I am setting about writing content for my personal website for a few reasons. First, this is an exercise of tools in which I may need to become more fluent. For example, I have not often used GitHub in undergrad, and so now I am using it a fair amount for setting up my site. I am also a little bit rusty in R, so now I am refreshing my memory of the language.
Second, I want to have a long term location for consistent writing. I am happy to write blog posts on the occasion that I find a topic which fits the purview of the blog, but I also found keeping a research-and-work journal helpful when working on projects so that I can remember the big ideas that I would otherwise forget while studying very small ideas. With the commencement of my graduate studies - and more projects on which to work than ever - this journal shall be my blessing and curse.
As should be clear enough, this “Journal” serves a different purpose than does my blog. This is not the place for me to document each original idea I have. Instead, I will record checkpoints in the projects on which I work. This includes the start or finish of a project and possibly publications and presentations. I will also record my motivations and the changes or developments to them.
For my GitHub Pages site, I have been writing all of my pages in RMarkdown with RStudio. This page I found online (by Garrett Grolemund, I think) has been a useful resource. I have also benefitted from some videos by Matthew Crump on his YouTube channel, “Programming For Psychologists.”
I have realized over this past week that I have begun a Git workflow for the record of this website project that is somewhat different from that being employed for my statistical computing course. This means, for example, that I have a GitHub Desktop app that I am not using as often as I thought I would be when I opened my GitHub account over the summer.
Right now, I am deciding to attend meetings for three reading groups this semester. These are in the subjects of causal inference, statistics pedagogy, and statistics for the physical sciences. I chose these groups in part because I think that they are largely independent in content, but also I have some interest in these topics, of course. I am not well read in any of the major domains of science in which statisticians tend to play - cosmology, environment, partical physics, neuroscience - but the themes of “high-dimensional statistics” and “uncertainty quantification” are common and have a rather concrete interpretation in the real world. For this reason, I think that I may find plenty of work to do that demands the exercise of engaging with unfamiliar domain knowledge, complicated methods, and careful reporting that is the triumvirate of skills that I need to build right now.
As for causal inference, this seems to me like a research niche that is inherently philosophical and difficult to interpret. Part of my interest in participating in this group is to understand what it actually is that a statistician is able to do to make a causal inference. It very well may form a pillar in my research program.
As for the pedagogy group, I am contemplating how lightly I should allow the projects that are discussed therein to guide my research goals. On the one hand, I may learn about experimental design among other things. However, the primary focus of the work conducted in this group is on how statistics is taught as a subject and is written as a technical genre. This interests me and I would like to follow the particular projects that were introduced to me during last week’s meeting, but I will let this be my “third group” in the list of three in the sense that it shall be treated as a useful hobby and not a pillar in my research program.
Besides reading groups and their particular activities, I am reading some literature on my own, both classic papers and new preprints on the arxiv. I found a few through a list by Nature of the most cited research of all time. Christian Robert had a poll on his blog back in 2010 asking for picks on the “most classic papers” in statistics. Andrew Gelman wrote a response on the Statistical Modeling, Causal Inference, and Social Science blog, considering some exclusions. I have downloaded a copy of every mentioned work and entered it into Mendeley for reading and notetaking.
Two classics that I have read are the following. “Inference and Missing Data” by Donald Rubin treats of the issue of missing data. Earlier works which depend on missing data assumed that the process which made the missing data values missing in the first place is random. Rubin specifies the weakest general conditions required to ignore the process causing missing data. “Improper Priors, Spline Smoothing and the Problem of Guarding Against Model Errors in Regression” by Grace Wahba shows that the method of spline smoothing is equivalent to Bayesian estimation with a partially improper prior. As Wahba summarizes, spline smoothing is a (the ?) natural solution to the problem of having several regression functions which may not contain among them the true model.
I could see myself finding classic papers such as these to report on for my blog. They are fairly dense reading, but the most important results are normally those which have broad applications (since so many researchers have reason to cite them). This could be a really good way to become well read in the literature over time or to ensure that I stop posting soon and forever.