To fit models like prospect theory or expected utility theory to choice data, stochastic model is needed to turn differences in values into choice probabilities. In these models, the parameter measuring risk aversion is strongly correlated with the parameter measuring the sensitivity to differences in value. We use dimensional analysis from the physical sciences to show that this is because the sensitivity parameter has units which depend on the risk aversion parameter. This means that comparing sensitivities across individuals with different level of risk aversion is meaningless and forbidden. We suggest a simple bug fix for prospect theory and other decision models which corrects this problem. The bug fix completely removes the correlation between sensitivity and risk aversion parameters in model estimations and allows the parameters to be interpreted as they were originally intended.