Factors explain relationships.
Features can be individually detected, such traits, preferences for individual books, buying habits, etc. Sometimes it makes sense but to reduce data and not to bring everything as detailed as possible, but instead to be able to form groups of features, explain the specific behavior. The purpose of the factor analysis - a statistical method, which can be calculated with many programs (eg SPSS).
The role of factor loadings in the factor analysis
Many individual features can be assigned to parent categories. If you enter, for example, read what books like certain persons, then you can use the book titles after the genre assign (eg textbooks, poetry, children's books, novels, etc.).
- These groupings are possible with very different characteristics. Just the question of how can we find out these major categories, dealt the factor analysis. This is seen, which correlate each variable is particularly high with one another, ie have high correlations.
- That may be about to mean that those who particularly enjoyed reading book A and book B, also book C preferred, but the books D, E and F have read rarely. Books A, B and C are then in the factor analysis probably (ie a dimension or feature group) associated with a factor and the books D, E and F of another dimension.
- How do you interpret this, however, remains up to you. In the example, it could be such as to obtain entertainment literature and non-fiction, thus two categories which can also be differentiated sense.
- Whether the result of the factor analysis but is also mathematically meaningful and can be all individual features with the calculated factors explain well, show the factor loadings. These correspond to the context of a feature by the factor and because it is said, as uploads the feature on the factor, the term factor loading has emerged.
How do you interpret a factor loading
- If the factor analysis, certain factors - ie groups of features - propose, then you can calculate how strongly associated the individual features, which are to be explained by the factors with these. Desirable of course is the highest possible connection (a high correlation) for each feature, so it can be really clearly attributed to one factor.
- The factor loadings can assume values between 1 and 0 and are usually somewhere in between. The higher the charge, the greater is the context where a value of 1 corresponds to a perfect connection. But also negative factor loadings are possible. This means that a high expression of the characteristic dimension (eg the preference for non-fiction) then that goes with it, that a specific individual feature is very pronounced (eg the novel Z is reluctant to read, if you have a high preference for non-fiction).
- Ideal it is when your characteristics that you have raised, particularly highly correlated with a factor and are the relationships with other factors close to zero, because the result is really unique. Otherwise, it may be useful to extract more factors (to be calculated) to use or less characteristic groups.