QUESTIONS & ANSWERS

Kristina	BD	Blau-Duncan model: association between father’s education and father’s occupation is bidirectional (line has two arrows). When we calculate indirect effect of father’s occupation on respondent’s first job through father’s education and respondent’s education (like in the lecture) do we have to presume that father’s occupation causes father’s education? HG: No, but you have to take the arrow into account when making the calculations. This bi-directional arrow then represents a confounding effect. BD address the issue on p166. If the focus is on occupational attainment, it is convenient to drop FED from the model, and use a total effect of 0.454 (0.270 + 0.516*0.310). Part of this is spurious, but who cares?	8
Ave	CMLR	Does CMLR require before the analysis the change of the structure of the data to the format described in the table 8 (Dessens et al 2003: 75)? HG: Yes, you need to redefine the data-matrix itself, even if you use stata to estimate the model. Stata has nice routines to do this, but you can do the same redefinition by saving multiple files and then add them in SPSS.	9
Ave	CMLR	Dessens et al (2003: 76) state, that "the uniform association between origin and destination categories diminishes over time". Where could i see it in the table 9 (Dessens et all. 2003: 77-79)? What shows that? HG: You see this in the significantly negative effects U*EYR on p15. A decline of -.06 would say that U goes down by approximately 1.3% each year.	9
Siim	Communism	What influence has communist period if we compare Western and Eastern Europe, is there some aspects what should take into account if we compare for example educational levels or occupational division? HG: Of course, this is a problem when analyzing or comparing with data from the communist period, but the problem is mostly with the awkward occupational classifications that were then used, not so much with the organization of work itself. If you use a non-communist classification, e.g. ISCO for classifying parents or first jobs, there is usually little problem. Like always and everywhere, the problems arise mostly with how farm work is organized. For education, I do not think there were very specific educational regimes that are unique to communist nations.	9
Lea	Communism	To whole course. Does ISCO or other classifications fit to post-communist countries like Estonia? In these countries all economic, social and other systems where destroyed in the soviet time, all persons where equally poor and I think that meanings or status of professions may be totally different. For example teacher is respected and well paid in Finland, but in Estonia it is despised profession. HG: This is not true, not now, and it was not true in communist times either. I do not believe that teacher is a despised profession in Estonia now, nor anywhere else. But I have heard the same complaints about teachers in the Netherlands.	7
Tarmo	Comparative mobility	I would like to know more about your future plans concerning this kind of research and, more generally, about the feasibility of such grand comparative study of stratification (or of social mobility, more specifically). HG: I am still working an a large scale comparison, but it needs to meet a few important criteria (A) real large scale, both in countries and time periods, (B) multivariate, (C) discrete data analysis. I am still working on it.	8
Tarmo	Comparative mobility	Continuing the same topic, there seem to be two opinions about cross-country differences in social mobility. There is the Goldthorpe-Erikson school that seems to believe that there are no between-country differences in social mobility (when mobility is appropriately defined). And then there is the Ganzeboom-Treiman school that seems to have discovered systematic differences between countries in terms of mobility. What is the present situation in this disagreement? HG: I think most researchers agree that GT had it right – in particular this is rigorously confirmed by Breen’s “Social Mobility in Europe” and acknowledged by Hout & DiPrete.	7
Siim	Education	Regarding article Measuring and Modeling Education Levels in European Societies. Would it be better to measure achieved education through real output (I mean what kind of skills graduates really got). HG: Skill assessment (see PIAAC) is not so easy either. Why not use directly observable outcomes such as occupation, earnings or spouses occupation?	8
Lea	Education	Article "Nominal comparability is not enough..." From this article comes out, that in Estonia educational attainment is less important for occupational status attainment than in most other countries. It seems to be true. Do you have some explanation, why it can be so? HG: I am not so sure that this is true. I believe that the measurement of education in the Estonian ESS, first round, contains errors.	8
Ave	Education	If education is measured in levels, is it ok to recode the variable in such a way that the levels will be substituted with number of years it typically takes to achieve that level of education and use it as numeric variable (independent variable in multinomial logistic regression). HG: this is a reasonable approach that ISLED tries to improve upon. The problem of ‘institutional duration’ is that is biases levels that take long, but have in fact lower levels – these are taken by dumb kids. This is often the case in highly stratified systems with vocational tracks. The error is small, but ISLED takes it out.	7
Ellen	EGP	In the method EGP you have called in the occupational sector group  a).... b) manual c) non-manual. Could you use Intellectual instead of  non-manual? If not, then why? HG: Yes, I sometimes do this in Dutch, where we also use “hoofdarbeid” (=”working with your head”) as phrase, but this has no acceptable equivalent in English. There may still be confusing about what you mean. E.g. routine sales work (cashier in a supermarket) is non-manual work, but is it intellectual?	7
Ellen	EGP	Why is skill level the last thing to consider in classification  methods? I have found it most valuable when evaluating jobs. HG: In fact, in the ISCOàEGP recodes skill is considered first (and derives from the occupation code itself), but then gets transformed by employment status. This means in practice that employment status (= self-employed or supervisor) take priority over skill requirements. I agree that this is a weakness in class typologies a la Goldthorpe. In fact, self- employed versions of sales workers and skilled manual workers look much more similar to salaried versions of these jobs than to each other/ HG: Or did you mean the skill priority rule in coding occupations? I thing it comes after the numerical dominance rule, and that is reasonable.	7
Kristina	EGP/ESEC	Rose and Harrison (2007) describe that ESeC measurement of socio-economic class bases on type of contract. How adequate is this measure in different European countries (especially since professionals might do project-based work and have short-term contract)? HG: I think that the whole contract business does not enter the actual construction of ESEC at all, it is just “theoretical background”. Moreover, everything they say about contracts (such as about pension rights) is particularly British and would not carry far in many European countries.	8
Kristina	Ethnicity	Hout and DiPrete (2006) mention that gender difference in the association between origins and destinations are absent (although occupational destinations differ). Could we expect something similar for ethnic groups? Would it make sense to look at ethnic groups separately in log-linear analysis (in one country)? HG: No, these seem to be very different things to me. Men and women have the same fathers, but blacks and whites have not.	6
Kristina	GTU	Ganzeboom, Treiman and Ultee (1991) mention opportunity for an institutional theory of social mobility. Is there ways to include institutional factors and different countries in the analysis of social mobility (including in the same model)? HG: We bring this up in the context of macro-theory, so how does status attainment vary across time and countries. One theory (modernization theory) would relate this to economic growth and related processes of modernization. Many people think modernization theory does not work (Hout & DiPrete). GTU observe that many researchers have pointed out that variation in status attainment of related to institutional factors, such political regime, welfare state regimes, educational structure, women’s rights, but we feel that no one has produced a theory on these factors that can really compete with modernization theory.	7
Siim	Income	Regarding the article about “How to measure Income“. We ask different types of income, but should we not give more attention to tax levels (taxes directly related to labour) and rather compare tax levels to income levels? HG: (A) I think it makes a big difference whether you are interested in incomes as outcomes, our income as asset. In the latter case, you want to adjust raw income measures for taxes, purchasing power, household equivalents, but not if your research question is about (e.g.) returns to education. HG: (B) Income measures are often composites of several underlying components. The composition in not always an addition or average. This makes income particularly vulnerable to measurement error, because the measurement errors of the components may multiply, not cancel each other out.	8
Ave	Income	Could you please explain more clearly what is Atkinson index and what are the ways of using it (Warner, Hoffmeyer-Zlotnik 2003 write about it)? HG: No, I cannot, I would have to consult wikipedia.	8
Lea	Income	How we can account persons who declared, that their household income is below 0? In my work I get very often declarations, that family’s income is much lower than rent and other expenditure for flat. They are not lying: grandmother gives potatoes, mother helps with some food, friends give sometimes something. How to declare this? HG: True household incomes cannot be negative – otherwise the respondent would have died and not answered your questionnaire. In practice it means that you have not included all forms of income. Put them all on a very low number (not zero, because log(0) in not defined). If this influences your results, omit the cases from the analysis.	8
Lea	Income	How to increase the respondent’s motivation to answer the sensitive questions about income? Is there a way to ask it indirect? How to understand, that respondent is not telling the truth? HG: I believe that the main reason why people do not answer income questions is that they do not know their incomes. Refusal is only a secondary problem. There are two general strategies to cope with the non-response: (1) present a show-card after an open question and say that an approximate answer is enough. (2) have them choose below of above the median, then present the above/below question again for the quartile and octile point – this gives you 8 categories.	7
Lea	Income	In Estonia we mostly use Gini Coefficient as indicator to compare income inequalities, is there some better indicator and in what cases? HG: Again, this is a very broad subject, about which many books have been written. Other than Gini I most often see more descriptively oriented measure, such as ratio of income of 80^th decile / 20^th deciles. Gini is a generalized version of this.	7
Leen	Income	Hoffmeyer-Zlotnik & Wolf point out that different type of research requires different type of questions of income. Could it happen that too detailed questioning of income leads to bigger ambiguity and errors in answers and finally leads to loss of information? HG: Yes, I believe this is a big problem in income measurement and researchers do not seem to pay attention. The income questions are too complicated and they are combined in non-additive ways. All of this creates low reliability and is also an important cause of missing values in income measurents. ESS has a wonderful one-shot question about household income that works well.	9
Lea	Income	For international comparisons of income there are taking account differences of systems of taxations and social security systems, but there are more important things to take account as expenditure to education, to day-care for children and so. How we can compare data from different countries? HG: this problem is generally treated by using PPI (purchasing power index), of with the most famous one is the “big mac” index. However, before you do this, think your problem trough: do you really need internationally calibrated measure, of is comparison to the local mean income enough. In this case, simply use log incomes.	8
Mauri	ISCO	Where is the most correct place to code assistant managers, vice presidents, etc? HG: Code assistants with those they are supposed to assist, unless a specific assistant code exist (which is sometime the case, e.g. in nursing and teaching).	8
Mauri	ISCO	If the occupation of a person consists of 2 different items, then should we apply the "skill level rule" by choosing the one requiring higher skill-level? HG: Yes, but first comes the numerical dominance rule.	8
Mauri	ISCO	Isn't "skill level rule" opposite to "production rule" (if person is supervisor and worker simultaneously, then by first rule we should code him as supervisor and by 2nd rule - as worker), or perhaps I missed something important about it? HG: Skill rule has priority.	8
Leen	ISCO	Can the ISCO type of measures be applied also to historical data? Maybe it has already been done? I think it might bring interesting insights to the understanding of status attainment. For instance all European countries have experienced large-scale urbanization (fed by demographic change and industrialization), but the timing of this change has been very different. What is characteristic (particularly to this period) is that new types of occupations and social ties emerge. This must influence the pattern of status attainment (social mobility) in a particular way. Is the need to be aware of the different timing of large-scale social changes like urbanization, when comparing different countries, a wider concern or it is not an issue? HG: Occupational stratification is fairly stable across time, so there is nothing particularly wrong with measuring historical stratification with ISEI. However, we cannot make an ISEI score for historical data (because such data typically lack occupational qualifications and occupational earnings scores, the two ingredients of ISEI. So it is hard to examine the historical constancy rigorously. Historians have turned to CAMSIS type scales (social distance) to measure occupational status, as this can be done with occupational homogamy data that are abundant in historical records. Marco van Leeuwen and Ineke Maas at Utrecht University work at this.	8
Ellen	ISCO	In my previous work I have noticed that ISCO is too detailed to use as a classifier and doesn't bring out the different levels of work. Is there going to be any changes? (I got an answer in the session this morning already). HG: The new ISCO-08 is even slightly more detailed. If it I too much, you can always reduce to 3 of 2 digits.	8
Mauri	ISCO	If code 1200 is for those managing at least 3 sub-managers, and 1300 for general managers for small businesses, then should all other mid-level managers be coded under 2000, 3000 or further in case it is evident that there are not enough sub-managers? HG: I am not sure I understand. If they are managing small firms with two departments (e.g. production and sales), they should be in1300.	7
Laur	ISCO	What are main advantages of ISCO-08 comparing with ISCO-88? HG: see: http://www.harryganzeboom.nl/isco08/qa-isco-08.htm and http://www.harryganzeboom.nl/isco08/index.htm. The short answer is: very little.	7
Ellen	ISCO	What will be the next-generation codifying system? HG: I will take some time for ISCO-08 to catch on and I would not expect it to change before 2028. As far a I am concerned, there is no need to change more often. The 2008 revision has not brought very much improvement. And most 2008 improvement have little to do with changing technological conditions.	7
Ave	ISEI	Estonian Labour Force Survey has several persons from 1 household (often 2). When analyzing this data, might it be a problem? Should I take only 1 person from 1 household? I see it as a problem when running regression analysis for example, where cases in analysis have to be independent, but if there is more than 1 person from 1 household, then these cases are not independent. HG: No, do not throw away data. The general answer to this type of problem are fixed-effects models (available in Stata, R, etc), where you introduce a dummy variable for each household. Another alternative is to develop an ‘efficiency’ weight, such as produced by Survey estimation or cluster correction in Stata (and probably other programs).	7
Ave	ISEI	Are there any other theoretical grounds to the occupational scales other than functionalist and marxian, weberian approach (Ganzeboom, De Graaf, Treiman 1992: 8 draws some attention to the underlying theories)? I mean are there other kind of scales based of other theoretical approaches? HG: Yes there is at least one, it is called relational scaling, which I would label as weberian (SEI as marxian, prestige as durkheimian). This derives from marriage or friendship tables and has traditionally been the sport in Cambridge, not Oxford (Camsis Scale). There is a 1970’s fight about this in British sociology that continues to this day. However, recently Goldthorpe (with Chan) has also turned in this direction. It works quite well.	9
Kristina	ISEI	If I analyse children school performance (with PISA data), does it makes sense to include into multilevel model parental ISEI, education and financial opportunities (since ISEI measures the way education is transferred into earnings)? Or would it be better to use parental ISCO if education and financial opportunities are also taken into account? HG: important question. I think it does make sense. Despite being based on education and earnings, ISEI is still a characteristic of the occupation, not of the person and his/her assets. You might argue that in such models individual earnings and education measure how people differ from what is normal in their occupation.	8
Laur	ISEI	According to Tartu University demographers about 10% of Estonian are living or working outside of Estonia. Great amount of them are working in Scandinavia as builders. Their average income might significantly influence their occupation position on socio-economic scale. Are this kind of context-based deviations problem in cross-national comparison (when using ISEI)? HG: 10%? Really? The problem would only arise if you would create an Estonian SEI and include emigrants somehow in your data. In ISEI they have the same status as all construction workers elsewhere, but enjoy higher income, at least relative to construction workers in Estonia. So that is the benefit of migration: higher income, but same status.	6
Marko	ISEI	The stratification models presented (for example BD model) don’t consider any cultural factors as influence on social status. Is this because they don’t matter or are they hard to measure in international researches? Or is there possibility of reverse causation maybe? HG: It is unclear to me what you would mean by these “cultural factors”. Education is often referred to as ‘cultural status’ (as opposed to income = economic status). This is certainly taken into account. Occupational prestige is also ‘cultural’ (relative to ISEI) and that has gone out of fashion no because of conceptual problems, but because research has shown it did not work as well as SEI measures. Other ‘cultural factors’ such as ethnicity are often addressed in status attainment models, but of course not part of occupational status measurement. Such reseach analyses whether ethnic groups have different occupations (with different status).	6
Marko	ISEI	Same goes for ISEI scale. It includes only formal determinants of social status. But social positions of occupations are at the same time defined by many other factors also. I think this is especially so with occupations with not very high requirements for education and with not very high incomes. HG: This has often been an argument in favor of prestige measures. E.g. many female occupations (nursing) enjoy relatively high prestige. This sounds conceptually plausible, but I now of no research that has shown the difference to be important, or even useful empirically. Empirically, prestige look much more like a weak indicator of occupational status than an independent dimension.	6
Laur	ISEI	Is it okay to use ISEI (and also ISLED) score in linear regression models? HG: Yes,	5
Mauri	ISLED	We spent quite a bit of time with correlation coefficients demonstrating how good is ISLED vs other measurement methods, but how exactly ISLED is calculated or derived, was touched rather briefly. HG: I can explain again, but it is not different from how ISEI was estimated.	6
Kristina	Loglinear	Could you explain the idea of Goodman’s model RCII in CLRM model framework? HG: I have tried in the lecture.	8
Siim	Missings	Second question is that for me is quite difficult to understand how one can be assured of the relevance of imputation in social surveys in case of lack of answers about income from respondent? HG: Agree. I do not think that we learn very much from imputation, however smart it is done. In the end it is just technical repair of the completeness of the data matrix, that works best when the data are missing completely at random (MCAR) and not when data are systematically missing. I think FIML (Full Information Maximum Likelihood – in LISREL) is the better way to go.	8
Marko	Missings	Holst & Christian (2003) and also Hoffmeyer-Zlotnik & Warner (2003) pointed out the problem of non-responses in income measurements. What is the standard policy with those cases, are they just removed from analysis (concerning stratification models)? HG: There are many possible strategies (and no ‘standard policy’) to go about missing values in any variable, also income, and their merits depend much on the situation. There are strategies that substitute missings, strategies that only use complete cases, and strategies that use available cases. All of these have merits, and most strategies can be done in various ways. I have found Paul Allison’s “Missing Data” in the Sage University Papers quite useful. It has convinced me that Full Information Maximum Likelihoog (FIML, implemented in Lisrel) and an interesting way to go. This is an avallable case strategy.	7
Marko	Missings	In relation with previous question, what would be the optimal way (if there is one) for treating missing values in attitude questionnaires? HG: Attitudes are often measured with multiple indicators and some form of additive scaling. I find Comp mean(items) in SPSS a powerful way to minimize missing values. If attitudes are analyzed as dependent variables (often the case), more elaborate forms of substitution hardly make sense to me.	7
Marko	Modernization	Hout and DiPrete, summing up RC28’s plenary session at the World Congress in Brisbane, state that modernization theory is wrong. But what about so called “post-modernization” theory (wide spread of communication technologies and internet, globalization, social and cultural pluralism etc), what is not so much economical, but cultural process? When social statuses are more blurred and family ties become weaker etc, doesn’t it influence stratification and models explaining it? HG: Sounds plausible to me, although much of what you say could also be subsumed under modernization itself. I believe that the international trends show convergence between countries and this would be in line with the globalization argument.	8
Leen	Mothers	Father’s education seems to be an important variable both in the original Blau&Duncan model and in its extensions. How important is the variable in contemporary research in general? Wouldn’t mother’s education be more important in this respect (at least in those societies where education among women is already widespread)? Assuming the educational homogamy of spouses (that has been shown in different studies) I would expect that the effect of fathers’ education only echoes the actual effect of mother’s education and both probably capture some kind of socialization effect. HG: This is an age-old question that is addressed on p.190 of BD. Their answer is a qualified NO. Introducing mothers changes little to the family effect if her education is closely related to father’s, which it is. If it would be not related to father’s (quod non), then it would not change the father’s effect. Introducing mothers gives you insight how fathers and mothers relate, but does not change the mobility pattern (=strength of intergenerational correlations).	8
Siim	Occupational mobility	Some thoughts after reading article „Occupational Attainment of First Jobs“. We analyze whether there is a relationship between father’s occupation and respondent occupation but how does that take into account structural changes (I mean growth of jobs related to services and for example diminishing in agriculture or manufacturing)? HG: Loglinear models are supposedly insensitive to structural mobility – and so is CLR. Results are not dependent upon the distribution of destinations or origins.	7
Siim	Occupations	Regarding socioeconomic classification, occupational structure is viewed as the backbone of the stratification system but how can we ensure comparability between countries or apply class schema same way, when occupations have different status in European countries? HG: The first generalization of Hout & DiPrete is that this is not the case: the ‘Treiman constant’. I subscribe to this largely: there is little evidence of systematic variability in evaluation of occupations and even less theory about it.	8
Leen	Structural change	How does the B&D type of modeling address the contextual change occurring over time (calendar period)? For instance when looking at the effect of education (or any other phenomenon that spreads over time) then people with higher education become less and less selective group. This means that when education spreads in a society the effect of it must become weaker not because education is not important anymore, but because it doesn’t distinguish people so well anymore. This also indicates how difficult it is to select variables for comparative longitudinal social surveys (it’s clearly not enough to base one's design solely in one particular social context, e.g. US in 1970s). HG: I think when educational differentials are shrinking and make less of a clear signal to employers, this will lead to less association. This is a real phenomenon, nothing artificial.	8
Leen	Theory	The course has addressed many social stratification indicators; we have seen several different ways to measure education, occupational status, social class, income, and discussed how to study social mobility. Some of those measures have clearly emerged from particular social (and historical) context. I already briefly commented BD model in this respect; also the British roots of the EGP were discussed in the class. My question is whether the issue of ethnocentrism has been consciously addressed while developing any of those models? Can ethnocentrism be avoided? Is it a bigger problem in some of the areas (education, occupation etc) compared to the others? HG: Theories should not be judged by whether they start from ethnocentric or idiosyncratic assumptions or not, but whether they pass empirical tests. The original Goldthorpe scheme for Britain that did not contain a specific category for farmers, is not problematic because that represents a specifically British image of stratification, but because it has been shown by other research that farmers have very peculiar mobility patterns, than can and will bias you results, if you do not keep them separate. This is also true in Britain, they just have fewer of them.	7
Leen	Women	In a last lecture you mentioned that in the US there is a tendency that women increasingly prefer staying longer home with kids instead of working (if I understood correctly). I wonder could this tendency appear due to the fact that the share of Hispanic women, who are more family-oriented in general and who started to emancipate later, has increased in the population. HG: I think I was giving a hearsay personal impression, I do not know the facts. Your explanation sounds plausible and could be tested.	9
Ave	Women	I have discovered, that mother's variables (occupation, education) is stronger related to their offspring's occupation than father's variables. Ellu Saar has mentioned the same (sorry, i have forgotten the details of this study) and research by Koucky, Bartušek, Kovarovic (2010) shows, that in Estonia mother's variables are stronger predictors of getting tertiary education than father's variables. I have tried to use variables which take higher education and occupation (mother & father combined) as parental variables, but in that case i found, that the effect is much smaller, than in the case, when i use them separately, so i would say mother's and father's effects are different. How would you handle mother's and father's variables in this case. In all those cases I used multinomial logistic regression predicting current occupation. HG: Interesting. I have published papers (with Sylvia Korupp) on this topic and I am working on a new one. Our general finding is that fathers are more important, but that this differs between men and women (sex role modeling). Another general finding is that the highest status is more important and that ‘dominance approach’ works well. There must be something special with Estonian mothers (or fathers)… Measurement error?	8