I like this example and might steal it. But I'd offer you a [bayesian] counter-argument: Say an HIV test is 99% accurate and responds positive on a randomly se

I like this example and might steal it. But I'd offer you a [bayesian] counter-argument: Say an HIV test is 99% accurate and responds positive on a randomly selected person. What's the chance they actually have HIV? About 5%! [0]. However, if you have prior reasons to believe you may have HIV, the chances are a lot higher than 5% given a positive test. To tie that back to lawsuits, your participants in the study have their priors set so that they don't expect to engage in frivolous lawsuits towards anyone. However they will likely recognize that they could fall victim to a frivolous lawsuit, and if they're normal, then the only type of lawsuits they could imagine being forced on them is a frivolous suit. Another way to say that is: probably 90% of the ridiculous lawsuits originate from 1% of the population. Then 99% of your participants will mentally model themselves as a sincere plaintiff (thus they shouldn't have to pay even if they lose), while a 1% frivolous rate is certainly high enough to make plausible the threat of being the defendant, and thus worthy of protection through legal fee reimbursement. [0]: http://cs.wellesley.edu/~cs199/lectures/34-bayes-rule.html http://cs.wellesley.edu/~cs199/lectures/34-bayes-rule.html