Derivation of standardised principal components
Data
Let be an matrix of values, where indexes age group and indexes some combination of classifying variables, such as country crossed with time. The values are real numbers, including negative numbers, such as log-transformed rates, or logit-transformed probabilities.
Singular value decomposition
We perform a singular value decomposition on , and retain only the first components, to obtain
is an matrix whose columns are left singular vectors. is a diagonal matrix holding the singular values. is a matrix whose columns are right singular vectors.
Standardising
Let be a vector, the th element of which is the mean of the th singular vector, . Similarly, let be a vector, the th element of which is the standard deviation of the th singular vector, . Then define where is an L-vector of ones. Let be a standardized version of ,
We can now express as
Furthermore, we can express matrix as $$\begin{align} \pmb{B} & = \pmb{U} \pmb{D}\pmb{M}_V ^\top \\ & = \pmb{U} \pmb{D} \pmb{m}_V \pmb{1}^\top$ \\ & = \pmb{b} \pmb{1}^\top. \end{align}$$
Result
Consider a randomly selected row from . From the construction of , and the orthogonality of the columns of $\color{cyan}{\text{TODO-spell this out a bit more}}$, we obtain and This implies that if set where then will look like a randomly-chosen column from .
$\color{cyan}{\text{TODO - illustrate with examples}}$