PATTERNS AND DETERMINANTS OF MULTIGENERATIONAL INEQUALITY

Much of the economic research on the intergenerational determinants of socioeconomic attainment derives its theoretical foundation from the highly influential model developed by Gary Becker and Nigel Tomes (Reference Becker and Tomes1986). Because socioeconomic attainment (e.g., earnings or occupational status) is partly the result of parental investments in the human capital of their offspring, this results in a positive correlation of socioeconomic status in a two-generational setting. This theoretical postulate has been repeatedly confirmed in empirical studies by positive correlations in earnings between two consecutive generations (typically between fathers and sons) (Solon Reference Solon1992; Zimmermann 1992; Björklund and Jäntti Reference Björklund and Jäntti2000, Reference Björklund, Jäntti, Salverda, Nolan and Smeeding2009; Black and Devereux Reference Black, Devereux, Card and Ashenfelter2011; Jäntti and Jenkins Reference Jäntti and Jenkins2013). Country differences in the strength of the correlation are, however, not trivial, indicating considerable heterogeneity in intergenerational earnings persistence across different contexts (e.g., Blanden Reference Blanden2013; Corak Reference Corak2013).Footnote ² Sweden, for example, appears to have higher earnings mobility, and thus lower intergenerational persistence in earnings than either the United States or Great Britain (Björklund and Jäntti Reference Björklund and Jäntti1997, Reference Björklund and Jäntti2000; Solon Reference Solon2002).

A process of regression to the mean should cause the influence of earlier generations to gradually diminish. In fact, according to Becker and Tomes (Reference Becker and Tomes1986), most ancestral influences are eliminated already after three generations in developed countries, giving little room for grandparental influence on their grandchildren's outcomes. This is consistent with a first-order Markov (AR1) process, where the outcome in one generation is only determined by characteristics of the parental generation (e.g., Hodge Reference Hodge1966). Several studies on class attainment based on occupation provide empirical support for this view, with no, or only a very limited, impact of grandparents' status on the status attainment of their grandchildren, once the status of the parents is controlled for (Hodge Reference Hodge1966; Warren and Hauser Reference Warren and Hauser1997; Erola and Moisio Reference Erola and Moisio2007; Braun and Stuhler Reference Braun and Stuhler2015).

In contrast, there are studies that provide evidence suggesting important associations between socioeconomic outcomes of grandfathers and grandsons, over and above the influence from the parental generation. In a study of earnings and education across three generations using data from a cohort of school children from 1938 in Malmö in southern Sweden, Lindahl and his colleagues (Reference Lindahl, Palme and Sandgren Massih2015) find a significant association between the education of grandparents and grandchildren, controlling for the impact of parental education. Grandparental influence on earnings is, however, much smaller. A first-order Markov process cannot fully describe the social mobility process in Sweden in the twentieth century.Footnote ³

Bitte Modin, Robert Erikson, and Denny Vågerö (2013) examine education and find that ninth-graders in contemporary Sweden are more likely to achieve top grades in Mathematics and Swedish if their grandparents had high grades in these subjects in third grade. They do not control for parental grades but include controls for parental educational level (as well as the educational level of all grandparents), and interpret their results as evidence of a direct influence from grandparents to grandchildren in terms of school performance. Martin Hällsten (Reference Hällsten2014) exploits cousin correlations in several outcomes, finding a grandparental influence above and beyond that of the parents in terms of grade point average (GPA), years of education, and attained occupational prestige. Allowing for heterogeneous effects across different socioeconomic origins, the findings suggest that the grandparental influence is greatest among individuals from wealthy origins.Footnote ⁴

For occupational attainment and mobility, studies of different contexts show significant associations between the attainment of grandfathers and grandsons, while controlling for the attainment of the fathers (Long and Ferrie Reference Long and Ferrie2012; Chan and Boliver Reference Chan and Boliver2013; Hertel and Groh-Samberg Reference Hertel and Groh-Samberg2014; Braun and Stuhler Reference Braun and Stuhler2015).

MECHANISMS OF MULTIGENERATIONAL PERSISTENCE

There are several reasons for the existence of a direct link between grandparents' and grandchildren's socioeconomic outcomes (Solon Reference Solon2014). From a two-generational setting, Zhen Zeng and Yu Xie (Reference Zeng and Xie2014) identify three main pathways of intergenerational influence: biological, economic, and socio-emotional. To the extent that genetics has an impact on ability and socioeconomic status (Beenstock Reference Beenstock2012; Black and Devereux Reference Black, Devereux, Card and Ashenfelter2011; Björklund, Jäntti, and Solon Reference Björklund, Jäntti and Solon2007; Björklund, Lindahl, and Plug Reference Björklund, Lindahl and Plug2006), this could explain some maternal grandfather influence, while the impact from paternal grandfathers should be lower because most of the effect is mediated by the inclusion of father characteristics. Thus, a stronger association in status, on average, between maternal grandfathers and grandsons than between paternal grandfathers and grandsons could be interpreted as support for this kind of biological pathway, especially when the two are not geographically proximate.

Grandfathers can transmit economic resources directly to their grandsons in the form of human capital, wealth, or networks, which in turn could provide access to higher quality education or occupations (Mare Reference Mare2011; Pfeffer and Hällsten Reference Pfeffer and Hällsten2012). Such direct influence through wealth and high-status networks should be most strongly felt at the upper end of the status distribution, and thus could be an explanation for a high degree of socioeconomic persistence among the upper classes (Zimmerman Reference Zimmerman1992). As with biological pathways, economic transfers do not require a close interaction between the generations through co-residence or geographic proximity. Networks and reputation could even remain important in cases when the grandfather is dead.

Socio-emotional factors or the transmission of cultural capital, on the other hand, require interaction between grandfathers and grandsons, which for most of the period covered in this study also presupposes residential proximity and that the grandfather is alive during the childhood of the grandson (Solon Reference Solon2014; Zeng and Xie Reference Zeng and Xie2014). Decreased morbidity and increased longevity enhances the opportunity for grandparent impact (Bengtson Reference Bengtson2001). At the same time, less intergenerational co-residence and the increasing importance of long-range migration work against this influence.

Random measurement errors in socioeconomic status variables can create a spurious association between grandfathers and grandsons in multigenerational regressions of the kind used here (Solon Reference Solon2014). Overall, we would expect such random measurement errors to be larger for income than for social class. However, because we will use estimates of life-time income based on multiple observations of both earnings and occupation, the impact of the measurement errors should be significantly reduced (cf. Solon Reference Solon1992). Variations in “market luck” could also create a similarly spurious grandparent effect (Stuhler 2012). This relates to the hypothesis that some latent factor determines socioeconomic outcomes, such as earnings, occupation or education, and this factor is reproduced between generations at a much higher rate than the outcomes we can measure (Clark Reference Clark2014; Clark and Cummins Reference Clark and Cummins2015). This could also explain why an association in observed outcomes between grandfathers and grandsons, net of the association between fathers and sons, can be found even in cases of a two-generational (Markov) transmission of the latent factor (Braun and Stuhler Reference Braun and Stuhler2015). This means that we cannot interpret a “grandparent effect” as proof of a direct impact from grandfathers to grandsons without further qualifications.

THE CONTEXT

This analysis is about long-term patterns of social mobility in Sweden, covering the period from the early nineteenth century to the early twenty-first century. As all other Western countries, Sweden underwent a radical transformation over these two centuries as a result of the agricultural and industrial revolutions, which together with the demographic transition, urbanization, the introduction of parliamentary democracy, and the development of a comprehensive welfare state, fundamentally changed the living conditions of its inhabitants. From being a poor agriculturally-based undemocratic monarchy in the early nineteenth century, Sweden transformed to one of the richest industrial democracies in the post-WWII period (Schön Reference Schön2000). In the first half of the nineteenth century, Swedish agriculture was in the midst of the agricultural revolution in which institutional and technical change contributed to increased productivity, both per acre and per employed (Schön Reference Schön2000; Svensson Reference Svensson2006; Olsson and Svensson Reference Olsson and Svensson2010). Intimately connected to this development and the rising income of farmers, early industrialization began in textile production (Schön Reference Schön1979). Later in the nineteenth century industrial production increased rapidly also in other sectors such as lumber, mining, and engineering leading to an industrial breakthrough (Schön Reference Schön2000).

This development continued in the twentieth century, which also saw the establishment of a multi-party parliamentary system. From the 1930s to the mid-1970s this party system was dominated by the Social Democratic (Labor) party, and the period especially after WWII saw rapid social reforms and the development of an extensive welfare state. One field of reform concerned education, resulting in a considerable increase in school participation in the post-war period. The percentage of all 16-year-olds completing at least nine years of schooling increased from about 4 percent in 1930 to about 26 percent in 1965 (Stanfors Reference Stanfors2007, p. 188). This was the first stage of a change which suggests that people were increasingly eager to take up white-collar jobs, as those destined for blue-collar jobs typically left school after the mandatory six to seven years (Folkskola). The increasing importance of school participation potentially allowed the use of more meritocratic procedures in recruitment and promotion in the white-collar sectors (Dribe, Helgertz, and van de Putte Reference Dribe, Helgertz and van de Putte2015). Since the 1960s higher education has continuously grown in importance.

Our analysis is based on data from a community of five parishes in Scania in southern Sweden (see Figure 1). The economic and social development in the area under study reflects the overall trends for the country as a whole (Dribe, Helgertz, and van de Putte Reference Dribe, Helgertz and van de Putte2015). During the examined period, a part of the area (Kävlinge) changed from being a typical rural area, primarily engaged in grain production and animal husbandry, into a small industrial town characterized by food and textile industries as well as being a local center for business and administration. Another part of the area remained rural, but was transformed as agriculture became more and more mechanized and farm sizes increased substantially, leading to a rapid depopulation of the countryside in the second half of the twentieth century.

Figure 1

STUDY AREA

Source: Map created by Finn Hedefalk, Lund University. Used with permission.

DATA AND METHOD

We use data from the Scanian Economic-Demographic Database (SEDD). These data, collected by researchers at the Centre for Economic Demography, Lund University and available at www.ed.lu.se/databases/sedd, consists of individual-level longitudinal information on socioeconomic attainment and origin for individuals in the five parishes (Bengtsson et al. Reference Bengtsson, Dribe and Quaranta2014). The database is one of very few that can follow individuals across multiple generations from preindustrial times up until the present, and with detailed information on occupation and income at the individual level. The parishes are of course not a representative sample of Sweden in a statistical sense, but the area is not atypical and reflects conditions shared by most rural and semi-urban areas of the time studied (see Dribe, Helgertz, and van de Putte Reference Dribe, Helgertz and van de Putte2015). In this analysis we study attainment during the period 1815–2011, including individuals born between 1777 and 1980. Until 1967, the data covers all individuals living in the five parishes. Information is provided from continuous population registers (a household-based register where information at the individual level is continuously updated), with information on demographic events, including migration to and from households for all individuals in the selected parishes. Birth and death registers have been used to adjust for possible under-recording of events in the population registers. Information on socioeconomic attainment is obtained from the population registers, vital event registers, poll-tax registers (mantalslängder), and from annual income and taxation registers which are available from 1902.

From 1968, individual-level information covering the entire country is available from various administrative registers, maintained by Statistics Sweden. This allowed us to extend the database along several dimensions. First, individuals who ever lived in the five parishes prior to 1968 and who were still alive in that year were followed until 2011, or until death or emigration, regardless of their geographic location in Sweden. Additionally, spouses, parents, grandparents, children and siblings of individuals belonging to the original population in Scania were added to the database, provided that they were alive and living in Sweden sometime after 1967. All individuals added to the sample population were similarly followed until 2011, death, or emigration from Sweden.

In the period before 1968, migration could potentially introduce selection bias, as it was both frequent and selective (Dribe Reference Dribe2003a, Reference Dribe2003b; cf. Abramitzky, Boustan, and Eriksson 2012). To deal with this all individuals in the sample have been traced back to their place of birth to get information on the occupation of their fathers.Footnote ⁵ As a result, we obtained information on father's occupation for in-migrants allowing us to include them in the analysis as well. There is of course a potential problem also with selective out-migration in the period prior to 1968. Migrants leaving the area before being recorded with occupation or income are lost to follow-up. However, migration tended to be circular as shown by the fact that in-migrants and out-migrants from the area were similar in terms of origins and destinations. About 30 percent moved to any of the four major cities in the area (Landskrona, Helsingborg, Lund, and Malmö), and about the same proportion originated in these cities (calculations from the same dataset as is used in the empirical analysis). About 45 percent of out-migrants moved to places in the immediate vicinity (same municipalities), while the corresponding figure for the origins of the in-migrants were 37 percent. In- and out-migrants also had fairly similar socioeconomic status. Mean occupational status measured on the HISCAM scale (discussed later) differed only by 0.6 index points between in-migrants and out-migrants (index ranges from 1 to 99). Similarly, the proportions in different classes (HISCLASS) were highly similar, differing by only 1–2 percentage points. Out-migrants had somewhat higher proportions in higher occupations (10 percent versus 8 percent for in-migrants) and somewhat lower proportions in the working class (26 percent versus 28 percent for in-migrants in the group unskilled workers).

These similarities in observed characteristics between in- and out-migrants lead us to believe that these groups were fairly similar in terms of unobserved factors related to status (e.g., ability), which implies that the two migrant groups should roughly balance each other out. What is important is that we have both migrants and stayers in the sample, which significantly reduces the risk of bias in the analysis stemming from selective migration. Nonetheless it is impossible to completely rule out the possibility that there are some systematic differences between in-migrants and out-migrants that we were unable to capture.

Figure 2 summarizes the linkage of the data across the generations and the attrition at different steps of the process. There are 132,400 males observed in the database in the period 1815–2011. We have 38,700 links from son to father and approximately 15,600 individuals linked three generations back in time. Because data are not available uniformly for all outcomes, the potential sample sizes for analyzing three generation occupational status and earnings are 8,500 and 5,500 individuals, respectively. More specifically, the earliest information on the attained occupational status for a (linked) grandfather is for those born in 1777, and the latest for those born in 1960. Earnings data only allows for the observation of linked grandfathers for cohorts born between 1892 and 1980. As a result of missing information on the outcome variable in any of the three generations, the realized samples used in the analysis are 4,300 (51 percent of the potential sample) for occupation and 2,200 (40 percent) for earnings.Footnote ⁶ The analysis of earnings spans the period 1902–2011, whereas the analysis of occupation covers the period 1813–1990. We conduct a sensitivity analysis using an identical, but smaller sample to ascertain that the different samples do not drive the results we base the conclusions of this article on.

Figure 2

OVERVIEW OF THE SAMPLE CONSTRUCTION (OBSERVATIONS ROUNDED UP/DOWN TO NEAREST ONE HUNDRED)

Source: See Table 1.

Individuals often have multiple observations on occupation (average about 3) and earnings (average about 13) in the registers. Having these multiple observations is especially important for the study of income mobility because single measures of income tend to exaggerate measurements of mobility (Solon Reference Solon1992).

Prior to 1968, the analyses of class and occupational status use data obtained from several sources: demographic events, population registers, as well as annual data from the poll-tax registers and income registers. After 1968, we rely on occupational information provided by the quinquennial censuses (1970–1990). For the period before 1968, occupational notations in SEDD are coded according to an internationally comparable coding scheme for historical occupations (HISCO) (Van Leeuwen, Maas, and Miles Reference Van Leeuwen, Maas and Miles2002). Occupations observed after 1968 were originally coded by Statistics Sweden in the NYK/SSYK classification, and were recoded into HISCOFootnote ⁷ after first being converted from ISCO-88Footnote ⁸ into ISCO-68 (Hendrickx Reference Hendrickx2002). Hence, all occupations are coded in HISCO. While an individual may have several occupational notations at different times and from different sources, we use observations for prime working ages between 30 and 50. When available, the occupation of the father is measured in the same way or at time of the birth of his son, which generally occurs within the age range 30–50.

In the analysis, we exploit socioeconomic information on both the maternal and paternal grandfather. Because information for both grandfathers is sometimes missing, we also conduct an analysis based on the highest attainment in the grandparental generation, regardless of whether it refers to the paternal or maternal side.

We look at three different dimensions of socioeconomic status: social class, occupational status, and earnings. The summary statistics for individuals by social class (HISCLASS), occupation (HISCAM), and earnings are given in Table 1. Social class is measured at about the same age for all generations (the observation closest to age 40) according to HISCLASS, which is a 12-category occupational classification scheme based on skill level, degree of supervision, whether manual or non-manual, and whether urban or rural.Footnote ⁹ We group the 12 classes into five which are described in Table 1. About 12 percent of grandfathers belonged to the top group of higher occupations, while this proportion increased to 30 percent for grandsons. At the same time, the proportion of farmers went down from 36 percent to 14 percent reflecting the changes taking place in social structure over the twentieth century in Sweden (Dribe, Helgertz, and van de Putte Reference Dribe, Helgertz and van de Putte2015).

Table 1

DESCRIPTIVE STATISTICS

Note:

Variable min and max values in brackets. Period effects not estimated for the income sample, as all G3 observations occur after 1943. The variable grandfather life status refers to the G1 grandfather with the highest observed SES, used in Table 7.

Source:

The Scanian Economic-Demographic Database, Bengtsson et al. (Reference Bengtsson, Dribe and Quaranta2014).

Occupational status is measured by the continuous HISCAM scale, which determines the position of an occupation in the overall hierarchy based on social interaction patterns, mainly using information on marriage and partner selection (Lambert et al. 2014). It relies on patterns of interaction between incumbents of different occupations, translating into a relative position in a social hierarchy. HISCAM is generated from the HISCO codes, standardized to have a mean of 50 and a standard deviation of 15 in a nationally representative population, ranging from 1 to 99 (19–99 in our sample).Footnote ¹⁰ We use the universal scale rather than the Sweden-specific version, due to the small sample size used in constructing the Swedish HISCAM scale.

Earnings are measured as the annual income from paid employment, including self-employment but excluding capital gains. We first estimate permanent lifetime earnings, to reduce the impact of measurement errors that could otherwise create artificial earnings mobility (Solon Reference Solon1992; Zimmerman Reference Zimmerman1992; Björklund and Jäntti Reference Björklund and Jäntti1997). The same procedure is used for the continuous measure of occupational status (HISCAM). Following Lindahl et al. (Reference Lindahl, Palme and Sandgren Massih2015), life-time earnings or occupational status is obtained by regressing the individual's observed earnings or occupational status in year t on year of birth, its square and cubic, as well as on a set of observation year dummies, following equation (1) and (2):

(1)

(2)

In the subsequent step, each individual and year-specific residual, , is computed, and the mean of this residual is included in the three-generation regressions, as a measure of lifetime earnings or HISCAM attainment. A value above zero indicates that the individual's performance is better than predicted by the individual's year of birth and period under observation. Essentially by construction, the lifetime measurements of HISCAM and earnings vary around zero for all three generations, as displayed in Table 1 (−0.007 to 0.017 for average residual log HISCAM and −0.060 to 0.002 for average residual log earnings).

EMPIRICAL RESULTS

Our empirical analysis initially focuses on describing overall patterns of attainment across three generations, using mobility tables of social class (HISCLASS). The mobility tables show absolute mobility, and mainly serve to illustrate general patterns of multigenerational mobility. Table 2 presents origin and destination social class for fathers-sons (G2-G3), grandfathers-fathers (G1-G2), and grandfathers-grandsons (G1-G3). The grandfather observations are based on the sample which maximizes the number of observations, and takes the highest status observed in the grandparental generation, regardless of this being observed on the paternal or maternal side.

Table 2

MOBILITY TABLES, HISCLASS (PERCENT)

Source:

See Table 1.

The figures on the diagonal represent the share of individuals remaining in the same class as their ancestors. More specifically, the panel displaying fathers and sons (G2-G3), indicate that 59.5 percent of the sons of fathers in higher occupations remain in this class. The figures off the diagonal display outflow percentages from each class, i.e., in which class people end up if they do not maintain the status of their ancestor, as indicated by the diagonal. Looking first at the top panel, displaying fathers and sons (G2-G3), there is a considerable inflow to the higher occupations from all classes, meaning that the sons achieved a higher class than their fathers. This is especially the case for upward mobility from individuals with skilled worker fathers, with 35.6 percent advancing into higher occupations. The same pattern is present, albeit to a slightly lesser extent, between grandfathers (G1) and fathers (G2), and of course also between grandfathers (G1) and grandsons (G3). Thus, even though there is a strong persistence in the highest-status group, as shown by the 50–60 percent immobile in this class, structural labor market changes implied a great increase in positions in higher occupations to allow individuals from lower class origins to move upwards in large numbers. In this way, class persistence and absolute class mobility occur at the same time. Moreover, downward class mobility is much less frequent than upward mobility, both between grandfathers and fathers (G1-G2) and fathers and sons (G2-G3). Overall, many people experience an important transition from manual, blue-collar jobs to non-manual, white-collar jobs, which often requires higher education.

Martin Dribe, Jonas Helgertz, and Bart van de Putte (Reference Dribe, Helgertz and van de Putte2015) look in more detail at temporal changes in two-generational mobility and link them to the industrialization process. As is also documented in Table 2, they find that mobility into middle-class, white-collar occupations, was a result both of changes in the occupational structure affecting absolute mobility rates, and increased societal openness increasing mobility net of these structural labor market changes, referred to as relative mobility. Thus, while changes in the class structure by definition affect mobility chances through the amount of available positions in the labor market, changing relative mobility refers to alterations in mobility chances due to recruitment methods, etc. While overall mobility rates began to change early in the industrialization process largely due to more downward mobility (out of farming), increased upward mobility was more connected to the increasing role of achievement over ascription in occupational recruitment (e.g., Blau and Duncan Reference Blau and Duncan1967; Treiman Reference Treiman and Laumann1970). The latter development did not become really important until the later stages of industrialization with increasing demand for formal education (Dribe, Helgertz, and van de Putte Reference Dribe, Helgertz and van de Putte2015).

Class Attainment

Next we examine the association between the grandfathers' social class and the class attainment of the grandsons, controlling for the fathers' class as well as a range of individual-level characteristics. Each regression model accounts for the fact that outcome variables for grandfathers aged 30–50 are observed with different frequencies.Footnote ¹¹

Due to the categorical, yet hierarchical, nature of the social class variable (HISCLASS), the grandparental influence is examined using an ordered logit model and the results are reported as unadjusted coefficients, or log odds ratios (see Table 3). The purpose of the analysis is not to assess the magnitude of the influence of prior generations, but simply to look at the associations, focusing on the log odds ratios for the parental and, in particular, grandparental influence. The models control for age of grandfather at the time of attainment, lifetime migration status, and whether residing in a metropolitan area. For each respective ancestral influence, the log odds ratios are calibrated against the reference category (farmers). Because lower numbers in HISCLASS mean higher class attainment, positive coefficients (log odds ratios) imply lower class attainment, and negative coefficients higher class attainment, compared to the reference category (farmers). For example, if having a father in the highest class is associated with a negative log odds ratio, this implies a lower likelihood of a son ending up in a lower class compared to having a father in the farmer class. Furthermore, the size of the log odds ratio indicates the strength of this association, allowing for comparisons between all classes in each respective ancestral generation.

Table 3

ORDERED LOGIT ESTIMATES (LOG ODDS RATIOS) OF G3 CLASS ATTAINMENT (HISCLASS)

Notes:

Models control for age, age squared, residing in metropolitan area and lifetime migration. Models include analytical weights based on the number of underlying observations in the data for each G3 individual.

* p<0.1, ** p<0.05, *** p<0.01

P-value for Wald-test of β_{PG1, higher occupations}=β_{MG1, higher occupations} (Model 4): 0.061

P-value for Wald-test of β_{PG1, unskilled workers} =β_{MG1, unskilled workers} (Model 4): 0.051

Source:

See Table 1.

Column 1 in Table 3 serves as a reference. The analysis disregards the influence from the grandparental generation. Unsurprisingly, the father's influence on son's attainment is strong, implying substantial class persistence across two generations. Having a father in the higher occupations or skilled worker group implies a significantly lower likelihood of ending up in a lower class compared to if the father is a farmer, as shown by the negative coefficients. The more strongly negative log odds ratio observed for sons with fathers in higher occupations (–1.3), however, indicates a higher likelihood for high status attainment compared to sons whose fathers were skilled workers (−0.5). There are also indications of strong intergenerational class persistence at the lower end of the status hierarchy, as indicated by the positive coefficients. In fact, the pattern indicates a rather linear gradient in terms of the intergenerational transmission of class. More specifically, the lower the attained class of the father, the higher the probability that the son remains in a lower class.

In column 2, we add paternal grandfather's (PG1) class attainment which does little to change the father-son (G2-G3) association. More interestingly, there is an association between paternal grandfather and grandson (PG1-G3), even if weaker than that between father and son. While coefficients are not statistically significant for each class, having a paternal grandfather in a higher class than the farmer class is associated with a lower chance of low class attainment. Furthermore, having a lower skilled or unskilled paternal grandfather is associated with a higher chance of low class attainment. In column 3, we look at associations with the maternal grandfather's class. The pattern is similar to the one observed for the influence of the paternal grandfather's class, but we find a stronger and statistically significant impact of having a maternal grandfather with an unskilled occupation. Again, we find a grandparent influence controlling for parental generation across some social groups.

Both paternal (PG1) and maternal grandfather's (MG1) class are included in column 4, which does not change the basic association between father and son, nor affect the association between grandfather and grandson. The substantive conclusion remains: having a maternal grandfather with an unskilled occupation significantly increases the likelihood of low class attainment for the third generation. Grandsons whose paternal grandfathers belong to the higher occupations enjoy a significantly higher probability of attaining this class.Footnote ¹² In column 5 we use the highest observed grandparental class, regardless of whether on the maternal or paternal side, which also maximizes the sample size. This does not change the results but further supports the conclusion of an independent influence of the grandfather on class attainment, in addition to any influence going through the father, at either end of the status hierarchy. Having a grandfather in the highest class is associated with higher class attainment, while a grandfather in the working class is associated with lower class attainment, when controlling for the attainment of the father.

Column 6 shows the changes over time in the influence of grandparents. Here we include a three-period interaction, with the post-WWII period as the reference category. Hence, the log odds ratios for the baseline refer to the post-WWII period and the interaction log odds ratios indicate the extent to which the associations are different in earlier periods. Overall, the results suggest few significant changes over time in the grandparental influence. The exception is the first half of the twentieth century, a time characterized by overwhelming labor market changes occurring in parallel with a rapid industrialization (Erikson Reference Erikson1983; Schön 2000). During this period, the disadvantage experienced by grandsons whose grandfathers belonged to the unskilled working class was accentuated, as indicated by the significant and positive interaction effect. A significant disadvantage is also observed among individuals with skilled worker grandfathers. Overall, however, the grandparental influence on class attainment does not appear to change that much over the nineteenth and the twentieth centuries.

Occupational Attainment (HISCAM)

The preceding analysis suggested the presence of a significant grandparental influence on class attainment controlling for the influence of the parental generation. We now examine the impact of grandfather occupational status, measured by the continuous scale HISCAM.Footnote ¹³ The baseline model is given in column 1 in Table 4. It shows that almost 40 percent of the parental generation's (G2) advantage or disadvantage is transmitted to the subsequent generation of grandsons (G3), as indicated by the statistically significant elasticity of 0.38. This is a sizeable father-son association in occupational status, exceeding those typically observed in previous research focusing on earnings (see next section).

Table 4

OLS ESTIMATES OF OCCUPATIONAL STATUS ASSOCIATIONS (G3 HISCAM AS DEPENDENT VARIABLE)

Notes:

Models control for residing in metropolitan area and lifetime migration. Models include analytical weights based on the number of underlying observations in the data for each G3 individual.

* p<0.1, ** p<0.05, *** p<0.01

P-value for Wald-test of β_PG1=β_MG1 (Model 4): 0.001

Source:

See Table 1.

Adding paternal grandfather occupational status (column 2) only slightly reduces the father-son association. The association between paternal grandfather and grandson attainment is substantially weaker—0.13 for grandfather-grandson, compared to 0.34 for father-son—but still statistically significant. Consequently, even controlling for parental influence, the paternal grandfather's occupational attainment significantly influences the grandson's attainment. Looking at the influence of the maternal grandfather (MG1), in column 3, the association is somewhat weaker than that observed for the paternal grandfather (PG1), yet still statistically significant (0.09 for MG1 compared to 0.13 for PG1).

Including the occupational status of both the paternal (PG1) and maternal grandfather (MG1) in column 4 does not change grandparental influence; both the maternal and paternal grandfathers matter for occupational attainment, but the influence of the paternal grandfather is larger (0.12 compared to 0.06).Footnote ¹⁴ Given the results so far, it is hardly surprising that the association between the combined grandfather-grandson occupational scores is similar to the paternal grandfather-grandson association (column 5) with a combined elasticity of 0.11 (statistically significant). The time period interactions in column 6, are not statistically significant. Hence, similar to class attainment, the grandparental influence on occupational status does not change markedly over the nineteenth and the twentieth centuries, despite major changes taking place in occupational structure and dominant mechanisms of recruitment.

Earnings

We now turn to the earnings association shown in Table 5. In the baseline model (column 1) we have the earnings association between fathers and sons. The elasticity is 0.18 and statistically significant, which is in line with similar two-generation estimates in other studies for Sweden, typically ranging between 0.2 and 0.3 (Björklund and Jäntti Reference Björklund and Jäntti1997; Björklund, Lindahl, and Plug Reference Björklund, Lindahl and Plug2006; Lindahl et al. Reference Lindahl, Palme and Sandgren Massih2015). This suggests that 18 percent of an earnings advantage, or disadvantage, in one generation is transmitted to the next generation. Thus, the simple intergenerational (G2-G3) transmission of occupational status appears to be a considerably stronger process compared to earnings, as suggested by the elasticity of 0.38 relative to 0.18.

Table 5

OLS ESTIMATES OF EARNINGS ASSOCIATIONS (G3 EARNINGS AS DEPENDENT VARIABLE)

Notes:

Models control for residing in metropolitan area and life time migration. Models include analytical weights based on the number of underlying observations in the data for each G3 individual.

* p<0.1, ** p<0.05, *** p<0.01

P-value for Wald-test of β_PG1=β_MG1 (Model 4): 0.3112

Source:

See Table 1.

Adding paternal grandfather's (PG1) lifetime earnings to the model (column 2) shows no significant grandparent effect, but attenuates the parental influence, which nevertheless remains statistically and economically significant. Similarly, there is no statistically significant association between the lifetime earnings of the maternal grandfather (MG1) and those of the grandson (G3), controlling for the attainment of the father (G2) (column 3). The magnitudes of the coefficients for maternal and paternal grandfathers are similar.

When including the lifetime earnings of both the maternal (MG1) and paternal (PG1) grandfather in the same model (column 4), both coefficients are of a similar magnitude (0.05 and 0.06, respectively), but statistically insignificant. When maximizing the sample size by using the highest observed lifetime earnings in the grandparental generation, regardless of whether this is observed on the maternal or the paternal side, the results remain similar. Column 5 shows only a very modest, and statistically insignificant, association in lifetime earnings between grandfathers and grandsons, when controlling for the father's earnings. While the elasticity for fathers' earnings is close to 0.2, it is only 0.03 for grandfathers. Thus, in terms of earnings we find very little to indicate an independent impact of grandparental earnings on grandchildren's earnings when controlling for the impact of parental earnings. This result is quite similar to the one for the Malmö cohort, reported in Lindahl et al. (Reference Lindahl, Palme and Sandgren Massih2015, Table 9), where the corresponding coefficient for grandfather was 0.08, and for father 0.28, with the grandfather coefficient only being statistically significant at the 10-percent level.

SENSITIVITY ANALYSIS

To ascertain whether the different results for earnings and occupational attainment are due to the different samples used in the two analyses, we estimated a model for both outcomes using the same sample (family lines for which we have information on both earnings and occupational status). The results are displayed in Table 6 and they are similar to the ones based on the larger samples. We continue to find a clear association between the rankings of the grandfather and the grandson for occupational attainment (HISCAM) even when controlling for the attainment of the father, but only an association between father and son for earnings. This confirms that the results for the occupational outcomes and earnings outcomes are indeed very different.

Table 6

OLS ESTIMATES OF G3 HISCAM AND EARNINGS ASSOCIATIONS, BASED ON IDENTICAL ANALYTICAL SAMPLES

Note:

Models control for residing in metropolitan area and lifetime migration. Models include analytical weights based on the number of underlying observations in the data for each G3 individual.

* p<0.1, ** p<0.05, *** p<0.01

Source:

See Table 1.

A common concern in historical mobility studies is that farmers drive the results by being a large and rather special group given ownership of land and its role for intergenerational inheritance (Long and Ferrie Reference Long and Ferrie2013; Hout and Guest Reference Hout and Guest2013; Xie and Killewald Reference Xie and Killewald2013). To assess the role of farmers, we estimated the same models excluding farmers in all three generations. The results are similar to those based on the full sample, both in terms of occupational status and earnings (detailed results available from the authors). We are thus confident that the patterns presented are not driven by the outcomes of farmers.

There are several possible explanations for a positive association between the outcomes of grandfathers and grandsons. One explanation for direct influence from grandfathers to grandsons would be overlapping life spans. We assume that while socio-emotional pathways require a certain level of proximity and interaction between grandparents and their children (see Zeng and Xie Reference Zeng and Xie2014; Knigge Reference Knigge2015), this is not necessary for the economic pathways, because resource transfers across generations do not depend on social interaction. To examine this in more detail, we estimate models including indicators of life status of the grandfather (G1) at the time of their grandson's (G3) birth. Table 7 displays the results for the lifetime measures of HISCAM and earnings, respectively. Naturally there is a considerable proportion for which the grandparental life status is unknown (65 percent in in the HISCAM sample and 14 percent in the earnings sample). Grandfather being alive is the reference category. The base coefficient for grandfather is statistically significant and of a considerable magnitude for occupational status, but small and insignificant for earnings. Even though the interaction effect for grandfather being dead is not statistically significant, it is negative and of a considerable magnitude (−0.150), and it is also similar for unknown life status (a model with grandfather dead/unknown merged into a single category yields an estimate of −0.128, p-value=0.10). The base coefficient, which refers to grandfather being alive, is also much larger than in the model without interactions (0.219 versus 0.108). This clearly points to the grandfather-grandson association being stronger when life spans overlap, but as the coefficient is not statistically significant it is difficult to draw a definitive conclusion.

Table 7

IMPACT OF GRANDPARENTAL LIFE STATUS AND RESIDENCE ON INTERGENERATIONAL ASSOCIATIONS (OLS)

Note:

Models control for residing in metropolitan area and life time migration. Models include analytical weights based on the number of underlying observations in the data for each G3 individual.

*p<0.1, ** p<0.05, *** p<0.01

Source:

See Table 1.