Wouldn't this be affected by how long ago the app was released or how recent a sale occurred?

Imagine day 1: You have 100 downloads... and 100 current installs. That's 100% ratio.

Perhaps "average number days it remained installed after downloading" would be more useful. Combined with "average number of times used while installed".

Though any time I think of metrics like that, I always see problems: What if you improved the retention rate, but you're not seeing it immediately because your average is still dragged down by the previous nine months of metrics?

Then I remember that graphs and averages can deceive me just as easily as they can inform me; so my plan is to gather raw data (userID 345294 installed game on 2015/08/12 at 12:17 pm; userID 345294 opened the game up, userID 345294 started a new game, etc...), and not try to force the data into any preconceived forms.

Then I could, when I want to look at the data from a different angle, easily generate graphs and averages when needed, to get the most information out of the data using the different viewpoints.

When you collect things like ratios and averages, you're taking actual data {2, 17, 4, 11, 16}, and compressing it {average: 10), losing data from the compression. Then if you want to look at the same data from a different angle in the future (such as only the averages of the past 41 days since you made a specific change), you can't!

We are informed not by data, but by the lense we use to look at the data. So to be more informed, we need to look at the same data through many different views.

That's my intention for user metrics in my (unfinished) games.