Statistically comparing distributions with small sample sizes (for dummies) Please assume, safely, that I know next to nothing about statistics. I do not have statistical computer programs and rely primarily on Excel, online calculators, and the good graces of smarter friends.
I am trying to compare distributions with generally small sample sizes, in the context of archaeological sites. For example, one analysis is looking at differences in the size of houses (i.e., floor area). Site A might have 10 houses, each with a known floor area. Site B might have 6 houses, each with a known floor area. If I assume that house size correlates with some form of inequality, what is the best method to compare such distributions? I initially used means and t-tests (via Excel). I also provided standard deviations (Excel) and Gini coefficients (online calculator) to give some indication of intra-site variability, acknowledging that variability could affect the sites' means.
Any dumbed-down advice would be sorely appreciated. Thank you!
 A: When data is continuous, a good method to compare the means is the t-test between two normal distributions.  A good method to compare the variances is the F-test for equality of variance between two normal distributions. These two methods are based on an assumption of normality which is hard to disprove for such small sample sizes. A priori, we might suspect that the distributions are scewed in a way that is not like the normal distribution. Transforming the data and doing the analysis on the transformed data might help. With house sizes I would presume that a log-transform is good if one or some houses are a lot larger than other. 
One could also take a more careful approach using, say ranks but the interpretation of a rejected hypothesis is more complex in the rank-framework.
When the sample size is this small, mean and variance is often the only meaningful information we can extract and hence compare meaningfully. 
To conclude, I believe the F-test and T-test are best here and you should consider transforming data beforehand.
