Current methods for estimating the disposition effect implicitly assume that all stocks are evaluated simultaneously in a single decision stage. Here we propose a two-stage model where investors first decide whether to sell a stock in the domain of gains or losses, and only then choose a stock to sell from within their chosen domain. As evidence, we show that the probability of individual gains being sold is inversely proportional to the number of gains in the portfolio, but is not associated with the number of losses. Similarly, the probability of individual losses being sold is inversely proportional to the number of losses in the portfolio, but is not associated with the number of gains. There are two consequences for the disposition effect: First, sell decisions are about the domain of gains versus losses, not just about individual stocks. Second, current regression methods must be refined to avoid substantial bias.