Gal, Rothschild, and Wagner (1990) interviewed students of ages 8, 11, and 14
about their understanding of how means were computed and what they were useful
for. They also gave the students nine pairs of distributions in graphic form and asked
them to decide whether the groups were different or not. Only half of the 11- and
14-year-olds who knew how to compute the mean of a single group (and, also, to
some extent, how to interpret it) went on to use means to compare two groups.
Hancock, Kaput, and Goldsmith (1992) and, more recently, Watson and Moritz
(1999) have reported similar findings.