The within-person encouragement design introduced here combines methodological approaches from three research traditions: (a) the analysis of within-person couplings using multilevel models, (b) the experimental manipulation of a treatment variable at the within-person level, and (c) the use of random encouragements as instrumental variables to induce exogenous experimental variation when strict treatment adherence is unrealistic. The proposed combination of these approaches opens up new possibilities to study treatment effects of a broad range of behavioral variables in realistic everyday contexts. We introduce this new research design together with a corresponding data analysis framework: instrumental variable estimation with two-level structural equation models. Using simulations, we show that the approach is applicable with feasible design dimensions regarding numbers of measurement occasions and participants and realistic assumptions about adherence to the encouragement conditions. Possible applications and extensions, as well as potential problems and limitations are discussed.