This paper develops a scaling procedure for estimating the latent/unobserva
ble dimensions underlying a set of manifest/observable variables. The scali
ng procedure performs, in effect, a singular value decomposition of a recta
ngular matrix of real elements with missing entries. In contrast to existin
g techniques such as factor analysis which work with a correlation or covar
iance matrix computed from the data matrix, the scaling procedure shown her
e analyzes the data matrix directly.
The scaling procedure is a general-purpose tool that can be used not only t
o estimate latent/unobservable dimensions but also to estimate an Eckart-Yo
ung lower-rank approximation matrix of a matrix with missing entries. Monte
Carlo tests show that the procedure reliably estimates the latent dimensio
ns and reproduces the missing elements of a matrix even at high levels of e
rror and missing data.
A number of applications to political data are shown and discussed.