VALIDATION AND EQUATING OF THE TWO VERSIONS OF THE SF-36 FIVE-ITEM MENTAL HEALTH INDEX

Abstract

Two versions of the Five-Item Mental Health Index (MHI-5) exist differing in their number of response options. Score sufficiency of the MHI-5 has not been evaluated yet. The aims of this study are to test metric properties of these two MHI-5 versions and to equate them using three different methodologies. The two versions of the MHI-5 were assessed in two Swiss surveys. These were equated with a linear rescaling approach and two Rasch-based score equating methodologies: a mean anchoring and a multi-group analysis. Metric properties and score invariance across methodologies are investigated with a stratified analysis by gender, age, and health conditions. The MHI-5 versions show reliability, unidimensionality, local item independence, and fit. Mean scores varied depending on the equating methodology applied and were consistently higher with linear rescaling. However, the relative differences in mean scores were comparable across strategies. MHI-5 has robust metric properties in the general and a disease-specific population. Although equating with linear rescaling may be used, a Rasch-based approach is generally superior with regards to the reliability of the resulting person ability estimates.