We theoretically show that agents with loss-averse preferences are more likely to lie to avoid receiving a financially bad outcome the lower the probability of the bad outcome. Intuitively, the lower the probability of receiving a bad outcome, ceteris paribus, the further the expected payoff from the bad outcome, and thus the greater the utility gain when lying to avoid the bad outcome that occurs further in the loss domain. To test this hypothesis, we first develop an econometric method to formally estimate the exact distribution of dishonesty in contexts in which agents privately observe the outcome of a random process but can report a different outcome (i.e., be dishonest) without detection. In this well studied context (e.g., using coin tosses and die rolls in experiments), inference has been essentially limited to whether reported outcomes differ from the distribution if agents report truthfully. In contrast, our method measures the full distribution of dishonesty which provides the proportion of lying, confidence intervals, and many other statistical inferences. Using this method, we find strong support for the loss aversion prediction by first comparing the estimated lying percentages across the existing literature including more than 50 papers and 200 treatments. However, given numerous variations across study designs, we next show within two new experiments in vastly different contexts that the loss aversion pattern holds even more dramatically once removing across-study variations.  Overall, our new studies, analyses of the extant literature and new method to estimate lying provide compelling evidence that lying to avoid bad outcomes robustly increases the lower the probability of the bad outcome. Our estimation approach to measure lying provides researchers with a simple yet powerful econometric tool to measure lying, and our findings suggest contexts where dishonesty is more likely to occur to which policy can address.