I have heard that normality of data is considered as a basic assumption for principle component analysis. Is it necessary to normalize data of food intake assessment in order to identify dietary patterns?
In order to compare amounts of intake that make clinical sense it would be advisable to standardize your measures. Perhaps you could consider an ordinal approach such as below, meeting, and above the required or recommended amount.
for PCA, data from food frequency questionnaires are used. Frequency of consumption data are most of the time not normal and this is what you might expect.
Yes, as explained by David above but if you use a program such as SPSS or Minitab, the software does this for you (to calculate the correlation matrix). PCA is then applied on this correlation matrix. Usually, one can choose between PCA on a correlation matrix or a covariance matrix. A PCA on the correlation matrix usually provides the better results because each variable is given equal weight and results are not influenced by the unit or size of the numbers (large number are generally accompanied with large variances).
Some statistical tests, for example the analysis of variance or principal components analysis (PCA), assume that variances are equal across groups or samples. Equal variances across samples are called homoscedasticity or homogeneity of variances. The Bartlett’s test can be used to verify that assumption and to demonstrate the applicability of PCA. Then, as explained by Marcel, both PCA and subsequent calculation of correlation matrix may be used to identify dietary patterns.
- Badges
- Science topic
- Similar topics
- Chemistry, Organic
- Organic