Unobserved Variables: Models and Misunderstandings

by David J. Bartholomew

​The classical statistical problem typically involves a probability distribution which depends on a number of unknown parameters. The form of the distribution may be known, partially or completely, and inferences have to be made on the basis of a sample of observations drawn from the distribution; often, but not necessarily, a random sample. This brief deals with problems where some of the sample members are either unobserved or hypothetical, the latter category being introduced as a means of better explaining the data. Sometimes we are interested in these kinds of variable themselves and sometimes in the parameters of the distribution. Many problems that can be cast into this form are treated. These missing data, mixtures, latent variables, time series and social measurement problems. Although all can be accommodated within a Bayesian framework, most are best treated from first principles.

94 pages, Kindle Edition

First published September 7, 2013

Reviews

[Usfromdk]

Not quite sure how to rate this - I'm probably at either 2 or 3 stars, but I'm not really sure which. I liked some parts and I thought less of other parts of the coverage (a few parts I won't mind admitting that I did not fully understand). Some observations and conceptualizations presented along the way were quite useful and interesting, but in other contexts I found it hard to follow why/how conceptualizing things the way the author suggested would really add much.

One thing to note here is that unless you have a quant/stats background of some kind, most of this book will be completely unreadable to you.