Methodology to classify occupations

After an individual’s occupation is captured from population registers, it must be transformed into a more comparable form. To make occupational classifications as meaningful as possible for studying homogamy and social mobility in the long run, we anticipate and account for varying methodological challenges. Fortunately, the methodology for classifying occupations into meaningful historical stratifications has developed significantly during the last few decades, enabling us to compare different countries and periods (van Leeuwen and Maas 2016, 2011; van Leeuwen, Maas, and Miles 2002). In articles III and IV, we took advantage of these commonly used methods (Table 1). First, we classified the occupational titles based on the historical international classification of occupations (HISCO), which consists of tens of thousands of occupational titles and coding (see more details about the HISCO system in van Leeuwen, Maas, and Miles 2002).²⁶ This process was rather time-consuming since no comprehensive HISCO-listings using Finnish occupational titles existed, although we did have access to the preliminary listings by van Leeuwen and Maas (see more details in footnote 5 in article IV). The second step involved recoding the HISCO classifications as larger meaningful occupational groupings. Prior studies have often used a historical international social class scheme (HISCLASS) that includes twelve classes as a starting point (van Leeuwen and Maas 2011). We made two additions to the HISCLASS scheme. We created a new labour class (13) category that accounts for types of labour difficult to assign to a particular a sector (urban or rural).

In addition, we created a crofter class (14) since our data enabled us to separate crofters from the class of farmers (see also the class division between crofters and farmers in Peltonen 1992).²⁷

The third classification phase involved establishing the final set of historical classes studied as part of the thesis project. In article III (homogamy), we utilised the traditional classes that have already been used in social history studies in Finland: (1) high-middle and upper classes, (2) farmers, (3) crofters and 4) labourers (Soininen 1974, p. 42; Alapuro 1985; Rasila 2003). In Finland, those in the high-middle and upper classes, which included estate owners, the nobility, the clergy, civil servants, army officers and merchants, only comprised roughly one to four per cent of the total population (Haapala and Peltola 2018). Notably, in this scheme other social classes can be divided by their relations to the land (owners, renters and landless). In fact, this type of stratification is

26 In addition, we made small changes to the HISCO system. First, highly educated persons were classified as having graduated. Second, individuals who lived in other people’s homes as renters or workers or because they were paid by the municipality were recorded as farmhands.

27 We assigned the great landowners (rusthollars / rusthollari) to the class of farmers. However, the crofter class includes, among others, arentimies, vuokraaja – arrendator – leaseholder / torppari – torpare – crofter / lampuoti – landbonde – tenant farmer.

40

arguably proper for studying homogamy patterns (article III). However, we decided to use additional class schemes in article IV (social mobility), which take into account diversity in the labour classes:

1) white collar, 2) farmers, 3) skilled workers and crofters, and 4) unskilled/semi-skilled workers.

Furthermore, we made international comparisons according to as similar classification scheme, whenever possible, which departed from prior international studies: 1) white collar, 2) farmers (including crofters), 3) skilled/semi-skilled workers and 4) unskilled workers (Modalsli 2017; Pérez 2019; Long and Ferrie 2013a).²⁸

28 In the international studies, there are no possibilities to separate crofters and farmers. Thus, we transformed the modern data (1950–2000) from the seven-class CASMIN version of the Erikson-Goldthorpe classification scheme into a four-class division that matches as closely as possible the previous periods: 1) white collar (I+II+ III+ IVa+b); 2) farmer (including crofters) (IVc); 3) skilled (V+VI); and 4) semi-skilled or unskilled (VIIa+ VIIb). This classification scheme differs from the previous one in that crofters are included in the farmers’ class. However, it is evident that the crofters’ role was small in the social stratification during the latter half of the 20^th century.

41

1. Occupation title2. HISCO3. Modified HISCLASS4. ‘Traditional’ Finnish classes 5. Finnish classes*6. International classes* . . .. . .1. Higher managers 1. High middle and upper classes (1–6)1. White collar (1–6)1. White collar (1–5) parish turner81230 2. Higher professionals2. Farmer (8)2. Farmer and crofter (8, 14) shoemaker (parish)80110 3. Lower managers 2. Farmer (8) 3. Skilled and crofters (7, 14) 3. Skilled/semi- skilled (6–7, 9) parish student -14. Lower prof and clerical, sales3. Tenant/crofter (14)4. Unskilled/semi-skilled (9–13)4. Unskilled (10– 13) parish tailor79100 5. Lower clerical and sales parish tailor's daughter79100 6. Foremen 4. Labour (7, 9–13) parish smith 83110 7. Skilled workers parish smith's daughter83110 8. Farmers parish turner81230 9. Lower skilled workers parish shoemaker 80110 10. Lower skilled farm workers parish cottage keeper51040 11. Unskilled workers parish cottage's janitor55130 12. Unskilled farm workers parish poor-113. Workers (new class) . . .. . .14. Crofters (new class) Table 1. The process of coding occupations into meaningful classes Notes: * In brackets: numbers refers to HISCLASS classes.

42 4.2.3. Methods

The starting point in prior studies for examining intergenerational occupational mobility or social heterogamy has been to observe transition tables (e.g. Pérez 2019). For example, the simplest transition table is where the occupations of parents and children can be divided into two classes: high and low. This yields the simplest transition table:

Parents’ occupation

Child’s occupation High Low

High High to High Low to High

Low High to Low Low to Low

This type of approach is useful for characterising the actual mobility that occurred between classes.

However, assessing absolute mobility is problematic in long-run studies since such a measure does not account for changes that occurred in the social structure. Therefore, it does no reveal how strongly occupation patterns persisted during the research period, which prevents us from comparing mobility estimates between research periods and countries due to varying social structures.

Prior studies have used one of two possible ways to examine relative mobility: 1) logits and the log-linear regression models and 2) the Altham statistic. Whereas the former strategy is more common in sociology literature, the latter is more common in economic history and economic literature. In practice, both methods rely on odds ratios, and therefore the results should be quite similar (see the discussions regarding methodological differences in Xie and Killewald 2013; Hout and Guest 2013; Long and Ferrie 2013a, 2013b; Xie 1992). We decided to utilise the regression framework in article III (homogamy), whereas we used the Altham statistic in article IV (social mobility).

To study spouses’ heterogamy, we utilised their social origin heterogamy, i.e. the social statuses of the spouses’ parents. This methodological practice is common in the field since wives were often assigned their husbands’ occupation when married (article III; van Leeuwen and Maas 2005). To capture the odds for heterogamous marriages, we utilised logistic regression models and odds ratios. Prior studies on heterogamy have used logistic regression models quite often since they make it possible to examine possible reasons behind the heterogamy (e.g. Bras and Kok 2005). The dependent variable is set as the heterogamy of the spouses’ parents’ occupational class (1 = heterogamy, 0 = homogamy). We modelled the impact of the grooms’ social status on heterogamous marriages. For example, we estimated the odds for a heterogamous marriage when the grooms’ social

43

status was farmer compared with non-farmer ( = 0 = 1). Furthermore, we added individual and family level variables, ²⁹, as well as regional and structural variables, ³⁰, to the model to observe the possible causes of heterogamy. Lastly, the odds ratios (above one means possible positive correlation with heterogamy, whereas below one means possible negative correlation) can be estimated using the following model:

( = 1)

( = 0)= + + + + (2)

The Altham statistic is commonly utilised in social mobility literature for several reasons. It is a relative measure, which enables us to estimate intergenerational associations regardless of changes in social structure. Moreover, it is a simple measure that does not rely on complex assumptions in the same way that regression frameworks do. Furthermore, it is possible to compare different transition tables directly and estimate the divergence of confidence levels. In addition, the Altham statistic is decomposable, which enabled us to estimate the various components of mobility. Lastly, it is a relative measure that does not rank occupations or classes. Thus, it is quite useful in long-run studies, where rankings and class position could change over time (Altham and Ferrie 2007; Long and Ferrie 2013b, 2013a; Pérez 2019).

The Altham statistic compares mobility based on an independence (I) table (no associations exist between rows and columns, practically every cell has a value of one) and transition tables, in this case tables P and Q. The following formula is used to calculate the difference between table P (a size of r x s), with elements ( ), and table Q, with elements ( ):

( , ) = log −log

/

(3)

The previous formula reveals the difference in association between tables P and Q; however, to capture its persistence in table P individually, the latter table (Q in the example above) is replaced with the independence table (I). In other words, the metric d(P,I) tells the strength of association in

29 The following variables are based on the 10Gen database: year of marriage; groom’s or bride’s first marriage;

the age of the spouses; the age difference between the spouses; the spouses’ position as an illegitimate child; the bride’s former position as a single mother; the spouses’ birth parishes and whether they differ.

30 Urbanisation and industrialisation (Gebhard 1908; Böcker collection 1835); social structure (Böcker collection 1835; Gebhard 1908; Kilpi 1913); people who emigrated (OSF 1882, 1885); the share of Finnish-speaking population (SYF 1885); the share of poor relief recipients (OSF 1890; PRC 1927–1936); the average lifespan in a parish (the 10Gen data, see text).

44

table P, while d(Q,I) captures the strength of association in table Q and d(P,Q) shows the difference of association in tables P and Q. The Altham statistic measures persistence, and it varies from zero to ∞: from independence to extremely thick persistence.

45