Initially we assign to each of the 30 answers a triplet
which means: "Among all people who chose this answer, v % are visual, a % are auditive and
k % are kinesthetic".
For that we preserve only the samples for which we can determine the category without too much of risk to be mistaken, i.e. those for which:
We assign 1 to the selected answers and 0 to the nonselected ones (table top). For each answer, we calculate the averages by categories (µ v, µa, µk). We deduces (v, a, k) by applying a coefficient such as v + a + k = 100 (formula opposite).
Three tables of 30 elements are thus obtained: v , a , k .
We check in the passing that they are negatively correlated one from each other . For our sample we find: þvk = -0.60, þva = -0.60, þak = -0.48.
We can then calculate the coefficient of correlation þ between each one of these tables and the table of your answers r . It is the formula opposite with the huge square roots.
A question remain open:
Do the experimental values of þ vk , þva and þak obtained by this method constitute an evidence of the validity of these three categories, or would we have found results comparable with questions based on whimsical categories such as the astrological sign?
In own way of brief reply, I delivered myself to the same calculation starting from a set of 50 samples generated by chance. (I could employ to classify them only the first of the two criteria described above), and I found the values þvk= -0.46, þva= -0.51, þak= -0.48. The fact that these results are more distant from -1 than the results drawn from the real samples tends to confirm, I think, the validity of these categories.