Chapter 3: Research Design: Development of Keyboarding Skills: Research: Theory & Research: Universal Design for Learning: CAST

A sample comprising the entire student population of four classrooms at the Graham and Parks Alternative Public School in Cambridge, Massachusetts was drawn. Two of these classrooms were combined grades three and four, and two rooms were combined grades five and six. The sample was initially 92 subjects, but 4 subjects were dropped due to measurement errors, reducing the final sample size to 88.

These 88 subjects came from a range of socioeconomic, ethnic, and racial backgrounds, including Anglo, African-American, (Asian) Indian , and Haitian Creole. The population of the school and of each individual classroom was constructed by school officials to reflect as closely as possible the population of the City of Cambridge as a whole with respect to these factors. Forty four subjects were females, and 44 were males, ranging in age from 7.8 to 13.4 years.

Thirty three subjects (37.5%) were receiving direct Special Education services at the time of the study, eleven subjects (12.5%) were being monitored to determine if direct services should be offered. The remaining 44 (50%) subjects were receiving no Special Education services and were not being monitored.

The school from which the sample was drawn had 372 students, of whom 64 (17%) were identified as Haitian bilingual and 138 (37%) were receiving Special Education support services. Of the students receiving Special Education Support, 131 (94%) were identified by the school as having some form of Specific Learning Disability. The remaining seven students (6%) who received Special Education support had been placed in other categories for service delivery. Precise data are not available, but there is no reason to believe that the percentage in the sample with identified learning disabilities differs significantly from the school-wide average of 94% of those students who were receiving Special Education support. All students in the school spent the majority of their time in regular mainstream classrooms; Special Education services were delivered during "pull out" sessions in a Resource Room or similar setting.

It should be noted that the level of identification of special needs within this sample (38%), and the level of identification in the school as a whole (37%), exceed the national norm of approximately 10% (Meyers & Hammill, 1990) by a substantial margin. This may be a result of several factors: The urban character of this school, Massachusetts' relatively liberal standards for the provision of Special Education support (Chapter 766), Cambridge's commitment to providing a high level of academic support to students who experience difficulties, and Cambridge's decision to provide as many support services as possible in the elementary grades in the hope that adequate support early in the educational process will obviate the need for support later on and result in better overall performance. In any event, this relatively large proportion of students with identified special needs can be expected to make any inference testing addressing differences between individuals with special needs and their normally developing peers more difficult because a relatively wide spectrum of students is identified as having special needs. This would militate towards a failure to reject null hypotheses associated with receipt of Special Education services.

Data Collection

Data were collected on each subject from several sources: Local school records, timed samples of handwriting speed and quality, timed samples of keyboarding speed and accuracy, computer generated records of computer assisted keyboarding instruction, individually administered bimanual alternating finger tapping tests, and personal reports. It was hoped that additional data could be gathered from Special Education records. However, it proved impossible to obtain permission from a sufficient number of families to allow access to these records, and one aspect of the study was seriously weakened.

Local records held in the Graham and Parks School Office yielded: Date of birth, sex, whether at least one parent either held a white collar job or was attending graduate school, and eligibility for Special Education Services. Date of birth and an arbitrary starting date of September 1, 1991 were used to calculate each subject's age at the onset of the study.

Timed samples of handwriting speed were drawn prior to the beginning of keyboarding instruction and again after the course of instruction by asking each subject to handwrite the poemTwinkle Twinkle Little Star from either memory or a printed model (Appendix A) as many times as possible in a three minute period. Speed was calculated by counting the number of letters and spaces in the written product. Information from these two samples of handwriting speed were summed to create a new variable, the total number of characters and spaces handwritten during both tests, which was converted to words per minute by simple division.

Quality of handwritten output was evaluated using a modified version of the scoring criterion provided by the Test of Written Language (Hammill & Larsen, 1978). Each writing sample was scored by four evaluators who received training as prescribed in the TOWL manual. The handwriting quality scores obtained from each of the four evaluators were recorded separately, then aggregated by calculation of arithmetic means of the scores given by each of the four raters. The handwriting quality measure was created by summing these two.

Four tests of keyboarding speed were also administered, using the same three minute time limit and the same poem,Twinkle Twinkle Little Star, from either memory or a printed model. These probes occurred once at the onset of instruction, twice during the course of instruction, and once at the end of the instructional period.

In order to make the typing task as similar as possible to the handwriting task, students were instructed not to backspace nor correct errors during any timed test, but rather to write or type to poem as rapidly and as accurately as possible. Despite the instructions not to delete errors, a number of students were observed making corrections during typing tests.

Determining the number of correct keystrokes performed on each keyboarding probe was not as straightforward an operation as expected, because of the need to separate errors of spelling and of memory from erroneous keystrokes or omissions. The printed results of each typing test were double scored by two different scorers, each using scoring criterion listed in Appendix B. When the scorers disagreed, the protocols were re-examined by both scorers simultaneously, and differences were resolved to arrive at scores for number of keys struck and number of errors. The number of correct keystrokes on each of the four typing probes was retained and converted to words per minute by simple division.

Elapsed time between typing tests was recorded. Measurement occasion one was established at the time of the first typing test and time was set to zero for each subject. Subsequent time measurements reflect the number of days that elapsed between typing tests. Finally, the metric of this variable was changed to months by simple division during data analysis. Elapsed time between the first typing test and the second was 1.47 months, between the second and the third was 1.51 months, and between the third and the fourth tests was 1.42 months. Every effort was made to insure that these times were relatively evenly spaced and approximately the same for each subject. However, because of student absences and scheduling difficulties involving the four classes that constituted the sample, it proved impossible to equalize the time between tests for all subjects. The times reported here are sample means, but actual elapsed time for each individual subject is used in the data analysis.

Data kept by the minicomputer used to provide keyboarding instruction were used to create an approximate measure of instructional time on task. The WICAT minicomputer was configured to keep records of the amount of time that each subject was signed in to a typing lesson. It was not possible to count keystrokes nor otherwise measure subjects' activity during these periods of engagement, so this measure is necessarily vague. However, it is included here because it arguably represents the maximum amount of time each individual could have been engaged in formal keyboarding instruction and practice. Total time on task is taken to represent the amount of time subjects spent using this minicomputer for in-class keyboarding instruction and practice.

The parents of 32 subjects, 9 of whom (28%) were receiving Special Education services, 1 (3%) who was being monitored by Special Education staff, and 22 (69%) of whom were not receiving Special Education services, granted permission for a bimanual alternating finger tapping test (Denkla, 1973; Kicpera, Wolff, & Drake, 1981; Wolff, Gunnoe, & Cohen, 1983; Wolff & Hurwitz, 1976). For convenience in data collection and to insure accuracy, a HyperCard� program was created and used. Instructions were given to each subject to alternatively strike the "a" and ";" keys of an Apple Macintosh Powerbook computer as rapidly as possible, using their left and right index fingers. The program discarded the first five sets of alternating strokes, then recorded the latency for the next ten alternations (twenty strokes) while keeping track of the actual keys struck in a separate file. This allowed protocol examinations to determine if slow times resulted from incorrect keystrokes or were simply slow performance. Three repetitions of this task were performed by each of the 32 subjects for whom permission was granted.

Data collected from the tapping task was examined, and the two slowest of each subject's trials was discarded. The fastest of the three trials was selected as the measure of tapping speed because it was judged to be the most reliable approximation of the subjects' true abilities.

Coincident with the last typing and handwriting tests, a questionnaire was completed by each subject (Appendix C). Individuals' reports of whether there was a computer (as distinct from a Nintendo or other home entertainment system) in their home, and whether they ever used it for homework were collected.

Time Frame of Data Collection

Data from local school records were collected during the months of October and November, 1990. Keyboarding and handwriting speed and accuracy were measured from February of 1991 through June of 1991. Personal report questionnaires were completed in June of 1991. Records of students' use of the WICAT minicomputer were generated in July of 1991. Bimanual alternating finger tapping tests were administered during October of 1992.

Data Analysis

In preparation for the formal data analysis, a comprehensive series of univariate analyses were performed on each of the individual measures. Then to investigate relationships between pairs of variables, bivariate ordinary least-squares regression models were fitted and bivariate plots were constructed. Regression, rather than correlation, was employed here because of the presence of a single conceptual variable representing Special Education Status comprised of two dichotomous variables: currently receiving Special Education support (yes or no) and being considered for Special Education support (yes or no). This sort of conceptual variable cannot be used in correlation analyses. The intention of these explorations was to develop familiarity with the distribution of data and to begin to identify individual cases of particular research interest.

Formal analyses of these data proceeded according to a method developed by Willett, utilizing multiple waves of data to model the growth of individuals' skills over time (Willett, 1989; Willett, in press). According to this approach "...the growth of each individual can be conceptualized as an ongoing, continuous process of which we have 'snapshots' of observed status on each of four occasions."

First, estimates of each subject's true baseline typing ability and rate of skill growth were obtained by fitting an ordinary least-squares regression model using time from the start of the study as a putative predictor of typing test score for each subject. In addition to a standard linear regression model, a quadratic growth model was also fitted to these data, but the explanatory power of the curvilinear term proved to be unimportant. Therefore, the quadratic growth model was abandoned.

Parameter estimates associated with the fitted within-person models, each individual's estimated baseline keyboarding skill and estimated rate of growth (intercept and slope), were obtained and output. Two of these new variables were designated outcome variables in subsequent between-person analyses and regression models were fitted to investigate their relationships with selected covariates.

For each between-person model regression that was investigated, an ordinary least-squares regression was fitted first and an estimate of the between-person root mean square error was obtained. Based on the between-person mean-square error and the estimated precision of the within-person intercept and slope estimates, weights were created and the between-person model was refitted by weighted least-squares regression analysis (Willett, 1989). The effect of this weighting scheme is to assign higher weights, and thus greater influence, to cases where the within-person slope and intercept estimates can be ascertained with relatively greater precision. This, in turn, can improve the precision with which the parameters of between-person relationships among the outcome variables and their predictors can be estimated in subsequent weighted least-squares regression analysis.

Guided by both substantive theory and the statistical adequacy of the models developed at each of several stages, a taxonomy of weighted least-squares regression models describing relationships between the outcome variables (actual baseline typing ability and rate of development of typing skills) and their background covariates were fitted to address the first two research questions.

At each stage of this analysis, results of regression modeling were examined and where supportable conclusions regarding relationships between variables could be drawn, these were noted. In the final stage of the main analysis two summary models were developed and fitted. These are believed to be the most parsimonious models that can adequately describe relationships between individual growth in typing skills and their covariates.

When final models had been created, variables representing Special Education Status were added and models including these terms were refitted to the data. This allowed me to create estimates of the baseline keyboarding skills and rate of development of these skills in students who were uninvolved with Special Education, were being considered for service by Special Education, or were being served by Special Education after the effects of control variables were statistically removed.

Finally, in an analysis that is substantially separate from the one performed to address the first two research questions, a similar but less exhaustive series of regression models was fitted and examined to investigate the third research question. These fitted models explored relationships between our obtained estimates of baseline keyboarding skill and the rate of development of these skills and two covariates: Special Education status and performance on the bimanual alternating finger tapping task.