How is data collapse done practically？

In statistical mechanics, people often do data collapse to prove a scaling relation. How is this done? How do you judge the goodness of the collapse? By eye or by a number? I do not see often discussion about this issue.