Steven Samuels wrote:
>>I'd like to amend one point in my previous post-that too many zero
values are cause for suspicion. I spoke to a friend who is familiar
with diagnostic psychological scales. She pointed out that the fewer
the items in a scale, the more likely zeros would be. In her
experience, mean scores were often small and lumps at zero were
common. I've found the same an eight binary item brief screen for
mental illness--most people had no "positive" items.<<
This is indeed a common problem but one for which I think the zero
inflation models are quite sensible contenders. The ZIP model is a
latent class model with latent class 1 being a degenerate Poisson and
latent class 2 being a non-degenerate Poisson, i.e., people with
genuinely no symptoms and people who reported no symptoms but . While
it's not a model available in common programs, a Zero Inflated Binomial
would be similar.
The problem of a diagnostic scale showing no symptoms for many
respondents and then some for others makes sense as a mixture problem
(potentially---it depends on the symptoms and how you think about them).
As it is common and sensible to use covariates to predict latent class
membership, I think for a lot of these kinds of diagnostic scale
problems, having covariates predict zero response would be very useful.
These covariates are predicting, in essence, the presence of symptoms at
all while the covariates for the scale (possibly distinct, possibly
overlapping) predict which latent class a subject was in.
JV
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
