Judging Significance from Error Bars

This paper explores and evaluates different ways to judge statisitical significance from a graph of estimates with error bars. The traditional bar-overlaps-bar test for judging significance is found to have low power. Better power is obtained from the bar-overlaps-point test.