We introduce DOSE - Dynamically Optimized Sequential Experimentation - and use it to estimate individual-level loss aversion in a representative sample of the U.S. population (N = 2,000). DOSE elicitations are more accurate, more stable across time, and faster to administer than standard...