search

parksw3 / epidist-paper

Other
10 stars 4 forks source link

So .... does it matter? #47

Open seabbs opened 1 year ago

seabbs commented 1 year ago

An angle we need to be sure to cover is the following comment:

So does this matter for decision making made using evidence based on these estimates?

This is something I have heard a few times and is a sensible enough point (though I rather think the burden of evidence is the reverse and biased methods should need to show that they don't negatively impact decisions).

An example, would be our Ebola case study. If estimates were slightly biased what does that mean for modelling that was done using them? What would the scale of the bias be for this to impact decision making?

Another example where this comes into to play is the growth rate based adjustment. How wrong would this need to be to matter (its obviously fairly complex and not sure we should actually cover this but worth keeping in mind).

I don't think we can or should fully cover this but I do think we need to reflect on it and highlight it as an area for further work. We may even want to make my argument above that the burden of evidence here is really on biased methods to show they don't cause an issue vs equally usable and less biased approaches.

parksw3 commented 1 year ago

Another example where this comes into to play is the growth rate based adjustment. How wrong would this need to be to matter (its obviously fairly complex and not sure we should actually cover this but worth keeping in mind).

I think it's worth keeping in mind that this is equivalent to dealing with truncation.

And I think this all goes to earlier comments from several people that we need to be clear about why we're doing this.

seabbs commented 1 year ago

I think it's worth keeping in mind that this is equivalent to dealing with truncation.

My point was that people may just say oh well its roughly okay so why does it matter if you have extended it to time-varying r or tell me to use a different formulation based on observation time.

And I think this all goes to earlier comments from several people that we need to be clear about why we're doing this.

Yes agree.

parksw3 commented 1 year ago

My point was that people may just say oh well its roughly okay so why does it matter if you have extended it to time-varying r or tell me to use a different formulation based on observation time.

Ah, good point.