Data that does not follow a normal distribution

Normality in dependent variables helps you say that the observed variation could be "random noise". If you have a variable that is dependent on other variables, you would not expect it to be normal because it is not random. For example, if you were to throw a ball a hundred times and measure the distance thrown, you might expect normality. If you, a five-year-old and an 80-year-old each threw a ball a hundred times and combine the data, you would not expect the distance data to be normal, since distance is a function of age, and you have different ages in your independent variable set. Do you model first, then look for normality on the residuals.