Can anyone suggest how to simulate data from a continuous, unimodal, positively skewed, two-tailed distribution with known mean and median and, preferably, a known variance? I am interested in comparing the power of some robust estimators and wish to generate data that resemble a known distribution with the following parameters: mean=5, median=2, SD=9, min=-7, max=80, mode(1)=5. I am aware of an approach used by Thomas Hettsmansperger to evaluate his test of symmetry and am attaching a description of the procedure. (I believe is not against posting rules to attach PDFs. Sorry if I am mistaken.) Unfortunately, this procedure yields data with a bimodal distribution, which does not reflect the distribution of the variable I am attempting to replicate in the simulation. Additionally, it does not constrain the variance of the data to take on a desired value. I would think that some form of contaminated normal (where observations from the contaminating distribution take on positive values only) would do the trick. Any suggestions would be greatly appreciated. -- Jim Dr. James W. Shaw 134 East 56th Street Westmont, IL 60559 Home: 630-324-6375 Mobile: 215-852-3045 -------------- next part -------------- A non-text attachment was scrubbed... Name: simulate data with known mean median difference.pdf Type: application/pdf Size: 42819 bytes Desc: not available URL: <https://stat.ethz.ch/pipermail/r-help/attachments/20110827/8d2a14cf/attachment.pdf>