I have responses to a questionnaire and I would like to divide the anwers into three groups to use them in comparisons for other variables. To stablish the limit values between clases I have found some published papers that calcule 1) tertiles and use these to stablish limit points, 2) others calculate mean +/- 2standard deviations and still others use 3) the K-means procedure. The distribution of my data is left skewed and have some very low values that maybe outlyers. What do you think would be the best option.