Computer‐assisted vocabulary instruction for students with disabilities: Evidence from an effect size analysis of single‐subject experimental design studies
Abstract
The purpose of this study was to synthesize the effectiveness of computer‐assisted instruction (CAI) studies aiming to increase vocabulary for students with disabilities in an effort to identify what type of CAI is promising for practice. An extensive search process with inclusion and exclusion criteria yielded a total of 13 single‐subject design studies to be included in the present study. Effect sizes were calculated using a percentage of nonoverlapping data (PND). We also analysed instructional features (e.g., visual supports, auditory supports, font/color selection, and corrective and interactive feedback) from the studies that examined effective instructional design features of CAI. Results indicated (a) the highest PND mean was for secondary school‐aged learners with disabilities; (b) both tablet‐assisted instructions and nontablet‐assisted instruction produced high PND (i.e., highly effective); and (c) although the majority of selected studies included visual and auditory supports in CAI for vocabulary, none of the studies provided opportunities for customization (e.g., student selection of colors and fonts). On the basis of the findings, implications for future research and practice are discussed.
Lay Description
What is already known about this topic:
- Technology as instructional aids provides students with disabilities with access to general education curriculum.
- Students with disabilities can benefit from explicit instruction combined with computer‐assisted instruction or intervention.
- Technology provides avenues for teachers to create learning opportunities and practices.
What this paper adds:
- The use of visual images, animations, and graphics led to positive academic outcomes of students with disabilities.
- The use of auditory supports was positively associated with the effectiveness of treatments.
- Providing students with disabilities with a choice can improve their academic outcomes and motivation.
- Customization makes computer‐assisted instruction or intervention adaptable for different levels of students with disabilities.
Implications for practice and/or policy:
- Technology and evidence‐based practice might reduce the gap in achievement levels between students with and without disabilities.
- Teachers should embed with explicit instruction that provides necessary support and structure for students with disabilities.
1 INTRODUCTION
Reading is much more than just looking and deciphering words on paper; reading is about gaining some sort of information from written text or symbols and being able to understand and react to it (Allor & Chard, 2011). Fundamentally, reading involves a process of recognizing words to extracting meaning from words and text (Detheridge & Detheridge, 1997; Samuels, 1988). The inability to read or to gain information from printed literature, thus, can negatively impact an individual's educational performance as well as cause limitations for future career opportunities. Many students with disabilities exhibit difficulties in a variety of reading subset skills including word recognition, reading fluency, reading comprehension, and vocabulary.
Vocabulary knowledge is a major predictor of reading achievement, and it is a crucial component of literacy that many students with disabilities struggle with due to lack of word recognition, oral language skills, and background experiences (Quinn, Wagner, Petscher, & Lopez, 2015). Further, there is a substantial increase in curricular demands in content‐area classes as students enter upper grades. These challenges are often exacerbated as children progress to upper elementary school, when the nature of their literacy instruction often shifts from learning to read to reading to learn and both quantity and quality of vocabulary is required (Chall & Jacobs, 2003). With these curriculum changes come increased exposure to unfamiliar vocabulary. As culminating effects of these skill deficits, students with disabilities exhibit below grade level skills in vocabulary and comprehension (Hock et al., 2009).
One of the most important facets that affect student learning, including learning to read, is an effective and conducive learning environment. The teacher plays a critical role in the learning process and effectively implementing technology in a practical fashion in order for the environment to conducive to student learning (Hakverdi‐Can & Sönmez, 2012). Technology as instructional aids provides students with disabilities with access to general education curriculum as well as vast amounts of information (Coleman & Cramer, 2015; Copley & Ziviani, 2004) such as digital libraries, virtual tours, museum websites, and many more interactive applications (Hakverdi‐Can & Sönmez, 2012). Teachers can utilize websites and applications to engage students and help make abstract facts and concepts come alive. Technology also provides avenues for teachers to create an inquiry‐based learning environment and learning opportunities, practices, and strategies that are student centered, motivating, and engaging for students with and without diverse learning needs (Hakverdi‐Can & Sönmez, 2012).
There is growing trend in education that includes a 1:1 technology initiative. In this initiative, each student and teacher is equipped with a laptop or tablet computer and internet access (Weston & Bain, 2010). The goal of this initiative is to use technology in the classroom to improve not only student learning but also teacher strategies and practices that equip students to compete in the workplace and life in the 21st century (Corn, Tagsold, & Argueta, 2012). Importantly, one overarching goal in education today is closing the gap of achievement between students with and without disabilities by using technology as instructional aid (U.S. Department of Education, 2010).
As seen through the previous attempts of systematic, intensive vocabulary acquisition interventions (e.g., Archer et al., 2014; Coleman, Cherry, Moore, Park, & Cihak, 2015), there is no doubt that students with disabilities can benefit from explicit instruction combined with computer‐assisted instruction or intervention (CAI). There are several CAIs that focus on vocabulary acquisition of students with disabilities, and the majority of vocabulary intervention research concerning students with disabilities alluded to teachers catering to the strengths of the students.
1.1 Purpose and research questions
Given the breadth of modern technology use paired with the benefits of technology‐based intervention for students with disabilities, there is a need to provide information that can be used to assist practitioners with implementing effective CAI for improving vocabulary‐related outcomes of those with disabilities. Thus, the primary goal of this study was to synthesize the effectiveness of CAI intervention studies aiming to increase vocabulary for students with disabilities in an effort to seek out what type of CAI is suggestive to be evidence based. This study, then, reviewed CAI studies that have focused on teaching sight word vocabulary or content vocabulary for students with disabilities. In particular, the research questions this paper addresses are as follows:
- What are the effects of CAI‐based vocabulary interventions for students with disabilities (Pre‐K to 12th grades)?
- What are the specific and critical instructional features of CAI‐based vocabulary interventions?
2 METHOD
2.1 Locating studies
A systematic search was conducted of peer‐reviewed literature published prior to December 2017. Electronic databases ERIC, PsycINFO, and Academic Search Premier (EBSCO) were queried using the following keywords: disabilities, at risk, special needs, and difficulties combined with the following terms that are related to computer‐assisted instruction: computer, computer‐assisted instruction, computer‐based instruction, tablet, tablet computer, iPad, assistive technology, tablet‐assisted instruction, tablet‐based instruction, iPad‐assisted instruction, and iPad‐based instruction.
The initial research based on the search terms generated 517 articles. Following the initial search, the first author examined the abstract of each resulting article to determine whether each study meet the inclusion criteria for further review. Age limits were not placed on participant selection; studies conducted with students at P‐12 level, young children and adults were included. Our search yielded 270 articles published in peer‐reviewed journals. Each of these were manually reviewed to determine if they met the following inclusion criteria:
- Participants were identified as having a disability or disabilities (e.g., intellectual disability, autism spectrum disorder, and learning disability); individuals at‐risk or with difficulties were also included when their difficulties were explicitly indicated in each article. At least in one study (e.g., Crowley, McLaughlin, & Kahn, 2013), only participants meeting this inclusion criterion were included—whereas some of the participants in this study were not included.
- The study used a single‐subject research design such as a multiple baseline, multiple probe, alternating treatments, reversal, and changing‐criterion design to examine the effects of CIA (i.e., CIA as the independent variable). The current review focused only on single‐subject design studies as (a) the current study intends to seek out practical implications for teachers, (b) the single‐subject design allows researchers to delve into specific changes of targeted outcomes or performance (Horner et al., 2005), and (c) the single‐subject design can lead researchers to strong conclusions about the factors that control the dependent variable without random assignment (Horner et al., 2005).
- The study provided sufficient quantifiable data in graphical format, thereby allowing the calculation of treatment effect as a percentage of nonoverlapping data (PND; Scruggs, Mastropieri, & Casto, 1987).
- The dependent variable was vocabulary‐related outcome(s) such as vocabulary acquisition, sight word learning, association of vocabulary with items, or name matching.
- Components of CAI were included throughout the intervention; computer‐assisted instruction combined with another instructional strategy (e.g., flashcards) were also included.
- Articles were published in English.
- Finally, only studies published in peer‐reviewed journals were included; studies from unpublished empirical studies, such as book chapters, conference papers, dissertations, theses, or research reports were not included to avoid overlap with peer reviewed.
Majority articles (over 240 articles) were excluded for the following reasons: (a) they are not an original research article; (b) they did not use single‐subject designs; (c) the dependent variable is not specifically vocabulary but other reading elements (e.g., phonemic awareness, phonics, and word reading fluency).
The final step of article search (#7) left 17 studies that met the six inclusion criteria, four studies (Forbes et al., 2013; Knight, Wood, Spooner, Browder, & O'Brien, 2015; Rivera, Mason, Moser, & Ahlgrim‐Delzell, 2014; Wood, Mustian, & Cooke, 2012) were further excluded because they did not provide sufficient information to compare PNDs (criterion #3); therefore, only 13 studies were qualified for PND calculation. For example, Knight et al. (2015) provided a graph for each participant but only combined results for vocabulary and literal comprehension were available. The other three ineligible studies (Forbes et al., 2013; Rivera et al., 2014; Wood et al., 2012) did not have baseline phase, which is essential to calculate PNDs. PNDs were incalculable in four studies (i.e., Forbes et al., 2013; Knight et al., 2015; Rivera et al., 2014; Wood et al., 2012) due to the lack of baseline phase.
2.2 Coding of identified studies
2.2.1 Calculating treatment effect
PND (Scruggs et al., 1987) was calculated as a method of synthesizing data from single‐case design studies. Several methods for data analysis in single‐case design studies such as various overlap methods (e.g., percentage of all nonoverlapping data and percentage of data points exceeding the median) have been proposed (Parker, Hagan‐Burke, & Vannest, 2007; Wolery, Busick, Reichow, & Barton, 2010); PND is “the most versatile and meaningful of the various methods proposed,” demonstrating the most sensible conclusions (Scruggs & Mastropieri, 2013, p. 17). PND is defined as the number of data points from Phase B (intervention) that exceed the highest data point from Phase A (baseline) divided by the total number of Phase B data points (Scruggs et al., 1987). The calculation involved a straight line through the highest baseline data point to determine the number of treatment data points that exceed line of the highest baseline data (Scruggs & Mastropieri, 1998).
There have been several data analysis methods suggested for single case research such as percentage of all nonoverlapping data and phi. While each nonoverlap indices has limitations (Parker et al., 2007), researchers (e.g., Parker et al., 2007; Scruggs & Mastropieri, 1998) agreed that PND is still a strong method for local decision‐making. More important, because of the relatively easy application and interpretation in single‐subject design studies, researchers agree that PND is the index that still the most often used to calculate effect sizes in the area of special education (Maggin, O'Keeffe, & Johnson, 2011; Scruggs & Mastropieri, 2013).
In the current study, PND was calculated by the comparing the highest baseline data point with all the treatment phases presented. When the study reported multiple graphs for each participant (e.g., Word Set 1, Word Set 2, and Word Set 3 for one participant), PND was calculated by averaging the multiple PNDs calculated from each graph. PND was interpreted as follows: PND scores >90% = very effective treatments, 70–90% = effective, 50–70% = questionable, and PND scores <50% = ineffective (Scruggs & Mastropieri, 1998). PND was calculated for all studies reviewed except those which examined comparative effects between two treatments in which the baseline phase was not included. If the study did not include a baseline phase, the improvement rate between the first and the last data points was calculated.
According to Schlosser, Lee, and Wendt (2008), it is not suggested to calculate PND by comparing the treatment 1 with the treatment 2 in the alternating treatments design studies. If there is a baseline, PND should be calculated for each treatment. For PND, the baseline data are essential. As for the multiple‐baseline or multiple‐probe designs, according to Schlosser et al. (2008), researchers should calculate nonoverlap of each pair of baseline and treatment and get the average PND from the pairs.
For three studies (Ganz, Boles, Goodwyn, & Flores, 2014; Herbert & Murdock, 1994; Rivera, Spooner, Wood, & Hicks, 2013) that used alternating treatment design, two studies (Herbert & Murdock, 1994; Rivera et al., 2013) included a baseline phase; Ganz et al. (2014) included nontreatment condition that was same as baseline. To represent the overall PND for each study coded, an arithmetic mean of all participants' PND was calculated.
2.2.2 Coding reliability
Interrater agreement was determined by the first and second authors independently for each of 13 studies selected. Initial coding for training was performed with randomly selected 20% of 13 studies (i.e., three studies) to ensure the consistency between the two interraters. Intercoder agreement was calculated by dividing the number of agreements by the total number of all coding items and multiplying by 100.
The coding items included number of participants, participant age, gender, disability, research design, independent variable, dependent variable, and PND calculations. Each rater read an article and recorded the coding items independently; the records were compared with each other. For example, a mean PND of each study was independently calculated by each coder and compared by two coders. For the current study, a total of six instructional features were coded by two coders: (a) visual supports, (b) auditory supports, (c) text highlighting, (d) customization (selecting fonts and colors), (e) interactive feedback, and (f) CAI with other learning strategies. The six features were organized as follows. First, the first coder analysed studies (e.g., Hall, Hughes, & Filbert, 2000; MacArthur, Ferretti, Okolo, & Cavalier, 2001; Seo & Woo, 2010) that examined effective and critical instructional design features of computer‐assisted interventions. For example, Seo and Woo (2010) selected and reviewed five features (i.e., visual representations, animations and graphics, text highlighting, selecting appropriate fonts and color, providing feedback, and having adaptive multimedia). The initial analysis yielded five instructional features of CAI; after reviewing articles included in the current study, authors added the last feature, (f) CAI with other learning strategies, to address research‐based strategies embedded in CAI for vocabulary.
Intercoder agreement was calculated as follows. First, two coders (i.e., the first and the second authors) coded each component (e.g., participant age, gender, disability, and instructional features) independently. Second, the number of ratings in agreement and the total number of ratings were counted to get a percentage of agreement. Third, the final intercoder agreement was calculated by dividing the number of agreements by the total number of all coding items and multiplying by 100. Intercoder agreement was 99% and two intercoders discussed until they reached 100% of agreement.
3 RESULTS
The extensive search and procedure to determine the eligibility of each study for inclusion identified 13 studies that aimed to increase word learning outcomes for individuals with disabilities thorough CAI. This section provides an overview of study characteristics, treatment effects, and instructional features as follows.
3.1 Overview of study characteristics
Table 1 provides an overview of participant characteristics (e.g., numbers, disabilities, and ages), research design, independent variable, and findings for each of 13 studies that met all inclusion criteria for this review. Study characteristics include participants, types of disabilities, and intended outcomes.
| Author (year) | Participant characteristics | Research design | Independent variable | Findings |
|---|---|---|---|---|
| Bosseler and Massaro (2003) | N = 6; 9.4–12.5 years old; ASD | Multiple baseline design | Computer software with vocabulary lessons using synthesized speech and images of the vocabulary items, written text, and captioning. | All students increased accuracy of vocabulary identification. |
| Cazzell et al. (2017) | N = 2; 9 and 12 years old; ID | Multiple baseline design | Computer‐based sight word instruction (e.g., reading aloud and repeating the word after hearing recorded sounds) that used flash card for unknown words | The correct identification of unknown sight words was increased during intervention. |
| Crowley et al. (2013) | N = 1; 7 years old (kindergarten); ASD | Delayed multiple baseline design | Teacher‐directed repeated reading of sight words using flash card and teacher feedback. iPad used for repeated practice for listening and speaking of each word. | The number of sight words read was increased during intervention. |
| Cullen et al. (2013) | N = 4; 4th grade; mild disabilities (i.e., LD, mild ID, and ADHD) | Multiple probe design across word sets | CAI using text‐to‐speech to type target sight words, highlight spoken words, read the words, and complete a cloze passage. | All students increased sight word recognition. |
| Ganz et al. (2014) |
N = 3; 8, 9, 14 years old; ASD |
Alternating treatment design | Tablet‐based instruction with least‐to‐most prompting and teacher modeling of each noun, verb, and picture presented on iPad; nontreatment: no use of tablets; | All students increased in the use of nouns and verbs. |
| Herbert and Murdock (1994) | N = 3; 6th grade; LLD. | Alternating treatments design | Baseline: CAI vocabulary program (with definitions, contextual sentences, and multiple choice tests) without speech; (a) CAI with synthesized (computer generated) speech, and (b) CAI with digitized speech digitized (prerecorded and human generated) speech |
All students increased during CAI with speech compared to without speech. Digitized > synthesized (2 of 3 students) |
| Hilton‐Prillhart et al. (2011) | N = 1; 7 years old (elementary); ADHD | Multiple baseline design | CAI sight word program that presented each sight word on screen and provided spoken word when students were asked to read aloud each sight word. | The number of sight words read correctly within 3 s were increased during intervention. |
| Lee and Vail (2005) | N = 4; young children with developmental disabilities | Multiple probe design across four word sets | CAI vocabulary intervention with video clips, sounds, text, and animations for each word. | Participants' percentage of correct responses increased during intervention. |
| Mechling and Gast (2003) | N = 3; 12–18 years; mild to moderate ID | Multiple probe design across three sets of word pairs | CAI vocabulary program with multimedia (text, photographs, and video recordings) for each target vocabulary (grocery items) | All students increased the percentage of locating correct grocery items. |
| Musti‐Rao et al. (2015) | N = 6; 1st grade (6–7 years old); at risk for reading failure (N = 3) and SLD (N = 3) | Multiple baseline design across word lists design | (a) Study 1 (N = 3): teacher‐directed iPad instruction where the students were asked to listen, say, write, and repeat a word with teacher feedback and error correction; (b) Study 2 (N = 3): self‐mediated iPad instruction where students worked independently with iPad by reading, listening, writing a word, recording their voice, and replay the recording. | Students increased sight word fluency during the iPad‐assisted instruction. |
| Rivera et al. (2013) | N = 2; 3rd grade Mexican–American students; moderate ID | Alternating treatments with an initial baseline | CAI vocabulary program with researcher‐developed stories with key words (students missed during pre‐assessment) underlined and sound effects, and picture; (a) English‐based intervention (materials presented in English) and (b) Spanish‐based intervention (materials presented in Spanish) | Students identified more vocabulary (pictures) during intervention. Results were mixed regarding (a) vs. (b). |
| Rivera et al. (2017) | N = 3; 6–8 years old; developmental delay and ID | Multiple probe design | Tablet‐based multimedia (e.g., Google Images and YouTube) intervention that provides stories where target vocabulary embedded | All students increased accuracy of vocabulary identification. |
| van der Meer et al. (2015) | N = 1; 10 years old; ASD | Multiple baseline across matching tasks design | iPad‐assisted instruction with words, pictures, and synthesized speech‐output along with teacher's instruction (e.g., time delay, prompting, and reinforcement) | The student's picture and word matching was increased during intervention. |
- Note. ASD = Autism spectrum disorder; CAI = computer‐assisted instruction; LD = learning disabilities; ID = intellectual disabilities; ADHD = attention‐deficit/hyperactivity disorder; LLD = language‐learning disabilities.
3.1.1 Participants
The current review includes 13 single‐case design studies, including 50 participants including K‐12 students. Majority of studies (11 studies, 69.23%) targeted on elementary school‐aged participants; one study (Lee & Vail, 2005) was conducted with young children with developmental disabilities, and the other three studies (Herbert & Murdock, 1994; Mechling & Gast, 2003) were implemented for secondary school‐aged students.
3.1.2 Types of disabilities
Five studies (Cullen, Keesey, Alber‐Morgan, & Wheaton, 2013; Herbert & Murdock, 1994; Hilton‐Prillhart, Hopkins, Skinner, & McCane‐Bowling, 2011; Mechling & Gast, 2003; Musti‐Rao, Lo, & Plati, 2015) included students with high‐incidence disabilities (e.g., learning disabilities) and students at risk. Students with low‐incidence disabilities (e.g., developmental disabilities, autism spectrum disorder, and moderate intellectual disabilities) were included in eight studies (Bosseler & Massaro, 2003; Cazzell et al., 2017; Crowley et al., 2013; Ganz et al., 2014; Lee & Vail, 2005; Rivera et al., 2013; Rivera, Hudson, Weiss, & Zambone, 2017; van der Meer et al., 2015).
3.1.3 Intended outcomes
All studies reported outcomes related to vocabulary such as sight word vocabulary and content vocabulary. Compared with vocabulary words in reading instruction that are mostly unknown or unfamiliar words, sight word is defined as the word “that can be both read and spelled within 2 seconds” (May, 2006, p. 113) such as no, stop, or come back; students are more likely asked to remember the sight words by sight as a whole. Sight word vocabulary was the dependent variable in five (i.e., Cazzell et al., 2017; Crowley et al., 2013; Cullen et al., 2013; Hilton‐Prillhart et al., 2011; Musti‐Rao et al., 2015) studies, and content vocabulary was the dependent variable in eight studies (i.e., Bosseler & Massaro, 2003; Ganz et al., 2014; Herbert & Murdock, 1994; Lee & Vail, 2005; Mechling & Gast, 2003; Rivera et al., 2013; Rivera et al., 2017; van der Meer et al., 2015). Rivera et al. (2013) measured the number of English and Spanish words correct after implementing CAI both in English and Spanish. The content vocabulary learning was examined more frequently than sight word acquisition; however, both sight word and content vocabulary interventions produced high PND scores (95.24% and 95.98%, respectively).
3.2 Treatment effects
3.2.1 PND results
Among the 13 studies (eligible for PND calculations), 10 studies (76.92%; Bosseler & Massaro, 2003; Crowley et al., 2013; Cullen et al., 2013; Herbert & Murdock, 1994; Hilton‐Prillhart et al., 2011; Mechling & Gast, 2003; Musti‐Rao et al., 2015; Rivera et al., 2013; Rivera et al., 2017; van der Meer et al., 2015) showed a PND score indicative of “highly effective” with a mean PND ≥ 90% for content vocabulary (i.e., Bosseler & Massaro, 2003; Herbert & Murdock, 1994; Mechling & Gast, 2003; Rivera et al., 2013; Rivera et al., 2017; van der Meer et al., 2015) and sight word vocabulary (i.e., Crowley et al., 2013; Cullen et al., 2013; Hilton‐Prillhart et al., 2011; Musti‐Rao et al., 2015). The other three studies (Cazzell et al., 2017; Ganz et al., 2014; Lee & Vail, 2005) could be categorized as being “effective,” demonstrating a PND of 70% to 90%. None of the 13 studies fell within the range of “questionable” treatments (PND = 50–70%) or “ineffective” treatments (PND < 50%). The mean PND calculation for 13 studies was 94.31%; overall, CAI for vocabulary approach can be considered a highly effective treatment for individuals with special needs.
Examination of studies grouped by school level (kindergarten, elementary, and secondary level) revealed the following mean PND calculations: kindergarten learners, 90.95%; elementary school‐aged learners, 94.41%; and secondary school‐aged learners, 90.53%. Although PND for kindergarten learners indicates “effective,” it is important to note that only one study (Lee & Vail, 2005) included kindergarten learners. When grouped by types of disability, following mean PND calculations were provided: high incidence disabilities, 98.58%; low‐incidence disabilities, 90.70%. See Table 2 for summarized study characteristics with the calculated PND values.
| NS | N | PND | |||
|---|---|---|---|---|---|
| M | Description | ||||
| School level | |||||
| Kindergarten | 2 | 5 | 90.95 | Highly effective | |
| Elementary | 9 | 27 | 94.41 | Highly effective | |
| (At and beyond) Secondary | 3 | 7 | 90.53 | Highly effective | |
| Disability | |||||
| High incidence | 5 | 19 | 98.58 | Highly effective | |
| Low incidence | 8 | 20 | 90.70 | Highly effective | |
| Independent variable | |||||
| Tablets | 6 | 16 | 96.22 | Highly effective | |
| Non tablets | 7 | 23 | 92.67 | Highly effective | |
| Dependent variable | |||||
| Sight word | 5 | 14 | 95.24 | Highly effective | |
| Content vocabulary | 8 | 25 | 95.98 | Highly effective | |
- Note. NS = number of studies; N = number of students; M = mean; PND = percentage of nonoverlapping data.
3.3 Instructional features of selected studies
Thirteen studies that were determined as eligible for PND calculations were analysed by the intervention characteristics (see Table 3). Additional details regarding school level, disability, independent variable, and dependent variable for each of the 39 participants from the 13 PND‐calculable studies are presented in Table 3.
| Study | Overall mean PND | Visual supports | Auditory supports | Text highlighting | Selecting appropriate fonts and colors | Interactive and ability/effort feedback | CAI with other learning strategies |
|---|---|---|---|---|---|---|---|
| Bosseler and Massaro (2003) | 96.3 | Y | Y | N | N | Y | N |
| Cazzell et al. (2017) | 79.0 | Y | Y | N | N | N | N |
| Crowley et al. (2013) | 100 | N | Y | N | N | N | N |
| Cullen et al. (2013) | 97.2 | N | Y | Y | N | Y | N |
| Ganz et al. (2014) | 77.3 | Y | N | N | N | N | N |
| Herbert and Murdock (1994) | 94.3 | N | Y | N | N | N | N |
| Hilton‐Prillhart et al. (2011) | 100 | N | Y | N | N | N | Y |
| Lee and Vail (2005) | 81.9 | Y | Y | N | N | Y | Y |
| Mechling and Gast (2003) | 100 | Y | Y | N | N | Y | Y |
| Musti‐Rao (2015) | 100 | N | Y | N | N | N | N |
| Rivera et al. (2013) | 100 | Y | Y | Y | N | N | Y |
| Rivera et al. (2013) | 100 | Y | Y | Y | N | N | N |
| van der Meer et al. (2015) | 100 | Y | Y | N | N | N | N |
| Total | 8/13 | 12/13 | 3/13 | 0/13 | 4/13 | 4/13 |
- Note. PND = Percentage of nonoverlapping data; CAI = computer‐assisted instruction
Six studies (Crowley et al., 2013; Ganz et al., 2014; Musti‐Rao et al., 2015; Rivera et al., 2013; Rivera et al., 2017; van der Meer et al., 2015) of the studies are tablet‐assisted interventions; the other seven studies (Bosseler & Massaro, 2003; Cazzell et al., 2017; Cullen et al., 2013; Herbert & Murdock, 1994; Hilton‐Prillhart et al., 2011;Lee & Vail, 2005 ; Mechling & Gast, 2003) used computer software programs using traditional computer that does not have touchscreen capabilities. Both of tablet‐assisted instructions and nontablet‐assisted instructions produced similar PND scores: 96.22% and 92.67%, respectively.
3.3.1 Visual supports
Eight studies (54.55%; Bosseler & Massaro, 2003; Cazzell et al., 2017; Ganz et al., 2014; Lee & Vail, 2005; Mechling & Gast, 2003; Rivera et al., 2013; Rivera et al., 2017; van der Meer et al., 2015) reported that the CAI for vocabulary included instructional components that are related to visual representations, animations, and graphics, and a mean PND of 91.81% was calculated for these interventions. Photographs and pictures aligned with vocabulary words were noted in the highly effective studies (PND of 100%) including pictures of target vocabulary words, iPad screen pages that show pictures/words, and animations in the format of a digital book (Mechling & Gast, 2003; Musti‐Rao et al., 2015; Rivera et al., 2013; Rivera et al., 2017; van der Meer et al., 2015). Specifically, visual images associated with the phrases or words (e.g., driving, jumping, night, ocean, when we got there I put on my goggles; Bosseler & Massaro, 2003; Cazzell et al., 2017; Ganz et al., 2014; Rivera et al., 2013; van der Meer et al., 2015) or visual supports in the form of video segments presenting actions or definitions of target words (Lee & Vail, 2005; Mechling & Gast, 2003). Interestingly, only one (Cazzell et al., 2017) of five sight word interventions (i.e., Cazzell et al., 2017; Crowley et al., 2013; Cullen et al., 2013; Hilton‐Prillhart et al., 2011; Musti‐Rao et al., 2015) included visual supports.
3.3.2 Auditory supports
Twelve studies (92.31%) included instructional components that were related to auditory supports, which were the most common components described in the studies. Text‐to‐speech (or voice output) was the most common example of auditory support described in highly effective studies that showed PND higher than 90% (e.g., Bosseler & Massaro, 2003; Cullen et al., 2013; Herbert & Murdock, 1994; van der Meer et al., 2015). In Lee and Vail (2005), which shows PND of 81.9%, the auditory support was provided in a form of verbal praise; for every correct responses, students received descriptive verbal praise such as, “Excellent, you clicked the word WALK.” Studies that reported auditory supports produced the average mean PND score, 95.72%. The one study (Ganz et al., 2014) that did not report auditory support reported the lowest PND score, 77.3%.
3.3.3 Text highlighting
Three studies (Cullen et al., 2013; Rivera et al., 2013; Rivera et al., 2017) incorporated text highlighting in their CAI‐based vocabulary instruction. For example, in Cullen et al. (2013), four 4th graders with special needs participated in CAI using the computer software (i.e., Kurzweil 3000 text‐to‐speech program) that provided several activities for sight word learning (e.g., typing target sight words and reading sight words into a microphone) including highlighting spoken words on the screen; this study produced the PND of 97.2%. Rivera et al. (2013) examined the effects of iPad‐assisted instruction using a multimedia‐shared story to improve the number of correct English and Spanish vocabulary words of a student with a moderate intellectual disability; in iBooks, the participant could highlight text using the iPad touch screen capabilities and the study produced the PND of 100%. In another study (Rivera et al., 2017), shared reading intervention was implemented via iPad and target vocabulary was embedded in each story. The interventionist stopped each target vocabulary, and the students were provided with related video, photographs, and systematic instruction; all students increased their vocabulary acquisition (PND of 100%).
3.3.4 Selecting appropriate fonts and colors
None of the studies included a component for selecting appropriate fonts and colors. Although customizing instructional environment according to each student's preference and reading level is one of the significant advantages of using tablets, none of the six tablet‐assisted interventions (i.e., Crowley et al., 2013; Ganz et al., 2014; Musti‐Rao et al., 2015; Rivera et al., 2013; Rivera et al., 2017; van der Meer et al., 2015) described the component. In Musti‐Rao et al. et al. (2015), six 1st graders identified as at risk for reading failure (three students in Study 1 and three other students in Study 2) participated in tablet‐assisted instruction; how their reading levels were reflected in choosing apps or programs were not described.
3.3.5 Interactive and ability/effort feedback
Four studies (i.e., Bosseler & Massaro, 2003; Cullen et al., 2013; Lee & Vail, 2005; Mechling & Gast, 2003) included providing interactive and ability/effort feedback within the tablets or computer software used during intervention. Studies that included mention of teacher feedback (e.g., Ganz et al., 2014; Musti‐Rao et al.‐2015; Rivera et al., 2017; van der Meer et al., 2015) or student self‐evaluation (e.g., Hilton‐Prillhart et al., 2011) were not counted. The specific types of technology‐generated feedback varied across the four studies, including visual feedback such as a happy or a sad face icon for correct and incorrect responses (Bosseler & Massaro, 2003), auditory feedback (Cullen et al., 2013; Mechling & Gast, 2003), and descriptive verbal praise (Lee & Vail, 2005). Overall, the four studies produced a mean PND of 93.9%, reflecting highly effective treatment.
3.3.6 CAI with other learning strategies
Four studies (Hilton‐Prillhart et al., 2011;Lee & Vail, 2005 ; Mechling & Gast, 2003 ; Rivera et al., 2013) combined CAI with learning strategies such as self‐monitoring, constant time delay, and model‐lead‐test strategy. Only studies that explicitly included mention about the use of learning strategies were included. Self‐monitoring (Hilton‐Prillhart et al., 2011) was included in one study; constant time delay was used in the combination of CAI in three studies (Lee & Vail, 2005; Mechling & Gast, 2003; Rivera et al., 2013). In Hilton‐Prillhart et al. (2011), students were asked to self‐evaluate their read aloud of words and record it in their self‐evaluation sheet. Constant time delay allowed students 5 s to initiate their response (Lee & Vail, 2005) or provided a 3‐second delay with pointing (i.e., controlling prompt; Mechling & Gast, 2003). In Rivera et al. (2013), students were given 4‐second delay to provide the correct response before controlling prompt (i.e., verbal model of the word) is delivered. A mean PND calculated for four studies was 95.5%.
4 DISCUSSION
The aims of this review have twofold: (a) to investigate the effects of CAI‐based vocabulary interventions that were designed for students with disabilities, and (b) to seek out suggestive instructional practices that can be incorporated into vocabulary instruction as evidence‐based CAI for individuals with disabilities. Thirteen single‐subject experimental studies were identified to analyse, and PND and instructional features suggested by existing literature revealed the findings as follows: (a) highest PND mean was found for secondary school‐aged learners with disabilities; (b) both of tablet‐assisted instructions and nontablet‐assisted instructions produced high PND (i.e., highly effective); and (c) although the majority of selected studies include visual and auditory supports in CAI for vocabulary, none of the studies provided opportunities for customization (e.g., student selection of colors and fonts).
Consistent with previous research findings regarding the use of visual supports of individuals with disabilities and the use of visual images, animations and graphics within CAI procedures (e.g., Bosseler & Massaro, 2003; Ganz et al., 2014; Lee & Vail, 2005) has also been associated with positive academic outcomes (mean PND of 91.81%). Of 13 studies, 5 studies did not include the components of visual supports; 4 of the 5 studies were sight word intervention studies. Several sight word studies suggest that visual supports can be applied to sight word instructions and increase outcomes (e.g., Eikeseth & Jahr, 2001; Hewett, 1964; Spector, 2011). As shown in the current review, the visual supports were more frequently used in content vocabulary instruction compared with sight word instruction. Because additional instructional component (e.g., games, tracing, and coloring) require visual presentations, the lack of visual supports means that the sight words were simply presented on a screen. Given the advantages of visual supports (e.g., increased motivation and engagement; Kao, Tsai, Liu, & Yang, 2016), the use of visual components should be considered.
Relative to the use of visual prompts, the review revealed that auditory supports have been used more frequently. Given the small differences between mean PND scores between the studies including visual (91.81%) and auditory supports (95.72%), the PND metric from the review indicates that the use of auditory supports was also associated with the effectiveness of treatments. In most of the studies reviewed, the auditory supports were provided in a form of sounding out each word or sentence. Very few studies provided auditory corrective (e.g., repeating the word that a student missed/misread). In addition to further examination of visual supports on word learning (e.g., sight word acquisition and content vocabulary acquisition), continued research on the effects of auditory supports is also recommended.
Compared with auditory and visual supports, feedback within CAI (i.e., feedback provided by a computer) was only offered in four (30.76%) of 13 studies. It is interesting that there were another three studies that provided teacher‐directed feedback. The low number of studies with computer‐assisted feedback indicates the lack of feedback features embedded in computers; however, it is also possible that teachers tried more specific and corrective feedback based on students' performance because we found that the computer‐assisted feedback is more likely basic feedback that simply include “yes or no,” “correct or incorrect,” or “smile or cry emoticon.”
Although six studies included the use of tablet computers during vocabulary instruction, none of the studies included selecting customization options such as options for fonts, colors, level, or pace; however, customization is a critical feature of tablet computers (Boyd, Hart Barnett, & More, 2015). Research has shown that allowing for providing learner customization options plays a critical role in improving self‐regulation of students with disabilities thereby facilitating reading outcomes (Wehmeyer, 2002). Providing students with disabilities with a choice can improve their academic outcomes and motivation (Tiger, Toussaint, & Roath, 2010; Ulke‐Kurkcuoglu & Kircaali‐Iftar, 2010). Furthermore, customization enables students to select the most appropriate level of difficulty and makes CAI adaptable for different levels of students with disabilities (Gerard, Spitulnik, & Linn, 2010; Jeffs, Behrmann, & Bannan‐Ritland, 2006; Maich, Hall, van Rhijn, & Henning, 2017).
The review revealed limited information on effective instructional components of CAI. Given the wide variety of individuals with disabilities can be served in classroom settings, providing individualized educational environment has been considered as a significant predictor for effective reading outcomes (Fraser & Fisher, 1982). It might be because the studies did not describe the design features of CAI (e.g., customization) even though the tablet apps included the features. As noted in Weng, Maeda, and Bouck (2014), the CAI research is still sparse and the detailed description on design features of CAI programs was rarely provided.
All of the tablet‐assisted instruction was conducted for elementary school students, given the unique benefits of tablet computers such as portability, easy navigation, and a touchscreen, it supports young children with their independent practice by helping them to navigate apps, move on to the next page, and learn with multimedia effects. In addition, tablets may help secondary students to promote their motivation and engagement (Kim, Blair, & Lim, 2014). Therefore, additional research is needed to examine effects of using tablets with students in kindergarten and secondary school levels to assist with their word learning.
Although there is emerging literature on tablet‐assisted instruction for students with disabilities, for many tablet‐based reading studies, how the apps were selected and how they were administered are still in question. Therefore, there is a need for further research targeting tablet‐assisted vocabulary instruction for individuals with this population of students.
4.1 Limitations
In this review, there are particular limitations that warrant further examination. First, although PND is considered as the most useful and versatile single‐case design methods that demonstrates a high reliability and can be very easily calculated (Scruggs & Mastropieri, 2013), the generic limitation of PND that the outlier during baseline may have a significant impact. In addition, individual PND scores were averaged to report the mean PND of each study despite the variability across participants. We found that the PND scores across the studies reviewed were high in overall; there were minimal differences found in each variable of analysis (e.g., feedback, auditory support, and visual support). In the study, PND was used to calculate the effectiveness of studies reviewed. Despite the long history of the use of PND in the area of single‐case design studies, PND has showed weaknesses that may lead to the misleading findings (Lenz, 2013; Parker et al., 2007; Parker, Vannest, & Davis, 2011).
In addition, despite the potential relationship between length and duration of CAI and treatment outcomes, the length and duration were varied across the studies, leading to significant differences in frequency and intensity of the CAI for individual with disabilities. Many studies lacked detailed information about length and duration of computer‐assisted vocabulary instruction.
Finally, the current paper included and reviewed only studies that were published in peer‐reviewed journals. As other forms of studies such as unpublished empirical studies, conference papers, research reports, and dissertations were not included, there is a possibility of publication bias.
4.2 Implication for practice and future research
4.2.1 Implication for practice
As revealed across the selected studies, students with disabilities were likely to benefit from vocabulary instruction that includes the use of CAI and its instructional components such as visual and auditory supports. Also, instructional features used in CAI for vocabulary learning incorporated an individualized, independent exposure to rich literary texts that supplements high quality teaching of vocabulary for students with disabilities. Evidence from CAI vocabulary studies commonly shows that the presentation of text with auditory and visual supports led to improved outcomes for students with disabilities. Students were given an opportunity to derive pleasure from reading text, build content, and increase vocabulary knowledge (e.g., Bosseler & Massaro, 2003; Lee & Vail, 2005; Mechling & Gast, 2003). Other benefits include but are not limited to as follows: students can access electronic texts above their identified reading level and use features of tablet or laptop computers to hear a word or text being read aloud while also seeing the words (e.g., Bosseler & Massaro, 2003; Cullen et al., 2013; Herbert & Murdock, 1994). Some CAI had a function that allowed students to re‐read/re‐listen to the highlighted with the click of a button (e.g., Cullen et al., 2013). When students use the highlighting and read aloud feature, they can see the relationship between the spoken word and printed text (e.g., Cullen et al., 2013; Rivera et al., 2013). As a result, CAI can be used to provide scaffolds for students with disabilities to access words and texts that could not have been read independently (Gonzalez, 2014).
Technology and evidence‐based practice might reduce the gap in achievement levels between students with and without disabilities. This gap broadens as students enter highschool classes where expectations and classroom demands rise (Kennedy & Deshler, 2012). Students need to be instructed in the appropriate uses and ways to interact with CAI in order for vocabulary instruction to be successful. Teachers should embed CAI with explicit instruction that provides necessary support and structure for students with learning disabilities to achieve and close the gap in achievement (Kennedy & Deshler, 2010). CAI approach in teaching vocabulary for students with disabilities is fairly new; therefore, research on longitudinal outcomes are limited. Finally, teachers' willingness to learn and prepare for the effective use of technology is vital.
4.2.2 Implications to research
Results of the current review suggest promise of CAI‐based vocabulary intervention. Yet the results also suggest the need for further research efforts to provide comprehensive evidence and practical implications to support the use of CAI to improve vocabulary‐related skills for a variety of special education populations.
First, additional empirical research is needed to add to the research base on CIA to improve vocabulary skills. According to the Human Activity Assistive Technology (HAAT) model (Cook & Hussey, 2002), in order to choose an appropriate assistive technology device (e.g., computer‐based instruction), educators need to consider the student's needs and abilities, the activity, and context for which it will be used. In order to assist teachers in choosing appropriate computer‐based vocabulary instruction, researchers should conduct systematic lines of research aimed at examining vocabulary outcomes across different types of students and examining effects associated with different types of instructional features. In these research lines, we recommend researchers to provide sufficient information on technology and instructional features to enable us an ability to develop a better understanding of critical and most effective instructional features.
Second, more single‐subject studies on vocabulary‐based CAI should be conducted. In the current review, only 13 studies met our inclusion criteria. In addition, there is considerable need for research on the psychometric properties of visual analysis—in particular, its reliability, validity, and accuracy. Despite over 50‐year history of research on the psychometric properties (e.g., reliability, validity, and accuracy) of visual analysis of single‐subject studies, there is a paucity of empirical evidence in the properties (Kratochwill & Levin, 2014). Therefore, future research using single‐subject designs should focus more on the psychometric properties of visual analysis to ensure the “functional relationship” between the CAI and the vocabulary outcome(s).
Third, in the current study, an effect size analysis was used to examine effects across types of instructional features. In the future, with greater numbers of studies, meta‐analytic approaches should include subgroup analyses (e.g., examining effect sizes across disability categories or other leaner characteristics) and investigations of potential moderators such as gender, different CIA features, or duration or dosage of instruction. Future meta‐analytic approaches should also consider publication bias.
Fourth, in this review, we calculated a total effect size for each study. However, this did not enable us to summarize effects for each individual student. Additional research is needed to allow for examination of individual learner responses to instruction (i.e., for whom and under what conditions is CIA most likely to improve vocabulary outcomes). This will help teachers make appropriate instructional decisions about CIA. To better apply conceptual ideas on computer‐based vocabulary instruction to practice, we recommend a comprehensive review approach including both qualitative synthesis and meta‐analysis to include diverse learners who have different patterns and processes in student learning (Vermunt & Vermetten, 2004).
