Descriptive statistics

If your outcome measure is not normally distributed, what can you do about it? If you simply ignore it, then your results are likely to be invalid. You could try transforming the variable to try and make the distribution more normal-looking. For example, if your variable has a long right-hand tail, often a natural log transformation will make it closer to a normal distribution. Many statistical packages have procedures that help you decide if a transformation is useful. In Stata, it is the ladder and gladder procedures. Ladder provides numerical suggestions and gladder, graphical.

ladder phy_b

The transformation with the smallest chi-square is the identity, i.e., no transformation. Let’s have a look at this in graphical form: