Garrido propensity article

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|

Garrido propensity article

Rich Ulrich

Melissa cites and Bruce gives an online reference for a pretty good article on

propensity scoring. I want to add that it does seem to be a bit light on the

theoretical side. Consider,

   "Exclude any covariates that predict treatment status perfectly,

   as distributions of covariates need to overlap between treatment

   and comparison groups (see Step 2)."

The need for overlap is perfectly true. However, if you have a fine candidate Var

for confounding which confounds perfectly, you have a FLAWED STUDY, and

this flaw deserves a prominent footnote. Yes, you have to drop the Var from the

propensity scoring; but you have to give recognition to a major outside

influence which must be considered in the discussion section of the study. 


Elsewhere, there's a related problem treated concretely. The authors suggest that

when age as a continuous variable results in too much non-overlap, a diagnostic

warning might be turned off by dichotomizing age. As solutions go, that's like un-

screwing a warning light when you can't fix what is wrong. "Throwing away information"

needs further consideration/justification, and discussion.


- - - - - - - -

Bruce posted --


That article appears to be freely available here:
   https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4213057/
I assume you are referring to Figure 1. 

Ives, Melissa L wrote
> Is there a way to get a graph in SPSS that is similar to the graph that
> results from Stata's 'psgraph' command?
>
>
> (example in: Garrido, M. M., Kelley, A. S., Paris, J., Roza, K., Meier, D.
> E., Morrison, R. S., et al. (2014). Methods for constructing and assessing
> propensity scores. Health Services Research, 49(5), 1701-1720.

...

--

Rich Ulrich



===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD