π

Comparing Typing Speeds of Mobile Input Methods

Show Sidebar

On march 17, I started a (german) call for volunteers in order to get data on typing speed of mobile input methods. I wanted to get accurate numbers to compare different input methods of nowadays mobile systems. The volunteers were asked to type a german text from Goethe (which consists of 107 words) on their mobile devices. They should measure the time spent while typing and send it along with some background data about them and their devices.

I've got submissions of 15 volunteers (2 women, 13 male) resulting in 38 measurements on approximately 20 different devices. Those values also contain hand-written samples as well as some full-size computer keyboard samples. Some samples provided comparable measurements of the same person on the same device but using different input methods.

Input methods

Here is a short list of some input methods according with linked photos:

The term "vertical" relates to the device orientation. Some devices can be rotated into horizontal mode and offer a slightly different on-screen keyboard layout resulting in different sized keys.

The identification string used in this study consists of various informations: for example the string "m HWh\_dict\_T9\_lrn7 SE\_750 (7w)" means:

**m stands for "mobile device" ("-" else)

**HWh: hardware keyboard with horizontal orientation of the device ("SW" for on-screen keyboard, "v" for vertical, orientation string omitted stands for default orientation)

**dict: a dictionary helped the user ("nodict" else)

**T9: T9 input was used

**lrn7: seven words were learned during the test

**SE\_750: short name for the device:

**7w: test person number seven, woman

Results

I tried to collect as much data as possible but few informations from the volunteers did not reach me in time. Those are marked with a question mark.

Note: there is a big version of each graph when you click on it.

">http://www.Karl-Voit.at/temp/suderei/2010-03-28_Typing_all_samples_time.png"

All samples are in this graph (ordered by time). The fastest samples are on top and consists only of non mobile inputs such as full size keyboards or hand written samples.

Except one very fast person (using a Nokia 6233 with T9), there is a notable difference (approx. 4:15) in time from non mobile to to mobile devices. Most persons were able to enter the sample text within ten minutes. Only two samples took longer than ten minutes.\\

">http://www.Karl-Voit.at/temp/suderei/2010-03-28_Typing_all_samples_words_per_minute.png"

This graph shows the values for words per minute accordingly to the previous graph.\\

">http://www.Karl-Voit.at/temp/suderei/2010-03-28_Typing_only_mobile_devices_time.png"

The main focus of this study is the typing speed of mobile devices. So this graph shows only the samples of mobile devices sorted by time. Following graphs do show samples of mobile devices without the samples of non mobile devices.

The average time was 7:51. Split up in the main categories hardware and software keyboards this results in 8:04 for HW and 8:12 for SW keyboards. Omitting the two slowest samples, this is 7:16 for HW and 7:44 for SW keyboards. This seems to prove a small advantage for users of hardware keyboards with less difference than I expected.

For the hardware keyboards only there are those two main categories: T9 keypad and QWERTY/QWERTZ. Omitting the extreme value of 19:22 (K750i with T9 set to off), the average T9 keypad time was 7:44 and the QWERTY/QWERTZ was 6:06. This is a big advantage for QWERTY/QWERTZ keyboards compared to T9 keypads.\\

">http://www.Karl-Voit.at/temp/suderei/2010-03-28_mobile_by_person.png"

Here are the mobile devices in context with the test persons. When a volunteer provided multiple samples on different devices or input methods, it is possible to compare input methods directly. Test person number one is me myself on all mobile devices I am used to. As you can see, my main text interface methods (Palm keyboard and Android Touch Input) are my fastest samples. Glad to know that ;-)

I have to mention that my subjective feeling of ShapeWriter was way faster than the objective numbers. So I'll stick to Touch Input on Android.

Test person number six is used to his Nokia 6310 and began using Nokia E52 recently. He reported that the keypad on his 6310 is way better than on the E52 because of mistyping characters which is caused by inferior haptic feedback while typing.

The difference of a small hardware keyboard on a mobile device is clearly visible at the samples of test person number seven. It is the same device for both samples: the faster one done with the hardware keyboard and the slower one using the on-screen input method of standard Android.

The two samples of test person number ten are showing the results of a person, who just switched from a Sony Ericsson K750i to an HTC Magic using Android. She is using her phones just for making calls and writing very few SMS messages.

The value of T9 as an input help is clearly visible from the samples of test person number twelve: there is a huge delay when T9 is disabled on the very same device.

The fastest mobile samples were recorded by test person number thirteen: she is very fast at typing on the keypad of her Nokia 6233. She writes many SMS per day and notes that the haptic feedback of the keypad is superior to many other phones. She is also using an iPod touch but on this device she is not as far as fast as on the phone.\\

">http://www.Karl-Voit.at/temp/suderei/2010-03-28_mobile_users_skills.png"

The test persons were asked which category they belong to: users or professionals. There is a small difference in terms of writing speed: users 7:19 and professionals 7:46. In my opinion, the device characteristics have way more influence on typing speed than IT/smartphone/phone skills.\\

">http://www.Karl-Voit.at/temp/suderei/2010-03-28_mobile_typing_speed_category.png"

All test persons were asked if they think, whether the currently used input method provides them a fast, normal, or slow input method. As you can see in the graph the users comments on that correlates to the actual numbers when being compared to the other time measurements although the difference is not that big.

The average times for slow/normal/fast are 10:01/7:59/6:08. Without the two extreme values (19:00 for slow and 13:47 for normal), the average times are 8:08 slow, 7:24 normal, and 6:08 fast. The test users were able to categorize their input speed accordingly.\\

">http://www.Karl-Voit.at/temp/suderei/2010-03-28_mobile_dictionary.png"

Some input methods provide help to the user by looking at the characters typed in and compare it (more or less fuzzy) to a dictionary. With this help the user might be able to type faster with more wrong characters but get a correct result anyhow because of spell correction and word proposal/completion.

Input samples that used a dictionary had an average time of 8:12 whereas input samples without a dictionary had 8:03. Without the two extreme values (19:00 for no dictionary and 13:47 with dictionary), the average times are 7:48 for dictionary and 6:33 without.

By further analyzing the data it might be the case that time spent on learning new words to the dictionary could be an additional reason for the poor result for dictionary supported input methods. When a user can type the sample text without having to learn new words to his dictionary, he is faster. But I do not have enough samples to be able to go more into this detail.\\

When I looked at the non-dictionary input methods only (without the extreme value of 19:22), I got 6:45 for software keyboards and 6:06 for hardware input methods. So once again an advantage for hardware keyboards on mobile devices.

Conclusio

I did this study mainly because I was curious. I wanted to know if a physical keyboard or keypad provides reasonable better typing performance than any kind of on-screen keyboards. This seems to be the case. Hardware keyboard input methods with QWERTY/QWERTZ performed better than the rest. T9 keypads (also hardware based input methods) could be found everywhere and did not show any special pattern to me.

This small study has only limited significance due to its limited number of samples. For example I got only one sample for ShapeWriter. So it is not legitimate to extrapolate many things on this page to a general rule of thumb. If you want the (anonymized) raw data, please do not hesitate to contact me. Leave any comment in the comment section of this page.

http://www.Karl-Voit.at/temp/suderei/2010-03-29_results_typing.png"

Note: this blog entry was originally authored using Serendipity and converted to Org-mode format for publicvoit via a dumb script. This may result in bad format or even lost content. Please write a comment if you want to get in touch with me so that I can try to fix things.

Comment via email or via Disqus comments below: