Use of ISCO 88 (COM) in Statistics Belgium

Lisbon 15 September

Astrid Depickere

Experience with ISCO at NIS Belgium

• Labour Force Survey

• SILC

• Structure of Earnings Survey

• Population Census

• Household Budget

• Vacancy Survey

Other examples in Belgium (not NIS)

• General Election Study Belgium (GES)• VRIND (Flemish Regional Indicators) • European Social Survey (ESS)• Survey on Living Flanders• European Values Study (EVS)• CIM (media)• Survey on Political knowledge• …

Surveys Belgium: different coding methods

Who? • Coding done by specialised coders, on the basis of job description given by

respondent (registered by interviewer in PAPI): LFS, Population Census, Houshold Budget

• Coding done by interviewers in CAPI: SILC, LFS (pilot)• Coding done by employers (SES)

Coding scheme? • Coding straight into ISCO: SILC, SES

• Coding into more detailed list, conversion afterwards into ISCO : LFS, Population Census, Household Budget

CAPI coding:• semantic search tool for job titles• code search in paper list

Job classification

Respondents job ISCO category ESeC class

Coding implies 4 steps:

• Job title & job description given by respondent

• Interviewers interpretation

• Registration by interviewer

• Coding (by interviewer or coder)

ISCO categories recoded into ESeC classes

Sources of bias4 steps: 1) Job title & job description given

bias due to incomplete or incorrect information provided by respondent

2) Interviewers interpretation

bias due to misinterpretation of interviewer

3) Registration by interviewer

bias due to incomplete registration of information given by respondent

4) Coding into ISCO or other classification (by interviewer or coder)

method bias (e.g. lookup system, order effect)

bias due to limited knowledge of coding scheme (& method)

Evaluation of bias

1) Double coding in GES95 (KUL), based on job description given by interviewers: intercoder-reliability = 1/3 => different codes assigned in 66% of all cases !!!

2) Comparison between percentages ISCO categories in LFS versus SES • LFS: expert coding• SES: coding by respondent (=employee, usually HR)

Coding in SES is considered to be better quality (objective, correcter)

Conclusion: - overrepresentation of management positions in LFS (=> overestimation of job level when

self reported)- overrepresentation of ‘others’ categories, ‘container’ categories in LFS (=> due to (bad)

coding habits)- underrepresentation of categories of ‘complicated’ jobs in LFS (coders are less inclined to

pick jobs that they don’t understand very well)

Specific points of attention1) Definition of Managers & treatment of supervisors: Problem of distinguishing between the two. Is a measurement problem.

- There is no generally accepted definition of ‘management level’ (kaders / cadres) in Belgium, is very company specific.- There is a link with leadership, but this is not sufficient as a criterion. - Linguistic difference within Belgium:

in French no distinction between manager and director (=‘directeur’)in Dutch the two exist

- Overestimation of management jobs: respondents overestimate themselves + companies are very creative with job titles that indicate responsality (manager, supervisor, responsable, coordinator,…). E.g. floor manager. Link between job title and job content can be very weak.

Solution? Not on the theoretical level, but on the measurement level. Split up job question (job title + job description), use additional questions (e.g. company size,

sector) + interviewer/coder training

Specific points of attention2) Nurses and Teachers

- In Belgium: two different levels of nursing, linked with education: higher education (not university) secondary education

- Teachers in secondary education: lower grade (12-14 year): higher education (not university). higher grade (15-18 year): university education

=> some categories of teachers and nurses are misclassified in ISCO- Similar problem with accountants in Belgium

3) Technicians- large variation in job titles (& job content)- coding difficulties (large frequencies in ‘other’ category)

Specific points of attention

4) Skill level issue- Skill level is a good proxy for complexity of information and has the advantage that it can be operationalised and taken into account in in the coding process in case of doubt. - But (in the Belgian context): link between diploma and job becomes weaker:

• Increased importance of LLL and on the job training• competencies gain importance over skills • increase in level of education

- Splitting skill levels 2 and 4? increased detail in skill level will make the link between job and diploma even weaker.

5) Missing in ISCOPlanner (production or logistic)