Validation studies : project using French data

1
Validation studies project using French data

• Assessing the consistency of ESeC with
  theoretical framework à la Goldthorpe
• Pointing out using ISCO as a common starting
  point is not necessarily going to be the best way
  to achieve comparability

2
What data sources can be used to test ESeC?

• The data 1998 working conditions survey
  (supplementing the LFS)
• allows to code national classification (PCS) and
  ISCO using PCS and NACE (classification of
  economic activity )
• Complementary data Adult Education Survey 2000

3
What data can be used to measure status of
employment?

•  
• employee / self-employed / employer

4
What data can be used to measure type of contract
(for employees) ?

• monthly wage
• indefinite duration contracts vs. temporary
  contracts, as well as life employment contracts
• working part-time

5
What data can be used to measure type of contract
(for employees) ?

• tenure
• whether the person was employed in the same
  establishment 2 years before, and if so, what was
  the wage growth over the period

6
What data can be used to measure autonomy/routine?

• supervisory responsibility, with or without power
  over the pay and career of subordinates
• production-line work
• job consists in repeating the same series of
  operations
• pace of work imposed by supervisors or machines
  or other technical constraints

7
What data can be used to measure autonomy/routine?

• person carries out instructions strictly
• instructions specify how to do the work as well
  as the work to do
• person deals with incidents on their own or calls
  to hierarchy

8
What data can be used to measure investment in
employee ?

• whether the person has benefited during the last
  12 months from training paid by his/her employer
  

9
What approach to test ESeC?

• Numerous ER variables selected for their a priori
  theoretical relevance
• Define ESeC as group of occupations similar in
  terms of ER within groups
• Desirable properties ESeC to be confronted to data

10
What approach to test ESeC?

• Simple solution tabulate all ER variables by
  ESeC category (to be tested) and compare means
  across groups, in terms of theoretical
  interpretation
• Problem ER is essentially multidimensional,
  comparing groups in terms of ER variables taken
  separately is not conclusive

11
What approach to test ESeC?

• One must construct one or several synthetic
  measures of ER, based on whole or part of subsets
  of the ER variables

12
Method used

• Consider any statistical unit that we want to
  group into categories as homogeneous as possible
  in terms of multidimensional ER for instance
  ISCO, PCS, CS

13
Method used

• Since we want to construct ESeC groups as a
  partition of the above statistical units, decide
  about the groupings on the basis of a proximity
  criterion in the space of ER

14
Method used

• Take the mean of every single ER variable across
  individuals belonging to each statistical unit
• The statistical analysis is then carried out on
  the set of statistical units , each of which is
  associated with a set of ER variables

15
Method used

• Question evaluate the distance between SU in
  terms of multidimensional ER
• gt CLUSTER ANALYSIS
• The closer two SU in terms of ER , the higher
  their proximity in the tree, which is the outcome
  of the analysis

16
Method used cluster analysis
17
Results for France

• Use two different statistical units
• gt CS because only information available in many
  if not most French surveys
• gt 2-digit ISCO because it can sometimes be
  coded using 4-digit PCS combined with NACE
• Note ISCO is necessary coded on the basis of
  PCS which involves coding error

18
Results Cluster analysis on CS
19
Results Cluster analysis on 2-digit ISCO
20
What now ?

• Consider (one or several) exogeneous grouping of
  SU providing one or several ESeC.
• Eg. A priori grouping of CS, of ISCO2,
  grouping of ISCO2 based on UK data and
  algorithm as V2.1 or V3
• Now use data (LFS and WCS98) to code CS and ISCO2
  (ie SU)
• Use groupings to code above ESeC in data (V2.1,
  V3 or any other definition)

21
What now ?

• As we are considering the same SU as before,
  using the same data
• QUESTION do we obtain comparable groupings to
  results of cluster analysis ?
• This comparison is our test of various ESeC
  definitions

22
What now ?

• Most simple criterion use LFS to cross-tabulate
  ER groups with ESeCV2, ESeCV3

23
What now ?

• Alternative approach compare tree in figure 1
  (CS) or figure 3 (ISCO) assume each SU can be
  interacted with any given definition of ESeC
• Eg. Consider SU ISCO234, according to
  cross-walk ISCO2 to V2.1 could belong to ESeC
  groups 2 or 3 (4, 5, 6, 7) according to some
  additional informations

24
What now ?

• Come back to individual data, divide group 34
  into 342, 343 (345, 346, 347) according to ESeC
  code. In practice this is the new SU
• Average out again ER variables by SU
• Do again cluster analysis
• Yield tree figure 4 (SUISCO2 x ESeC V2.1)

25
What now ?

• Test the proximity of new SUs
• Expect for instance 132 to be closed to 342
• Additional empirical test no very conclusive for
  ESeC

26
Outcome of the statistical analysis

• Concerning heads of business, the threshold of 10
  employees may not be most appropriate it should
  be discussed whether 20 or 50 should be used
  instead
• Concerning salespersons, our results suggest that
  they are closer to routine occupation than to
  clercks

27
Outcome of the statistical analysis

• Drivers belong to the group of blue-collar
  workers according to the French PCS, but also
  according to our statistical analysis, their
  position in ESeC should therefore be in group 8
  rather than 6
• We also mentioned technicians, closer to
  administrative and service intermediate
  occupation than to foremen and supervisor