We construct a joint distribution of fertility and children's activities treating the Poisson process generating the number of children as being endogenous in the multinomial logit process generating children's activities using a latent factor structure. Latent factors are incorporated into the equations for number of children and children's activities to allow for unobserved influences on fertility to affect children's activities, thus enabling us to make a distinction between selection on unobservables and selection on observables. We apply maximum simulated likelihood (MSL) techniques to estimate the parameters of our models. We find that the effect of fertility has a large downward bias in naive models. The marginal effect of fertility on child labor and schooling are twice as large when common unobserved heterogeneity in fertility and activity decisions are accounted for.