Rasch analysis and item reduction of the hypomanic personality scale

https://doi.org/10.1016/j.paid.2008.02.009Get rights and content

Abstract

The aim of the current study was to reduce the number of items in the 48-item hypomanic personality scale (HPS) and determine whether a unidimensional scale of the hypomanic trait could be derived. Previously collected HPS data from University students (n = 318) were applied to the Rasch model (one-parameter item response theory). Overall scale and individual item fit statistics were used to judge fit to the model and item maps employed to determine coverage of the trait. Cronbach’s Alpha and correlations with other questionnaires pre- and post-item reduction were evaluated. Rasch analysis indicated that the original HPS was not unidimensional, had significant redundancy and differential item functioning by age and gender. An iterative process of item reduction produced a 20-item HPS (HPS-20) that retained the concepts of the original HPS and had excellent fit to the Rasch model (χ2 p = 0.27). Unidimensionality of the HPS-20 was confirmed. The traditional psychometric properties of the HPS-20 and coverage of the underlying hypomanic construct were similar to the original. It was possible to derive a unidimensional measure of the hypomanic trait. Further use of the HPS-20 is encouraged as it may increase understanding of the risk factors for affective disorders.

Introduction

People experiencing hypomania are characterised by optimism, happiness, extraversion, sociability, and increased energy but can also be irritable, impulsive, risk-taking, irresponsible and controlling (Eckblad & Chapman, 1986). It has been suggested that individuals exhibiting hypomanic personality traits may be at an increased risk of developing bipolar disorders (Kwapil et al., 2000), which has been estimated the fifth most important cause of disability and third most important mental illness worldwide (World Health Organization, 2001).

The hypomanic personality scale (HPS; Eckblad & Chapman, 1986) is one of the most commonly used questionnaires to assess hypomanic personality traits and has been used to identify those most at risk for bipolar disorders (Meyer, 2002). The HPS is a self-report questionnaire capturing sociability, high ambition and self-esteem, positive affect, increased energy levels, and self-perceived individuality and creativity.

A number of studies have provided encouraging evidence of the construct validity of the HPS. In the development study, Eckblad and Chapman (1986) found that high scorers on the HPS had more mood disorders and hypomanic episodes, more psychotic-type symptoms, higher rates of substance abuse and lower psychosocial functioning compared with low scoring controls. These findings were corroborated by Kwapil et al. (2000) and similar results were found by Klein, Lewinsohn, and Seeley (1996). Using the German version of the HPS, Meyer and Hautzinger (2003) also found a group scoring highly on the HPS had a significantly higher lifetime risk of manic episodes (20.8%) and hypomanic episodes (41.7%) than a low scoring control group (1.3% and 6.6%, respectively). High scorers on the HPS have also been found to share similar traits and biases to individuals with bipolar disorder (Bentall & Thompson, 1990; Johnson, Ruggero, & Carver, 2005).

Perhaps most importantly, the study by Kwapil et al. (2000), which followed-up HPS high scorers after 12 years, found that the HPS scores predicted the onset of bipolar disorders. A disorder prevalence of 25% was reported for the (high scoring) risk group compared to 0% in the (low scoring) control group. The authors also found that 36% of the high scorers had experienced a major depressive episode in the follow-up period compared to 10% of the control group.

Despite this encouraging evidence about the utility and validity of the HPS, very little work has been conducted assessing its dimensionality. Research on the dimensionality of questionnaires in this field has been generally limited to the application of factor analysis (e.g. Eckblad and Chapman, 1986, Hantouche et al., 2003, Meyer et al., 2007).

Item response theory (IRT) is a general statistical theory about item (question) and scale (questionnaire) performance and how that performance relates to the factor(s) that are measured by the items in the scale. The simplest logistic latent trait IRT model is the Rasch one-parameter model (Rasch, 1960).

Rasch analysis places questionnaire response data for each individual and each question on the same spectrum of person severity and item severity (i.e. every person and item is given a location). The Rasch model assumes that the probability a particular individual will respond in a certain way to a particular item is a logistic function of the relative distance between the item location and the person location and only a function of this. In the case of dichotomous items this is formally expressed bypi(θ)=e(θ-bi)1+e(θ-bi)where pi(θ) is the probability that persons with severity θ will affirm item i, and b is the item severity parameter.

For the HPS to conform to the Rasch model, the probability that an individual would affirm an item on the scale must be a function of the level of hypomanic trait exhibited by the individual and the level of hypomanic trait represented by the item being completed and only a function of these factors. For the scale and items to be working consistently, there should be evidence that those people who (according to their scores on the whole scale) possess a high level of the hypomanic trait tend to have a high probability of affirming items representing low levels of the trait. Assuming that the data fit, the Rasch model transforms them from ordinal scores into interval level measurement with the logit (log odds unit) as the unit of measurement. The use of the Rasch model informs whether items and scales can be considered unidimensional. Once unidimensionality is confirmed it is then justifiable to claim that the items measure one construct and it is acceptable to add the scores of these items together to obtain a total score for that construct.

The use of Rasch analysis in the development and analysis of tests and questionnaires is increasing in psychology and psychiatry (e.g. Betemps and Baker, 2004, Merrell and Tymms, 2005, Pallant and & Tennant, 2007, Schultz-Larsen et al., 2007). Several authors (Nijsten et al., 2006, Prieto et al., 2003, Wright, 1996, Wright, 1999) have argued that the use of Rasch in assessing questionnaire scaling properties is preferred to the use of classical test theory (CTT) and factor analysis. Factor analysis does not necessarily provide a conceptual linear assessment of the construct, even if there is a high loading onto one factor (Wright, 1999, Waugh and Chapman, 2005), and may provide misleading evidence that a scale is working well when it is not (Waugh & Chapman, 2005).

The HPS may be a useful tool for gaining an understanding of the aetiology of and risk factors for affective disorders. However, the limited research conducted into the dimensionality of the scale and the underlying hypomanic construct together with the scale’s length (48-items) may act as barriers to its increased use. The aim of the research was to apply the Rasch model to HPS data to determine whether a unidimensional measure of hypomania could be derived and to reduce the number of items to produce a short and more easily administered scale.

Section snippets

Data source

The data used for the analysis had been previously collected as part of a PhD project. 318 undergraduate students at the University of Manchester completed a battery of online questionnaires as part of their course requirements. Two hundred and nineteen (68.9%) were female and 99 (31.1%) male. The mean age of the sample was 21.5 years (SD = 4.0, range 18.0–48.0). The following measures were completed.

HPS

The HPS comprises 48 statements with dichotomous response options (‘True’ [scored 1] or ‘Not

Rasch analysis

Rasch analysis of the 48-item HPS indicated that it did not fit the model (overall χ2 p < 0.001). The residual mean value for the items was 0.003 with a SD of 1.66. Since the SD is much higher than the expected value of 1 there is evidence that the items do not fit the model. The residual mean for persons was −0.038 with a SD of 0.932 indicating that respondents did not misfit the model. Two items (item 20: ‘I am no more self-aware than the majority of people’ and item 38: ‘When I feel very

Conclusions

This study is one of the few to apply modern test theory in the form of Rasch analysis to a mania-related scale. The use of Rasch has enabled the derivation of a significantly shortened HPS that will lend itself more readily to inclusion in studies of the general population. The HPS-20 retains the scope of the original scale both in terms of the level of trait captured and the ideas covered. The indicators of hypomanic personality included by Eckblad & Chapman in the original scale are also

Acknowledgement

The authors would like to thank Sara Melo who collected and supplied the data which made the research possible.

References (35)

  • R.P. Bentall et al.

    Emotional stroop performance and the manic defence

    British Journal of Clinical Psychology

    (1990)
  • E. Betemps et al.

    Evaluation of the Mississippi PTSD scale--revised using Rasch measurement

    Mental Health Services Research

    (2004)
  • B. Birmaher et al.

    Course and outcome of bipolar spectrum disorder in children and adolescents: A review of the existing literature

    Development and Psychopathology

    (2006)
  • G.S. Claridge

    Origins of mental illness

    (1985)
  • G. Claridge

    Schizotypy: Implications for illness and health

    (1997)
  • D.P. Crowne et al.

    A new scale of social desirability independent of psychopathology

    Journal of Consulting Psychology

    (1960)
  • M. Eckblad et al.

    Development and validation of a scale for hypomanic personality

    Journal of Abnormal Psychology

    (1986)
  • Cited by (43)

    • Hypomanic Tendencies and Lifetime Aggression

      2019, Neurology Psychiatry and Brain Research
      Citation Excerpt :

      Despite sample and factor dimension differences, these three models seemed functionally equivalent in their ability to account for variance in the aggression indicators examined in this study. The HPS-20 (Meads & Bentall, 2008) continues to facilitate research on the affective disorders as a brief hypomania screening measure. The HPS-20 was derived from the same item content and the two measures are highly correlated.

    • A Rasch analysis of the Revised Study Process Questionnaire in an Australian osteopathy student cohort

      2018, Studies in Educational Evaluation
      Citation Excerpt :

      Lamoureux et al. (2007) summarise the Rasch model as ”… the probability of a given response affirming an item is a logistic function of the relative distance between the item’s location and respondent’s location on a linear scale.” The location of the item or respondent is measured in log odds units called ‘logits’ (Meads & Bentall, 2008; Rodger et al., 2013). When compared to item response theory (IRT) models, Rasch analysis allows for the evaluation of the summated scale/raw score as a sufficient statistic (Andrich & Hagquist, 2012; da Rocha, Chachamovich, de Almeida Fleck, & Tennant, 2013; Hagquist & Andrich, 2004), that is, “…no other statistic that can be calculated from the same sample provides any additional information as to the value of the parameter (p. 310)” (Fisher, 1922).

    • Measuring the validity and psychometric properties of a short form of the Hypomanic Personality Scale

      2015, Personality and Individual Differences
      Citation Excerpt :

      The present results indicate that although the overall pattern of correlations and effect sizes are relatively comparable for the short and original HPS, the short form seems to be tapping slightly more pathological characteristics of hypomanic personality than the original measure (an issue that potential users of the short scale should take into consideration). Meads and Bentall’s (2008) derivation of the short form is limited by the use of a one-parameter IRT model that assumes equal discrimination across all items. A two-parameter IRT model offers advantages for examining measures of personality and psychopathology in that such models consider both item difficulty and discrimination.

    View all citing articles on Scopus
    View full text