Received: 30 January 2017

Revised: 10 May 2017

Accepted: 12 May 2017

DOI: 10.1002/mpr.1574

ORIGINAL ARTICLE

Psychometric investigation of the specific phobia of vomiting inventory: A new factor model Danielle J. Maack1

|

Chad Ebesutani2

|

Todd A. Smitherman1

1

Department of Psychology, University of Mississippi, Mississippi, USA

Abstract

2

Well‐validated, standardized measures are lacking for the assessment of emetophobia, the

Department of Psychology, Duksung Women's University, Seoul, Korea

Correspondence Danielle Maack, Department of Psychology, P.O. Box 1848, University, MS 38677‐1848, USA. Email: [email protected]

specific phobia of vomiting. The Specific Phobia of Vomiting Inventory (SPOVI) was recently developed and shows promise as a useful measure of emetophobia. The goal of the present study was to further examine and investigate the psychometric properties of the SPOVI in a large student sample (n = 1626), specifically focusing on its factor structure, measurement invariance across gender, and convergent/divergent validity. Confirmatory factor analysis results provide support for a one‐factor model of the SPOVI, in contrast to the previously proposed two‐factor model. Internal consistency of the SPOVI was good (α = 0.89) and measurement invariance across gender invariance was supported. The SPOVI also demonstrated good psychometric properties with respect to convergent and divergent validity. The present study's demonstration of the reliability and validity of the SPOVI suggests that the instrument may be a valuable tool for assessing emetophobia symptoms based on its one‐factor structure. KEY W ORDS

CFA, emetophobia, measure invariance, specific phobia of vomiting

1

|

I N T RO D U CT I O N

are likely underestimates as emetophobia symptoms may mistakenly be attributed to other anxiety disorders such as panic disorder, social

Emetophobia, a specific phobia of vomiting, is a relatively under‐

phobia, obsessive–compulsive disorder, and health anxiety (see

researched anxiety disorder that has recently begun to attract

Boschen, 2007, for a review). Emetophobia can cause significant

increased attention (Boschen, 2007; Marks, 1987; van Hout &

distress (e.g. panic attacks; fear of somatic symptoms related to

Bouman, 2012; Veale & Lambrou, 2006). As a specific phobia,

vomiting; Lipsitz et al., 2001) and functional impairment (avoiding

emetophobia is currently subsumed under the “other” category of spe-

social situations; being significantly underweight from dieting restric-

cific phobias (DSM‐V; American Psychiatric Association [APA], 2013;

tions; Veale et al., 2013; Boschen, 2007), particularly among women,

ICD‐10; World Health Organization [WHO], 1992). As a basic defini-

some of whom delay pregnancy (Lipsitz et al., 2001; Maack et al.,

tion, emetophobia is the excessive fear of and preoccupation with

2013; Veale & Lambrou, 2006) for fear of morning sickness and/or fear

vomiting. The presentation of this disorder is largely diverse and poorly

of the baby vomiting.

understood (Boschen, 2007). From what has been assessed,

Despite a growing body of research examining emetophobia, only

emetophobia appears to be characterized by an early age of onset

recently have two measures been developed to scientifically assess

(Lipsitz, Fyer, Paterniti, & Klein, 2001), a chronic course (Lipsitz et al.,

and standardize evaluation of symptoms. The Emetophobia Question-

2001; Maack, Deacon, & Zhao, 2013) without successful treatment,

naire‐13 (EmetQ‐13; Boschen, Veale, Ellison, & Reddell, 2013) and the

and few (if any) periods of remittance (Lipsitz et al., 2001).

Specific Phobia of Vomiting Inventory (SPOVI; Veale et al., 2013) have

Emetophobic symptoms can be triggered by both internal stimuli (i.e.

been presented as promising measures of severity of emetophobia

catastrophic cognitions, physiological sensations) and external stimuli

symptoms. The initial measure development paper of the SPOVI

(i.e. sights/sounds related to the experience of vomiting; Maack

assessed its psychometric properties among a clinical sample and

et al., 2013). Although previously considered a rare phobia, prevalence

included a control group with no known vomiting fears; however, the

estimates range between 1.7 and 3.1% for men and 6 and 7% for

control group was noted to have insufficient variance in item

women (Hunter & Antony, 2009; Phillips, 1985), and these numbers

responses to allow for factor analysis (Veale et al., 2013). In this initial

Int J Methods Psychiatr Res. 2017;e1574. https://doi.org/10.1002/mpr.1574

wileyonlinelibrary.com/journal/mpr

Copyright © 2017 John Wiley & Sons, Ltd.

1 of 7

2 of 7

MAACK

ET AL.

study, the authors proposed a two‐factor structure underlying the

emetophobia, and distinguish such symptoms beyond general duress

SPOVI representing (1) avoidance behavior (i.e. avoiding/trying

(anxiety and depressive symptoms) and/or sensitivity to such (anxiety

to control stimuli due to a fear of vomiting), and (2) threat monitoring

sensitivity).

(i.e. worry/self‐focus on monitoring symptoms of being ill) based on Horn's parallel factor analysis. Of note, the initial study did not provide the correlation between the two proposed factors or provide an ethnic

2

METHOD

|

breakdown of participants. The SPOVI has been implemented in several studies as a

2.1

Participants

|

measure of emetophobic symptoms, including two case studies

Participants were 1626 students recruited from a large south‐

detailing emetophobia treatment where SPOVI scores were the

eastern University. The mean age of the sample was 19.04 years

main clinical outcome (Fix, Proctor, & Gray, 2016; Paulus & Norton,

(standard deviation [SD] = 2.12, range = 18–48), and the group

2016), a group design pilot study of a cognitive behavioral

consisted of 1038 females (63.8%) and 588 males (36.2%). The

intervention for emetophobia (Riddle‐Walker et al., 2016), and as

ethnic makeup of the sample was 77% Caucasian, 17.2% Black,

an analogue measure to assess clinical correlates of emetophobia

1.8% Asian, 1.8% Multiracial, 1.7% Hispanic or Latino, 0.2% Pacific

in a college sample (Wu, Rudy, Arnold, & Storch, 2015). Addition-

Islander, and 0.2% Native American or Alaskan Native. Informed

ally, it has been used to distinguish between individuals who

consent was obtained from all individual participants included in

restricted food on the basis of fear of vomiting and those exhibiting

the study.

similar eating pathology who did not endorse such fear (Veale, Costa, Murphy, & Ellison, 2012). To date, however, the only study of the SPOVI assessing its

2.2

Measures

|

2013). Further psychometric investigation of the instrument in a

2.2.1 | Specific phobia of vomiting inventory (SPOVI; Veale et al., 2013)

normative sample would enhance research in this specific area.

The SPOVI is a 14‐item, self report questionnaire of emetophobia

In particular, the distinction between “normal” and “abnormal” is a

symptom severity. Each item is scored on a Likert‐type scale ranging

fundamental element of understanding the psychometric validity and

from 0 (not at all) to 4 (all the time). Total score ranges from 0 to 56,

clinical utility of psychological measures (Nunnally & Bernstein,

with higher scores reflecting greater endorsement of symptoms/

1994). For example, establishing normative values through rigorous

severity. A cutoff score of 10 is considered a positive screen of

research facilitates comparison of subsequent findings across settings

emetophobia (Veale et al., 2013). From the initial development

(i.e. community versus clinical versus university samples) in terms of

study, internal consistency of the overall scale ranged from

symptom intensity, frequency, duration, and resultant functional

α = 0.81 (community sample) to α = 0.91 (clinical sample), with

impairment. A nuanced view of these distinctions can, in turn, lead to

subscale consistencies (only assessed previously in the clinical

greater understanding of etiology, course, and treatment, particularly

sample) of α = 0.85 (avoidance) and α = 0.88 (threat monitoring;

to the degree programmatic research establishes instruments that are

Veale et al., 2013). In the present sample the overall internal consis-

psychometrics is the original scale development paper (Veale et al.,

sensitive to treatment change. As such, the present study aimed to further examine and

tency was α = 0.89 (avoidance subscale α = 0.85; threat monitoring subscale α = 0.78).

investigate the psychometric properties of the SPOVI among a student (as most anxiety disorders disproportionately affect women; Seedat

2.2.2 2007)

et al., 2009), and convergent/divergent validity. Given the SPOVI's

The ASI‐3 (Taylor et al., 2007) is an 18‐item measure that assesses

iterative development based on interviews with individuals seeking

the degree to which participants fear perceived negative conse-

emetophobia treatment (Veale et al., 2013), the transdiagnostic nature

quences of anxiety symptoms. Items are rated on a 5‐point Likert‐

of anxiety disorders more generally, and the limited research to sug-

type scale from 0 (very little) to 4 (very much) with total scores rang-

gest otherwise, a priori hypotheses suggested that the factor structure

ing from 0 to 72. The ASI‐3 measures three theoretically derived

in a normative sample would emulate that shown in the initial instru-

facets of anxiety sensitivity: physical (e.g. “It scares me when my

ment development study. Similarly, given a lack of evidence for differ-

heart beats rapidly”), cognitive (e.g. “When I feel ‘spacey’ or spaced

ential responses to items across genders, and inclusion of both men

out, I worry that I may be mentally ill”), and social concerns (e.g. “It

sample, focusing specifically on factor structure, gender invariance

|

Anxiety sensitivity index‐3 (ASI‐3; Taylor et al.,

and women in iterative interviews conducted in the original study

is important not to appear nervous”). Initial validation demonstrated

(Veale et al., 2013), it was also hypothesized that the instrument would

that the ASI‐3 possessed sound psychometric properties as examined

exhibit gender invariance in the current sample. Finally, to assess

across a number of sites using diverse participants (Taylor et al.,

convergent and divergent validity for the SPOVI, the anxiety and

2007). However, the total ASI‐3 score was used in the present anal-

depression subscales of the 21‐item Depression, Anxiety and

ysis as it has been found to be a better predictor (accounting for

Stress Scale (DASS‐21) and the Anxiety Sensitivity Index‐3 (ASI‐3)

50% of the variance) of anxiety sensitivity than use of the separate

were used. Theoretically, and in line with the initial development

subscales (Ebesutani, McLeish, Luberto, Young, & Maack, 2014;

study, these measures were selected to help determine if the

Osman et al., 2010). In the present sample, overall internal consis-

SPOVI is able to assess and discern symptomatology related to

tency was good (α = 0.89).

MAACK

3 of 7

ET AL.

2.2.3 | Depression, anxiety and stress Scales – 21‐item version (DASS‐21; Lovibond & Lovibond, 1995a)

procedures available in Mplus, which is recommended when using

The DASS‐21 is a 21‐item self‐ report questionnaire designed to

sages/11/1535.html).

the WLSMV estimator (see www.statmodel.com/discussion/mes-

assess the core symptoms of depression, anxiety, and stress. Items are measured on a Likert‐type scale ranging from 0 “Did not apply to

2.4.3

me at all” to 3 “Applied to me very much, or most of the time” with sub-

To compare nested models, the chi‐square (χ2) difference test (Bentler

scale scores ranging from 0 to 21. The DASS‐21 has demonstrated

& Bonett, 1980) was performed. The “difftest” function in Mplus was

adequate test–retest reliability (Brown, Chorpita, Korotitisch, &

used to compute the χ2 difference test (as opposed to hand calculating

Barlow, 1997), and there is extensive evidence for its construct and

the χ2 difference test). This method was chosen because limited

discriminant validity (Antony, Bieling, Cox, Enns, & Swinson, 1998;

information estimators (such as the WLSMV estimator) necessitate

Brown et al., 1997; Clara, Cox, & Enns, 2001; Lovibond & Lovibond,

estimation of degrees of freedom and because the differences

1995a; Lovibond & Lovibond, 1995b). Both the anxiety and depression

between χ2 values are not distributed as χ2 values when limited estima-

|

Model comparisons

symptom severity subscales of the DASS‐21 were examined in

tors are employed. The Mplus Technical Appendices (at http://www.

this study for assessment of convergent/divergent validity. Internal

statmodel.com/ download/webnotes/webnote10.pdf; Asparouhov &

consistency estimates (α) in this sample were acceptable for the

Muthen, 2006) and the Mplus User's Guide (Muthen & Muthen,

anxiety subscale = 0.78 and good for both the depression = 0.83 and

2010) outline how to execute the difftest function in Mplus and

stress subscales = 0.82.

indicate how degrees of freedom are estimated.

2.3

2.4.4

Procedure

|

|

Haberman analyses

The current study was part of a larger psychology research screening.

The relative reliability and utility of the subscales compared to the total

All procedures performed in studies involving human participants were

score (when appropriate) was examined via Haberman analyses

in accordance with the ethical standards of the institutional and/or

(Haberman, 2008). Haberman Analyses examine whether a subscale

national research committee and with the 1964 Helsinki declaration

score is a better predictor of an individual's “true score” of the trait

and its later amendments or comparable ethical standards. After

targeted by that subscale compared to using the total score. The

informed consent was obtained, participants completed all measures

Haberman procedures involve computing the proportional reduction

via an online survey system. Students received research or course

in mean squared error based on the total score (PRMSEtotal) and

credit for participation.

comparing that value to the proportional reduction in mean squared error based on the subscale score (PRMSEsubscale), the latter of which

2.4 2.4.1

Data analytic strategy

| |

Confirmatory factor analysis (CFA)

is typically estimated by the reliability alpha coefficient (see Reise, Bonifay, & Haviland, 2013). If PRMSEsubscale > PRMSEtotal, then the subscale provides a more reliable indicator of the subscale's true

Using confirmatory factor analysis (CFA) via Mplus version 7.11

score relative to the total score. If, however, PRMSEtotal >

(Muthen & Muthen, 2010), the fit of competing factor structures of

PRMSEsubscale, then the total score captures at least the same

the SPOVI was examined. Polychoric correlations and the robust

amount of variance as the subscale, and thus scoring and interpreting

weighted least‐squares with mean and variance adjustment (WLSMV)

subscales is not necessary (see Reise et al., 2013).

estimator were used because the analyses were based on Likert‐ response options that produce categorical (ordinal) data. The WLSMV

2.4.5

|

Measurement invariance across gender

estimator is one of the most recommended estimators to use when

Multi‐group CFA (MG‐CFA) and the MG‐CFA procedures outlined by

conducting CFA with categorical data (Flora & Curran, 2004;

Brown (2006) were employed to examine the measurement invariance

Holgado‐Tello, Chacon‐Moscoso, Barbero‐Garcia, & Vila‐Abad, 2010;

of the best‐fitting model across gender. First, single‐sample solutions

Muthen, du Toit, & Spisic, 1997). The following fit indices and cutoffs

were examined for good fit within each of the male‐only (n = 588)

were used to examine model fit of the tested structures: the root mean

and female‐only (n = 1038) subsamples separately. The tested model

square error of approximation (RMSEA; Steiger, 1990); Comparative

should fit well (based on the aforementioned fit indices and cutoffs)

Fit Index (CFI; Bentler, 1990), and Tucker–Lewis Index (TLI; Tucker &

in each sample separately before proceeding to further MG‐CFA

Lewis, 1973). CFI values greater than 0.90 (Bentler, 1990) and 0.95

invariance tests. Following the examination of single‐sample solutions,

(Hu & Bentler, 1999) indicated acceptable and good model fit, respec-

configural invariance (also known as the test of “equal form”) was

tively. RMSEA values lower than 0.08, and 0.05 indicated acceptable

examined. This test is conducted on the full sample and tests whether

and good fit, respectively (Browne & Cudeck, 1993).

the overall number of factors and the item‐to‐factor grouping patterns across all factors are the same across genders. Support for configural

2.4.2

|

Missing data

Relatively few participants had missing data on the SPOVI: 1578

invariance is based on whether the fit indices achieve the aforementioned benchmarks for good model fit (cf. Brown, 2006).

(97.0%) had no missing data, 44 (2.7%) had only one missing item,

If configural invariance is supported, metric invariance can then be

three (0.2%) had two missing items, and one (0.1%) had four

tested. Metric invariance (also referred to as the test of “equal factor

missing items. Missing data were addressed using pairwise present

loadings”) refers to whether the factors have the same meaning (and

4 of 7

MAACK

ET AL.

thus the same factor loading strengths) across groups. This is tested by

WRMR = 1.674). The fit indices relative to the 1‐factor model

imposing the equality constraint for factor loadings between sub-

(RMSEA = 0.069; CFI = 0.961, TLI = 0.953; WRMR = 1.717) were

groups (e.g. across males and females). Support for metric invariance

not substantially different. All factor loadings of the 2‐factor model

was determined using the recommended ΔCFI difference test (Chen,

were significant (p < 0.001), ranging from 0.65 to 0.96. When com-

2007). If the difference in the CFI fit index of the metric invariance

pared to the 1‐factor model using the χ2 difference test, the 2‐factor

and configural invariance model is less than 0.01 (i.e. ΔCFI