

employment part- and full-time

wages and salaries

occupation

capital stock

investment (enterprise level)

research and development (enterprise level)

Chart 3.1 Components of job turnover

a) Absolute values.

Source:OECD Employment Outlook, September 1987.

18. There follows an extensive survey of country sources for the G-7 countries and Denmark and Sweden mainly displayed in three tables:

Table l. .Definitions of enterprises and establishments

Table 2. Definitions of opening and closing

Table 3. Variables available

An annex provides further details of country data sources for 12 countries and a bibliography.

19. EO96 Ch.5 is entitled Employment adjustment, workers and unemployment and looks at data on job turnover and labour turnover and unemployment inflows and outflows. It defines job turnover as 'the sum of over-the -year changes in employment levels across all establishments' and says it is a 'reasonably comprehensive measure of employment adjustment in that it incorporates both the reallocation of employment across industries as well as the reallocation of individual establishments across productive units within industries.' While it leaves out the turnover within establishments it is still regarded as an important tool as policymakers tend to focus on hiring and firing at the establishment level.

20. A panel (p.165) uses a simple numerical example to illustrate the difference between labour turnover and job turnover, adding that 'labour turnover is the sum of job turnover, which relates to the expansion or contraction of establishments of firms, and the movement of workers into and out of ongoing jobs ...Workers leave firms and firms hire other workers to replace them , regardless of whether the firm itself is growing or declining.'

21. On page 166 it says that 'labour turnover measures changes in individuals among jobs regardless of whether the jobs themselves are newly created, ongoing (and subsequently filled by others) or whether the job themselves disappear' and goes on to outline the relationship between labour turnover, job turnover, job creation and job destruction.

22. Discussing the links between labour turnover and unemployment, the article says that job turnover is a component of labour turnover, measured from the perspective of employment change that led to a hiring or a separation. Flows in both directions between employment and unemployment are also components of labour turnover. But new job creation or the filling of a job vacancy does not lead to changes in either the stock of the unemployed or to an increase in outflows from unemployment. Explanations include:

• new jobs may be taken by those already in work who change jobs;
• some new jobs may be taken by people previously inactive;
• people can drop out of the labour force into inactivity.

23. The article concludes by reiterating the point that 'the development of more comparable data sources on job creation and destruction ... is crucial. Understanding how adjustment takes place and the implications for unemployment requires ... compatible data on labour turnover, in particular, accurate measures of flows in and out of unemployment and flows from one job to another.' An annex provides sources and methods for seven countries and discusses the difficulty of comparing them.

24. Three other recent articles cover other aspects of labour market gross flows.

25. EO95 Ch.1 is a review of recent labour market developments and prospects and includes in Section D an examination of the duration of unemployment, and unemployment flows. It justifies the latter by pointing out that 'the dynamics of labour markets are only partially captured by an analysis of changes in the stock of unemployment and that part of it that is deemed 'long-term unemployment'. Underlying these changes are much larger flows between the three labour force states ...'. It does this by constructing 'proxy' flows in and out of unemployment for 19 countries (using a method introduced in previous issues of EO see box below) and, for seven countries, mainly using matched flow data, looking particularly at the probability of leaving unemployment and the destinations of those who do.

The proxy data for unemployment inflows are stock data for those unemployed for one month or less. The proxy data for outflows are calculated by an identity linking changes in the stock of unemployed to inflows and outflows; they are estimated as the difference between the average monthly level of inflows (as calculated above) and the monthly average change in unemployment over one year, as follows:

outflows = [ I (t) + I ( t - l ) ] - [ C (t) - C (t-1) ]

2 12

where I (t) and I ( t - l ) are the monthly inflows and C (t) and C (t-1) the level of unemployment for years t and t-1 respectively.

A note on p. 26 points out reasons why the proxy inflow data might be biased either way and also says that the proxy outflows though clearly different from the true outflow are nevertheless 'useful measures to capture some of the dynamics of unemployment and broad differences between countries'.

The same note, discussing the matched data from labour force surveys, lists a number of pitfalls in comparing and interpreting these data, including differences over the period the data are matched, sample attrition, errors and inconsistent answers, and changes and unwindings of labour force status between interviews.

26. EO97 Ch. 2 is a study of earnings mobility and Annex 2.A describes data sources and definitions for the six countries used for the longitudinal analyses presented in the chapter. It mentions the worry that an analysis restricted to people for whom a continuous earnings history is available over the period of analysis may well not be representative of the workforce as a whole. Their exclusion may be due to sampling attrition (in which case it could be compensated for by statistical means) or because some workers are not employed for all of the period of analysis; these 'intermittent' workers may not be typical. The treatment of part-time workers is also crucial to the usefulness of the analysis.

27. EO97 Ch. 5 focuses on measures of job security and, among other indicators, examines patterns of employer tenure, retention rates, and job-losers and job-leavers who left jobs in the last six months and are currently not in employment. It presents tenure data for 23 countries, retention rates for ten countries and job-loser and leaver data for 18 countries. Annex 5.A provides details of sources and definitions.

Employer tenure data generally come from household surveys; the exceptions are Japan (employer survey) and Finland (administrative source). The data refer to the time a worker has been continuously employed by the same employer. A boxed panel on p.147 discusses some of the issues concerning the measurement of job-losers and leavers.

28. In an OECD book on LMD in the Russian Federation there is a useful and full recapitulation of OECD's earlier work on the concepts and definitions of labour turnover.

In a paper by Gimpleson and Lippoldt labour turnover is defined as the sum of hirings and separations during a specified interval e.g. quarterly or annual. It thus measures the total number of individual employment transitions, whether they involve individuals getting or leaving a job. The paper quotes from OECD EO96 (p.166) as follows 'labour turnover measures changes in individuals among jobs, regardless of whether the jobs themselves are newly created, ongoing (and subsequently filled by others) or whether the jobs themselves disappear'. Labour turnover provides insights into the behaviour of both enterprises and employees.

29. Separations may be voluntary (e.g. quits) or involuntary (e.g. layoffs). Hirings may be aimed at filling a vacated job or a new opening. Dividing the absolute numbers of separated and hired by the average annual level of employment gives the corresponding percentage rates. For an individual, a separation may be associated with a transition to any of the three basic labour market statuses; and a hiring may involve the transition from any of them.

30. Labour turnover (i.e. the sum of hirings and separations or H+S) is a measure of gross labour reallocation. The difference between hirings and separations (H-S) shows the net change in employment. These indicators can be shown as percentages of annual average employment to give an indication of their relative size. Moreover, the difference between labour turnover (H+S) and the absolute value of net employment change (/H-S/) shows the volume of 'additional labour turnover' not accounted for by the net change in employment (see box on next page). Among other factors, this portion is said to include:

• turnover due to attrition (e.g. some workers separate due to retirement and must be replaced just to maintain a desired level of employment);
• mismatch between labour supply and demand ( e.g. due to the particular distribution of skills across the labour force, in some cases two or more transitions may be required to achieve a one person change in employment level);
• job-to-job transitions (which by definition involve two transitions: a separation and a hire);
• friction (e.g. problems in scheduling that may result in a temporary hire to fill a gap in staffing pending a permanent hire).

Thus, additional turnover includes movement related to restructuring as well as other personnel considerations.

The proportion of 'additional labour turnover' in total labour turnover may be calculated as follows:

(H+S)- /H-S/ or l - /H-S/

(H+S) (H+S)

This indicator varies from 0 if labour turnover is caused only by net changes in employment, to 1.0 in the case where no net changes in employment occur (H =S). The closer its value is to 1.0 the less is the contribution of net employment changes to the gross labour reallocation.

'Additional labour turnover' may be considered a measure of the upper bound of 'churning' or the amount of labour turnover beyond the absolute minimum that is necessary to accommodate a given employment change. This measure provides only a rough approximation of churning in the sense of excess turnover, in that it assumes that each hire or separation required for a particular net adjustment in employment can be accomplished by one labour market transition. Thus, in practice, the true 'excess' labour turnover (e.g. the portion due to friction in the labour market or a lack of information for labour market participants) would be somewhat lower. Moreover, this measure of churning does not take into account the necessity of reallocation of labour within existing employment levels (i.e. job-to-job movement). Such transitions may be a legitimate part of restructuring; but a single job-to-job movement increases labour turnover by a factor of two (not counting any hire to fill the vacated post). Thus, with respect to churning, this measure should not be interpreted as positive or negative, except in conjunction with other evidence.

31. Job turnover provides similar information with respect to the actual work positions, as opposed to the number of individuals. It is the net change in the number of jobs (i.e. occupied positions) at the enterprise level during an interval. In other words, it is equal to the net amount of jobs created in opening and expanding enterprises plus the net number of jobs lost in contracting or closing enterprises. Job turnover can be expressed as a proportion of employment to give an indication of the reallocation of working places.

32. Another way to follow changes in worker mobility is to look at job tenure. The share of workers with less than one year of job tenure represents the share of jobs which have been open for at least one hiring over the past year. Thus there is a direct relationship between job tenure and labour turnover, a phenonomen noted in international comparisons.

33. EO98 Ch. 3 is a study of the transition from education to the labour market but was not available in time to be used in Section 3's database analyses.

Country data used in OECD studies

34. In compiling this section I have examined 15 OECD LMD studies from 1989 to 1997 inclusive involving international comparisons mainly in Employment Outlook. The following table shows, for each of the present 29 member countries, the number of times their LMD data have featured in these studies. The countries are listed according to the year they joined the OECD. (Data considerations aside, countries joining in the 1990s will not feature very prominently.) In nearly all cases, OECD provides good quality information on the sources used in each country.

Country data used in OECD international comparison LMD studies