Transformations

Data transformation is the process of changing the data in some way. More formally, a transformation involves creating a new variable or set of variables from an existing variable or set of variables.

Objectives of transformation

Data transformation is undertaken with the following objectives:

• Making it easier to see patterns in the data (e.g., the Log transformations and Principal Components Analysis).
• Making it easier to communicate patterns in the data (e.g., the Net Promoter Score).
• To address violations of the assumptions of statistical tests (e.g., Ranks, Log transformations).
• To improve the validity of regression models (e.g., Basis Functions).
• To reduce the amount of data (e.g., Principal Components Analysis).

Standard transformations of a categorical variable

A categorical variable can be transformed in one of two ways:

1. It can be turned into a numeric variable, by coming up with some rules about the numeric interpretation of categories. For example:
   • Replacing the category 18 to 24 with 21 and 25 to 29 with 27 (this is a type of Recoding known as Midpoint Recoding.
   • Computing the Net Promoter Score.
2. The categories of a categorical variable can be combined. Most commonly, small categories are merged into larger categories. For example:
   • When a question asks for reasons for a particular behavior, any reasons that are selected by a small number of respondents can be classified as Other.
   • Variables that collect data on Rating Scales may be converted to Binary Variables to make further analysis simpler.

Standard transformations of numeric variables

Univeriate

• Ranks
• Log transformations
• Trimming
• Windsorizing

Multivariate

• Principal Components Analysis
• Cluster Analysis
• t-SNE
• Basis functions, such as:
  • Dummy Variables
  • Polynomials
  • Orthogonal polynomials

See also

• Coding
• Recoding