For developers I would strongly recommend to support iOS version
(latest - 1), unless the latest version has been available for 6 months already, then just support the latest version. Also, make the choice not considering "now" but consider when the app will be released at the earliest, because that's what matters and is often 3+ months away.
Today I would recommend to not support iOS 7 anymore. The only device you're missing out on by dropping iOS 7 support would be the iPhone 4, which is comparatively very slow. It's the first Retina device with essentially the 3GS hardware given a 30% speed bump but needs to render 4x as many pixels and also is the last device to have a single-core CPU. For anything but a trivial game the iPhone 4 is useless (as in: too slow).
Lastly, older devices give you a greater potential market, but this is a weak market of users who may just have dropped their device in a drawer and forgot about it, or dropped it for real and just threw it away, or who gave it away to their grandparents, etc - essentially any remaining users won't buy apps nearly as much as "latest iPhone model" owners do.
So: "number of devices sold != potential customers". The number 1 mistake made my developers making the "which version of iOS should I support" choice is thinking that the ratio of devices sold to active users remains constant over time.
These numbers will become more disproportionate the older the device is. So when a new device comes out, each sale equals an active user. After 1 year it may be 85%. After 2 it may be 60% and after 3 it might just be 25%. I made those numbers up but this is the general way of how "sold devices" over time transform into an ever decreasing number of "users who might be inclined to buy my app". To the point where you may have millions of devices but no potential customers anymore. That's typically around
(latest version - 2) so you should strive to go one higher as soon as you have a reason to do so (useful new API for instance).
As a framework developer I tend to be more inclusive, because it's usually relatively simple to support the older OS versions (the gist of the workload is on individual developers). However for TilemapKit I'm probably going to cut off support for iOS 7 and go for iOS 8 as minimum target instead - mainly because the SpriteKit in iOS 7.0 is essentially broken and there are a ton of great features only available in iOS 8 (normal maps & lighting, more actions, better physics, convenience methods).