For our quantitative investigation, we use event history techniques of analysis. In the following paragraphs we highlight both the advantages and problems related to this procedure.
Especially in population studies, though not exclusively, scholars increasingly point out the importance of measuring variables of interest on a continuous basis. This is realized by an event-oriented research design, where, ideally, individuals are followed throughout their life, and the occurrences of the events under study are recorded. An event is considered as a “qualitative change that occurs at a specific point in time,” and this “change must consist of a relatively sharp disjunction between what precedes and what follows” (Allison 1984: 9). Event history data may display, for instance, events in the process of family formation, such as union formation and childbirth, or changes in other life trajectories, such as education or employment.
One important aim of this research strategy is to study causes of events. Thus, the dataset should not only include full histories of the event under study, but also complete information about possible explanatory variables. Some of these variables, such as region of birth or in some cases religion, may be constant over time; others, such as educational attainment or employment, may vary (Allison 1984). Blossfeld et al. (2007) emphasize that “the major advantage of event history data is that they provide the most complete data possible on changes in qualitative variables that may occur at any point in time” (Blossfeld et al. 2007: 19). In contrast, other research designs, such as cross-sectional samples or panel designs, seldom offer information about the exact time of occurrence, not only of the event under study, but also as regards explanatory variables (Blossfeld et al. 1989). Thus, especially explanations based upon cross-sectional data are inappropriate whenever there are changes in causal variables – as is true for most cases (Tuma and Hannan 1984).
According to Blossfeld et al. (1989), the event history research design offers several advantages. It explicitly takes change and the dynamics of empirical phenomena into account; it gives information about prior history that might help to improve the explanatory and prognostic capacity of statistical models; it permits the reconstruction of a continuous process; and it allows for investigating complex and/or parallel processes.
There are various ways to record event-oriented information. Data might be collected retroactively or through a long-term panel study, which is, of course, much more cost- intensive. However, event history datasets recorded retrospectively suffer from several limitations. It has been argued, for instance, that respondents face difficulties in recalling the timing of changes accurately; that retrospective designs are inappropriate to record information on attitudes; or that such designs are based on survivors and ignore those individuals who have died or migrated (Blossfeld et al. 2007). Further, retrospective studies might suffer from misrepresentations of specific populations. In a study on educational homogamy, Blossfeld and Timm (2003) showed, for instance, that persons who were single at the time of the interview were generally excluded.
Nonetheless, the analysis of event history data offers the possibility for a causal understanding of social processes. According to Aalen (1987) causality in dynamic modelling has to be understood as follows: “The cause has to precede the effect in time. A factor is only called a cause if variation in this factor produces changes in consecutive parts of the process” (Aalen 1987: 185). Or as put by Blossfeld et al. (2007): “Event history models … relate change in future outcomes to conditions in the past and try to predict future changes on the basis of past observations” (Blossfeld et al. 2007: 21).
4.3.2 Data and Model Description
For the event history analysis of this study, we use the Indagine longitudinale sulle famiglie italiane of 1997 and 1999 (Longitudinal Survey of Italian Households, ILFI).8 The ILFI is
one of the few existing panel surveys in Italy. It was first conducted in 1997 and carried out by the universities of Milan (Bicocca), Trento, Bologna, ISTAT, and others. The survey was
8 Università degli Studi di Milano Bicocca, Università degli Studi di Trento, Università degli Studi di Bologna,
Indagine longitudinale sulle famiglie italiane, 1997 – 1999. File dati su supporto magnetico. Responsabile scientifico: A. Schizzerotto.
continued every two years up to 2003. In the first wave, 9,770 members of 4,404 Italian families were interviewed (for further information, see Schizzerotto 2002).
Using retrospective data from the first two waves, we calculate multiplicative intensity models in order to analyze women’s risk of entering cohabitation and direct marriage as a first partnership in Italy. We concentrate on women born between 1940 and 1974. We decided to exclude women born before 1940, firstly because entry into marriage was not only the prevailing but nearly the exclusive practice of entering a couple relationship, and secondly because of the very low number of cohabiting women in these cohorts as a result of this behavior. Women born between 1900 and 1939 or after 1974, as well as foreigners and women who lived abroad during their childhood, were not considered. The original dataset consists of information on 5,313 men and 5,819 women. After having cleaned and restricted the data, information on 3,2339 women was used – 81 of them entered
cohabitation as a first relationship and 2,436 entered marriage directly.
To estimate first cohabitation and direct marriage intensities of women in Italy, we use multiplicative intensity models. The observation starts at age 15. The corresponding baseline hazard is modeled as a piecewise function that will be divided into 0-60, 60-120, 120-180, 180-240, and 240-300 months (from exact age 15-20, 20-25, 25-30, 30-35 and 35- 40, respectively). Censoring will occur at entry into direct marriage (or at entry into cohabitation as a first relationship for the model considering direct marriage), upon reaching the age of 40, or at the month of the interview, whichever occurs first. We control for a number of time-constant and time-dependent covariates. The following formula describes the main effects model:
µ(t)jklmnopqr = aj(t) × bk × cl × dm × en × fo(t) × gp(t) × hq(t) × ir(t)
Factor a represents the effect of time, i.e. time from the exact age of 15 until entry into cohabitation or censoring, whereas j(t) denotes the time segments, which are assumed to be piecewise constant. Factors b to e indicate the time-constant covariates and factors f to i represent the time-varying covariates. Cohort, region of residence at age 15, education of both parents, and religion are used as time-constant covariates. Cohorts are subdivided into women born in the periods 1940-49, 1950-59, 1960-69, and 1970-74. For region of
residence at age 15, we distinguish between Northwest, Northeast, Center, South and Islands regions. We are aware of the fact that the islands of Sicily and Sardinia are quite different from each other in many aspects and especially as far as family formation is concerned, but the low number of cohabiting women in these areas did not allow for a more detailed categorization. The ILFI offers full migration histories – to identify the region of residence during socialization, we calculated the region of residence at age 15. The education of the father and the mother was classified into low (illiterate person, no degree, or primary degree), medium (lower secondary) and high (higher secondary, university) level of education. In addition, we used a missing category. In the final model, we matched the education of both parents together, using the following classification: Both parents have a low level of education, both have a medium or both have a high level of education, the mother’s education is higher than the father’s, and the father’s education is higher than the mother’s. The missing category was then deleted. For religion, we argue that religious affiliation is relatively stable over the life-course, so we use it as a time- constant covariate. Moreover, the ILFI provides information on religion only for women interviewed in 1997. Women who entered the survey in 1999 were not asked to provide information on their religious affiliation. The categories of religion are Catholic and not Catholic (no religion, Christian without church affiliation, other, or missing – including those interviewed only in 1999). (See Appendix A, Table A.1 for more detailed information.)
As time-varying covariates, we use educational level, educational attendance, having a first conception, and employment status. For educational level, we distinguish no degree/primary, lower secondary, higher secondary, and university. Educational attendance was calculated according to time spent in education, independently of whether a woman acquired a degree or not. Time periods with less than five months between one exit from education and the next entry into education were ignored since summer vacations or the time between A-level school examinations and entry into university are normally not perceived as being out of education. Since birth occurs nearly always within marriage and contributes to entry into union, we decided to use first conception instead of first birth. Having a first conception was calculated by subtracting eight months from the month of birth, as most women are not aware that they are pregnant during the first few weeks of pregnancy (Baizán et al. 2001). For employment status, we distinguish between being in the labor market (active) and being out of the labor market (inactive). Although we are aware
that a more detailed distinction, such as working part-time or full-time, would be more appropriate, we decided to use a simple categorization (active–inactive). This decision was mostly driven by the low number of cases we had when we tried to use thinner levels of employment.
In our analysis we include all current factors, which means that in all sections of the analysis we control for the impact of the factors specified in this sub-section.