Precision vs Accuracy

As I pulled into the parking lot earlier this month, I noticed something that falls into the category of "why did you even see that much less remember it"... a car parked on the line.

Those who know me well know that, well, that's just how I'm wired. I'm going to notice that kind of thing. The off by one pixel diagram. The crooked picture on the wall. It's a blessing and a curse.

Anyway, over the course of a couple weeks, I began to realize that I see this particular vehicle almost every day as I pull into the parking lot, and that pretty much every day, without fail, the front passenger tire is on the line.

That, my friends, is amazing precision. Consistency. Repeatability. A trait that we look for and desire, especially in the IT world. But it bugs me, just a little, every day. And it got me thinking: is high precision enough on its own?

I started scrolling my mental ledger back to some of my statistical analysis coursework (taken somewhere between the Stone Age and the advent of the Internet Search Engine, depending on which of my kids you talk to)... and remembered there are really two factors, not one, that we consider to be "on target".

Precision

Precision is a measure of repeatability and consistency. How often in a repeated process do we end up with the same or extremely similar result? OK, awesome. We see low variability and minimal deviation, which means we have a good chance at predicting the result of the next run of the same process. I like it.

Again, that car is parked precisely. But it's consistently off target.

Accuracy

Accuracy describes how close a result is to the intended target or the "true" value. Put another way, how correct are the results? Being able to "hit the target" is also awesome, but just being accurate doesn't mean we're being consistent... which means we might have lower confidence that we can be on target next time. Eww...

Together

As the chart depicts, and as the parked car demonstrates, we can be precise but not accurate, and alternatively, we can be accurate but not precise.

Obviously, the goal is to be precisely accurate. But what if we're starting out with a process that has imprecise, inaccurate results? Which do we try to rectify first?

Your mileage may vary, but my default position would be to first try to solve for precision. Dial in the steps to our process and minimize the number of variables to be able to better predict the result, even if that result is not the desired result.

With our newly consistent process, now we can adjust key inputs to our process to arrive at a more accurate result. As long as we haven't completely nerfed the process itself, we should be able to move closer to the target, and do so consistently.

Does that mean the other sequence is wrong? Absolutely not, and depending on the problem at hand it may make a lot of sense to solve in the reverse order.

Or maybe to just repaint the lines in the parking lot, but that's a whole different topic!

In summary...

A lot of times in our daily work, we strive for consistency. We also strive for accuracy. Sometimes, when we have one of those results, we forget about the importance of the other. At the end of the day, a good process will be both, and that way we can reasonably predict accurate results... the best of all worlds.