This is a descriptive, quantitative study with a cross-sectional design conducted to assess the internal structural validity of the WC-MAL using Rasch analysis, a probabilistic model grounded in item response theory. Additionally, the reliability and criterion validity were assessed. This study was conducted according to the Declaration of Helsinki and approved by the Ethical Committee of the Universidade Federal de São Carlos (3.710.548). All participants provided informed consent.

This study used a non-probabilistic convenience sample. Convenience sampling was chosen because the instrument is specifically designed for the assessment of individuals with SCI. Participants were recruited through social media, radio advertisements, and project flyers. Inclusion criteria were adults with SCI, chronic injury duration (>6 months), and use of a wheelchair as the primary means for daily mobility. Exclusion criteria included not being enrolled in any specific wheelchair skills or mobility training program.

A sample size of 60 was targeted for the internal structural validity analysis to ensure sufficient statistical power for Rasch analysis. This sample was deemed appropriate based on the general assumption that the minimum sample size for polytomous observations is at least 50 participants. For Rasch modelling, a sample size of 60 provides useful and stable estimations of item and person locations irrespective of scale targeting. [18, 19] This sample size allows adequate testing of dimensionality of the instrument and ensures that the model can accurately evaluate the relationships between items and the underlying construct.

For data collection, two independent physiotherapists (TRS and CLD, raters #1 and #2 respectively) administered the WC-MAL via video calls, with sessions conducted two weeks apart. This time was deemed adequate once the participants were considered clinically stable and were not enrolled in any specific training to improve their mobility in wheelchairs.

The raters read the manual of WC-MAL and were trained for its application by the senior author. They were blinded to each other’s scores to ensure unbiased evaluations. The data from rater#1 were used for the internal structural validity (Rash analysis) and intra-rater reliability through concordance analysis with the data from rater #2.

Following the second WC-MAL interview, a home visit was arranged to install a tachometer on the wheelchair wheel in those individuals who were living in São Carlos City (SP, Brazil), enabling the measurement of movement and wheelchair use over three days.

The WC-MAL is a semi-structured interview that comprises 23 activities (items) assessed by three scales: the Frequency Scale, Performance Scale, and Assistance Scale. Respondents are asked about their wheelchair use for each activity listed in the instrument. If the respondent answers “yes” to having performed the activity within the past week, the activity is then scored using the three scales. Each scale ranges from 0 to 5, with 0 indicating non-performance and 5 indicating the highest frequency, best performance, and greatest independence for the activity, respectively. If the response is “no,” the respondent is asked to provide a reason, which can be selected from a predefined list of general possible causes.

At the conclusion of the interview, individual scores are obtained for each of the three scales, and a composite score is calculated by averaging the scores from each scale. Each of the three WC-MAL scales ranges from 0–5, with higher scores indicating greater frequency, better performance, or greater independence. The composite score, calculated as the mean of the three scale scores, also ranges from 0 to 5, reflecting an overall measure of mobility across the assessed activities [9].

The Rasch measurement model was employed to assess the structural validity of the WC-MAL. For this analysis, the Winsteps program (Version 3.92.1, Winsteps, Beaverton, Oregon, USA) was used, focusing on the Frequency Scale, Performance Scale, and Assistance Scale scores. This model estimates a person’s ability relative to item difficulty, expressed in log-odds units (logits) on a single continuum scale. Participants with higher ability and more difficult items are positioned on the same end negative side of the continuum scale, and vice versa. Item fit statistics are expressed using infit and outfit mean square (MNSQ) statistics, based on the chi-square statistic with each observation weighted by its statistical information (model variance). A range of 0.7–2 is used as a criterion for good fit.

Targeting refers to how well the difficulty of items matches the abilities of the study sample. The standard error of the person measures was used for assessment. Cutoff points were defined as follows: fair targeting (1–2 error), good targeting (<1 error), and very good targeting (<0.5 error). Differential Item Functioning (DIF) analysis, including both uniform and non-uniform DIF, was performed to identify significant differences in item responses by subgroups based on demographic characteristics. DIF was assessed by gender, type of lesion (complete or incomplete), level of lesion (cervical, thoracic, and lumbar), and presence or absence of shoulder pain. Non-uniform DIF indicates unstable behavior in item response probability due to an external factor, suggesting potentially biased responses. Notable DIF was defined as a difference of >1.0 logits [20, 21].

Local dependency was identified through paired standardized residual correlations between items exceeding 0.30 [21]. If local dependency occurs, it is recommended to combine the dependent items into one. Measurement precision was assessed by person separation reliability (PSR) and the separation index (PSI). Person separation is used to classify individuals, and low person separation in a relevant person sample implies that the instrument may not be sensitive enough to distinguish between higher and lower performers. A PSR of ≥0.80 (PSI ≥ 2.00) indicates that the instrument can distinguish the study population into two to three levels of disability [21].

The unidimensionality of the Rasch model is assessed by independent t-tests for each person, with less than 5% of tests outside the ±1.96 range indicating a unidimensional scale. In principal components analysis of the residuals (PCA), 60% of the variance explained by the raw data is considered evidence of unidimensionality. An eigenvalue in the first contrast of the residuals >2.0 suggests the need to measure a second construct [22].

The intra-rater reliability of the composite WC-MAL score was analyzed using the random model Intraclass Correlation Coefficient (ICC [1,1]) with 95% confidence intervals (95% CI), considering the following levels of agreement: weak (ICC < 0.40); moderate (ICC = 0.40- 0.75) and excellent (ICC > 0.75) [23]. Additionally, the Standard Error of Measurement (SEM) was calculated to provide an estimate of the precision of individual scores in the assessment.

The internal consistency of the WC-MAL was evaluated using Cronbach’s alpha coefficient, a commonly used measure to assess the reliability of scales. Cronbach’s alpha quantifies the extent to which the items within a scale are interrelated, indicating the instrument’s internal reliability. An alpha value of 0.70 or higher was considered acceptable.

For the concurrent criterion validity analysis, the data from the tachometer and the WC-MAL Frenquency Scale scores were compared. The concurrent criterion validity was evaluated using Pearson’s correlation coefficient (r) between the tachometer data and the WC-MAL Frequency Scale scores. Correlation magnitudes were classified as moderate to good (r = 0.50–0.75) or strong (r ≥ 0.75) [23]. A strong positive correlation was hypothesized to exist between both assessments.