Hi,
Is anyone aware of research that has used expert domain knowledge of potential outcomes (prior qualitative knowledge) and incorporated it into (logistic) regression coefficients? I am running a simulation study and would like to see whether I can create an outcome function incorporate a domain experts quantification of events into the coefficients.
Any research shared would be really appreciated.
Many thanks,
Janine
Please check this one:
Article A regression framework incorporating quantitative and negati...
Thanks! Matthew Kerry could you please elaborate as to how intercepts in multilevel models (assume MLM stands for that) would help creating coefficients incorporating domain expertise?
Noha A. Mostafa Thanks for sharing the article, really appreciate it! However, the authors restrict the overall regression value of the semiSVR given prior knowledge but not the coefficients only. As I create functions with unknown parameters and unknown outcomes, I'd like to use the prior qualitative knowledge incorporated into the coefficients. Thanks anyway!
Hi Janine,
Combining prior beliefs with new data to create an updated belief (posterior) sounds like something that could be done most easily within a Bayesian framework.
You are looking for a way to incorporate "qualitative knowledge", but to incorporate the knowledge in any way, you will need to quantify it somehow: in a Bayesian framework, that means making assumptions about the probability distribution.
I'm not sure how one could frame this within a logistic regression, but within a regular linear regression it is not very complicated (see https://en.wikipedia.org/wiki/Bayesian_linear_regression). To apply it to logistic regression, you could try to shoehorn Bayes into logistic regression somehow (probably someone already did, so you might get a free ride if you search the literature). Otherwise, you could just, as a first approximation, use Bayesian linear regression on a linear probability model.
Hope this made sense, and good luck!
Ian
Thanks Ian Michael Trotter ! You are absolutely right, and already saw where I am heading with this! I am actually looking for a way to quantify the qualitative knowledge from domain experts to create a space from which I can sample my logistic regression coefficients for my simulation. To do so, the mean vector of coefficients and the corresponding precision matrix have to be estimated which then can be incorporated within then Bayesian framework. So I am stuck at the very first step at the moment. But many thanks for your input!
Janine,
Have a look at
https://en.wikipedia.org/wiki/Grey_box_model
and references there.
Regards
- Badges
- Science topic
- Similar topics
- Methodology
- Research Methodology
- Research Papers