Statistical tests for comparing a skewed clinical sample I recently surveyed 350 low-income families -- they were randomly split into two groups: control and treatment. One of the variables I am very interested in is the amount of savings of each family. The sample is highly skewed (see below), as the vast majority of the sample has no savings ($0), and then there are few with modest savings. 
I'd like some input on what kind of stat test I can use to compare the two groups between each other pre-treatment to show that the savings rate and amounts are the same, and post-treatment to compare the change in savings between the groups after treatment has been applied.

Thanks in advance for any input you guys may offer.
 A: 
I'd like some input on what kind of stat test I can use to compare the two groups between each other pre-treatment to show that the savings rate and amounts are the same

You can't show they're the same, only that you can't identify a difference (which is not at all the same thing, since it might just indicate a lack of power). You might (possibly) be after something more like equivalence testing.
Assuming you just want the usual kind of hypothesis test, on the comparison of the proportion of zeroes, you'd be looking at a straight two-sample test of proportions (or perhaps a 2x2 chi-square, the outcome should be the same). To compare the amounts conditional on them being positive, you'd either looks at a parametric test suitable for a skewed distribution (e.g. a GLM for a Gamma or inverse Gaussian), or perhaps a nonparametric test, like a permutation test of the means, or a two-sample Mann-Whitney-Wilcoxon.  
The change in savings could either be done (similarly to the before comparison) as a two-stage thing (do they differ on the change in proportion of zeros? does the average amount of non-zeros change?) or you might looks at some kind of zero-inflated model.
(You might even consider comparing change-in-mean across the whole, say with a permutation test or something.)
Your big problem is the very small sample sizes (for those that are non-zero) involved.
