Academic journal article Demographic Research

Biological and Sociological Interpretations of Age-Adjustment in Studies of Higher Order Birth Rates

Academic journal article Demographic Research

Biological and Sociological Interpretations of Age-Adjustment in Studies of Higher Order Birth Rates

Article excerpt

Abstract

Several studies of the effect of education on second or third birth rates (e.g. Hoem et al. (2001)) have used the concept of relative age at previous birth (B. Hoem (1996)). B. Hoem's idea was to focus on the social meaning of age at previous birth by redefining it according to the woman's educational attainment. We broaden the discussion by considering other interpretations of the explanatory power of the age at previous birth, particularly via known trends in biological fecundity. A mathematical analysis of the approach reveals side effects that have not been taken sufficiently into account. Our recommendation is not to use the relative age approach without supplementing it with the more traditional approach which includes the actual age at previous birth.

1. Introduction

In recent years, several studies of the effect of education on second or third birth rates (Hoem et al. (2001); Oláh (2003); Kreyenfeld (2002); Köppen (2006)) have used the concept of relative age at second (or first) birth which was originally introduced by Britta Hoem (1996).

The idea is that when comparing the third (second) birth rate between mothers with different education levels but the same age at second (first) birth, one does not take into account the different "social meanings" of the woman's age at her previous birth. Hoem (1996) suggested that in order to take this problem into account one could include the age relative to the educational level of the mother at her previous birth instead of her actual (biological) age at previous birth. She made the point that "it is important to account for the very different distributions of age at second birth that women have at different educational levels, otherwise incorrect conclusions may be drawn from the analysis".

With this note we wish to clarify from a more formal (i.e. mathematical) point of view what is going on in regression models where this suggestion is employed. But first of all, let us put forward some arguments why it is important to take age into account when modeling the effect of education on fertility3.

There are obvious biological reasons why age should be taken into account when modeling fertility, because age can be seen as a proxy for the woman's biological fecundity which indeed becomes smaller with age. However, as well as a biological component, fertility also has an important behavioural component. This perspective is emphasized in Hoem's approach, as we will argue later on.

Kreyenfeld (2002) made the point that some of the effect of education on fertility might be transmitted through a later age at previous birth because women who have a higher education become mothers later than other women. She also noted that "given that the positive effect of women's educational attainment was primarily transmitted through a late age at first birth, it should disappear if one holds the age at first birth constant". Köppen (2006) argued similarly that "education influences fertility also indirectly". This was in fact what Hoem (1996) was aiming at: that age at previous birth is an intermediary variable between education and the birth rate that we are modeling.

An illustration is shown in Figure 1. The broken arrow represents the so-called direct effect of education on fertility whereas the two solid arrows represent the indirect effect of education on fertility, i.e. the effect that is mediated through the woman's age at previous birth. The sum (in the linear case) of the direct and the indirect effect is referred to as the total effect4.

The tradition in the epidemiological literature is that an intermediate variable (such as age at first birth in our example) should not be taken into account (see eg. Chapters 8 and 21 of Rothman and Greenland (1998) and Chapter 12 of Diggle et al. (2001)) since this blocks the part of the effect of the variable of interest which is mediated through the intermediate variable. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.