Statistics for Testing for a common latent variable in a linear regression: Or how to "fix" a bad variable by adding multiple proxies for it