The total marital fertility rate and its extensions

What we will call the age-based TMFR is computed conventionally by adding up agespecific marital-fertility rates in the hope of estimating the number of children ever born to a woman who is married throughout her childbearing years. Demographers have long been strongly skeptical about this quantity because it normally indicates implausibly many children. Our analysis of data from the Romanian GGS confirms this finding, and we propose an alternative duration-based TMFR computed in the spirit of parity progression ratios. At the same time we extend the method to cover any type of living arrangement (cohabitation, marriage, nonpartnered arrangement, and so on). Because each resulting Total Union-type Fertility Rate (TUFR) explicitly accounts for the living arrangement, it improves on the conventional Total Fertility Rate, which does not. We embed the investigation in an event-history analysis with fixed and time-varying control covariates and find patterns of relative risks for such variables that reveal interesting features of childbearing behavior in the Romanian data, which we use to illustrate the method. In most cases these patterns are quite robust against model re-specification, including the shift from the age-based to the duration-based approach. Since the number of female respondents is 'only' about 6,000 (minus records that cannot be used for the current purpose) in a normal single-round GGS, there is considerable inherent random variation in the data set, but we show that simple few-term moving-average graduation suffices to overcome this problem. 1 The authors have contributed equally to this paper.