We are passionate about developing statistical models and tools to assess demographic and population-level health trends and differentials in countries around the world. Ultimately, we hope that the insights from our research help to improve reproductive, maternal, and child health worldwide.

Our research lies at the intersection of statistics and demography. We have developed statistical methods for the assessment of key indicators pertaining to the primary areas of

• Family planning, abortion, and fertility;
• Stillbirths, maternal mortality, and mortality among newborns, children, and youth.

Examples include the global assessment of unmet need for contraceptives, abortion incidence, the maternal mortality ratio, and the under-five mortality rate.

Do you wonder why we need statistical models to monitor demographic and population health indicators?
Models are needed when data alone are not sufficient, for example when ...

• Data are missing for years of interest, i.e. for recent years or years in between survey rounds. This is a common issue when assessing trends in demographic outcomes in low and middle income countries without well-functioning registration systems.
• Data are subject to data quality issues, i.e. bias or random measurement error, which may be substantial. Examples of data quality challenges in my research area include bias in the self-reporting of stigmatized outcomes such as abortions, recall bias associated with birth or sibling survival histories, and misclassification of maternal cause of death in vital registration systems.
• Data are available that are related to the outcome of interest, as opposed to measurements of the outcome of interest itself. For example, when interest is in the elevation in the sex ratio at birth due to son preference or the percentage of pregnancies that are unplanned, available data typically does not measure the outcome directly.

What are some modeling contributions?
Data limitations impose challenges for the development and evaluation of statistical models, and for communication of findings and limitations. Our research has addressed these challenges for various indicators via the conceptualization, development, and validation of context-specific Bayesian models, and the communication of the value and limitations of such models. In model development for a given outcome of interest, whether it is child mortality rates or abortion incidence, we find it helpful to distinguish between challenges related to capturing

• the dynamics of the underlying outcome of interest, i.e., how it varies across time and populations, referred to as a process model, versus
• the relationships between the data and the outcome of interest, referred to here as data models.

Motivated by real-world problems, we have spearheaded advances in both areas. For example, we have developed process models with

• Demographic accounting equations to relate outcomes to one another and construct estimates that are internally consistent and informed by related outcomes. Examples include the estimation of abortion and unplanned pregnancies (Bearak et al. (2020a)) and subnational population reconstruction and projections ( Alexander and Alkema (2020));
• Mathematical descriptions to capture transition processes. Our work in this area includes the modeling of contraceptive use (Alkema et al. (2013), Cahill et al. (2018)), inflation of the sex ratio at birth (Chao et al. (2021b)), and the total fertility rate (Alkema et al. (2011)).
• Hierarchical and temporal model components to allow for exchange of information across populations and time. Hierarchical (or multilevel) models are used throughout our work in settings where parameter estimates are needed for data-poor(er) units. Examples of temporal smoothing are the Bayesian B-spline bias-reduction (B3) method for estimating the under-five mortality rate using spline-based temporal smoothing, and the addition of ARIMA processes to hierarchical (sparse) regression models, for estimating maternal mortality and stillbirth rates.

For data model development, our work includes:

• The development of data models for using birth history data that account for recall bias, non-sampling variance, and survey effects (Alkema and New (2014) );
• Data models for compositional data as used for family planning indicators (Alkema et al. (2013), Cahill et al. (2018)) and maternal cause of death fractions (Say et al. (2014));
• Estimation of misclassification of maternal deaths in CRVS data (Peterson et al. (2019)) ;
• Definition adjustments for stillbirth rates that are not reported in the reference definition of 28 weeks of gestational age (but instead, based on an alternative gestational age or birth weight), Wang et al. (2020).

A summary of contributions by substantive area is given in Highlights.

We aim to make available improved estimation methods and resulting estimates to diverse international audiences. To that end, we collaborate with various United Nations agencies including UNICEF, the World Health Organization, and the United Nations Population Division, as well as the Guttmacher Institute and the the global Family Planning 2020 (FP2020) initiative. Statistical methods and tools that we have developed have been used for international reporting for key indicators.

We gratefully acknowledge ongoing grant support from the Bill and Melinda Gates Foundation.