Demography and sociology have produced a sophisticated body of work on fertility outcomes, i.e., whether, why, and when people have children. Such research has uncovered many characteristics underlying fertility. Unfortunately, traditional ways of statistical analyses means that results are difficult to compare across studies and that we have little understanding on the (relative) importance of these characteristics. In light of the credibility crisis that has taken hold of many other fields and disciplines, one might similarly worry about the robustness of findings in demography. In this talk I will argue that a focus on predictive ability and on microsimulation will advance fertility research and make it more reliable. In the first part of the talk, I will make a plea for reporting out-of-sample predictive ability. I argue that this form of predictive ability i) is the most relevant effect size we should be interested in, because it is a measure of how well our theories do in practice, ii) prevents overfitting and is less susceptible to researcher degrees of freedom, iii) facilitates comparison between theory and data-driven approaches, and iv) allows for prediction benchmarks which have facilitated cumulative knowledge in other disciplines. In the second part of the talk, I will discuss how microsimulation models of fertility can advance existing research. A simulation model that includes findings from reproductive medicine on the ability to have children (e.g., the age at sterility, fecundability) is developed to quantify to what extent higher levels of childlessness among highly educated Dutch women are explained by preferences and partnership trajectories. I conclude by arguing that a shift towards prediction in combination with theory-driven microsimulation models increases the robustness of findings and allows a unique perspective on the state of fertility research.