Month: September 2018

I went to a conference last week, more of a meet up really, and they presented the results of the work that we’ve all been doing, indicating that there were several statistically significant improvements in the expected direction.

I’m sure the analysis is well intentioned and basically correct, so I didn’t really have any problem with it, but my arm shot up anyway, because I wanted to see more details of the analysis. The results were so good, I was just curious to see how they were so good really- what were the 95% confidence intervals, the sample sizes, alpha levels, just the nitty gritty of the analysis so I could really get all the detail of it.

But they didn’t have it. They didn’t have much on the slides, and they hadn’t brought any supplementary materials.

I don’t have a problem with that either. I think probably the reason why they didn’t is they don’t usually have stats wonks sitting at the back asking awkward questions. So they didn’t feel the need to be prepared.

I cut my teeth (whille doing my PhD) at academic conferences. Asking awkward questions about statistical analyses is a spectator sport there. And I love it. Everyone’s a black hat, just waiting to crawl inside your work and blow it apart. It’s like a duel, like a competition. And of course that’s how science works. Everyone’s desperate to prove you wrong, and if you can stay at the top, then fair play to you. Probably something in it.

There isn’t enough of that where I work, in the NHS. It really reinforced to me the need to keep going with what I’m doing. I want to train, face to face, everyone in my whole organisation who works with data. Two hours or fifteen, I want to equip them to ask awkward questions.

And one day, I want to sit at the back of the room with my arms folded and watch a whole roomful of arms shoot up.

THEN I can relax.

A couple of days ago on Twitter I said the following:

“Increasingly, RStudio’s products are so good that I feel a lot of my advice to my organisation is “buy a lot of RStudio products”. I love RStudio (I have a tattoo of their logo on my arm, even!) and they clearly give a lot of stuff away (we used their products for nothing for *years*). But I wish I could at least acknowledge some competition in this arena. As far as I can see, if you want to develop a cutting edge data science team with R, it’s RStudio all the way.

I just feel like a brand ambassador rather than someone giving solid, independent advice. I think it’s good advice, don’t get me wrong, but what are the alternatives if you want to interact with R over an authenticated connection other than Shiny with Server Pro?”

I’ve been thinking about it more and more over the last few days and I think my perspective has shifted because my role in my organisation has changed a little. For quite a long time I’ve been churning out Shiny code (and PHP/ MySQL) to run our patient experience data portal. But now I’ve zoomed out a bit and although I’m still doing that I’m working on a couple of other projects and have started to think about how we build our team- using Git, coding collaboratively, enforcing a code style. I’m starting to think about tools.

Now I love RStudio. As I say in the tweet I have a tattoo of their logo on my arm. This is not a metaphor. An actual tattoo on my actual arm. But I’m starting to get concerned that I’m so focused on R and Shiny that I’m starting to miss the big picture. It’s my job to know about other approaches and to understand the benefits and drawbacks of each. Even if I end up saying “There are two other ways of doing this that don’t involve Shiny, but they’re both too difficult/ expensive/ unreliable/ whatever” then fine. But I just don’t feel comfortable advising my organisation without a wider view.

So I’m going to veer off a bit, test the water. I feel sure this will involve Python, Jupyter, and whatever reactive type programming thing Python people use (see? I’m clueless). I know a bit of Python so it shouldn’t be too arduous. And I’ll poke around the rest of the space too. Julia. Other ways of interacting with R that don’t involve Shiny.

As always, I’ll report back once I’ve made head or tail of it, which could be a while.