May 172004
 

The arrangement of keys on a keyboard traditionally follows the QWERTY layout (named for the first 5 keys on the left hand: q, w, e, r, t, y). The QWERTY layout was designed before computers existed, and was used as the layout on Remington typewriters. However, the layout has survived for more than a century with very few, if any, improvements. It is the layout that almost everyone uses, but is it really the best choice?

Another keyboard layout, DVORAK (named after the inventor), was patented in 1932, but has never been widely adopted. Many typists who have tried this layout have been adamant that DVORAK is more efficient than QWERTY, but others have claimed that DVORAK is not really much of an improvement. Additionally, switching to a new layout would be quite an inconvenience for most typists. Is a DVORAK layout more efficient, and is it worth making the transition from QWERTY?

Experimental Design

Although the QWERTY and DVORAK layouts have fairly standard implementations, almost every keyboard is slightly different. For this reason, I chose to test the particular layout used on my personal keyboard[1], which allows for both QWERTY and DVORAK layouts. The differences between my DVORAK layout and other layouts are fairly minor, and primarily involve different positions of the { and } characters. While these differences are important, their positions are the same for both layouts being tested and thus won’t alter the results. The layouts being examined are diagramed below, with the home rows lightened:

Map of QWERTY keyboard layout

QWERTY Keyboard Layout

Map of the DVORAK keyboard layout

DVORAK Keyboard Layout

To compare the efficiency and usefulness of each keyboard layout, a scoring system was devised. Each keystroke performed is associated with a particular numerical score which indicates the difficulty of performing the keystroke. These scores are calculated for each word in a text document, and the totals divided by the number of keystrokes required to type the document. The final result is an average keystroke difficulty for the entire file, where a lower score indicates more efficient typing.

The keystroke difficulty score was calculated for each character as (Finger Penalty) * (Key Penalty) * (Reuse Penalty), where:

  • Finger Penalty: Some fingers on the hand are less dexterous than others. The index, middle, ring and pinky fingers were given penalties of 1.0, 1.1, 1.2 and 1.3, respectively.
  • Key Penalty: Keys which are not on the home row are more difficult to press, as the typist must move her fingers to the appropriate key. Keys were given a penalty ranging from 1.0 to 3.0, depending on their distance from the home row.
  • Reuse Penalty: It is difficult to reuse the same hand or finger for two consecutive keystrokes. Consecutive characters which used the same hand were assigned a penalty of 3.0, while characters which used the same finger were assigned a penalty of 4.0.

While care was taken when choosing each penalty, their proper values are not particularly intuitive. Further research could certainly be done to determine if different penalty values are appropriate, or even if new penalties ought to be introduced. However, initial research indicated that modifying the penalties did alter the final scores, but did not change the final conclusions of the study.

Test Cases

Text samples were randomly chosen from the categories outlined below. Thirteen samples were gathered in each category, with no two texts in a category being written by the same author. Texts were selected through a random search of the web for a particular keyword, such as “java code.”

  • Ancient Texts: Works by authors like Aristotle, Plato, Homer and Virgil.
  • Blogs: Online journals published by ordinary people.
  • C++ Code
  • E-mail: Plaintext e-mails from a variety of authors.
  • HTML Code: Webpages such as Google, SourceForge, CNN and CNET.
  • Java Code
  • Movie Scripts: Full length scripts from popular movies like Contact, The Fifth Element, Cast Away and Twelve Monkeys.
  • News Articles: News reports from various national and international newspapers.
  • Perl Code
  • PHP Code

The samples were divided in this manner because it seemed likely that different types of text would score quite differently. Computer code often contains many symbols, in addition to letters and numbers, which would increase scores. There was also the possibility that traditional texts, such as news articles and books, would differ in score.

Results

The diagrams below summarize the data gathered by the software analysis[2]. Scores are reported as the average score per character in a file, with 13 text files in each category. (Note that plots will represent QWERTY with blue and DVORAK with red.)

Plot of typing difficulty on a QWERTY keyboard layout

Plot of typing difficulty on a DVORAK keyboard layout

The first conclusion to be drawn from these plots is that computer code is much more variable than regular texts. The range of score per character is much wider for each of the computer languages, particularly Perl. However, the range for each of the regular texts is quite small, with the exception of a few outliers. This suggests that the different styles of code used by each of the programmers significantly increases or decreases the difficulty of their typing. Perl programmers, in particular, seem to have a very wide variety of programming styles.

Next, it is quite clear the DVORAK scores are lower than QWERTY scores. In each category, the DVORAK scores appear about one point lower than QWERTY scores. A summary of all samples shows that the lowest observed improvement from QWERTY to DVORAK was 0.18 points per character, while the highest improvement was 1.28. Unfortunately, most statistical tests will not be appropriate, or will provide misleading results, because the set of scores for each keyboard layout is not normally distributed (Shapiro-Wilk Normality Test returns p-value 0 for both layouts).

However, there are several distribution-free tests that might be used. The scores for each layout are not independent, as two tests (QWERTY and DVORAK) were run on the same subject, so a paired test would be quite appropriate. The Wilcoxon Paired Sign Rank test is a distribution-free test, and reports a p-value 0 for difference in score between QWERTY and DVORAK. At any significance level, we must reject the null hypothesis and conclude that the mean difference between QWERTY and DVORAK scores is not zero. The Wilcoxon test also reports a 99% confidence interval of (0.72, 0.85) for difference in scores. A typist converting from a QWERTY to DVORAK layout should see an improvement of at least 0.72 points per character, on average.

Now, the differences between the computer language and traditional text samples suggest that they be compared separately. Conclusions may then be drawn about users who typically write emails, blogs, etc., or about users who typically write computer code. Users may then make decisions based on their specific needs.

Plot of typing difficulty for traditional text documents

Plot of typing difficulty for computer documents

Clearly, those who type mostly traditional documents will observe a larger improvement than those who type mostly computer code. Again using the Wilcoxon test, the 99% confidence intervals for traditional text and computer code are (0.90, 0.97) and (0.53, 0.69), respectively. A typist who writes very little computer code should expect to gain almost one point per character when switching to DVORAK. On the other hand, those who write a lot of computer code may not gain more than half a point per character.

Normal users will see an improvement of at least .90 points per character. Given that the average difficulty for traditional texts on a QWERTY layout is 3.5 points per character, a .90 point decrease indicates a 25% improvement in typing efficiency. On average, the typical user will find each character 25% less difficult on a DVORAK keyboard than on a QWERTY keyboard! Similarly, a user who types mostly computer code will experience a 10% improvement. These are significant improvements in efficiency, and seem sufficient for computer users to consider converting to DVORAK.

QWERTY vs DVORAK Conclusions

No matter what type of document being typed, DVORAK seems to offer a reasonable improvement in typing efficiency. This improvement ought to manifest itself through a less frustrating typing experience and decreased hand/wrist discomfort. Consecutive characters will less frequently require the use of the same hand or finger and the fingers will need to leave the home row less often. Typing speed may also improve, as each word will be less difficult to type.

Unfortunately, the task of switching from QWERTY to DVORAK is hard. Users will need to retrain their hands and brains to think in DVORAK instead of QWERTY. They will also experience a temporary decrease in typing efficiency as they adjust to the new layout. Finally, it may be quite some time before they are able to successfully move between QWERTY and DVORAK keyboards without being thoroughly confused. Nevertheless, in the long run the user will find typing less frustrating and more efficient if a DVORAK layout is in use.

References

[1] Kinesis Corporation. Online. http://www.kinesis-ergo.com/
[2] Software analysis performed using custom Java application. E-mail author to obtain sources.

Acknowledgements

[1] Review and assistance from Professor Tom Scofield, Calvin College Mathematics Department.
[2] Plots and statistical tests performed with R.

 Leave a Reply

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>

(to be displayed above your comment)

(for moderation; your e-mail address will not be made public)