A study has been performed about the error rates of mobile keyboards. The three keyboards picked were the iPhone virtual software keyboard, QWERTY keyboards (like those on BlackBerries and Treos), and T9 numeric keypads. iPhone users erred at a rate of 5.6 errors per message, T9 users erred at a rate of 2.6 errors per message, and QWERTY users erred at a rate of 2.1 errors per message. The test was done with 20 folks in each group, and the iPhone owners had to have used their device for at least a month to qualify as eligible.

There are some doozies in the press release:

While the iPhone's corrective text feature helps, this data suggests that iPhone users who have owned the device for a month still make about the same number of errors as the day they got it," said Gavin Lew, Managing Director.

Compared to hard-key QWERTY devices, the iPhone may fall short for consumers who use on their mobile device heavily for email and text messaging. The iPhone was clearly associated with higher text entry error rates than a hard-key QWERTY phone. The finding that iPhone owners made more texting errors on iPhones than their hard-key QWERTY counterparts (on their own QWERTY phones) suggests that the iPhone may have a higher fundamental error rate. Specifically, the high rate of false alarms for iPhone keys adjacent to high frequency letters is troubling. The iPhone’s predictive and corrective text features do alleviate some of the errors users make while texting, but it does not catch them all.

"iPhone is a great switch from a numeric phone. But if you're switching from a hard-key QWERTY phone, try the iPhone in the store first."

I'm not sure I agree with him on that last bit. The rate difference from QWERTY to T9 vs. the error rate on the iPhone is basically insignificant. What's another .5 errors per message amongst friends when you're talking about an average of 5+ EPM anyway? But what's most interesting to me is that software and hardware keyboards were just as fast. It's just that software keyboards are prone to more errors.