Evaluation in Ayres Sensory Integration ® Praxis Tests: Construct Validity and Internal Reliability

Participants

We recruited a convenience sample of children from the continental United States. In total, 308 children, ages 3 to 12 yr, participated: 163 children with no known medical or educational concerns (control group); 76 boys and 87 girls; M age = 7.87 yr) and 145 children with sensory integration difficulties (SI group); 49 boys and 96 girls; M age = 7.91 yr). Testers recruited the control group from their social and professional networks (e.g., churches, extended family, neighbors, schools), and testers recruited the SI group from therapy practices. Groups differed significantly for gender, with the SI sample including significantly more male children, χ²(1) = 10.43, p = .001; groups did not differ significantly for age, t(306) = 0.14, p = .88. We did not record race, ethnicity, or socioeconomic status.

Inclusion criteria were (1) chronological ages 3 yr, 0 mo to 12 yr, 11 mo and (2) English as the primary language. We included children in the control group if their parents reported that they performed within age expectations and had no medical or educational diagnosis. We excluded children with medical concerns, with behaviors suggestive of sensory integration deficits, or who had received prior sensory integration intervention. We also excluded children with physical or intellectual disabilities or uncorrected vision or hearing deficits. We included children in the SI group who were identified with sensory integration difficulties on the basis of clinical records that included assessment by an occupational therapist, physical therapist, or speech-language pathologist using the SIPT, Sensory Processing Measure (Parham et al., 2007), or Sensory Profile (Dunn, 1999). Children in the SI group had a variety of medical or educational diagnoses (e.g., autism spectrum disorder, attention deficit hyperactivity disorder, speech or language delay, anxiety disorder, or a regulatory disorder), but we excluded children with physical or intellectual disabilities or uncorrected vision or hearing deficits.

Procedures

Examiners included 67 occupational therapists, 1 physical therapist, and 1 speech-language pathologist. All were trained to competence in the ethical conducting of research and EASI test administration via online training modules. Examiners obtained parental consent (and assent from participants age 7 yr or older).

Instruments

The EASI Praxis tests are detailed in Table 1.

Data Analysis

We used WINSTEPS (Version 4.5.2) to conduct iterative Rasch analyses that led to revisions to items and scoring structure. The Rasch model is a latent-trait psychometric model that constructs linear measures from ordinal raw data. It generates item-difficulty and person-ability measures along a single, unidimensional line, expressing measures in “logits” (i.e., log-odds probability units; for a complete description, see Bond et al., 2020).

Results of Rasch iterative analyses led us to eliminate 20 items from Praxis: Positions, 17 items from Praxis: Sequences, and 24 items from Praxis: Following Directions. We removed most of these items because they had similar difficulty levels and, therefore, added little valuable content. For Praxis: Following Directions, we initially scored children on both accuracy and time to complete each item; however, our analyses suggested that time was not a strong aspect of the construct, so we removed these scores. On Praxis: Ideation, we originally included a quality category called “initiation” (i.e., the amount of time to begin acting). We removed this category. Additionally, we converted the total tally score to a 3-point scale on the basis of the total actions demonstrated. Notably, we did not remove all items that failed to fit the model (see the Results section for details on final item fit). Some misfit is expected, and any misfitting items did not degrade or detract from the measurement system (Linacre, 2002).

Here, we report evidence for construct validity and internal reliability of data collected with the retained items. We complemented Rasch procedures with classical statistical methods (e.g., Cohen’s d, Cronbach’s α). We did not correct p values for multiple comparisons because of the exploratory nature of this study.

Construct Validity

Table 2 contains summary Rasch statistics for the four tests. Table A.1 in the Supplemental Appendix (available online with this article at https://research.aota.org/ajot) provides item measures and fit statistics. All tests had uniformly positive point-measure correlations. For Praxis: Positions, 95.6% of items fit the model (i.e., one item failed to fit). For Praxis: Sequences, 96.3% of items fit the model (i.e., one item failed to fit). For Praxis: Following Directions, all accuracy items fit the model. For Praxis: Ideation, all items fit the model. Rating scales for all tests progressed appropriately; all categories were used at least 10 times for each item. Visual examination of Wright maps (see Figures A.1–A.4 in the Supplemental Appendix) indicated adequate item spread and match between sample ability and item difficulty.

Table 3 contains the results of known-groups analyses for accuracy and time scores. The control group scored significantly higher on each test, suggesting stronger praxis ability. Effect sizes for total test accuracy across all tests were mostly medium (i.e., >0.50). Effect sizes were somewhat smaller for Praxis: Ideation, although still ≥0.30.

Internal Reliability

Table 4 provides the results of reliability analyses. For all tests, Cronbach’s α values were ≥.70. Total scores for each test exceeded our desired α threshold of .80. Rasch person-reliability indices were ≥.87; strata ranged from 3.78 to 4.65.