Cass Sunstein at Bloomberg writes on the estimates which are required for new regulations. A study shows there's no systemic error (bureaucrats underestimating costs or overestimating benefits), although the estimates probably aren't very accurate.
What would be more interesting to know is how often the analysis results in changes to the regulations or dropping the effort altogether. I'm still waiting for a thorough redo of the regs on paperwork and regulations to make them fit the 21st century. Not holding my breath though.