## 3. Covariates

For each of the 695 respondents in our data set we calculate the number of pre-union children of the respondent and his or her partner. We consider a child as belonging to the second union if it is born at union formation or later [Note 11].

In contrast to the Swedish data used in [Vikat et al. 1999], the Austrian FFS includes very detailed information on the partner's pre-union children and whether and how many pre-union children the respondent and the partner have brought into the new household, but the information on the partner's pre-union children proved to be very inconsistent.

Consider the variable age of the youngest child at the start of the second union. To construct this variable we first verified the matching between stepchildren in the respondent's birth biography and the number of the partner's pre-union children indicated in the partner's biography. Fifty persons reported that their partner brought at least one pre-union child into the household at the time of the formation of the second union, but only for twelve (among those 50) respondents did these children show up as stepchildren in the birth biography of the respondent him-/herself. Hence, for 38 respondents we do not have any information on the birth date of the partner's pre-union children and consequently we cannot take into consideration these children in constructing the age of the youngest child at the start of the second union [Note 12]. As a solution to this data inconsistency we include the variable age of the youngest child only in regressions where we base our calculations on female respondents only. This practice is justified by the fact that the youngest child is the child that is most likely to be brought into the new household and women are more likely than men to bring their pre-union children into their second union. The age of the youngest child varies across the duration variable, but for our analyses we code the variable as a fixed covariate by calculating the age of the youngest child at the start of the second union. The number of resident and non-resident pre-union children of either partner is also a fixed covariate and is only recorded at the time of the formation of the second union. Since we do not have any further information on changes of residence of the partner's children, we cannot re-code the variable for the point of time of the start of the second union. But one can argue that it is usually foreseeable at the start of a union if a pre-union child will join the joint household or not.

We control for the age of the youngest child since we expect women whose youngest child at the start of the second union is already of school age or older to have lower intensities of a conception of a first child in the second union. As was done in [Griffith et al. 1985], we organize the variable according to the number of children. The most common hypotheses why mothers of children who are older at the start of the second union are less likely to have another child are (a) that mothers do not want to prolong or start the period of childbearing again, and (b) the child is already too old to benefit from a half sibling. But as noted in the introduction, the age of the youngest child may have the opposite effect. The parenthood effect, which may apply for the partner who does not have any pre-union children, may be stronger in the case of older children, since they are less likely to be considered substitute for a biological child. The distribution of occurrences and exposures across the different levels of the variable age of the youngest child is summarised in Table 2 of the Appendix.

Instead of explaining the intensity of the conception of a first common child in terms of the respondent's and partner's characteristics and including the sex of the respondent as a further covariate (see [Vikat et al. 1999]), we directly distinguish between a woman's and a man's characteristics within the set of alternative covariates [Note 13]. We kept the gender in our models to be able to check whether our results differ between couples depending on whether the respondent is male or female. As we only know that it is the second-order union for the respondent, the argument would be that women who are partners of men for whom it is the second union are different from women who are themselves already in their second union. Similarly, the reverse argument holds for men, i.e., men in second unions may be different from male partners in unions that are the second unions for women.

Among the second unions considered in our study, 43.7 per cent of all female partners indicate at least one pre-union child, while only 26.8 per cent of male partners have had at least one pre-union child (Appendix, Table 1). Distinguishing between pre-union children living in the household and those not living in the household at the time of the formation of the second union emphasises our previous discussion. Female partners are more likely to bring their pre-union child(ren) into the newly formed household. 41.6 per cent of all female partners indicate that they have brought at least one of their pre-union children into the household, while the corresponding number for men is only 5.3 per cent (Appendix, Table 1). Even when we take into account that male partners indicate on average about 40 per cent fewer pre-union children than female partners, this difference is significant.

Some obvious differences between female respondents and female partners and male respondents and male partners, respectively, are clearly evident in our data set (Appendix, Table 2 and Appendix, Table 3). About 54 per cent of all female respondents have pre-union children, as compared to 19 per cent of all women who are partners of male respondents. For men the difference is less pronounced. 36 per cent of all male respondents have pre-union children, while only 23 per cent of men who are partners of female respondents indicate having pre-union children.

Besides pre-union children, we have also experimented with a set of covariates that might pick up some individual characteristics (Appendix, Table 1).

For analysing the effect of the age of both partners we chose the representation woman's age at the start of the second union and man's age at the start of the second union. 77.1 per cent of female partners, but only 59.6 per cent of male partners, were below age 30 at the start of the second union (Appendix, Table 1). Controlling for the age of each partner (and in particular for the woman's age at the start of the second union) is particularly important since our results would otherwise be confounded by the argument that there is a negative age effect on fertility in second unions. Another argument for lower fertility could also be greater age-heterogamy (large age differences between the partners) in second unions.

To capture other characteristics of the first union in addition to any pre-union child, we also include the dissolution form of the first union and whether the respondent has been formerly married. Since information on the characteristics of the preceding union is only available for the respondent, we shall run gender-specific regressions in the next section once we have included these factors in our analysis [Note 14]. Before considering the distribution of occurrences and exposures across these covariates we have to explain briefly the relevant information contained in the Austrian FFS.

For each union the following dates and characteristics are recorded: union status (married vs. cohabiting) at time of union formation, date of marriage, the reason for union disruption if one had occurred (whether the respondent broke the union or his/her partner or both partners, or whether the partner died). To arrive at a compact representation for the covariate `dissolution form of the preceding union', we have used only three alternative levels: separation (regardless of who ended the union), widowed and no answer. In our data set 91.1 (95.5) per cent of all female (male) respondents recorded a union disruption, 4.2 (1.5) per cent recorded that their first partner had died, and 4.6 (3.0) per cent gave no answer (Appendix, Table 2 and Table 3). We include the dissolution type of the preceding union in our analysis since the fact whether second unions have come about because of divorce or of widowhood is likely to influence their character [Burgoyne and Clark 1981].

Our interest in the effect of the variable `formerly married' stems from the different hypotheses proposed in the literature as regards the family status at the date of formation of a second union. [Heekerens 1986] focuses on reproductive behaviour in combination with remarriage behaviour. He offers two hypotheses: (a) The marriage order of the male partner (first-married versus re-married) influences the completed fertility. On average the number of children is lower for marriages that are already higher-order marriages for the male partner. (b) "According to the second hypothesis, the lower fertility of remarried women (by comparison with first-married women) is at least partially due to the fact that a higher percentage of these (remarried) women is married to remarried men" ([Heekerens 1986], p. 515). Both hypotheses rest on the observation that "a substantial proportion of men with divorce experience tend to bring into marriage a preference for not becoming a parent (again) during their marriage" ([Furstenberg 1980], p. 470). This lower preference for childbearing on the part of re-married (divorced) men is investigated in [Rosenstiel 1984] in greater detail. Not only is the normative pressure from society lower once you have children, but having pre-union children also implies a higher financial burden (laws require fathers to support children) and might even make a further child not `feasible'. And the experience of a divorce might also negatively influence the attitude towards further childbearing.

Additionally we control for marital status (pre-cohabitation - cohabiting - married) and the calendar time period. The distinction between three levels of marital status allows us to control for the fact that the start of the second union differs across the respondents, i.e., the period of observation starts for the majority nine months before the time of the formation of the union. Only for about a fourth of all respondents does the period of observation begin less than nine months prior to union formation or at union formation. Since the variables union status and calendar time period are characteristics of the second union and common to both partners we can include these factors in regressions where we include all respondents in our selected data set. Couples in their pre-cohabitation time account for 13.9 per cent of the time of exposure to the "risk" of the conception of a first common child in our data. Cohabiting couples account for 54.3 per cent and married couples for 31.9 per cent of the total time of exposure (Appendix, Table 1). The covariate calendar time period is split up according to changes in maternity leave periods and economic trends and is aimed to include changes in childbearing behaviour as influenced by family policies (cf. [Hoem et al. 1999]). In fact, in our data set the calendar time period seems to take into account the increasing prevalence of second unions in more recent years. More than half (56.1 per cent) of the total exposure time is contributed by months after 1987.

 Fertility in second unions in Austria: Findings from the Austrian FFS Isabella Buber, Alexia Prskawetz © 2000 Max-Planck-Gesellschaft ISSN 1435-9871 http://www.demographic-research.org/Volumes/Vol3/2