The H-method

The psychiatrist that wants to measure intelligence at young persons knows that there are many measurements (variables) needed for each persons. He/she might want to work with 150 variables or more in order to get reliable data. At the same time he/she knows that that there are relatively few mental factors that are deterministic for intelligence. If the data are collected in a matrix X, the number of variables (columns) are 150 or more and the number of samples, young people may be say 50 (the rows of X). Geometrically the column are located in a low dimensional space, which means that he/she will only select say, 5 or 7 factors (score variables) to represent the results of analysis. A similar situation is observed, when we are working with many variables (where many means more than 5 to 10 variables). It is advantageous to identify the subspace of the column space of X, where the relevant variation of X is located. By building the solution up by parts, the algorithms of the H-method both identify the part of the column space of X that should be used and also secures the prediction aspect of the model.