Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

Australian Journal of Psychology 2017

doi: 10.1111/ajpy.12192

Rasch analysis of the Frost Multidimensional Perfectionism Scale

Kate Mitchell-Parker ,1 Oleg N. Medvedev ,2 Christian U. Krägeloh ,1 and Richard J. Siegert1


1
Department of Psychology, Auckland University of Technology and 2Centre for Medical and Health Science Education,
School of Medicine, Faculty of Medical and Health Sciences, University of Auckland, Auckland, New Zealand

Abstract

Objective: The Frost Multidimensional Perfectionism Scale (FMPS) is widely used to measure the construct of perfectionism.
Previous studies evaluating the factor structure of the FMPS have reported inconsistent findings. The study objective was to
examine the psychometric properties of proposed four, five, and six factor solutions of the FMPS using Rasch analysis. Method:
Using the responses from a sample of New Zealand athletes and exercisers (n = 425) from a variety of sports, we subjected the
dataset to Rasch analysis. The overall and individual item fit, unidimensionality, local independence, and person separation
reliability were evaluated by treating factors of each model as subtests. Results: After disordered thresholds were uniformly
rescored, the factor Organisation displayed significant misfit to the model across all three solutions tested. Removal of Organisa-
tion and combining two parental facets into one resulted in adequate model fit and good item discrimination ability for all three
solutions. The modified six-factor model was the best in terms of reliability, with no differential item functioning and the high-
est person separation index. Conclusion: The present results support the psychometric properties and internal structural validity
of the rescored FMPS with the parental factors combined and without the Organisation facet. Psychometric properties of the
instrument can be further enhanced by using the ordinal-to-interval conversion table presented here, which does not require
modification of the original response format. These findings will be of interest in different areas where accurate assessment of
the overall perfectionism trait is important and open new avenues for perfectionism research.

Key words: Frost Multidimensional Perfectionism Scale, item response theory, measurement, perfectionism, psychometrics, Rasch
analysis

What is already known about this topic What this topic adds
• Perfectionism has been associated with high achieve- • Rasch analysis provided a unique and powerful
ment but also psychological problems. method for evaluating psychometric properties and in
• Psychometric evaluations of the FMPS have been particular internal construct validity of the FMPS.
unable to establish a stable factor structure. • The current Rasch analysis suggests that Organisation
• Methodological issues have affected the research to may not be psychometrically integrated with the con-
date, limiting the ability of researchers to resolve the struct of perfectionism and two parental factors reflect
debate around the factor structure of the FMPS. a single factor of Parental Influence.
• The psychometric properties of the FMPS are
enhanced by using the ordinal-to-interval conversion
Interest in the personality trait of perfectionism has table generated by this study, which does not require
increased markedly in recent years. Perfectionism was origi- modification of the original response format of the
nally conceptualised as a unidimensional concept, with the scale.
prominent characteristic being the setting of excessively
high standards accompanied by overly critical self- Rosenblate (1990) proposed that perfectionism is multidi-
evaluation (Burns, 1980). Frost, Marten, Lahart, and mensional with the overarching latent perfectionism trait,
which was supported empirically, but its basic components
remain debated (Anshel, Weatherby, Kang, & Watson,
Correspondence: Oleg N. Medvedev, Centre for Medical and Health
Science Education, School of Medicine, Faculty of Medical and
2009). Early research has linked perfectionism with a num-
Health Sciences, University of Auckland, Auckland 1142, ber of psychological conditions such as the development
New Zealand. Email: o.medvedev@auckland.ac.nz and maintenance of eating disorders (Bardone-Cone et al.,
Received 12 January 2017. Accepted for publication 21 2007), depression (Hewitt, Flett, & Ediger, 1996), anxiety
November 2017. (Flett, Greene, & Hewitt, 2004), and obsessive compulsive
© 2017 The Australian Psychological Society
2 K. Mitchell-Parker et al.

disorders (Martinelli, Chasson, Wetterneck, Hart, & Björg- the parental factors have been shown to fit consistently
vinsson, 2014). Perfectionism has also been related to anger within the perfectionistic strivings and perfectionistic con-
and rumination (Besharat & Shahidi, 2010), low self-esteem cerns framework (Stoeber & Otto, 2006).
(Gotwals, Dunn, & Wayment, 2003), and suicidal ideation
(Flett, Hewitt, & Heisel, 2014). THE MULTIDIMENSIONAL PERFECTIONISM SCALE

CONCEPTUALISATION OF PERFECTIONISM Drawing on the original descriptions of perfectionism, a


multidimensional scale was developed by Frost et al. (1990)
At the heart of the current perfectionism debate lies a dis- to measure five hypothesised dimensions of this construct.
agreement over the concept of perfectionism—what exactly Existing measures of perfectionism, all of which were sub-
is meant by the word ‘perfectionism’. Early clinical descrip- scales of larger scales, were examined and relevant items
tions of perfectionism by Burns (1980), Hamachek (1978), from these subscales were included in the initial item pool
Hollander (1965), Pacht (1984), and Reed (1985) emphasise of the FMPS. Frost et al. (1990) used exploratory factor
excessively high self-imposed personal standards as being analysis (EFA) and reliability analysis with a sample of
central to perfectionism. Accompanying self-evaluation is 232 female undergraduate students to conclude that perfec-
characterised by a tendency to be overly critical of one’s tionism consists of six factors: Personal Standards, Concern
behaviour or performance. Perfectionists are also proposed over Mistakes, Doubts about Actions, Parental Criticism,
to place a high value on their parents’ expectations and Parental Expectations, and Organisation. The factor Personal
evaluations of them. Failure to perform to the highest levels Standards can be defined as the setting of high standards for
of perfectionism will potentially result in a loss of parental evaluation of one’s own behaviour. Concern over Mistakes
love and approval (Frost et al., 1990). Burns (1980), Hama- is described as perceived failure to comply with set stan-
chek (1978), and Reed (1985) identified a self-evaluative dards, reflecting a negative reaction to errors. Doubts about
tendency to experience uncertainty or doubt regarding the Actions refers to uncertainty about the quality of one’s per-
quality of one’s performance. A desire for order and neat- formance. Parental Criticism refers to a perception that
ness was described as a feature ‘sometimes’ used to charac- one’s parents are or were overly critical, and Parental
terise perfectionists (Hollander, 1965). Expectations is a perception that one’s parents set extremely
Hamachek (1978) argued that some aspects of perfection- high standards. Lastly, Organisation refers to the need for
ism might be adaptive since they foster excellence and striv- order and neatness. Of the 35 items, some address current
ing to meet important goals. Hamachek termed this more beliefs (e.g., ‘People will think less of me if I make a mis-
adaptive form of perfectionism ‘normal perfectionism’, take’), whereas others refer to past situations (e.g., ‘My par-
while an individual who sets high standards but allows little ents wanted me to be the best at everything’). Using the
to no room for error was termed ‘neurotic perfectionist’. FMPS, a total perfectionism score is calculated by adding
Almost 40 years on from Hamachek’s original article, a the responses from all of the factors except for Organisation,
large body of research supports the idea that two basic which Frost et al. (1990) did not include in the calculation
forms of perfectionism can be distinguished, widely known of an overall perfectionism score. All items are in the form
as perfectionistic strivings and perfectionistic concerns of a statement, and responses are made by way of a 5-point
(Stoeber, 2017). Likert-type scale ranging from ‘strongly disagree’ to
An important question regards whether the various ‘strongly agree’. The data obtained from the FMPS is ordinal
hypothesised dimensions of perfectionism are core elements in nature with potential scores ranging from 35 (lowest) to
or causes of perfectionism. The four major contributors to 175 (highest).
the original perfectionism literature (Burns, Hamachek, This six-factor measure of perfectionism as proposed by
Hollander and Pacht) identified parents’ expectations and Frost et al. (1990) has been critiqued by a number of
evaluations as a cause of perfectionism. As Shafran, Cooper, researchers including Rhéaume, Freeston, Dugas, Letarte,
and Fairburn (2002) suggest, the impact of not meeting and Ladouceur (1995), Shafran et al. (2002), and Stöber
your parents’ expectations (resulting in a negative self-eval- (1998). Initial problems were identified with the construc-
uation) is the feature which is relevant to the construct of tion of the FMPS. When attempting to replicate the initial
perfectionism, rather than the parental approval or disap- EFA results with a second sample of participants (Frost
proval itself. A similar debate is raised for the element of et al., 1990), the second EFA produced the same six-factor
Organisation, which is not believed to be related to setting solution but several of the 36 items from the initial solution
high standards or self-evaluation. Rather, Organisation is a loaded higher on a different factor than what was observed
way in which individuals go about achieving personal stan- for the initial sample. The final assignment of items to the
dards (to reach high personal standards one must be orga- subscales was based on the desire for conceptual integrity of
nised) (Hollander, 1965). Of note, neither Organisation nor the subscales and to maximise reliability. This resulted in
© 2017 The Australian Psychological Society
Rasch analysis of the FMPS 3

the replacement of two items and elimination of a third Concern over Mistakes, High Standards, Organisation, Par-
item to attain the final 35-item version; however, several enting, and Doubts about Actions using both EFA and CFA.
items from the Parental Expectations scale were still cross- Hawkins et al. (2006) also used both EFA and CFA to pro-
loading on the Parental Criticism factor. pose a four-factor 33-item solution (items 16 and
The second issue with the FMPS lies in the pattern of low 18 excluded due to cross-loadings identified), with an ade-
correlations with some subscales of the instrument when quate model fit. The final four-factor 33-item solution con-
compared with other measures of perfectionism (Frost sisted of Organisation, Parental Expectations and Criticism,
et al., 1990). The Organisation subscale was excluded from Concern over Mistakes and Doubts, and Personal Standards.
calculation of the total score because of its minimal or non- With increasing use of the FMPS to assess perfectionism
significant correlations with other measures of perfection- for research and clinical purposes, a close inspection of psy-
ism (Frost et al., 1990). The factor Concern over Mistakes chometric properties is imperative. Because the FMPS is a
appears most central to the Frost concept of perfectionism, multidimensional measure, its factorial structure is espe-
given that this subscale has stronger correlations with cially relevant. Evaluation of the current models of perfec-
related constructs and contributes the most variance (25%) tionism will advance theoretical understanding of ‘what’
to the FMPS. The other subscales have been proposed as perfectionism is, as measured by the FMPS. Moving closer
being correlates of perfectionism, such as being doubtful towards agreement on what components make up perfec-
about actions, or personality traits, such as being highly tionism will have important consequences for perfectionism
organised (Shafran et al., 2002). It has therefore been sug- research and treatment approaches. If the FMPS is not sup-
gested that current multidimensional measures—including ported as being stable with six factors, yet the six-factor
the FMPS—do not assess perfectionism per se, but assess FMPS is still widely used for research purposes, then there
related constructs (Shafran & Mansell, 2001). is a need to consider revising the original FMPS and encour-
Despite the reservations noted above, the FMPS was aging researchers to use the modified FMPS instead. Theo-
adopted by researchers in personality and clinical psychol- retical and practical implications of revising the FMPS
ogy. The widespread use of the FMPS has led to the accep- would include (1) new understandings of the perfectionism
tance of perfectionism as a multidimensional construct. construct as measured by the FMPS, and (2) scale refine-
Shafran et al. (2002) suggest that this acceptance has ment which may result in the FMPS demonstrating ade-
resulted in the concept of perfectionism being too readily quate psychometric properties for research purposes.
equated with its method of measurement. Rather than Removal of factors which are not core components of per-
being an independent theoretical construct, ‘perfectionism’ fectionism would have further implications for our under-
is now usually equated with the score on the multidimen- standing of how perfectionism relates to psychopathologies,
sional self-report instrument designed to assess it. The major individual performance (e.g., in sports, musical, and aca-
difficulty with this regards whether the measurement tool is demic contexts), and other constructs of interest.
actually measuring the concept of perfectionism accurately.
Empirical investigations into the factor structure may thus RASCH ANALYSIS
help advance this debate.
To date, no reports are available investigating the psycho-
FACTOR STRUCTURE OF THE MULTIDIMENSIONAL metric properties of the FMPS using modern item response
PERFECTIONISM SCALE theory (IRT); specifically the Rasch model (Rasch, 1960).
Rasch analysis can be useful for investigating psychometric
Studies reporting on the factor structure of the FMPS have properties of an instrument, given its advantages over
produced inconsistent results. Stallman and Hurst (2011), approaches of classical test theory (CTT) such as EFA and
using both EFA and confirmatory factor analysis (CFA), CFA, as documented below. The current study selected
provided evidence that a five-factor model of the FMPS pro- three best supported models to be subjected to Rasch
vides a stable factor structure. Using a similar methodology, analysis by treating factors as subtests based on Lundgren-
a three-factor model has been proposed (Purdon, Antony, & Nilsson, Jonsdottir, Ahlborg, and Tennant (2013) approach.
Swinson, 1999), along with a four-factor model (Hawkins, The six-factor model of Frost et al. (1990) was chosen as it
Watt, & Sinclair, 2006; Stöber, 1998), a five-factor model is the original model of perfectionism, and the original
(Cox, Enns, & Clara, 2002; Stallman & Hurst, 2011), and a FMPS continues to be widely used in research (Flett &
six-factor model (Frost et al., 1990). Only a few studies Hewitt, 2015). The five-factor 29-item model of Stallman
were able to support some of these proposed factor struc- and Hurst (2011), and the four-factor 33-item model pro-
tures of the FMPS using CFA. Stallman and Hurst (2011) posed by Hawkins et al. (2006) were chosen for analysis as
reduced the number of items to 29 due to low factor load- both research teams used a combination of EFA and CFA to
ings (less than 0.45) and identified five factors including support their relevant factor structures.
© 2017 The Australian Psychological Society
4 K. Mitchell-Parker et al.

The Rasch model is an advanced measurement approach, Furuya, & Altenmüller, 2016; Wirtz, Ehlert, Kottwitz, La
which is able to overcome some limitations of CTT such as a Marca, & Semmer, 2013). The present study enables
lack of control over the difficulty level of scale items and researchers to analyse FMPS data using parametric statistics
appropriate ordering of ordinal response categories. For and to make valid comparisons with physiological data, as
instance, in CTT an overall score is calculated as a sum of all ordinal-to-interval conversion tables have been produced.
the scale items but individual items explain different Rasch analysis examines several parameters such as local
amount of information about the construct, and therefore independence assumptions, item bias, unidimensionality,
cannot be treated as equal contributors to the total score and the ordering of item and response options in polyto-
(Stucki, Daltroy, Katz, Johannesson, & Liang, 1996). This mous items (Kersten & Kayes, 2011). When fit to the Rasch
fact is also reflected by CTT where individual items have dif- model is achieved, these parameters satisfy the model
ferent loadings on their respective factor and hence contrib- requirements and the participants can be ordered according
ute different values to a latent trait; however, CTT does not to their level of ability on a scale measuring the latent trait
account for this when estimating a total score. In Rasch (e.g., perfectionism), and the items ordered by the level of
analysis, both the item and person parameters are placed difficulty (e.g., the level of perfectionism assessed by an
along the same log-odds interval scale, which displays the item). The overall global perfectionism score suggested by
spread of item thresholds. Unlike CTT, Rasch analysis pro- Frost et al. (1990) implies that there is an overarching latent
vides a person-item-threshold map, which can be presented trait of perfectionism. Therefore, the structure representing
graphically to illustrate how well the range of item difficul- this latent trait should satisfy expectations of the unidimen-
ties covers the abilities or qualities of a study sample sional Rasch model if individual factors are treated as
(Tennant & Conaghan, 2007). An additional benefit of subtests, as was demonstrated by Lundgren-Nilsson
Rasch analysis is the identification of item bias, or Differen- et al. (2013).
tial Item Functioning (DIF). DIF is observed if participants Theory and research on perfectionism continues to
with the same level on the latent trait (i.e., perfectionism), evolve (Stoeber, 2017). Given that perfectionism is evi-
but from different groups (e.g., male vs female) respond dif- denced to play an important role in the aetiology, main-
ferently to an item. This bias can be consistent across a trait tenance, and course of certain psychopathological states,
(uniform) or not consistent across a trait (non-uniform). it is vital to advance knowledge around which dimen-
Rasch analysis can correct for uniform bias, but generally sions underpin this phenomenon. Comparing three previ-
items with a non-uniform bias should be removed from the ously proposed models using a new technique—namely
scale (Kersten & Kayes, 2011). Furthermore, while the psy- Rasch analysis—will contribute unique evidence to the
chometric properties of the FMPS are generally acknowl- ongoing debate of which factors should be included in
edged to be acceptable, the use of CTT may not discriminate the FMPS.
accurately between individual perfectionism levels. Rasch
analysis enables precise measurement of individuals at
METHOD
extremes of the scale range (Hobart & Cano, 2009).
Increased measurement precision is relevant to a construct
Participants
such as perfectionism, where researchers are interested in
extreme scores at the top and bottom ends of the scale. The target population for this study was athletes and people
A notable advantage of Rasch analysis over CTT is the who exercise regularly (at least two times per week). We
ability of Rasch to transform ordinal scores into interval- expected that these individuals will have higher levels of
level data that increase precision of measurement, which perfectionism than non-exercising people (Gill & Dzewal-
has been demonstrated empirically (Norquist, Fitzpatrick, towski, 1988; Picard, 1999). Participants numbered
Dawson, & Jenkinson, 2004). 425 (129 male, 296 female) individuals who were either
The original FMPS is ordinal in nature, with the measure regular exercisers or competitive athletes dwelling in
providing a ‘manifest score’ of perfectionism. Also as a result New Zealand. The following age groups were created based
of the FMPS being ordinal, arithmetic calculations such as on age distribution for the purpose of DIF testing:
addition and subtraction are impracticable (Stucki et al.,1996). 16–25 years (n = 111), 26–30 years (n = 92), 31–39 years
The FMPS should not be used with parametric statistics such (n = 108), 40+ years (n = 114). In terms of their current
as analysis of variance without violating fundamental statisti- level of competitive sport involvement, 169 participants did
cal assumptions of these tests. Validity of comparisons with not compete in a sport but exercised regularly; 132 com-
neurophysiological data (e.g., skin conductance, heart rate) is peted at a local level; 36 competed at a regional level; 50 at
compromised. This is an important consideration as a national level; and 38 competed at an international level.
researchers are more frequently correlating the FMPS with The current sample size is above the recommended sample
physiological data as interest in perfectionism grows (Ioannou, size estimates for Rasch analysis (Linacre, 1994).
© 2017 The Australian Psychological Society
Rasch analysis of the FMPS 5

Measures the overall person and item fit residuals should have a


mean close to 0.00 and a standard deviation (SD) close to
The FMPS (Frost et al., 1990) is a 35-item self-report ques-
1.00. Individual items should have fit residuals within the
tionnaire that contains six subscales: Concern over Mis-
range of −2.50 and +2.50. All items are required to be
takes, Doubts about Actions, Personal Standards, Parental
locally independent, which can be tested by examining the
Expectations, Parental Criticism, and Organisation. Partici-
residual correlations between items. A residual correlation
pants respond to a 5-point Likert type scale with anchors of
with a magnitude more than 0.20 compared to the mean of
‘Strongly disagree’ (1), ‘Disagree’ (2), ‘Neither agree nor dis-
all residual correlations is a sign of local dependency
agree’ (3), ‘Agree’ (4), and ‘Strongly agree’ (5). The internal
(Christensen, Makransky, & Horton, 2016; Marais &
consistency of the subscales range from 0.77 to 0.93, and is
Andrich, 2008). The items which are locally dependent are
0.90 for the total perfectionism score. Participants also com-
combined into a subtest (Wainer & Kiely, 1987), and the
pleted a demographic questionnaire which collected data on
overall model fit is re-tested. Checks are made to ensure
gender, age, exercise frequency (during a typical week),
that there is no significant DIF for age, gender, and other
and current level of competitive sport involvement (non-
personal factors. If any of the examined items shows signifi-
competitive, local, regional, national, or international).
cant DIF for a specific personal factor, that item can be split
into relevant categories, without the need to remove the
Procedure item in question (Wainer & Kiely, 1987). This is then fol-
An email was sent to local New Zealand sports clubs and fit- lowed by re-testing of the overall model fit.
ness facilities inviting interested members to participate in RUMM2030 software has inbuilt function to test dimen-
an online survey assessing perfectionism in exercisers and sionality based on the method developed by Smith (2002).
athletes. The sports clubs were selected as a convenience The person locations for two groups of items with the high-
sample based on the researcher’s contacts. Inclusion criteria est positive and the highest negative loadings on the first
was age (16 years and over) and exercise habits (must usu- principal component are compared using an independent
ally exercise at least once per week). Participating individ- samples t-test, after a latent trait factor is removed. The per-
uals were directed, via an email from their sports club or centage of significant t-tests will be computed together with
fitness facility, to click on a URL hyperlink to access the the 95% binominal confidence interval (CI) by consider-
online survey. The participant information sheet appeared ing a sample size. If the number of significant t-tests is less
on the computer screen before the questionnaire com- than 5% or the lower bound of the binominal CI calculated
menced. All questionnaires were administered online. All for this percentage of significant t-tests is less than 5%
instructions and items in the questionnaires were in (e.g., 4%), then unidimensionality is supported.
English. Ethical clearance was received by the authors’ uni- Reliability of the scale is assessed by the person separation
versity ethics committee. index (PSI), the Rasch numerical equivalent of Cronbach’s
α. The PSI indicates how well the scale (represented by item
thresholds) can discriminate among different trait levels in a
Statistical analysis
sample. Ordinal data are converted to interval-level data
Rasch analysis was conducted using the software using the software RUMM2030 (Andrich et al., 2009).
RUMM2030 produced in RUMM Laboratory in Perth,
Australia (Andrich, Sheridan, & Luo, 2009). Generally, the RESULTS
Rasch analysis followed sequential steps, as has been
described by Siegert, Tennant, and Turner-Stokes (2010). Likelihood ratio test was conducted and confirmed suitabil-
This included investigation of the performance of individual ity of the data for the Partial Credit model
items through DIF analysis or threshold ordering, but also (χ 2(101) = 329.07, p = .001). Table 1 summarises the over-
the structure of the scale through testing for all Rasch model fit statistics for the initial and modified ana-
unidimensionality. lyses of tested models. Initial analysis of the full FMPS scale
In Rasch analysis, a number of fit indices are considered showed an overall poor fit with significant item-trait inter-
to decide whether the data show acceptable fit to the model. action (χ 2(210) = 652.72, p = .001) and multidimensional-
The item-trait interaction, which is represented by an over- ity (33% of significant t-tests), but overall reliability of the
all and individual item chi-square fit statistic, should not be scale was good as indexed by a PSI of 0.92 (Table 1).
statistically significant (p > .05, Bonferroni adjusted). A sta- Table S1, Supporting Information, provides the individual
tistically significant result on the chi-square fit statistic indi- item fit statistics for the initial analysis. At the individual
cates that scale/items work differently at different levels of item level, 3 of the 6 Organisation items (items 7, 27, and
the latent trait, which violates the assumption of the Rasch 31) were misfitting, which was the highest ratio (50%)
measurement model. In the case of an excellent model fit, compared to the other factors of perfectionism. The factor
© 2017 The Australian Psychological Society
6 K. Mitchell-Parker et al.

Table 1 Summary of fit statistics for the initial and the modified Rasch analyses of the FMPS models
Independent t-test
Item fit residual Person fit residual Goodness of fit
Analyses Value/SD Value/SD χ (df )
2
p PSI ST < 5% %LB
Full scale −0.11 2.05 −0.46 1.89 652.72 (210) <.001 .92 32.71 >5
Frost model
Initial −0.42 2.60 −0.43 1.13 154.87 (54) <.001 .75 9.88 >5
Modified −0.25 2.81 −0.44 1.07 39.91 (36) .30 .78 4.71 2.63
Stallman model
Initial −0.67 2.91 −0.44 1.02 85.79 (45) <.001 .64 10.82 >5
Modified −0.47 3.04 −0.42 0.98 8.45 (16) .93 .73 4.24 2.16
Hawkins model
Initial −0.97 3.22 −0.50 1.04 74.63 (36) <.001 .67 10.82 >5
Modified −0.29 2.67 −0.49 1.07 21.44 (27) .77 .76 3.30 1.22
Note. LB, lower bound of the 95% confidence interval; ST, significant t-tests in %.

Concern over Mistakes had 3 misfitting items (items 9, 21, further improved the overall model fit, but the chi square
and 25). Parental Expectations had 1 misfitting item (item was still significant (χ 2(54) = 62.74, p = .04). At this stage
15), as did Personal Standards (item 16) and Doubts about the residual correlation matrix was examined, and a strong
Actions (item 33). Also, 27 of 35 items displayed disordered residual correlation that exceeded the 0.20 cut-off point
thresholds around the middle category (neither agree nor above the mean of all residual correlations (Christensen
disagree), and the remaining 8 items had only marginally et al., 2016) was found between subtests 3 (Parental Expec-
ordered thresholds. Table S1 provides the individual item fit tations) and 4 (Parental Criticism), indicating local depen-
statistics for the initial analysis before rescoring. Visual dency between these subtests. Therefore, these subtests
examination of response probability curves revealed that were combined into one subtest to solve local dependency
the thresholds of the middle category are consistently issue, and the best model fit was achieved (χ 2(36) = 39.91,
located close to the ‘slightly disagree’ category, which is p = .30). The scale also showed acceptable reliability and
illustrated in Fig. 1a (upper panel, initial) and representative
was strictly unidimensional (Table 1, Frost model modified).
for the majority of items. Therefore, it was decided to
No DIF was identified for personal factors age, gender, or
rescore all items uniformly by combining the category
level of sports performance. The modified Rasch solution
‘slightly disagree’ with ‘neither agree nor disagree’. Please
based on the Frost et al. (1990) model consisted of four fac-
note that the Rasch software, RUMM2030, converts Likert-
tors: Concern over Mistakes, Personal Standards, Doubts
scale responses to a format where the lowest response
about Actions, and Parental Influences.
option is scored as 0. Fig. 1 thus shows 5-point Likert-scale
responses ranging from 0 to 4 instead of from 1 to 5.
After uniform rescoring, none of the items displayed disor- Hawkins, Watt and Sinclair four-factor model
dered thresholds, which is illustrated by Fig. 1b (bottom panel, The four-factor model of Hawkins et al. (2006) was fitted to
modified). A slight improvement of the overall model fit was the Rasch model with factor items combined into subtests
observed but it did not satisfy the model expectations and the two excluded items deleted. The overall model fit
(χ 2(210) = 553.29, p = .001). At this stage the residual correla- for the four-factor solution was not satisfactory. The chi
tion matrix was examined and showed patterns of local depen- square was significant (χ 2(36) = 74.63, p = .001), and unidi-
dency between six groups of items, which were generally mensionality was not confirmed (Table 1, Hawkins model
consistent with the six-factor Frost model (Frost et al., 1990). initial). Reliability was low with a PSI of 0.67. At the indi-
vidual item level, subtest 2 (Organisation) displayed signifi-
Original six-factor Frost model cant model misfit with item fit residual of 3.37 and chi
The original six-factor Frost model (Frost et al., 1990) was square of 48.04 (df = 9) and was removed from the scale.
tested by combining the items of each factor into six sub- After removing the Organisation subtest, the best overall
tests (Lundgren-Nilsson et al., 2013). The overall fit and individual item model fit was attained with non-
improved substantially but it was still unsatisfactory significant chi square (χ 2(27) = 21.44, p = .77), better reli-
(χ 2(54) = 154.87, p = .001), and the unidimensionality ability (PSI = .76) and strict unidimensionality (Table 1,
assumption was violated (Table 1, Frost model initial). At Hawkins model modified). No DIF was identified for per-
the individual item (subtest) level, subtest six sonal factors. The modified Rasch solution for the Hawkins
(Organisation) displayed misfit with an item fit residual of et al. (2006) model consisted of three factors: Concern over
3.58 and significant chi square (χ 2(9) = 79.66, p < .001). Mistakes and Doubts, Personal Standards, and Parental
The misfitting subtest six (Organisation) was deleted, which Influences.
© 2017 The Australian Psychological Society
Rasch analysis of the FMPS 7

Figure 1 (a) Initial. Item cate-


gory probability curves illustrat-
ing disordered thresholds for
FMPS item 2, which is represen-
tative for the majority of the
items. Response options are
labelled here from 0 = ‘Strongly
disagree’ to 4 = ‘Strongly agree’.
(b) Final. Item category proba-
bility curves illustrating rescored
(lower panel) thresholds for
FMPS item 2. All items have
been uniformly rescored by
combining ‘Disagree’ with ‘Nei-
ther agree nor disagree’.
‘Strongly disagree’ is thus scored
as 0, both ‘Disagree’ and ‘Nei-
ther agree nor disagree’ as
1, ‘Agree’ as 2, and ‘Strongly
agree’ as 3.

Stallman and Hurst five-factor model Model comparison

The 5-factor model of Stallman and Hurst (2011) was Fig. 2 shows the person-item threshold distributions for all
tested with the Rasch model. In the same way, subtests three modified solutions of the analysis. It shows that the
were created after the six excluded items were deleted, sample abilities are well covered by the item thresholds in
and uniform rescoring was conducted to accommodate the all solutions, with no signs of ceiling or floor effects. For all
proposed five-factor structure. The overall model fit for the solutions the distribution of person thresholds was close to
five-factor solution was not satisfactory. The reliability was normal. The modified Frost et al. four-factor model dis-
low with a PSI of 0.64, and the overall fit was poor played good item coverage, which is reflected by good dis-
(χ 2(45) = 85.79, p = .001) (Table 1, Stallman model ini- criminability (PSI = 0.78). When Organisation is removed
tial). Unidimensionality was also not confirmed, and subt- from both the Hawkins et al. (2006) and Stallman and
est 2 (Organisation) displayed significant misfit to the Hurst (2011) models, the modified three-factor model of
Rasch model with a chi square of 46.89 (df = 9), p = .001. Hawkins et al. (2006) appears slightly better than the modi-
Therefore the Organisation subtest was deleted, which pro- fied four-factor Stallman and Hurst (2011) model because
duced a better model fit (χ 2(16) = 8.45, p = .93), PSI = there are no gaps between individual item thresholds. The
0.73. No DIF was identified for personal factors. The mod- modified Stallman and Hurst (2011) four-factor model
ified Rasch solution for the Stallman and Hurst (2011) shows large gaps between item thresholds on the high and
model is comprised of four factors, which are the same fac- low ends of the scale and consequently shows a lower PSI
tors as the modified Frost et al. (1990) model: Concern (0.73). For all solutions the mean of person item thresholds
over Mistakes, Personal Standards, Doubts about Actions, is slightly below the item mean, suggesting that difficult
and Parental Influences. items might be over-represented in this scale.

© 2017 The Australian Psychological Society


8 K. Mitchell-Parker et al.

Figure 2 Person-item threshold distribution for the three modified factor models of the FMPS.

© 2017 The Australian Psychological Society


Rasch analysis of the FMPS 9

All solutions appeared adequate in measuring perfection- obtain a precise perfectionism score when using the FMPS.
ism without the Organisation subtest. However, the modi- After accounting for the disordered thresholds, the modified
fied four-factor Frost solution had the highest reliability four factor solution based on the Frost model had the high-
(PSI = 0.78), which justified the use of this model to gener- est reliability (PSI = 0.78), although notably all three
ate an ordinal-to-interval conversion table. models showed acceptable reliability levels at PSI of 0.73
and above. The item thresholds for the modified Frost
Ordinal-to-interval scale conversion table model are well spread along the measurement continuum
of the perfectionism construct, and well-targeted levels of
Table S2 includes conversion scores from an ordinal-to- perfectionism experienced by the study participants. Com-
interval level scale for the modified FMPS, with the modi- bining the factors Parental Expectations and Parental Criti-
fied model presenting as four factors (Concern over Mis- cism into one factor is consistent with findings from other
takes, Personal Standards, Parental Influences, and Doubts researchers (Stallman & Hurst, 2011; Stöber, 1998). The
about Actions) and 29 items. Similar to Frost et al. (1990), exclusion of Organisation as a factor is also consistent with
Organisation items (items 2, 7, 8, 27, 29, and 31) should findings and recommendations from Frost et al. (1990) and
not be included in the calculation of a final score. Table S2 Stoeber and Otto (2006). Evidence shows that a person
includes conversion algorithms that allow users to increase may be a perfectionist in some areas of his or her life, but
the precision of the FPMS questionnaire without needing to may not uphold the same perfectionist tendencies in other
adjust the original response format of the scale. areas of life (Slaney & Ashby, 1996). In a study with inter-
collegiate varsity athletes (Dunn, Gotwals, & Dunn, 2005),
DISCUSSION participants completed two domain-specific (i.e., sport and
school) versions of Hewitt and Flett’s (1991) Multidimen-
The current study reported on Rasch analysis conducted to sional Perfectionism Scale (HF-MPS), as well as one mea-
advance the psychometric properties of the FMPS, a widely suring perfectionist tendencies in general. The results
used multidimensional measure of perfectionism. We dem- indicated that athletes were significantly more perfectionists
onstrated that the modified FMPS has good internal struc- in their sport than in their studies, and significantly more
tural validity once all items are rescored, the parental perfectionists in their studies than in general. Although the
factors are combined, and the Organisation factor is idea of domain-specific perfectionism is in contrast with
removed. Precision of the FMPS can be further enhanced by Flett and Hewitt’s (2002) claim that extreme perfectionists
using the proposed ordinal-to-interval conversion table pre- want to be perfect in all aspects of their lives, it is plausible
sented here, without the need to modify the original that athletes, for example, feel that having a neat and
response format. Our results support the psychometric orderly house is of less importance compared to setting high
properties and internal structural validity of the modified standards for performance and striving to avoid mistakes
FMPS without the Organisation facet, which did not satisfy and failure. Further research is needed in this area to
the Rasch model expectations across all three tested models. advance knowledge on how the different factors of perfec-
The best model fit was achieved when disordered thresholds tionism are related to different domains of perfectionism. It
were uniformly rescored and the two parental facets were is also possible that Organisation is more relevant to a dif-
combined into one subtest, which suggest a single factor we ferent, but related, construct such as conscientiousness
have called Parental Influences. The modified Rasch model (MacCann, Duckworth, & Roberts, 2009).
of the best fit suggests a four-factor solution for the FMPS, The uniform rescoring of all items was conducted by com-
comprising of Concern over Mistakes, Parental Influences, bining the category ‘slightly disagree’ with ‘neither agree nor
Personal Standards, and Doubts about Actions. The results disagree’ to correct the disordered thresholds. This decision
of this analysis demonstrated that the overarching perfec- was made on the basis that the thresholds of the middle cate-
tionism trait can be measured with high precision using the gory were consistently located close to the slightly disagree
ordinal-to-interval conversion table generated by this Rasch category. Issues with a ‘neutral’ category when conducting
analysis. Rasch analysis are not uncommon, and the tendency to pre-
The Rasch analysis confirmed the unidimensionality of all fer a middle response category has been found in a number
three models tested once Organisation was excluded, mean- of self-report questionnaires (Wetzel, Lüdtke, Zettler, &
ing that a psychometrically supported total perfectionism Böhnke, 2015). We followed the recommendation by Ker-
score can now be calculated for the modified scale. This sten and Kayes (2011) and combined the category options
finding is in line with the scoring instructions for the total (‘neither agree nor disagree’ and ‘slightly disagree’) that were
FMPS score proposed by the Frost et al. (1990), which also associated with the disordered thresholds.
excludes Organisation items. Therefore, researchers can The modified four-factor 29-item model supported by
now use the conversion algorithms from Table S2 to trans- Rasch analysis was comprised of the factors Concern over
form ordinal responses into interval-level data, in order to Mistakes, Doubts about Actions, Parental Influences, and
© 2017 The Australian Psychological Society
10 K. Mitchell-Parker et al.

Personal Standards. Other researchers have also found evi- Table S2 Converting from ordinal-to-interval-level scores
dence for a four-factor model using CTT methods (Hawkins for the modified Frost Multidimensional Perfectionism scale
et al., 2006; Stöber, 1998); however, this is the first study (Frost et al., 1990).
that used Rasch analysis to demonstrate that Organisation is
not psychometrically fitting within the construct of perfec- REFERENCES
tionism as defined by the FMPS. The two parental factors
have frequently been combined in the previous investiga- Andrich, D., Sheridan, B., & Luo, G. (2009). RUMM 2030. Perth:
tions of the FMPS factor structure (Stallman & Hurst, 2011; RUMM Laboratory.
Stöber, 1998), suggesting that the Parental Expectation and Anshel, M. H., Weatherby, N. L., Kang, M., & Watson, T. (2009).
Parental Concern items are better treated as one factor. Fur- Rasch calibration of a unidimensional perfectionism inventory
for sport. Psychology of Sport and Exercise, 10(1), 210–216. https://
ther research is needed to determine whether parental
doi.org/10.1016/j.psychsport.2008.07.006
influences are a core component of perfectionism or a pre- Bardone-Cone, A. M., Wonderlich, S. A., Frost, R. O., Bulik, C. M.,
dictor of perfectionism. From a psychometric viewpoint, the Mitchell, J. E., Uppala, S., & Simonich, H. (2007). Perfectionism
Parental Influence factor was supported as fitting within a and eating disorders: Current status and future directions. Clinical
Psychology Review, 27(3), 384–405. https://doi.org/10.1016/j.cpr.
four-factor perfectionism model. Based on the current
2006.12.005
Rasch analysis findings, the authors recommend to consider Besharat, M. A., & Shahidi, S. (2010). Perfectionism, anger, and
Organisation as potential predictor rather that an integrated anger rumination. International Journal of Psychology, 45(6),
component of the perfectionism trait. The recommendation 427–434. https://doi.org/10.1080/00207594.2010.501336
from this study is the use of a four-factor 29-item FMPS, Burns, D. D. (1980). The perfectionist’s script for self-defeat. Psychol-
ogy Today, 14(6), 34–52. https://doi.org/10.1007/978-94-010-
with the scores converted from ordinal to interval level for 9329-3
greater measurement precision. The results of this study Christensen, K. B., Makransky, G., & Horton, M. (2016). Critical
also have important theoretical and clinical implications for values for Yen’s Q 3: Identification of local dependence in the Rasch
understanding perfectionism, especially in athletes and peo- model using residual correlations. Applied Psychological Measurement,
41(3), 178–194. https://doi.org/10.1177/0146621616677520
ple who exercise. Given that much of our understanding of
Cox, B. J., Enns, M. W., & Clara, I. P. (2002). The multidimensional
the construct of perfectionism derives from the FMPS, it is structure of perfectionism in clinically distressed and college stu-
vital that the measure is supported as having adequate psy- dent samples. Psychological Assessment, 15(3), 365–373. https://doi.
chometric properties for the sporting population, which had org/10.1037//1040-3590.14.3.365
Dunn, J. G., Gotwals, J. K., & Dunn, J. C. (2005). An examination
not previously been achieved. Removal of the Organisation
of the domain specificity of perfectionism among intercollegiate
subscale and combining of the two Parenting subscales student-athletes. Personality and Individual Differences, 38(6),
would render the modified FMPS as being suitable for use 1439–1448. https://doi.org/10.1016/j.paid.2004.09.009
with this population, as supported by our findings. Flett, G. L., Greene, A., & Hewitt, P. L. (2004). Dimensions of per-
The following limitations should be noted. The three fectionism and anxiety sensitivity. Journal of Rational-Emotive &
Cognitive-Behavior Therapy, 22(1), 39–57. https://doi.org/10.1023/
models were tested with a sample of athletes and exercisers, b:jore.0000011576.18538.8e
which limits the generalisability of our findings to these Flett, G. L., & Hewitt, P. L. (2002). Perfectionism: Theory, research, and
populations. Replication with more diverse samples includ- treatment. Washington, DC: American Psychological Association.
ing clinical samples, general population, and a variety of Flett, G. L., & Hewitt, P. L. (2015). Measures of perfectionism. In
Measures of personality and social psychological constructs.
ethnic and cultural groups could be beneficial. Additionally,
(pp. 595–618). London: Academic Press.
for all solutions tested the SD for the item fit residual is Flett, G. L., Hewitt, P. L., & Heisel, M. J. (2014). The destructiveness
above the expectations of Rasch model. This can be of perfectionism revisited: Implications for the assessment of sui-
explained by over-representation of difficult items in the cide risk and the prevention of suicide. Review of General Psychol-
ogy, 18(3), 156–172. https://doi.org/10.1037/gpr0000011
scale. Given that much of our understanding of perfection-
Frost, R., Marten, P., Lahart, C., & Rosenblate, R. (1990). The
ism is derived from this measure, replication of these find- dimensions of perfectionism. Cognitive Therapy and Research,
ings with different samples using both CTT and IRT 14(5), 449–468. https://doi.org/10.1007/BF01172967
methods is needed to provide more robust support for the Gill, D. L., & Dzewaltowski, D. A. (1988). Competitive orientations
structure of the FMPS. among intercollegiate athletes: Is winning the only thing? The
Sport Psychologist, 2(3), 212–221. https://doi.org/10.1123/tsp.2.
3.212
SUPPORTING INFORMATION Gotwals, J. K., Dunn, J. G. H., & Wayment, H. A. (2003). An exam-
ination of perfectionism and self-esteem in intercollegiate ath-
letes. Journal of Sport Behavior, 26(1), 17–38. https://doi.org/10.
Additional Supporting Information may be found in the 1016/j.paid.2004.09.009
online version of this article at the publisher’s web-site: Hamachek, D. E. (1978). Psychodynamics of normal and neurotic
perfectionism. Psychology: A Journal of Human Behavior,
Table S1 Individual item fit statistics for initial analysis of 15(1), 27–33.
the Frost Multidimensional Perfectionism Scale including Hawkins, C. C., Watt, H. M. G., & Sinclair, K. E. (2006). Psychomet-
item locations, fit residuals and chi-square. ric properties of the Frost Multidimensional Perfectionism Scale

© 2017 The Australian Psychological Society


Rasch analysis of the FMPS 11

with Australian adolescent girls: Clarification of multidimension- Reed, G. F. (1985). Obsessional experience and compulsive behaviour: A
ality and perfectionist typology. Educational and Psychological Mea- cognitive-structural approach. New York, NY: Academic Press.
surement, 66(6), 1001–1022. https://doi.org/10.1177/ Rhéaume, J., Freeston, M. H., Dugas, M. J., Letarte, H., &
0013164405285909 Ladouceur, R. (1995). Perfectionism, responsibility and
Hewitt, P. L., & Flett, G. L. (1991). Perfectionism in the self and obsessive-compulsive symptoms. Behaviour Research and Therapy,
social contexts: Conceptualization, assessment, and association 33(7), 785–794. https://doi.org/10.1016/0005-7967(95)00017-r
with psychopathology. Journal of Personality and Social Psychology, Shafran, R., Cooper, Z., & Fairburn, C. G. (2002). Clinical perfec-
60(3), 456–470. https://doi.org/10.1037/0022-3514.60.3.456 tionism: A cognitive-behavioural analysis. Behaviour Research and
Hewitt, P. L., Flett, G. L., & Ediger, E. (1996). Perfectionism and Therapy, 40(7), 773–791. https://doi.org/10.1016/s0005-7967
depression: Longitudinal assessment of a specific vulnerability (01)00059-6
hypothesis. Journal of Abnormal Psychology, 105, 276–280. https:// Shafran, R., & Mansell, W. (2001). Perfectionism and psychopathol-
doi.org/10.1037//0021-843x.105.2.276 ogy: A review of research and treatment. Clinical Psychology
Hobart, J., & Cano, S. (2009). Improving the evaluation of thera- Review, 21(6), 879–906. https://doi.org/10.1016/s0272-7358(00)
peutic interventions in multiple sclerosis: the role of new psycho- 00072-6
metric methods. Health Technology Assessment, 13(12), 1–200. Siegert, R. J., Tennant, A., & Turner-Stokes, L. (2010). Rasch analy-
Hollander, M. H. (1965). Perfectionism. Comprehensive Psychiatry, 6, sis of the Beck Depression Inventory-II in a neurological rehabili-
94–103. https://doi.org/10.1016/s0010-440x(65)80016-5 tation sample. Disability and Rehabilitation, 32(1), 8–17. https://
Ioannou, C. I., Furuya, S., & Altenmüller, E. (2016). The impact of doi.org/10.1080/09638280902971398
stress on motor performance in skilled musicians suffering from Slaney, R. B., & Ashby, J. S. (1996). Perfectionists: Study of a crite-
focal dystonia: Physiological and psychological characteristics. rion group. Journal of Counseling and Development, 74(4), 393–398.
Neuropsychologia, 85, 226–236. https://doi.org/10.1016/j. https://doi.org/10.1002/j.1556-6676.1996.tb01885.x
neuropsychologia.2016.03.029 Smith, E. V. (2002). Detecting and evaluating the impact of multidi-
Kersten, P., & Kayes, N. M. (2011). Outcome measurement and the mensionality using item fit statistics and principal component
use of Rasch analysis, a statistics-free introduction. New Zealand analysis of residuals. Journal of Applied Measurement, 3, 205–231.
Journal of Physiotherapy, 39(2), 92–99. Stallman, H. M., & Hurst, C. P. (2011). The factor structure of the
Linacre, J. M. (1994). Sample size and item calibration stability. Frost Multidimensional Perfectionism Scale in university stu-
Rasch Measurement Transactions, 7(4), 328 www.rasch.org/rmt/ dents. Australian Psychologist, 46(4), 229–236. https://doi.org/10.
rmt74m.htm 1111/j.1742-9544.2010.00010.x
Lundgren-Nilsson, Å., Jonsdottir, I. H., Ahlborg, G., & Tennant, A. Stöber, J. (1998). The Frost Multidimensional Perfectionism Scale
(2013). Construct validity of the psychological general well being revisited: More perfect with four (instead of six) dimensions. Per-
index (PGWBI) in a sample of patients undergoing treatment for sonality and Individual Differences, 24(4), 481–491. https://doi.org/
stress-related exhaustion: A Rasch analysis. Health and Quality of 10.1016/S0191-8869(97)00207-9
Life Outcomes, 11(1), 1. https://doi.org/10.1186/1477-7525-11-2 Stoeber, J. (2017). The psychology of perfectionism: Critical issues,
MacCann, C., Duckworth, A. L., & Roberts, R. D. (2009). Empirical open questions, and future directions. In J. Stoeber (Ed.), The
identification of the major facets of conscientiousness. Learning psychology of perfectionism: Theory, research, applications. London:
and Individual Differences, 19(4), 451–458. https://doi.org/10. Routledge.
1016/j.lindif.2009.03.007
Stoeber, J., & Otto, K. (2006). Positive conceptions of perfectionism:
Marais, I., & Andrich, D. (2008). Effects of varying magnitude and
Approaches, evidence, challenges. Personality and Social Psychology
patterns of response dependence in the unidimensional Rasch
Review, 10(4), 295–319. https://doi.org/10.1207/
model. Journal of Applied Measurement, 9(2), 105–124. http://
s15327957pspr1004_2
scseec.edu.au/
Stucki, G., Daltroy, L., Katz, J. N., Johannesson, M., & Liang, M. H.
Martinelli, M., Chasson, G. S., Wetterneck, C. T., Hart, J. M., &
(1996). Interpretation of change scores in ordinal clinical scales
Björgvinsson, T. (2014). Perfectionism dimensions as predictors
and health status measures: The whole may not equal the sum
of symptom dimensions of obsessive-compulsive disorder. Bulle-
of the parts. Journal of Clinical Epidemiology, 49(7), 711–717.
tin of the Menninger Clinic, 78(2), 140–159. https://doi.org/10.
https://doi.org/10.1016/0895-4356(96)00016-9
1521/bumc.2014.78.2.140
Tennant, A., & Conaghan, P. G. (2007). The Rasch measurement
Norquist, J. M., Fitzpatrick, R., Dawson, J., & Jenkinson, C. (2004).
model in rheumatology: What is it and why use it? When should
Comparing alternative Rasch-based methods vs raw scores in
measuring change in health. Medical Care, 42(1), I–25. https:// it be applied, and what should one look for in a Rasch paper?
doi.org/10.1097/01.mlr.0000103530.13056.88 Arthritis and Rheumatism, 57(8), 1358–1362. https://doi.org/10.
Pacht, A. R. (1984). Reflections on perfection. American Psychologist, 1002/art.23108
39(4), 386–390. https://doi.org/10.1037//0003-066x.39.4.386 Wainer, H., & Kiely, G. (1987). Item clusters and computerized
Picard, C. L. (1999). The level of competition as a factor for the adaptive testing: A case for testlets. Journal of Educational Measure-
development of eating disorders in female collegiate athletes. ment, 24(3), 185–201. https://doi.org/10.1111/j.1745-3984.1987.
Journal of Youth and Adolescence, 28(5), 583–594. https://doi.org/ tb00274.x
10.1023/A:1021606710398 Wetzel, E., Lüdtke, O., Zettler, I., & Böhnke, J. R. (2015). The sta-
Purdon, C., Antony, M. M., & Swinson, R. P. (1999). Psychometric bility of extreme response style and acquiescence over 8 years.
properties of the Frost Multidimensional Perfectionism Scale in a Assessment, 23(3), 279–291. https://doi.org/10.1177/
clinical anxiety disorders sample. Journal of Clinical Psychology, 1073191115583714
55(10), 1271–1286. https://doi.org/10.1002/(sici)1097-4679 Wirtz, P. H., Ehlert, U., Kottwitz, M. U., La Marca, R., &
(199910)55:10<1271::aid-jclp8>3.0.co;2-a Semmer, N. K. (2013). Occupational role stress is associated with
Rasch, G. (1960). Probabilistic models for some intelligence and attain- higher cortisol reactivity to acute stress. Journal of Occupational
ment test. Copenhagen: Danish Institute for Educational Health Psychology, 18(2), 121–131. https://doi.org/10.1037/
Research. a0031802

© 2017 The Australian Psychological Society

You might also like