Introduction
Should health anxiety be conceptualized as a dimensional (continuous) or a categorical (taxonic) construct? Although this question has been debated, to date no firm conclusions have been reached (see Warwick & Salkovskis, Reference Warwick and Salkovskis1990; Hiller et al. Reference Hiller, Rief and Fichter2002; Ferguson, Reference Ferguson, Maj, Akiskal, Mezzich and and Okasha2005; Noyes, Reference Noyes, Maj, Akiskal, Mezzich and Okasha2005; Marcus et al. Reference Marcus, Gurely, Marchi and Bauer2007). Therefore, it has been argued that taxometric methods should be applied to address this question directly for health anxiety (Ferguson, Reference Ferguson, Maj, Akiskal, Mezzich and and Okasha2005; Marcus et al. Reference Marcus, Gurely, Marchi and Bauer2007). Taxometric methods represent a set of procedures designed to demonstrate the latent structure of a construct as either dimensional (i.e. distributed as a continuous variable, with individuals varying quantitatively from each other) or taxonic (i.e. individuals are differentiated into non-arbitrary groups or categories) (Meehl & Yonce, Reference Meehl and Yonce1994; Meehl, Reference Meehl1995; Waller & Meehl, Reference Waller and Meehl1998; see Ruscio & Ruscio, Reference Ruscio and Ruscio2004a for a review).
Dimensionality: implications for theory, research and practice
Ruscio et al. (Reference Ruscio, Haslam and Ruscio2006) argue that whether or not a construct is dimensional or taxonic has important implications regarding the types of theoretical explanations offered for clinical and individual differences, as well as for the types of research methods and clinical practice used. Theoretically, a dimensional account implies the existence of additive multi-causal agents. By contrast, a categorical approach indicates the existence of either a single causal factor (e.g. single gene or threshold model) or multi-causal effects that produce a qualitative disjunction between taxons and compliments (e.g. emergenesis or developmental bifurcation). Clinically, a dimensional approach would mean a move towards empirically derived multiple divisions as a basis for differential diagnosis, rather than the use of arbitrary cut-offs (cf. Widiger & Trull, Reference Widiger and Trull2007). For research, a dimensional account implies using the full range of scores without splitting samples into ‘cases’ and ‘non-cases’ (cf. Marcus et al. Reference Marcus, John and Edens2004, Reference Marcus, Lilienfeld, Edens and Poythress2006) and developing further dimensional psychometric assessments (Ruscio et al. Reference Ruscio, Haslam and Ruscio2006). Given these different implications, a direct test of the latent dimensional structure of health anxiety is required.
The dimensional debate in health anxiety
There is strong indirect evidence to support the position that health anxiety is dimensional. First, the same pattern of effects for health anxiety is observed on a variety of outcomes in both clinical and non-clinical samples (Marcus et al. Reference Marcus, Gurely, Marchi and Bauer2007). This indicates that the theoretical models apply across the range of scores observed for health anxiety. Second, a number of individual differences (worry, depression, anxiety, attachment style and impulsivity) that are theoretically and empirically related to health anxiety (Kellner, Reference Kellner1986; Ferguson, Reference Ferguson2000, Reference Ferguson2004) have been shown to have dimensional structures (Ruscio et al. Reference Ruscio, Haslam and Ruscio2006).
The indirect evidence supporting a categorical approach is much weaker. There are taxometric data showing that the tendency to over-report symptoms, a characteristic of health anxiety, is taxonic (Strong et al. Reference Strong, Greene and Schinka2000, Reference Strong, Glassmire, Frederick and Greene2006). However, the work reported by Strong et al. (Reference Strong, Greene and Schinka2000, Reference Strong, Glassmire, Frederick and Greene2006) focused on psychiatric symptoms rather than somatic symptoms and may reflect an impression of management style rather than a tendency to over-report due to somatic awareness (Barsky, Reference Barsky1979). There is evidence that aspects of health anxiety symptomatology have a distinct boundary with other somatoform disorders (Fink et al. Reference Fink, Ornbol, Toft, Sparle, Frostholm and Olesen2004). However, the existence of a boundary between disorders does not preclude that the disorder itself may be dimensional (Ruscio & Ruscio, Reference Ruscio and Ruscio2004b; De Boeck et al. Reference De Boeck, Wilson and Acton2005).
Importantly, although the evidence for a categorical model is weak and a dimensional model is strong, it is still common practice in the literature to split samples into cases and non-cases using either psychometric measures or clinical schedules (e.g. Barsky et al. Reference Barsky, Wyshak and Klerman1986; Brown et al. Reference Brown, Kosslyn, Delmater, Fama and Barsky1999; Hadjistavropoulos et al. Reference Hadjistavropoulos, Hadjistavropoulos and Quine2000; see Marcus et al. Reference Marcus, Gurely, Marchi and Bauer2007 for a review). The use of both categorical and dimensional approaches to health anxiety has been advanced (Hiller et al. Reference Hiller, Rief and Fichter2002). Given the very different implications for theory development, research and clinical practice afforded by dimensional and categorical conceptualizations, a direct test of the dimensionality of health anxiety is required (see Ruscio et al. Reference Ruscio, Haslam and Ruscio2006).
Health anxiety versus anxiety about actual illness
Health anxiety represents the fears and worries about illness in the absence of objective illness (APA, 1994). People may report anxiety, depression and general psychiatric distress co-morbidly with actual illness (Haug et al. Reference Haug, Svebak, Wilhelsen, Berstad and Ursin1994; Gupta et al. Reference Gupta, Mitchell, Giuffre and Crawford2001; Baune et al. Reference Baune, Adrian and Jacobi2007), either during or following a diagnosis (Katon et al. Reference Katon, Russo, Ashley and Buchwald1999; Tedstone & Tarrier, Reference Tedstone and Tarrier2003). There is evidence that a diagnosis is associated with increased somatic awareness (see Katon et al. Reference Katon, Russo, Ashley and Buchwald1999). In fact, there is evidence that, of those expressing worry about illness, approximately 50% have actual illness (see Looper & Kirmayer, Reference Looper and Kirmayer2001; Noyes et al. Reference Noyes, Carney, Hillis, Jones and Langbehn2005). The psychological distress and somatic awareness associated with actual ill health needs to be distinguished from anxiety about health in the absence of actual illness. Therefore, taxometric analyses of health anxiety were conducted on individuals who were not currently receiving medical treatment.
Method
Sample and sampling
Following recommendations, the sample was not specifically targeted for health anxiety or designed to be a mixture of high, medium and low health anxious individuals (Ruscio & Ruscio, Reference Ruscio and Ruscio2004a, Reference Ruscio and Rusciob; Ruscio et al. Reference Ruscio, Haslam and Ruscio2006). Instead, a convenience sample of 711 UK employees from three private sector industries and two public (local government) sector services was recruited (see Table 1 for details). Complete data on indicators of health anxiety were available from 699 (98%) of these participants. This sample was screened for those who were currently receiving medical treatment, and a sample of 501 subjects who were not currently undergoing medical treatment was classified as healthy. Although the same pattern of results was observed for the full sample of 699 participants, the results for the healthy 501 participants are reported as these meet a classic definition of health anxietyFootnote 1 Footnote †.
Table 1. The sample
Values in parentheses are s.d.
Measures
Health anxiety
The nine-item version of the Whiteley Index (WI; Barsky et al. Reference Barsky, Wyshak and Klerman1986), scored on a five-point Likert-type scale (Barsky et al. Reference Barsky, Wyshak and Klerman1990), was used to assess health anxiety and it has been found to have a test–retest reliability of 0.73 (Ferguson, Reference Ferguson2004). This index was chosen for the following three reasons. First, it is a standard, reliable and valid index of health anxiety that covers the conceptual range of the construct (see Barsky et al. Reference Barsky, Wyshak and Klerman1986, Reference Barsky, Wyshak and Klerman1990; Ferguson, Reference Ferguson2000, Reference Ferguson2004; Ferguson et al. Reference Ferguson, Moghaddam and Bibby2007). Second, items (referred to as indicators in taxometric analyses) need to be able to distinguish cases from non-cases (Meehl, Reference Meehl1995). When indexed to range from 1 to 5, the WI has been shown to differentiate cases from non-cases (see Barsky et al. Reference Barsky, Wyshak and Klerman1990). Third, each item on this index is scored on a five-point Likert-type scale. Likert-type scaling, compared to dichotomous scoring, increases the power of taxometric analyses (Ruscio & Ruscio, Reference Ruscio and Ruscio2004a, Reference Ruscio and Rusciob). This version of the WI was adopted as it covered the conceptual nature of the construct, can be scored to differentiate cases from non-cases and is scored on Likert-type scales. The wording of the nine items was taken directly from Barsky et al. (Reference Barsky, Wyshak and Klerman1986). Items 1–3 reflect the highest three loading items for disease conviction, items 4–7 reflect the four highest loading items for disease phobia and items 8 and 9 reflect two of the three highest loading items for bodily preoccupation as originally described by Pilowsky (Reference Pilowsky1967). The nine items therefore include the main indicators of health anxiety as defined by Pilowsky (Reference Pilowsky1967) Footnote 2.
Screening current medical status
There is evidence to support the use of self-reported medical histories as a reliable way to establish health status (Bradford et al. Reference Bradford, Graham and Reinert1993; Biossonnault & Badke, Reference Boissonnault and Badke2005) as well as physician and out-patient utilization (Cleary & Jette, Reference Cleary and Jette1984; Ungar & Coyte, Reference Ungar and Coyte1998). For example, Cleary & Jette (Reference Cleary and Jette1984) reported an average discrepancy (reported minus actual) of 0.05 visits over a 1-year period. Participants therefore answered two dichotomous items (YES or NO) derived from Ungar & Coyte (Reference Ungar and Coyte1998) to indicate their current out-patient and general practitioner (GP) utilization: (1) Are you currently receiving any out-patient treatment? and (2) Are you currently being treated by your GP for any illness? An answer of NO to both of these questions meant that the individual was defined as healthy.
Taxometric procedures
The taxometric procedures adopted conformed to the recommendations made by Ruscio & Ruscio (Reference Ruscio and Ruscio2004a, Reference Ruscio and Rusciob) and Ruscio et al. (Reference Ruscio, Haslam and Ruscio2006). A large sample, not selected on predefined criteria (e.g. scoring high on measures of symptom reporting), was recruited. Selecting on predefined criteria may lead to false taxa (known as pseudo-taxa) that reflect the method rather than the existence of a true taxon (see Ruscio & Ruscio, Reference Ruscio and Ruscio2004a, Reference Ruscio and Rusciob). To increase statistical power, indicators were selected that were scored as continuous scales. The following sequence of analyses were undertaken to ensure that the indicators: (1) covered the conceptual range of the construct, (2) had low nuisance covariance, (3) were valid and demonstrated good ability to discriminate any potential taxon (disordered group/cases) from complement (non-disordered group/non-cases) and (4) had minimal skew (Ruscio & Ruscio, Reference Ruscio and Ruscio2004a; Ruscio et al. Reference Ruscio, Ruscio and Keane2004). Initially, indicator validity was examined. Indicators should be able to distinguish taxonic cases from their complement and Meehl (Reference Meehl1995) has suggested that valid indicators should have a mean separation of 1.25 standard deviations (s.d.). Second, indictors that had high item–total correlations were chosen to represent the most valid indicators of the construct (Ruscio & Ruscio, Reference Ruscio and Ruscio2004a). Based on the indicators identified through these two steps, evidence for nuisance covariance was explored by examining the mean inter-indicator correlations in the upper and lower quartiles of the construct and in any potential complement and taxonic groups. For nuisance covariance to be tolerable, this should be less than 0.30 (Meehl, Reference Meehl1995). Positive skew can lead to the identification of pseudo-taxa with a small base rate, whereas negative skew can misidentify a high base rate taxon (see Ruscio et al. Reference Ruscio, Ruscio and Keane2004). Therefore, it is necessary to report values of skew to enable interpretation of the data. Furthermore, the programs used compare taxometric curves derived from the actual data to simulated curves. The simulated curves are derived from the actual data characteristics, including skewness, allowing a contextualized interpretation of the results and helping to avoid the identification of pseudo-taxa (Ruscio et al. Reference Ruscio, Haslam and Ruscio2006).
Taxometric analysis
Taxometric analysis was implemented using the programs developed by Ruscio and colleagues (see Ruscio & Ruscio Reference Ruscio and Ruscio2004a, Reference Ruscio and Rusciob; Ruscio et al. Reference Ruscio, Haslam and Ruscio2006; see also www.taxometricmethod.com/). The appropriately identified indicators were submitted to mean above minus below a cut (MAMBAC; Meehl & Yonce, Reference Meehl and Yonce1994), maximum eigenvalue (MAXEIGEN; Waller & Meehl, Reference Waller and Meehl1998) and L-mode factor analysis (L-MODE; Waller & Meehl, Reference Waller and Meehl1998). Details of these techniques are described in the references above and in general by Ruscio & Ruscio (Reference Ruscio and Ruscio2004a, Reference Ruscio and Rusciob) and Ruscio et al. (Reference Ruscio, Haslam and Ruscio2006). The basic rationale behind these procedures is to divide the sample along successive divisions of an input indicator and then calculate the mean difference in scores (based on the remaining indicators) above and below a cut (MAMBAC) or dividing cases into successive overlapping (90%) windows and calculating the first eigenvalue from the covariance of the remaining variables. Cuts along the indicator are presented on the x axis and mean differences or eigenvalues on the y axis. For dimensional solutions curves will be either concave (MAMBAC, MAXEIGEN) or flat or irregular (MAXEIGEN) but will be peaked for both techniques for a taxonic solution. L-MODE uses factor analytic procedures on all of the indicator variables to calculate scores on the first principal component and the distribution of these is plotted. A unimodal distribution indicates a dimensional solution and a bimodal distribution a taxonic solution. The MAMBAC models were based on 50 evenly spaced cuts. The MAXEIGEN analyses were based on 50 windows with 90% overlap.
Data analyses were conducted using taxometric programs developed by Ruscio et al. (Reference Ruscio, Haslam and Ruscio2006). These programs have two important features to aid the interpretation of the data: (1) graphical comparison of observed versus simulated data and (2) calculation of fit indices to distinguish dimensional from taxonic solutions. Based on the distributional and correlation structures observed in the actual data, these programs generate simulated taxonic and dimensional data. The programs compare the curves for the simulated and actual data to produce a root mean square residual (RMSR). The smaller the RMSR, the better the observed curve fits the curve generated from the simulated data. By producing multiple simulated taxonic and dimensional data sets, it is possible to calculate the average RMSR and its s.d. An index similar to Cohen's d (Fitd) is calculated by subtracting dimensional fit from the taxonic fit. Positive values indicate a dimensional structure and negative values a taxonic structure. Recently, Ruscio et al. (Reference Ruscio, Haslam and Ruscio2006, Reference Ruscio, Zimmerman, McGlinchey, Chelminski and Young2007) have developed the curve comparison fit index (CCFI), which is based on the RMSR fits. The CCFI varies between 0 and 1 and is symmetric around 0.5. Values above 0.5 indicate a taxonic solution and <0.5 a dimensional solution. Ruscio et al. (Reference Ruscio, Zimmerman, McGlinchey, Chelminski and Young2007) reported that Monte Carlo studies show that the CCFI more accurately distinguishes taxonic from dimensional solutions than traditional indices such as the Goodness of Fit Index (GFI; Waller & Meehl, Reference Waller and Meehl1998) and the base rate s.d. As such, the Fitd and the CCFI are reported to aid interpretation along with the visual representations derived from the MAMBAC, MAXEIG and L-MODE curvesFootnote 3.
Results
Representativeness of the sample
The taxometric analyses were applied to the sample of 501 participants who were classified as healthy (answered NO to both of the screening questions). The basic sample demographics for the whole sample and the 501 participants used in these analyses were generally representative of the UK population, which has a mean age of 39 years and is 51% female (Office for National Statistics) (see Table 1). There is inconsistent evidence with respect to the relationship between health anxiety and age or sex (Barksy et al. Reference Barsky, Frank, Cleary, Wyshak and Klerman1991; Noyes, Reference Noyes, Maj, Akiskal, Mezzich and Okasha2005; Noyes et al. Reference Noyes, Carney, Hillis, Jones and Langbehn2005). For these data, health anxiety (as indexed by both the original nine-item WI and the six-item version of the WI used in the taxometric analyses) was unrelated to age (both r's=0.04, both p's=0.69) and only marginally associated with sex (all t's >2.5, all p's <0.01), with women, compared to men, scoring marginally lower on both the full (22.6 v. 21.6) and six-item (14.4 v. 13.4) versions.
Barsky et al. (Reference Barsky, Wyshak and Klerman1990) reported that patients with clinical hypochondriasis scored on average 3.31 when the WI is averaged to range from 1 to 5. The healthy sample of 501 participants was split into ‘cases’ (those with a score ⩾3.31) and ‘non-cases’ (those who score <3.31). This split indicated that 6.6% were classed as ‘cases’, a figure that is consistent with previous reports for health anxiety in non-clinical samples (Noyes et al. Reference Noyes, Hartz, Doebbeling, Malis, Happel, Werner and Yagla2000; Looper & Kirmayer, Reference Looper and Kirmayer2001; Rief et al. Reference Rief, Hessel and Braehler2001).
Selection of indicators
For the sample of 501 healthy participants, the coefficient α for the full nine-item version of the WI was 0.73. Initially, all nine items of the WI were screened for indicator validity using Ruscio et al.'s (Reference Ruscio, Haslam and Ruscio2006) indicator.dist program. For these analyses the sample was split into cases and non-cases using the cut of 3.31 on the averaged nine-item version (cf. Barsky et al. Reference Barsky, Wyshak and Klerman1990). The indicator validities (expressed in standard units as Cohen's d), item–total correlations and skew for each indicator are presented in Table 2. Indicators showed a degree of skew that was generally within the ranges reported for other taxometric studies (Ruscio et al. Reference Ruscio, Ruscio and Keane2004, Reference Ruscio, Zimmerman, McGlinchey, Chelminski and Young2007). Three indicators had validities <1.25 (indicators 1, 7 and 9) and two (indicators 7 and 9) had unacceptable item–total correlations. When these three indicators were removed, the α for the remaining six indictors increased to 0.81. Based on these six indicators, the mean inter-indicator correlations in the bottom and top quartiles were 0.02 and 0.03 respectively and 0.42 for the full sample. The mean correlation in the putative taxon was −0.02, and 0.17 in the putative compliment. Thus, there was no evidence for nuisance covariance. This six-item index correlated 0.95 with the full nine-item scale, indicating that the six-item index covered the majority of the construct.
Table 2. Descriptive statistics for indicators
Standard error of skew=0.11.
Taxometric analyses
The taxometric curves for the MAMBAC, MAXEIG and L-MODE analyses are shown in Figs 1–3 respectively. These figures show that the curves produced by the actual data are closer to the simulated dimensional rather than taxonic data curves, supporting a dimensional interpretation. Confirmatory support for the dimensional representation is provided by the positive Fitd values for the MAMBAC (Fitd=4.15) and MAXEIGEN (Fitd=4.35) analyses and CCFI values <0.50 for the MAMBAC (CCFI=0.37) and MAXEIGEN (CCFI=0.41) analyses.
Fig. 1. MAMBAC (mean above minus below a cut) curves. Lighter smooth lines (––) show simulated data, the darker dotted line (
) the actual data. The two lighter smooth lines represent the distribution of the simulated data that are +1 s.d. and −1 s.d. from the average of all 10 simulated curves.
Fig. 2. MAXEIGEN (maximum eigenvalue) curves. Lighter smooth lines (––) show simulated data, the darker dotted line () the actual data. The two lighter smooth lines represent the distribution of the simulated data that are +1 s.d. and −1 s.d. from the average of all 10 simulated curves.
Fig. 3. L-MODE (L-mode factor analysis) curves. Lighter smooth lines (––) show simulated data, the darker, thicker line (––) the actual data. The two lighter smooth lines represent the distribution of the simulated data that are +1 s.d. and −1 s.d. from the average of all 10 simulated curves.
Discussion
The pattern of results reported in this study indicates that health anxiety is better represented as a dimensional rather than a categorical construct (Warwick & Salkovskis, Reference Warwick and Salkovskis1990). Within a dimensional conceptualization of health anxiety, individuals differ quantitatively across a dimension, with hypochondriasis marking an extreme of this distribution. These findings have a number of implications for theory and clinical utility.
Theoretical implications
A dimensional conceptualization of health anxiety has three main theoretical implications. The first implication concerns the development of theoretical models, suggesting that these should focus on additive multi-causal agents or risk factors (Ruscio et al. Reference Ruscio, Haslam and Ruscio2006: see also Meehl & Golden, Reference Meehl, Golden, Kendall and and Butcher1982). As such, integrative models of health anxiety should be developed that include: (1) a number of potential risk factors (e.g. life stress, family medical history, and attachment styles: see Noyes et al. Reference Noyes, Stuart, Langbehn, Happel, Longley, Muller and Yagla2003, Reference Noyes, Watson, Carney, Letuchy, Peloso, Black and Doebbeling2004), (2) cognitive/perceptual processing (Barksy et al. Reference Barsky, Coeytaux, Sarnie and Cleary1993; Pauli & Alpers, Reference Pauli and Alpers2002; Ferguson et al. Reference Ferguson, Moghaddam and Bibby2007), (3) traits including somatosensory amplification, neuroticism, extraversion, health attributions and attachment style (Barsky, Reference Barsky1979; MacLeod et al. Reference MacLeod, Haynes and Sensky1998; Ferguson, Reference Ferguson2000; Waldinger et al. Reference Waldinger, Schulz, Barsky and Ahern2006) and (4) biological processes (Ferguson, Reference Ferguson, Maj, Akiskal, Mezzich and and Okasha2005, Reference Ferguson2008). Such models should take a longitudinal developmental perspective examining the reciprocal relationship between potential triggers (risk factors and traits) and maintenance (cognitive processes and biological mechanisms) factors. This work should be supplemented with experimental and quasi-experimental work to clarify causal relationships. Second, as health anxiety is continuously distributed within a population, these theoretical models should be examined using large unselected samples without having to specifically examine clinical ‘cases’ (Ruscio et al. Reference Ruscio, Haslam and Ruscio2006). This also has the advantage of increasing statistical power (Cohen & Cohen, Reference Cohen and Cohen1983). Third, psychometric and clinical assessments of health anxiety should be further developed based on continua. Such measures should include items that reflect as wider a range as possible of potential symptoms and be capable of making fine-grained discriminations (Ruscio et al. Reference Ruscio, Haslam and Ruscio2006).
Clinical implications
Clinicians argue that a dimensional approach does not have clinical utility. Specifically, they argue that dimensional scores cannot be used to make a diagnosis and, as such, a dimensional approach does not have clinical utility over existing categorical systems. Recently, Widiger & Trull (Reference Widiger and Trull2007) have addressed this issue with the example of IQ. IQ is a dimensional construct that can be used to define individuals with scores <70 and direct therapeutic interventions at this group. Similarly, subtests from IQ batteries can be used to identify potential early memory and attention problems, relative to accepted norms, for those at risk of dementia. Indeed, Ruscio et al. (Reference Ruscio, Haslam and Ruscio2006) argue that it is appropriate to draw distinctions within a latent dimensional construct as long as these are systematic and empirically justifiable based on the distributional nature of the measure. This approach is similar to the use of continuous diagnostic signs in physical medicine (e.g. blood pressure, heart rate, temperature, blood sugar) that are reasoned with using a number of meaningful distinctions to reach a final differential diagnosis. The challenge, which is beyond the scope of this paper, is to develop useful clinical diagnoses treating health anxiety as a continuum. One way to achieve this with a dimensional construct is to identify ‘inflection points’ where health rapidly worsens (Kessler, Reference Kessler2002) or to conduct sensitivity and specificity analyses to identify clinical cut-offs (Fink et al. Reference Fink, Ewald, Jensen, Sorensen, Engberg, Holm and Munk-Jorgensen1999: see Ruscio et al. Reference Ruscio, Haslam and Ruscio2006 for other strategies).
Finally, the dimensional conceptualization means that it is possible to perform early stage clinical trials (stages 1 and 2) using regression approaches with large unselected samples that represent the range of scores on an index of health anxiety. In this way, early potential therapeutic effects for health anxiety can be identified prior to larger scale trials.
Caveats
The sample was based on a working population. This may lead to questions about the generalizability of the findings. However, the sample is generally representative of the UK general population. It may also be argued that the indicators used do not represent the whole construct of health anxiety. With respect to this, the measure used is a widely used standard, reliable and valid index of health anxiety and shown to be able to differentiate cases from non-cases (Barksy et al. 1986) and the six indicators used showed a high correlation with the full scale. Finally, this study examined the dimensionality of health anxiety in those without illness. However, it has been noted that in those who worry about illness, approximately 50% will have an actual illness (Looper & Kirmayer, Reference Looper and Kirmayer2001; Noyes et al. Reference Noyes, Carney, Hillis, Jones and Langbehn2005). This study, therefore, did not address the dimensionality of health anxiety in those with a current illness. However, the taxometric analyses of the whole sample (including both those with and without current illness) supported the dimensional model. The sample size for those with illness (n=198) was too small to conduct a taxometric analysis. Therefore, replications of these findings using different indicators and across sample of those with and without a current illness would be welcome.
Acknowledgements
This research was funded by a grant from the UK Health and Safety Executive award to Eamonn Ferguson and Dr Helen J. Cassaday. Thanks to Jane Ward for help with the initial data collection and data input and Peter A. Bibby, Ellen Townsend and Claire Lawrence for comments on an earlier draft of the paper. The views expressed in this paper are those of the author and do not reflect those of the UK HSE or any one involved in the research project.
Declaration of Interest
None.
