Confessions of a very applied statistician (a thread)

I try to think of a model for almost everything I observe. Some things bend naturally to this practice, you pick a model off your shelf and it’s perfect. Other times it’s a mess of complexity and there’s a puzzle for you to figure out. Fun either way.

Models are containers that you conjure from thin air, inside you place phenomena as small as a coin flip or as large as infinity. They fit neatly in containers that capture their essence: in which don't need to be shoved and there’s no extra space leftover once they’re in place.

Before you have data, a model is just an empty container. Data breathes life into it. Some data has more life, and more information, and you covet it. Having more data is like having a larger, clearer window into the machinery of the universe.

You're always wondering if your model is the right container for your data. There are so many options and you'd love to try them all and then see which ones fit snugly. It’s overwhelming, so sometimes you just ignore the options and pick one that's handy, or your favorite.

Models tell stories that are hidden from plain sight. Using them is like focusing a weirdly specific lens on a concentration of light generated by the universe. Looking at the models from various perspectives focuses that light into your retina and brain.

Models don’t always need data to tell stories. The data-free stories may be interesting, but they are fiction. Data inside the wrong model is a different kind of fiction, things may be out of focus. Data inside the right model is non-fiction but you need to consider the source.

It's best when the stories are interesting. An interesting story is surprising AND you’re confident it's true. Often the stories aren’t true despite being interesting. The first joy is finding an interesting story in your model/data, then coming to believe that it is true.

The second joy is telling an interesting story to someone else. That person can understand the story without the model, but if they want trust the story is true, they’ll need to also understand the model. This can take some time, and you'll need to convince them.

The third joy is using the model to do something useful: making a decision and feeling confident about it, making many good decisions quickly, all over the world, sending bits of information to the right places at the right time. Something better happened because of your model.