Whether to log transform variable when untransformed variable has positive skew and transformed has negative skew with additional missing data? I have performed a log transformation on my skewed data, however on my DV it went from positive skew to negative skew after the (log) transformation, further data was missing from my DV after the transformation. Please help
 A: Additional missing data after log transformation
If you have additional missing data after log transformation, it is likely that you have data that is less than or equal to zero. (i.e., log(0), log(-1), etc. is not defined). So if you want to use a log transformation on data with negative numbers, you need to add a constant to the raw variable so that the minimum of the resulting variable is greater than zero. So your transformation could be
$$\log(x + c)$$
where $x$ is your untransformed variable and $c = 1 - \textrm{min}(x)$.
Transformation flips the skewness
There is plenty of discussion on this site about when and whether transformations are useful. You might also like this discussion of issues surrounding transformations. In general, if a log transformation is flipping the direction of your skewness, then there is a good chance that you did not have very much skewness to begin with. To test whether the transformation makes a substantive difference with the context of multiple regression, examine your correlations, R-squares, and standardised betas before and after transformation, and see what changes you observed. In many cases you will see that it makes little difference.
Another point, is that the assumption pertains to the residuals of a multiple regression and not the dependent variable itself.
If you really care about optimising the transformation to make the variable approximate a normal distribution, then you can use the Box-Cox transformation. Or a simpler approach is just to try a range of transformations. A common set of transformations from greater to less change is:
-1/x^2
-1/x
log(x)
sqrt(x)

So if log(x) is transforming too much, you could try sqrt(x).
