Research Journal: Oct 30 - Nov 5
Pivoting a project out of a dead end
October 30: Today was a good day on the writing front. I finished up the Methodology section, which was a big dose of effort. I’m on Results now ,which is much easier and will probably be the subject of much debate between Sonia and I. I still think there’s some meat to be extracted in terms of operating characteristics, but there’s so much data to wade through that I need to figure out how this is done. That Pharmaceutical Statistics journal will probably be the point of inspiration.
October 31: Today I got as far as I could with the writing of the current results for the Platform paper. The operating characteristics part of the paper needs some more thinking through. The benefits need to be on an individual level, so that’s what’s different for this design.
I also did some literature review on incorporating time information into a Bayesian network. I did this blindly during first year, but now I need the proper knowledge for what assumptions are needed for the structure to still be valid.
N-of-1 data has elements of both subject heterogeneity and changes in time, so I checked to see if there was a model for this. There was a review by an economist on the random autoregressive model, which is like the love child of mixed-models and AR(p) models. That’s what’s being used in the model so far, and it’s actually appeared in the Platform paper too. Possibly a good research topic for saving as a note.
November 1: Today I focused entirely on the Bayesian network project to try to incorporate the autoregressive terms, based on yesterday’s research. This ended up being a lot quicker than I anticipated, and I got a result that was at least a little interesting.
I realized today too that I was a little too invested in a result that I thought I should be getting. The result has conclusions we can make too, it’s just not what I originally wanted. I think I can make a presentation off of it, which takes a small load off my shoulders.
November 2: Today I talked with some people about potentially getting some data for my research, but the data was not what I envisioned to being helpful. Even making the slides seems a little difficult with no results to really glean from.
November 3: Today I ran into more deadends with the simulation approaches that I tried to devise. The benefit is in finding potential mediators and confounders, but this doesn’t really make sense if the randomization is done effectively. There needs to be a new angle that allows the benefits of the Bayesian Networks to be emphasized.
November 4-5: Today was a set of rest days from the busy thinking from this week. I think I hit a nail on the head by placing the methodology in a different context. The Bayesian Networks are too restrictive in the interventional setting, but have use in an observational, exploratory setting. I need to revise the introduction to account for this.