Model Inputs \(\rightarrow\) Family Planning Parameters \(\rightarrow\) Desired Number of Children


We assume that each woman in the model has certain fertility preferences, operationalized by her stated ideal family size, or desired number of children. This variable determines what ‘mentality’ of family planning a woman is currently operating under: ‘spacing’ or ‘limiting’. While women have less than their desired number of children they are considered to be ‘spacing’. Once they have met (or exceeded) this number they are considered to be ‘limiting’.


We assume that a woman’s desired number of children remains constant over her reproductive life-cycle. We did not find any studies that looked at changes in desired number of children over the course of a woman’s life. However, we did find four related studies that examined consistency in reporting fertility preferences over time, ranging from five months to four years. Two longitudinal studies from India and Morocco specifically assessed consistency of women’s reported desired number of children,[1,2] and two studies from rural Ghana assessed women’s desire to stop childbearing over time.[3,4] A review of this literature did not reveal any consistent trends in how desired fertility may change throughout a woman’s life.

We analyzed data on the reported ideal number of children from the DHS (v613). We removed missing or “Don’t know” responses. Data were available for 4,102,452 women from 254 surveys in 74 countries. We assumed that non-numeric responses indicated fatalistic attitudes towards family size (i.e. no preference). These responses (n=212,933) occurred on average for 5.2% of women, ranging from near 0% in Lesotho and Vietnam to around 20% in Afghanistan and Chad and nearly 30% in Yemen. We re-coded these responses to 10 desired children, and top-coded the number of specified desired children at 20 (500 responses were top-coded). Here we plot the global distribution of desired number of children (by subgroup) and explore potential secular trends in the mean and variance of this distribution.


We assume that the distribution of number of desired children follows a Poisson distribution (\(\text{Pois}(\lambda)\)), with an additional point mass (\(\pi\)) at \(p\). This allows us to account for high concentrations of particular answers, for example, a large spike at 2 desired children:

\[ P(K=k)=(1-\pi)\frac{\lambda^k e^{-\lambda}}{k!} + \mathbb{I}(k=p)\pi \]

We fit hierarchical models to estimate priors for these parameters, stratified by subgroup (urban/rural and level of education). We used upper middle income priors for high income countries due to lack of DHS data in high income countries.