A regression fit on failure ~ 1 + smart would've presented this data nicely and showed how the various SMART stats interact. If our goal is just to forecast, fitting that regression with a penalized method would've likely provided a great result.
As I received an email about this comment the other day, I wanted to add: this is certainly not the only (or even close to the best) model you could've fit, my point was only to spawn discussion here. There are a tremendous number of ways this data could've been presented that would communicate more than how it was.
