The percentages are percentages of the 58% of failing devices. Of the devices that failed, 29% were iPhone 6, 23% were the 6s, and 14% were the 6s Plus. Add those together and we're missing the final 33% of failed devices but it's safe to assume that a random collection of 6 Plus, 5SE, 5s, 5c, etc. make up that final 33% of the 58%.
So let me see if I understand this epic math fail correctly. Given n devices, there were k devices that were brought in for repair. Of those k devices, 58% were iOS devices, and of those 58%, 29% were iPhone 6 devices.
Which tells us absolutely nothing about the actual failure rate without knowing how the makeup of those n devices relates to the makeup of those k devices. It tells us nothing about the actual failure rate without knowing what percentage of each model within k were junked and replaced without notifying the service center in question. It tells us nothing about whether the Android and iOS users have similar levels of self-sufficiency in terms of figuring out how to solve their own problems. And there are probably at least three or four other fairly fundamental errors that make this data essentially pure noise.
Arguments over minor methodology points, such as whether to count specific types of failures in the reliability numbers, are basically moot, because the "data" is purely anecdotal and is not mathematically related to the actual rate of failure to begin with. This isn't statistically any better than saying, "Of my friends, more people have had problems with Android phones than iOS phones" or vice versa. If you know nothing about whether the sample population has similar distribution to the general population and you know nothing about whether the data is even an accurate measurement of the sample population itself, then these numbers are quite literally no better than a random number generator with a Gaussian distribution. You might as well arrive at the results by throwing darts at a dartboard. It will be approximately as meaningful.
Am I missing something?
Trust me, if even 1% of iPhone hardware failed during its warranty period, heads would roll, much less 58%.