| Summary: | 碩士 === 國立臺灣大學 === 職能治療研究所 === 106 === Background: There have been many studies based on classical test theory (CTT) which validated the psychometric properties of the Stroke Rehabilitation Assessment of Movement (STREAM). But there were at least 3 psychometric issues that needed to be investigated, including (1) the unidimensionality of upper extremity (UE) / lower extremity (LE) movements subscales of the STREAM; (2) the direct transformation of ordinal raw STREAM total scores into interval scores, which was not mathematically valid; and (3) the estimation of the item difficulty parameters was sample dependent. The Rasch analysis enables examining unidimensionality and Rasch reliability. Moreover, Rasch analysis could transform the STREAM from an ordinal-level measure into an interval-level measure, and estimate both person’s ability and item difficulty. However, so far there was only one study using the Rasch analysis to validate the STREAM, which couldn’t provide sufficient and solid evidence.
Purpose: To cross-validate the psychometric properties of UE / LE movements subscales of the STREAM with the Rasch Analysis, including (1) the unidimensionality of construct validity; (2) the Rasch reliability; and (3) the correlation of the order of item difficulty parameters between previous study and this study. Besides, researchers validated the correlation of latent trait scores based on previous parameters and this study’s.
Methods: Researchers employed secondary data analysis. First, 302 stroke patients during subacute period were derived by screening secondary data source. Then all items were examined with rating-scale model of Rasch analysis. For
example, the unidimensionality was examined by each item’s goodness-of-fit index; and the Rasch reliability was estimated by standard error of measurement, which was assumed that it varied with different latent trait level. Last but not least, we cross-validated the outcomes between previous study and this study, and compared the consistency and the correlation of the psychometric properties.
Results: (1) Only 6 items from UE subscale possessed the unidimensionality, including “Raises hand to touch top of the head”, “Places hand on sacrum”, “Raises arm overhead to fullest elevation”, “Supinates and pronates forearm”, “Closes hand from fully opened position”, and “Opens hand from fully closed position.” While 8 items from LE subscale possessed the unidimensionality, including “Flexes hip in sitting”, “Extends knee in sitting”, “Flexes knee in sitting”, “Dorsiflexes ankle in sitting”, “Plantar flexes ankle in sitting”, “Extends knee and dorsiflexes ankle in sitting”, “Flexes affected knee with hip extended”, and “Dorsiflexes affected ankle with knee extended.” (2) The Rasch reliabilities of UE / LE subscale of this study were 0.92 and 0.93 respectively, which were both slightly higher than Hsueh’s study (0.86 and 0.91 respectively). Besides, 243 patients’ person reliabilities of UE / LE subscale of this study were higher than 0.90, and remaining 59 patients’ person reliabilities were between 0.80 ~ 0.90. (3) The order of item difficulty parameters from UE / LE subscale between previous study and this study was modestly / moderately correlated,with Spearman''s rho equaled to -0.31 and 0.48 respectively. (4) The latent trait scores from UE / LE subscale estimated by previous parameters and this study’s was highly / moderately correlated,with Pearson''s r equaled to 0.96 and 0.62 respectively.
Conclusions: With regard to UE subscale, although the outcome of the
unidimensionality was inconsistent, and the order of item difficulty parameters was modestly correlated, the latent trait scores estimated by previous parameters and this study’s was highly correlated. On the contrary, LE subscale had more consistent outcome of the unidimensionality as well as moderately correlated order of item difficulty parameters, but the latent trait scores estimated by previous parameters and this study’s was just moderately correlated. The difference between analysis results and expectation was probably related to sample properties. But the limitation was that it’s still difficult to identify which parameter was better since the comparison was only between two studies. It’s suggested to revalidate the psychometric properties of STREAM with Rasch Analysis and clarify the problems in the future.
|
|---|
