Text Entry Rate Of Access Interfaces Used By People With Physical Impairments

Heidi Horstmann Koester, PhD1 and Sajay Arthanat, PhD, OTR/L, ATP2

1Koester Performance Research, Ann Arbor MI and 2University of New Hampshire, Durham NH


This study systematically reviewed the research on assistive technology (AT) access interfaces used for text entry, and conducted a quantitative synthesis of text entry rates associated with common interfaces. We searched 10 databases and included studies in which:  typing speed was reported in words per minute or equivalent; the access interface was available for public use; and individuals with physical impairments were in the study population.  We used only the text entry rates reported for individuals with physical impairments; studies also had to report the sample size, and the average and standard deviation for the text entry rate data. Thirty-nine studies met the criteria and involved 7 interface types. Automatic speech recognition (ASR), physical keyboard, on-screen keyboard (OSK) with cursor selection, and OSK with scanning selection had at least 4 studies and 30 subjects each, with text entry rates averaging 15.4, 12.5, 4.2, and 1.7 wpm, respectively.


            The purpose of this study is to perform a systematic literature review, in order to develop a better understanding of the typing speed provided by text entry methods for people with physical impairments.  This is a companion study to Koester and Arthanat (2016) in which we presented results for physical keyboard typing. Here, we focus primarily on the remaining access methods.


Search Strategy

            To search the literature, we used the approach outlined by Schlosser et al. (2005).  The guiding search question was: “What are the reported speeds of text entry methods relevant to people with physical limitations?”  We generated keywords based on the Person, Intervention, and Outcomes components:

  1. Person: categories related to disability such as “Disabled Persons” and “Motor Skills Disorders” and specific diagnoses that typically produce physical impairments.
  2. Intervention: categories related to computer access methods, such as “Assistive Technology,” “Communication Aids for Disabled” and keywords specific to particular types of access (e.g., “mouthstick”).
  3. Outcome: keywords related to text entry rate, speed, and accuracy, as well as a general “Outcome” category.

The full search string was formed by combining keywords as Person AND Intervention AND Outcome.  For the complete search string, see the supplementary material for this paper (Koester and Arthanat, 2017).

            We searched 10 databases: PubMed, PEDro, OTseeker, ERIC, DARE, Cochrane, google scholar, ACM, CINAHL, and PsychInfo, yielding 3687 records.  Another 30 articles were identified by reviewing citations in major assistive technology textbooks. We manually screened all 3717 records based on the title and abstract, keeping those that focused on at least one assistive technology to enhance computer use or communication, yielding 635 articles.

Study Selection

            We performed a second round of abstract screening to exclude abstracts where the dependent variables mentioned did not include some measure of text entry performance.  (Abstracts that didn’t include specific dependent variables were retained at this point.)  We also excluded abstracts that focused exclusively on pointing performance rather than typing.  This stage retained 380 articles.

            We reviewed the full text of those 380 articles, excluding articles for the following reasons:

  1. Text entry rate (TER) was not reported, either in words per minute (WPM) or equivalent.
  2. None of the access interfaces was available for consumer use, either via commercial sale or as a free download.
  3. The study included only able-bodied subjects.
  4. The method of measuring TER did not follow conventional techniques.
  5. The results reported were anecdotal, duplicated in another article, or unclear.

This process left 43 articles for further analysis.

Organization of Evidence

            We extracted data from the 43 studies into a spreadsheet based on the critical review form of the McMaster University Evidence-Based Practice Group (Law et al., 1998), extracting the study purpose and design, participant characteristics and sample size, access method(s), text entry measurement procedures, and quantitative typing speed data.  We organized the data by the interface types used in the study.  Interface types were physical keyboard, automatic speech recognition (ASR), Morse code, OSK with cursor selection, OSK with scanning selection, brain-computer interface, and other. 

Quantitative Analysis

            The purpose of the quantitative synthesis was to combine the text entry rates across studies for each access interface, in order to get an overall estimate of the TER for that access interface. For each interface, we created a spreadsheet tabulating the data from all studies that involved that interface. Four studies were removed because they didn’t report standard deviation, which is required for mathematically combining results across studies.  This process yielded 56 entries (i.e., data rows in the spreadsheets) from 39 articles across the seven interface types.

            The final step in the quantitative synthesis was to combine the extracted data across all studies for each interface type.  Using the sample size and the average and standard deviation of text entry rate for each study, we computed the combined average and standard deviation of TER for each interface type, using the calculator at statstodo.com (Chang, 2016).

            The data used in this review are publicly available at kpronline.com/ter-review.php.


Summary across Interfaces

            Table 1 summarizes the results for each main interface, in order of average text entry rate.  Three interfaces provided text entry rates above 10 wpm (speech recognition, physical keyboard, and two-switch Morse code), while the remaining four interfaces yielded text entry rates below 5 wpm.

Cursor Selection On-screen Keyboard.

            These interfaces present a virtual keyboard on the screen.  To enter text, the user moves the mouse cursor to the desired target and either clicks or dwells there. This category does not include the tap-to-type on-screen keyboard that is used on a tablet or smartphone.  Table 2 groups the TER data for these 11 studies across four motor sites. The average TER across all 52 subjects in these studies was 4.24 wpm.

Scanning Selection On-screen Keyboard

            Scanning selection interfaces also present a virtual keyboard on the screen, but selections are made using one or two switches.  Table 3 groups the TER data for these 14 studies across main categories of typing method (letter-by-letter or letters + word prediction) and number of switches.  The average text entry rate across all 34 subjects in these studies was 1.67 wpm.

Automatic Speech Recognition.

            Automatic speech recognition (ASR) allows the user to enter text by speaking. Table 4 groups the text entry data for the four qualifying studies involving ASR.  The average text entry rate across all 50 subjects in these studies was 15.42 wpm.


            This systematic literature review is a unique and novel effort in organizing and understanding what is known about text entry performance for individuals with physical impairments.  Given the intricacies of control interface selection and the absence of established benchmarks, the findings of this review can serve as external evidence for AT control interface selection. The evidence lends a baseline expectation of the TER that may be achieved by a given individual with a physical impairment, when using a specific interface. At the same time, client experience, clinician reasoning, interface placement, configuration, and rate enhancement also need to be factored in to optimize TER specific to the client.

Future Work

            One need is to work toward more well-defined and replicable methods in computer access research.  While many of the basic elements were mentioned in most articles, important details were almost always omitted.  A common structure for performing text entry studies would provide a stronger platform for cumulating results across studies over time.

            This database of studies is a good foundation to address a variety of questions about computer access methods and to identify gaps in the literature base to guide future research. We hope to find effective ways of maintaining and leveraging this database over time, possibly by allowing others to add studies and use the database to explore their own questions.


Table 1. Data summarizing overall statistics for each interface. 95% CI is the 95% confidence interval for the mean.
Interface N Mean SD 95% CI Min Max
Speech recognition 50 15.42 9.99 [12.65, 18.19] 3.5 32.2
Standard keyboard 164 12.47 8.9 [11.11, 13.83] 1.2 48
2-switch Morse 1 12.39 - - - -
1-switch Morse 1 4.88 - - - -
Cursor selection OSK 52 4.24 2.82 [3.47, 5.01] 0.6 11.82
Scanning selection OSK 34 1.67 1.22 [1.26, 2.08] 0.51 6.51
Brain-computer interface 4 0.66 0.08 [0.58, 0.74] 0.56 0.72


Table 2. Data for cursor selection OSK interface, showing all studies divided into four body-site categories.  Statistics are for words per minute text entry rate. Text entry is letter-by-letter without word prediction, except for Wobbrock 2006, in which prediction enhanced TER by 3 wpm for this subject. *Study also reported word prediction data; TER was very similar to letters-only text entry.  SCI = spinal cord injury, CP = cerebral palsy, DMD = Duchenne’s Muscular Dystrophy, GBS = Guillain-Barre Syndrome, UE = Upper Extremity.
Upper Extremity Control of Cursor OSK          
Study Diagnosis & Setup N Mean SD Min Max
Lancioni 2011 CP, hand joystick 1 0.6 0
Smith 2009* CP & C5-6 4 7.47 2.75 3.48 9.72
Wobbrock 2008 Hi SCI, trackball 1 4.86 0
Wobbrock 2006 Hi SCI, trackball, WP 1 11.82 0
Wobbrock 2004 6 CP, 1 MS, w/c joystick 7 0.84 0.36
Total for Upper Extremity Control: 14 3.79 4.06 0.6 11.82
Head Control of Cursor OSK          
Study Diagnosis & Setup N Mean SD Min Max
Pouplin 2016* C4-5 6 6.74 1.20
Pouplin 2015 C4-5 6 3.78 0.93 2.52 8.82
Pires 2012 C3-4, CP 2 1.68 0.34 1.44 1.92
Devries 1998 C5, GBS 2 4.55 0.92 3.9 5.2
Lau 1993 2 SCI, 2 DMD 4 4 1.43 1.8 5.02
Total for Head Control: 20 4.58 1.9 1.44 8.82
Chin Control of Cursor OSK          
Study Diagnosis & Setup N Mean SD Min Max
Pouplin 2016 C4-5, chin + trackball 9 3.3 2.8
Mixed Control of Cursor OSK          
Study Diagnosis & Setup N Mean SD Min Max
Pouplin 2014* 4 UE, 4 head, 1 eye 9 5.13 2.26 1.4 7.6
OVERALL TOTAL: 52 4.24 2.82 0.6 11.82

Table 3. Data for scanning selection OSK interface.  Statistics are for words per minute text entry rate. Diagnosis, switch site, letter layout (e.g., Frequency, Alphabetical), and scan speed (in seconds) are also shown, where available. Mx = varied values, but reported individually for each subject; dash = unknown.  *Data are for revised configuration, where WP outperformed LO, 2.72 to 1.51 wpm; in original configuration, WP and LO were no different. **WP better than LO (1.48 wpm) for this individual.  LIS = locked-in syndrome, CP = cerebral palsy, GBS = Guillain-Barre Syndrome, ALS = Amyotrophic Later Sclerosis.
Letters-Only 1-switch Scanning          
Study Dx Site Layout Speed N Mean SD Min Max
Roark 2015 LIS Chin Freq 1.0 1 1.2 0
Pouplin 2014 LIS - Freq dyn - 1 1.1 0 3.48 9.72
Mankowski 2013 CP - Mx Mx 5 1.07 0.1 0.98 1.22
Chiapparino 2011 CP Hand - 2.0 1 1.33 0
Lancioni 2011 CP Mx - Mx 3 0.94 0.13 0.86 1.09
Lancioni 2010 CP Hand - Mx 2 1.05 0.06 1.0 1.09
Chan 2010 CP Throat Dvorak 1.5 1 0.51 0    
Simpson 2007 CP Mx Freq Mx 4 1.59 0.43 1.08 2.05
Pires 2012 CP Head - - 1 1.36 0    
Lancioni 2009 CP Mx - - 2 0.99 0.08 0.94 1.05
Koester 1990 GBS Hand Freq 0.6 1 3.92 0    
Total for Letters-Only 1-switch: 22 1.27 0.67 0.51 3.92
Letters + WP 1-switch Scanning          
Study Dx Site Layout Speed N Mean SD Min Max
Koester 2014* Mixed Mx Mx Mx 9 2.72 1.8 1.12 6.51
Lancioni 2013** ALS Mouth - - 1 1.85 0
Blain 2010 C1-4 Tongue - 1.5 1 1.1 0
Total for Letters+WP 1-switch: 11 2.49 1.7 1.1 6.51
Letters-Only 2-switch Scanning:          
Study Dx Site Layout Speed N Mean SD Min Max
Lancioni 2010 CP Voice+hand - - 1 1.33 0
OVERALL TOTAL: 34 1.67 1.22 0.51 6.51


Table 4. Data for automatic speech recognition interface.  Statistics are for words per minute text entry rate. SCI = spinal cord injury, RSI = repetitive stress injury.
Automatic Speech Recognition          
Study Diagnosis N Mean SD Min Max
Pouplin 2015 C4-5 SCI 5 18.8 3.83
Alcantud 2006 - 15 13.81 14.5
Koester 2004 10 SCI + 3 RSI + 10 other 23 16.90 7.88 3.5 32.2
Sears 2001 C6 SCI and above 7 11.61 6.78
OVERALL TOTAL: 50 15.42 9.99 3.5 32.2



1. Chang A. (2016). Utility to combine statistics from multiple groups into one group. statstodo.com/ CombineMeansSDs_Pgm.php.

2. Koester H, Arthanat S. (2016).  Systematic Review Of Text Entry Methods For People With Physical Impairments. RESNA 2016 Conference Proceedings.

3. Koester H, Arthant S. (2017). Citations for the 43 reviewed studies and additional search details can be found at kpronline.com/pubs.php.

4. Law M, Stewart D, Pollock N, Letts L, Bosch J, Westmorland M. (1998). Guidelines for Critical Review Form. srs-mcmaster.ca/research/evidence-based-practice-research-group/.

5. Schlosser RW, Wendt O, Angermeier KL, Shetty M. (2005). Searching for evidence in augmentative and alternative communication: Navigating a scattered literature. Augmentative and Alternative Communication, 21:4, 233-255.