Empirical findinigs for western countries Results
4 Data and method

In this study, standardized data from 17 European countries with Family and Fertility Surveys (FFS) are used to investigate whether parents prefer one sex over the other, or a mixed sex composition of their offspring [Note 1]. The FFS database allows a unique crossnational analysis, applying the same methodology to highly comparable data.

For all countries, our analysis is based on women who are 25-39 years old, currently live in a partnership, and have already two or more children [Note 2]. We have decided to focus on the transition from the second to the third child, as in the one-child family the main decision is probably whether to have a second child or not, with less room for an influence of sex. Furthermore, we assume that having more than two children is beyond the ‘standard’ of contemporary Western societies and progression to higher parities needs additional explanation, for example a couple’s gender preferences [Note 3].

There is no direct question in the FFS asking for the parents’ gender preferences. This need not be of harm for the analysis, however. The mere expression of a son preference, for example, is no guarantee that the respondent’s fertility behavior will actually change. Also, couples tend to state preferences in accord with the actual sex of their children already born [27]. Therefore, indirect measures of gender preferences may even be more advantageous than direct approaches.

In our analysis, we particularly investigate into ‘manifested’ gender preferences. We estimate an ordered probit model with the question of whether a couple either has, or desires a third child, being the dependent variable. If the respondent has two children and reports to have no desire for additional children, the dependent variable equals zero. It equals one if the respondent has two children and reports the desire to have more children. Finally, the dependent variable equals two if the respondent either has already more than two children, or has two children and reports a current pregnancy (see [Table 1] for descriptive statistics of the dependent variable).

In the absence of information on completed family size, the ordered dependent variable used here has the advantage of capturing both, the desire to progress to the third child as well as the actual progression to the third child. We are aware of the fact that the intentions for a third child and the actual progression to the third child are conceptually different and subject to different sources of error. First, the occurrence of a third birth may be unintended and therefore uninformative about gender preferences or other reasons affecting the desire to have a third child. Second, intentions for a third child may not have materialized in an actual birth by the time of the survey due to intended birth-spacing, delays in conception, etc. The ordered dependent variable used in our analysis therefore treats fertility intentions as a precursor of future births. Especially for relatively young couples with a recent second birth, we regard the intention to have another child as an important indicator for a third-child preference [24]. Although such an interpretation may be problematic in situations with a large fraction of unintended births, we feel that the advantages of using the information on fertility intentions outweighs its potential disadvantages. In particular as long as the intention to progress to the third child is positively associated with a higher probability of actually progressing, our ordered dependent variable should be a proper measure of the desire for children. In summary, since unintended births and contraceptive failures are probably uncorrelated with the sex of the first two children, our measure of fertility desires is appropriate even in the case of different contraceptive regimes and different levels of unintended births in the FFS countries used in our study.

We estimate two models, both with the same dependent variable, and the same set of standard explanatory variables. However, in Model 1 a binary sex-composition variable is used. It equals one, if the first two children are of the same sex, zero otherwise. A significantly positive coefficient of this variable indicates a preference for a sex-mix in a country. This model does not provide any information yet, however, whether there is any preference for a specific gender. This is investigated in Model 2, where we insert three dummies for the sex combinations of previous children in the equation. Regarding the sex composition of children already born, we differ between boy-boy, girl-girl and sex mix, where the latter is used as reference category. If none of the sex-combination dummies shows a significant effect on the parents’ propensity to have another child, we interpret this as an indicator for no gender preferences. If the first two children are of the same sex, either boys or girls, and both respective dummy variables have a significant positive effect on the dependent variable, we regard this as pointing to a preference for a sex-mix. This means, a couple continues childbearing, hoping for their offspring to be of the opposite sex as compared to the children they already have. If the two children born first are boys (girls, respectively) and this has a significant and positive effect on the dependent variable, we assume a preference for girls (boys, respectively). Eventually, it is tested, whether the coefficients for boy-boy and girl-girl are significantly different from each other. Only then, we will speak of a significant boy-, or girl-preference in a specific country.

We hypothesize that a couple is more likely to curtail its fertility, when the actual sex composition of their two first-born children reflects their gender preferences.

In addition to the variables for the sex combination of the first two children, we include among the independent variables the age of the woman (and its square), the woman’s age at first birth, the interval between first and second birth, the religiosity of the woman (where available), whether the woman grew up in an urban region (where available), and the educational level of the woman..

Empirical findinigs for western countries Results

logo70.gif (2450 bytes)

Gender Preferences for Children in Europe: Empirical Results from 17 FFS Countries
Karsten Hank and Hans-Peter Kohler
© 2000 Max-Planck-Gesellschaft ISSN 1435-9871