The Devil Is in the Tail: Manchester Short Assessment of Quality of Life (MANSA) test percentiles and normalized T-scores.
In measurement-based care or routine outcome monitoring self-report questionnaires are used to monitor how a patient fares in therapy. Interpretation and utilization of test results are improved and communication facilitated when raw scores are converted to common metrics, such as standard scores (T-scores) and percentile rank scores. Both types of common metrics are described and reviewed and a warning is issued that percentiles may be misinterpreted at the tails of the scale.
Data from the Manchester Short Assessment (MANSA) was used to investigate various approaches to obtain T-scores. The study analyzed cross-sectional data from two normative samples: a representative sample of the Dutch general population (N = 11,789) and a clinical sample (N = 9987) of patients with substance use disorder. Linear, normalized, and IRT-based T-scores were compared.Results revealed that T-scores derived from a linear conversion were biased at the lower end of the scale. Normalizing raw test scores through either Rankit normalization or by an IRT approach yielded improved and quite similar T-scores. For all possible raw scores on the MANSA, corresponding normalized T-scores are presented, as well as Percentile Rank scores for the two reference groups. Finally, cut-off values for reliable change and clinically significant change are presented for raw scores and T-scores.
Practical guidance is offered for converting raw test scores into two common metrics: normalized T-scores and Percentile Rank scores. For T-scores, simple linear conversions yielded biased results. These findings have implications for test developers, practitioners, and researchers who want to express test results in valid and unbiased common metrics.
The online version contains supplementary material available at 10.1007/s10862-025-10266-0.
Data from the Manchester Short Assessment (MANSA) was used to investigate various approaches to obtain T-scores. The study analyzed cross-sectional data from two normative samples: a representative sample of the Dutch general population (N = 11,789) and a clinical sample (N = 9987) of patients with substance use disorder. Linear, normalized, and IRT-based T-scores were compared.Results revealed that T-scores derived from a linear conversion were biased at the lower end of the scale. Normalizing raw test scores through either Rankit normalization or by an IRT approach yielded improved and quite similar T-scores. For all possible raw scores on the MANSA, corresponding normalized T-scores are presented, as well as Percentile Rank scores for the two reference groups. Finally, cut-off values for reliable change and clinically significant change are presented for raw scores and T-scores.
Practical guidance is offered for converting raw test scores into two common metrics: normalized T-scores and Percentile Rank scores. For T-scores, simple linear conversions yielded biased results. These findings have implications for test developers, practitioners, and researchers who want to express test results in valid and unbiased common metrics.
The online version contains supplementary material available at 10.1007/s10862-025-10266-0.