We would like your opinion on whether GAMMs are a good option and how best to go about implementing for the following:

During a period of patient care, a nurse will accrue $Y$ colonies of bacteria on their hands. During every hand contact ($n$) (with surface area $A$) with a surface this number may increase a percentage ($\lambda$ %) of what she touches ($V$). However potentially this may also decrease as some percentage ($\beta$ %) of the bacteria are removed. After they finish the care episode, they may wash their hands with efficacy $h$.

The model is hence dependent on $Y=Y(n,A,V,\lambda,\beta,h)$

We think:

$n$=empirical, different for every nurse, $h=lognormal~(1.5,0.1)$, $\lambda=\Gamma(15,3)$, $\beta=$empirical non-negative. $A=lognormal~(7,1.9)$ and $V$=empirical, $h=\Gamma(5.91,0.4)$.

We'd like to know if GAM fitting is an appropriate way of estimating $Y$. Please let us know if you require further clarification.


  • 1
    $\begingroup$ I'm a little confused -- you you have data on all of your explanatory variables? It seems like not, if you're specifying a distribution for them. Or are you doing something Bayesian, and sampling from a multivariate prior? If so, do your unobserved variables correlate with each other, or with your observed variables? $\endgroup$ Commented Apr 16, 2013 at 18:26
  • $\begingroup$ @ACD The only known input variables are $n$ (the number of surfaces each nurse touches) and $V$ (the contamination level of the surfaces). During each surface contact $\lambda$,$A$, and $\beta$ vary according to the given distributions. $\endgroup$
    – HCAI
    Commented Apr 16, 2013 at 20:29

1 Answer 1


I think the model is incomplete. Why not consider her hands as a forest of SIR models? There are some things that "die on the vine" and for that "R" applies.

Here are links on SIR:

It seems that lognormal distributions may apply to some limited set of special cases of disease, but they are not general enough for unilateral applicability.

I have my head in random forests and a forest of SIR models (see IJAM link) is going to account for variation, give you robust answers, and have a basis that is generally applicable to the problem.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.