QUESTIONS AND ANSWERS ABOUT ISEI-08
This document contains questions and answers on the new International Socio-Economic Index of occupational status [ISEI-08 or new ISEI] estimated for the International Standard Classification of Occupations [ISCO-08]
Last revision: October 6 2010 [HG]
How was new ISEI-08 constructed?
The new ISEI was constructed on a database of 198500 men and women with valid education, occupation and (personal) incomes derived from the combined 2002-2007 issues of the International Social Survey Programme [ISSP]. The methodology used is similar to the construction of ISEI for ISCO-68 and ISCO-88 described by Ganzeboom et al. (1992, 1996, 2003). It involved the calculation of an optimal score for detailed occupations in the following elementary status attainment:
EDUCATION à OCCUPATION à EARNINGS
The optimization aims at minimizing the direct effect of education on earnings and maximizing the indirect (‘mediated’) effect of education on earnings via occupation.
Note that while the aim of the new ISEI was to produce socio-economic status scores for ISCO-08, it was in fact calculated on occupations classified in ISCO-88 and then converted to ISCO-08 occupations.
What are the differences with previous ISEI constructions?
The main differences with the previous ISEI construction are:
· A new database was used. This database is more recent, larger and cross-nationally more diverse than the one used earlier. It contains occupations that were coded into ISCO by the original researchers, while previously conversions from country specific classifications were used.
· The new ISEI was constructed on a data on women and men, while previously only men were used to estimate the scale. The earnings data were corrected for hours worked to adjust the different prevalence of part-time work between men and women in many countries.
Minor differences derive from empirical contingencies of the databases used. In order to obtain some level of reliability, a cutting point of 21 valid observations for each occupational unit was used and this limit is reached in different instances in the two databases. Due to its larger size, the new ISEI scale tends to be more detailed than the previous one.
Why is the new ISEI constructed on men and women, and not on men only, like previously?
The construction of the scale on men only has created feelings of discomfort among many users and raised suspicions that the scale may be invalid or less valid for women (mothers). Ganzeboom et al. (1992, 1996, 2003) were not on agreement with this position and argued that a scale estimated on men and women combined would downplay the fact that women receive less pay than men for the same occupation, as this pay difference becomes an ingredient of the occupational status.
The reasons to move now to a scale estimated on women and men combined are of a pragmatic nature. First, it expands the available data and thus produces more reliable estimates. Second, this is all more true for heavily female dominated occupations (such as nurses and preprimary teachers), for which the previous scales could produce only estimates using the rare men that are found in these occupations (who may be not so representative of these occupations to begin with).
The inclusion of women in the construction of the scale is responsible for the somewhat lower scores that some female-dominated occupations now obtain.
Why is the new ISEI (to be used for ISCO-08) constructed on data that only contain ISCO-88 occupations?
It would be best to construct the scale on a dataset with occupations that are coded in ISCO-08. However, a dataset that meets the requirements (large, cross-nationally representative, detailed occupation codes, education and earnings data) seems not to be available at this time. Moreover, the non-existence of ISEI and other derived measures for ISCO-08 may actually be a reason for cross-national projects that could produce such data (like ESS or ISSP) choose not to adopt ISCO-08 as their coding standard.
ISCO-08 is strongly related, but far from identical to ISCO-88. The changes at the major and sub-major group levels (1- and 2-digit occupations) are indeed minor, but with exceptions. At the more detailed 3 and 4-digit level, the changes are more profound, although much of if involves a reordering of 4-digit occupations between minor (3-digit) groups. In as far as this is the case, little information should be lost in translating the ISCO-88 results to the ISCO-08 scaling.
However, there remain parts of the ISCO-08 classification, for which no new ISEI can be derived this way, as there are no fitting ISCO-88 counterparts. An example is the new sub-major group 9400 (Food Preparation Assistants), to which no specific parts in ISCO-88 correspond. Fortunately, in this case (and some others) the ‘new’ ISCO-08 actually reconstitutes distinctions that were also made in ISCO-68, for which an ISEI score was calculated by Ganzeboom et al. 1992. We have taken scale scores from ISEI-68 to fill these gaps.
Of course, it remains to be seen how the present results hold up against databases that would be coded in ISCO-08, as they will become available in the future.
Is there any evidence on the quality of the new ISEI scale relative to the previous ones?
The old and new scales are strongly related. In the ISSP database that was used for developing the scale, the old and new scale correlate 0.92. In the ESS database (round 1-4 combined) the correlation is 0.91. The ESS database provides evidence that the new scale is slightly better than the old one. In a latent variables model that uses both respondent’s and spouse’s occupation, the above correlation is decomposed in a 0.98 loading for ISEI-08 and a 0.94 loading for ISEI-88. This suggests that any association between occupation and a related variable is 4% less attenuated when one uses the new scale.
Of course, all of this is evidence on data that were initially coded only in ISCO-88. It remains to be seen how this will hold up in databases with double coded occupations.
What is the relationship between ISEI and occupational prestige?
There is a widespread misunderstanding that SEI (socio-economic) measures of occupational status are the same or ‘approximate’ occupational prestige (such as measured by the Standard International Occupation Prestige Score [SIOPS or ‘Treiman-scale’]) scales. While the two are obviously related (being continuous indicators of occupational status) and empirically correlated (around 0.75), they are conceptually distinct entities. Prestige scales are constructed using popular evaluations, socio-economic index scores refer to an occupation’s main antecedent (education) and main consequence (earnings) as its formative parts. An SEI score can thus be conceived as the cultural and economic resources that are typical of the incumbents of a certain occupation. Similarly, SEI conceives of occupations as the main mechanism that transfers education into earnings, and tapping this mechanism is used in constructing the ISEI scale.
The misunderstanding arises primarily because the original Duncan SEI score for the USA (Duncan, 1961) used occupational prestige as a criterion variable to derive weights for occupational earnings and occupational education to produce its score. While the original intention was to ‘approximate’ prestige scores for all occupations, the Duncan procedure in effect purged the prestige score altogether (even for the occupations for which they were available).
The correlation between prestige and SEI is around 0.75, but there is much evidence that the primary source of the differences between the two is in the weak measurement properties of prestige measures. Prestige measures tend to attenuate associations with occupations rather strongly.
How is ISEI related to socio-economic classifications such as EGP or ESEC?
In practical use, ISEI differs from socio-economic classifications such as EGP or ESEC, because ISEI is a continuous hierarchical scale, while these classifications present a limited (up to 9/11) amount of discrete classes that are often used as nominal variables.
In deriving the measures, there are two important differences:
· ISEI is empirically constructed, while EGP and ESEC are a-priori constructions that are at best validated on empirical data.
· ISEI only uses occupational data (i.e. occupation as classified by ISCO), whereas EGP and ESEC heavily use status-in-employment (self-employment and supervision) indicators to construct socio-economic classes.
Conceptually, the basis of ISEI is in how occupations operate in the status attainment process, which the discourse around EGP and ESEC most often refers to work situation and contractual conditions.
Despite these differences ISEI and EGP (and presumably ESEC) are empirically strongly correlated (0.90), much more strongly than ISEI and SIOPS. If optimally scaled using a single dimension, EGP scores are virtually identical to their ISEI means.
Ganzeboom, Harry B.G.; Treiman, Donald J. (2003). "Three Internationally Standardised Measures for Comparative Research on Occupational Status." Pp. 159-193 in Jürgen H.P. Hoffmeyer-Zlotnik & Christof Wolf (Eds.), Advances in Cross-National Comparison. A European Working Book for Demographic and Socio-Economic Variables. New York: Kluwer Academic Press. Pp. 159-193. (6.1Mb)
Ganzeboom, Harry B.G.; Treiman, Donald J. (1996). “Internationally Comparable Measures of Occupational Status for the 1988 International Standard Classification of Occupations”. Social Science Research (25), pp. 201-239. (3.9 Mb)
Ganzeboom, Harry B.G.; De Graaf, Paul; Treiman, Donald J.; (with De Leeuw, Jan) (1992). "A Standard International Socio-Economic Index of Occupational Status", Social Science Research (21-1), pp. 1-56. (3.5Mb)