Response categories and anger measurement: do fewer categories result in poorer measurement

Background Anger is a key long-term outcome from trauma exposure, regardless of trauma type, and it is implicated as a moderator of response to treatment. It therefore seems important that anger is assessed in both epidemiological studies of trauma sequelae and in intervention evaluation research. This study explored the measurement properties of a recently investigated anger scale, the Dimensions of Anger Reactions (DAR) Scale. In our previous study, the DAR was found to be a measure of trait anger, but although brief, the nine response categories per item may have confused respondents, suggesting fewer response categories may work equally well. Additionally, our previous analysis suggested there were two redundant items within the DAR. Methods Three samples of Australian veterans were used to investigate the psychometric properties associated with alterations to the response categories of the DAR; veterans who participated in the DAR validation study, those participating in group therapy programmes for post-traumatic stress disorder, and veterans participating in lifestyle programmes. Item response theory analysis was used to explore the internal properties of competing DAR models, and models were assessed against external criteria. Results The results showed that the number of item responses in the DAR exceeded channel capacity, and that response bias occurred in the second half of the instrument. We hypothesized that this was due to respondents not discriminating among the many response categories. Based on a modelling exercise in which we reduced the number of DAR items from 7 to 5 and the number of response categories from 9 to 5, validation tests showed that there was no loss of sensitivity, reliability or validity. To avoid confusion with the DAR, we have referred to the revised version of the DAR as the DAR5. Conclusions We conclude that the DAR5, which abbreviates the original DAR to half its original length, has similar psychometric properties and is therefore to be preferred especially for use with persons who are under stress, cognitively impaired or less mature. The study findings regarding the optimum number of response categories have implications for the development of other instruments