Thank you for this comment. I added the following explanation to the article: “For example, suppose you have a variable “state”, and the first state in an alphabetical order is “AK” that has three observations. Many languages such as R use the first alphabetical level as the reference level, so the “AK” becomes the reference level. However, it has only three observations in your current dataset! It may be less than three observations if you repeat the same data drawing process. A better and common practice is to set the most frequent level, or a level that has sufficient number of observations.”