A Bifactor and item response analysis of the geriatric anxiety inventory

Int Psychogeriatr. 2017 Oct;29(10):1647-1656. doi: 10.1017/S1041610217001004. Epub 2017 Jun 20.


Background: Due to previously reported mixed findings, there is a need for further empirical research on the factorial structure of the commonly used Geriatric Anxiety Inventory (GAI). Therefore, the psychometric properties of the GAI and its short form version (GAI-SF) were evaluated in a psychogeriatric mixed in-and-out patient sample (n = 543).

Methods: Unidimensionality was tested using a bifactor analysis. Rasch modeling was used to assess scale properties. Sex, cognitive functioning and depressive symptoms were tested for differential item functioning (DIF).

Results: The bifactor analysis identified an essential unidimensional (general) factor structure but also specific local factors. The general factor comprises all the 20 items as one factor, and the results showed that the variance in the general and specific factors (subscale) scores is best explained by the single general factor. These findings were demonstrated for both versions of the GAI. Furthermore, the Rasch models identified extensive item overlap, indicating redundant items in the full version of the GAI. The GAI-SF also seems to extract much of the same information as the full form. Test scores and items have the same meaning for older adults across different demographic status.

Conclusion: The findings support the use of a total sum score for both GAI and GAI-SF. Notably, when using the GAI-SF, no information is lost, in comparison with the full scale, thus, supporting the option of choosing the short form (version) when considered most appropriate in demanding clinical contexts.

Keywords: Rasch; anxiety; bifactor; geriatric; item-response; psychiatry; psychometric; unidimensional.

Publication types

  • Validation Study

MeSH terms

  • Aged
  • Aged, 80 and over
  • Anxiety Disorders / diagnosis*
  • Anxiety Disorders / epidemiology
  • Female
  • Geriatric Assessment / methods*
  • Humans
  • Language
  • Logistic Models
  • Male
  • Norway
  • Psychometrics / instrumentation*
  • Reproducibility of Results
  • Surveys and Questionnaires