See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/44685959

Measuring Patient Anxiety in Primary Care: Rasch Analysis of the 6-item

Spielberger State Anxiety Scale

Article in Value in Health · September 2010

DOI: 10.1111/j.1524-4733.2010.00758.x · Source: PubMed

CITATIONS READS
41 6,052

3 authors:

Helen Court Katy Greenland

University of Glasgow Cardiff University
13 PUBLICATIONS 644 CITATIONS 35 PUBLICATIONS 672 CITATIONS

SEE PROFILE SEE PROFILE

Tom Margrain
Cardiff University
93 PUBLICATIONS 2,308 CITATIONS

SEE PROFILE

All content following this page was uploaded by Katy Greenland on 19 February 2021.

The user has requested enhancement of the downloaded file.


Measuring Patient Anxiety in Primary Care: Rasch Analysis of
the 6-item Spielberger State Anxiety Scale
Helen Court, PhD, Katy Greenland, PhD, Tom H. Margrain, PhD
Cardiff University, Cardiff, UK
A B S T R AC T
Objectives: The 6-item Spielberger State Anxiety Scale has been used as a
replacement of the original version in many health-care studies. The
purpose of this study was to evaluate the measurement properties of the
shortened 6-item Spielberger State Anxiety Scale using Rasch analysis in
general medical practice patients (N = 297).
Methods: Participants (aged 16 years or above) were recruited on a con-
secutive basis from three general medical practices. Prior to their appoint-
ment, participants were asked to complete a 6-item Spielberger State
Anxiety Scale.
Results: The results of the study showed that the scale is unidimentional,
and each item measures a different level of patient anxiety. The rating scale
Introduction
Heightened patient anxiety is a detrimental factor to health-care
outcomes. Anxiety has been associated with disrupted recall of
information [1], poor attention [1], reduced satisfaction [2], and
can be a barrier to effective patient–practitioner communication
[3]. Patient anxiety also contributes to wasted health-care
resources, because of patient noncompliance [4], nonattendance
of appointments [5,6], and in other cases, excessive utilization of
health-care services [7,8]. Furthermore, severe or prolonged
anxiety can also represent a psychiatric disorder that is a focus
for treatment [9].
The Spielberger State-Trait Inventory (STAI) has been widely
used to measure patient anxiety in primary health care [10–18].
The original scale incorporates 20 items that measure “state”
anxiety and 20 items that measure “trait” anxiety. Trait anxiety
is a stable personality trait that influences a person’s “anxiety
proneness” [19]; whereas, state anxiety is a transient experience
caused by a person’s cognitive appraisal of a potential threat or
danger [19,20]. In other words, state anxiety is that dimension of
anxiety which may be reactive to the health-care experience. A
shortened 6-item version of the state scale has been developed
[21] to help reduce the associated respondent burden of the
full-length version. The 6-item state anxiety scale has good inter-
nal reliability (Cronbach alpha 0.82) and correlation with the full
STAI is high (r = 0.95) [21]. The shortened scale has been used in
many health-care settings, including dental [12,22], medical [16],
and general medical practice [23,24].
The original and shortened versions of the Spielberger state
anxiety scale utilizes a Likert scale. Each item has four response
categories (“not at all,” “somewhat,” “moderately,” and “very
much”) which are assigned numerical values (1–4). These values
are added together to produce an anxiety score (Likert scoring).
At best, this approach yields ordinal-level data. This is because
Address correspondence to: Helen Court, School of Optometry and
Vision Sciences, Cardiff University, Cardiff CF24 4LU, UK. E-mail:
waltersh1@cardiff.ac.uk
10.1111/j.1524-4733.2010.00758.x
© 2010, International Society for Pharmacoeconomics and Outcomes Research (ISPOR)
Volume 13 • Number 6 • 2010
VA L U E I N H E A LT H
vhe_758 813..819
operated well and item and person reliability was good. Furthermore,
principal-components analysis of the residuals confirmed the scale mea-
sures a unitary concept. A scoring key was generated to allow conversion
of raw scores to a continuous measurement.
Conclusion: The 6-item Spielberger State Anxiety Scale is shorter than the
original version and has good psychometric properties. This would suggest
the scale is a valid alternative to the full version for use in primary
health-care practice and research.
Keywords: anxiety, primary health care, questionnaire, Rasch analysis.
the true relationship between response categories is unknown.
For example, the difference between “not at all” and “some-
what” may be different to the difference between “somewhat”
and “moderately.” In addition, it is not known if all the items
measure equal levels of anxiety. This approach limits the inter-
pretation of the anxiety score because the difference between a
score of 6 to 8 on an ordinal scale may not represent the same
distance as a score between 8 and 10. This precludes mathemati-
cal operations such as the calculation of change scores, or effect
sizes. However, Rasch analysis can be used to overcome many of
the limitations associated with Likert scales [25,26]. Rasch
analysis strengthens the measurement quality of a questionnaire
by weighting individual items based on their contribution to the
underlying trait, allowing transformation of raw scores into con-
tinuous data. Additionally, items which are redundant for precise
measurement can be identified and removed from the scale i.e.,
misfitting items. Rasch analysis has become a popular method to
improve the design, sensitivity, and validity of questionnaires in
health care [27–29].
Previous Rasch analysis studies have identified a number of
misfitting items in the full STAI 20-item state scale [30,31].
Misfitting items can indicate that an item measures something
different from the rest of the scale. Furthermore, such items can
introduce off-variable noise and degrade the measurement. To
date, there are no studies that report the psychometric properties
of the shortened 6-item version using Rasch analysis.
Therefore, the main aim of this study was to evaluate the
measurement properties and unidimensionality of the shortened
6-item Spielberger State Anxiety Scale using Rasch analysis. The
secondary aim was to evaluate construct validity. This will provide
further evidence about the validity of using the shortened scale as
a replacement of the full version in primary health-care studies.
Methods
Study Design and Population
Three general medical practices in the Vale of Glamorgan agreed
to distribute questionnaires to patients. These were located in a
1098-3015/10/813 813–819 813
814 Court et al.

busy city center, a town, and a rural location. The sample was Table 1 Characteristics of the patient sample
drawn from consecutive patients attending for an appointment
N 297
with a general practitioner. Practices were asked to ensure that
every patient (16 years and over) was given an information sheet Gender (%)
to read and a 6-item Spielberger State Anxiety Scale [21] to Female 194 (65.3)
complete before they had their appointment. Reception staff Male 103 (34.7)
Age 44.4 (⫾18.9 years)
were asked to keep note of the number of patients refusing to Education (%)
complete a questionnaire. All procedures adhered to the tenets of GCSE or equivalent 97 (32.7)
the Declaration of Helsinki and ethical approval was obtained A levels or equivalent 70 (23.6)
from the South East Wales Research Ethical Committee. Degree 55 (18.5)
No qualifications 72 (24.2)
Missing 3 (1.0)
Reason for appointment (%)
Statistical Analysis Routine 57 (19.2)
All data were entered into the statistics package Statistical Problem 144 (48.5)
Treatment 31 (10.4)
Package for the Social Sciences Ver. 12 (SPSS Inc., Chicago, IL, Emergency 60 (20.2)
USA) and recoded such that all items had a consistent valance. Missing 5 (1.7)
Rasch analysis was undertaken according to the Andrich Last appointment (%)
Rating Scale model [32] using Winsteps ver. 3.58.1 [33]. The <6 months 232 (78.1)
6–12 months 41 (13.8)
Rasch model is distinctive from most statistical modeling because 1–2 years 18 (6.1)
the aim was not to describe a set of data, rather, it was an “ideal” >2 years 5 (1.7)
that the data should meet in order to provide successful measure- Missing 1 (0.3)
ment. Rasch analysis is a probabilistic logistic model, which First language (%)
English 281 (94.6)
produces Logit values describing item difficulty, person ability, Welsh 9 (3.0)
and determines threshold values for each response category for Other 3 (1.0)
the items. In this way, Rasch analysis can provide questionnaire Missing 4 (1.3)
scores which are on a true interval scale.
Firstly, Rasch analysis was performed to evaluate the
operation of the response categories. This analysis identifies how
well respondents can discriminate reliably between response considered absent if below 0.5 logit between item calibrations
categories. [36,37].
Secondly, Rasch analysis fit statistics were used to identify The reliability of the final questionnaire was measured using
how well each item contributed to the underlying unidimensional person and item reliability estimates.
measure [25]. Fit statistics describe how both items and person Construct validity examines whether a measurement tool
responses fit the predicted responses of the Rasch model. Rasch (e.g., questionnaire) has the relationships with other variables
analysis provides two chi-square statistics, infit and outfit, which which we would expect [38]. Construct validity was examined by
are calculated from the mean square of the residuals. These range independent t-test of the differences in scores between subjects
from zero to infinity. Items fitting perfectly with the unidimen- attending for a routine or emergency appointment. We hypoth-
sional scale have an expected infit or outfit statistic of 1. esized that patients attending for an emergency appointment
Values less than 1 indicate that the item overfits the model. would report significantly higher levels of anxiety.
Substantially overfitting items add little extra information to the
scale and as such, they are redundant; whereas, values higher
than 1 suggest misfit to the model, and these items may be Results
measuring something different from the rest of the scale. Infit Questionnaire responses were received from 297 patients. The
statistics are weighted to give more importance to those people demographic details of the patients are shown in Table 1. Person
who are closer to the item mean. Outfit statistics are not fit statistics provide information about how closely people are
weighted and so, are more sensitive to outlying scores. Therefore, responding according to the prediction of the Rasch model. Poor
items with poor fit statistics compromise the validity of the fit statistics highlight people who may not be responding in a
measurement. Smith et al. (1998) suggest that interpretation of consistent way, in other words, rogue responders. Of the 26
item fit statistics can be aided by defining infit and outfit cut-off people identified as substantially misfitting the model (outfit and
values based upon the sample size (infit: 1 ⫾ 2/√N and outfit: infit mean square >1.40), the individual questionnaire responses
1 ⫾ 6/√N) [34,35]. Interestingly, unlike item fit statistics, sample were examined by the authors. The questionnaires were exam-
size has little influence upon person fit statistics because ques- ined to identify any in which respondents had selected same
tionnaires tend to be short in length (100 items or less) [34]. response category. Three of the items had reversed scales, making
Unidimensionality was also assessed by principal- it easy to identify responses which were completely contradic-
components analysis (PCA) of the residuals. Two criteria were tory. None of the participants had responded this way. Therefore,
used to assess unidimensionality. Firstly, the proportion of the all questionnaires were retained for analysis.
variance explained by the measures for the empirical calculation
should be similar to the model [25]. Secondly, the unexplained
variance explained by the first contrast should be less than 2 Response Scale Analysis
eigenvalue units [33]. Winsteps provides category diagnostic statistics that describe
Item estimates should be invariant to the group assessed. how well the response categories (e.g., “not at all”, “somewhat”)
Differential item functioning (DIF) occurs when items have dif- operate. Categories that operate well should have ordered
ferent item difficultly estimates across groups [25]. In the present structure calibration thresholds. This indicates that every cat-
study, DIF was evaluated for age (<50 years as younger, ⱖ50 egory has a distinct probability of being selected more than any
years as older) and gender (male, female) using t-tests. DIF was other category for a particular person difficulty. The structure
Rasch Analysis of the 6-item STAI
Figure 1 Probability curve to show the operation
of four response categories (Cat 1 = “not at all”,
Cat 2 = “somewhat,” Cat 3 = “moderately,” Cat
4 = “very much”).
calibration thresholds are visually identified in a probability
curve and show that the response categories function well
(Fig. 1). Furthermore, the step difficulties advance between the
recommended values of 1.4 and 5.0 logits [39], ensuring mea-
surement stability.
Person and Item Estimates
In Figure 2, the spread of each item calibration is visualized and
is compared with the range of person ability estimates. The range
of the items is -1.01 to 1.86 Logits. Items located at the bottom
of the map, e.g., Item 4 “right now I am relaxed,” help discrimi-
nate between those people with lower anxiety. Conversely, items
located at the top of the map, e.g., Item 3: “Right now I feel
upset,” are high-level anxiety discriminating items.
Item calibration estimates are shown in Table 2 i.e., the esti-
mated level of anxiety measured by each item. Winsteps pro-
vides statistics to describe the precision of these estimates. The
high item separation reliability coefficient (0.99) of the items
indicates the stability of the item estimates. The root mean
square error (RMSE) is a further measure of the accuracy of the
item estimates. Values range from 0–1 and values close to 0
indicate good accuracy. The RMSE over all the items is 0.11 for
this analysis.
Figure 1 shows the mean anxiety level that each item mea-
sures. The mean of the person estimates is -1.06 Logits (standard
deviation ⫾ 1.58), with a range from 5.32 to -5.20 Logits.
Inspection of the person–item map indicates that the items are
marginally targeted toward the higher end of anxiety. However,
when the influences of the categories are accounted for, the items
measure over a larger range (Fig. 1) i.e., the items span virtually
the complete range of person anxiety levels.
Winsteps provides a statistic called the person separation
reliability coefficient that describes the reliability of person order-
ing and is similar to the conventional Cronbach alpha coefficient.
It is 0.78 for this sample. The person separation ratio expresses
815
the reliability of the scale to discriminate between people of
different abilities. It is defined as the ratio of the adjusted person
standard deviation to the standard error of the measurement (i.e.,
the variance not accounted for by the Rasch model), measured in
standard error units [25]. The person separation ratio (signal-to-
noise ratio) was 1.89, which is only slightly lower than the
recommended value of 2 [40].
None of the items exhibited DIF for either gender or age
(Table 3).
The estimates obtained with Rasch analysis were used to
generate a scoring key which recodes the raw questionnaire
scores into continuous data (Table 4).
Evaluating Unidimensionality
According to the guidelines suggested by Smith et al. (1998),
the infit and outfit cut-off values for a sample size of 297 are
0.88–1.12 and 0.65–1.35 for infit and outfit, respectively [34].
Table 2 identifies that only one of the six items (item 6) had
both infit and outfit mean squares which were outside these
criteria. Infit values were just outside these criteria for a further
three items (items 1, 3, and 4). However, while aspects of these
items do not exhibit “perfect” fit to the Rasch model, Linacre
indicates that items with infits/outfits of up to 1.5 are still pro-
ductive for measurement [33] i.e., these items add to the scale
in a meaningful way. Therefore, although the fit of these items
is questionable, we decided to retain these items and further
evaluate the unidimensionality of the scale based upon PCA of
the residuals.
PCA of the residuals identified that the variance explained by
the measures for the empirical calculation (75.6%) was almost
identical to the model (75.7%). The unexplained variance
explained by the first contrast was 1.7 eigenvalue units (i.e., <2.0
eigenvalue units). Taken together, these results suggest unidimen-
sionality of the scale.
816 Court et al.

Figure 2 Person item map for the 6-item Spielberger State Anxiety Scale.
Rasch Analysis of the 6-item STAI 817

Table 2 Fit statistics and item calibration measures for the 6-item State sionality of the scale was further supported by PCA of the
Anxiety Scale residuals. The validity of the scale as a measure of state anxiety
was also supported by the expected results that patients attend-
Mean square Item calibration ing for an emergency appointment reported significantly higher
Item Infit Outfit (SE) scores.
1. Right now I feel calm 0.70 0.75 -0.42 (0.10) Inspection of the person–item map (Fig. 1) reveals that the
2. Right now I am tense 1.12 1.02 0.23 (0.10) items are targeted toward the higher levels of anxiety. However,
3. Right now I feel upset 1.21 0.89 1.86 (0.13) when the category structure of each item is considered, the items
4. Right now I am relaxed 0.86 0.98 -1.01 (0.10)
measure almost the complete range of patient anxiety for this
5. Right now I feel content 0.96 1.06 -0.83 (0.10)
6. Right now I am worried 1.37 1.45 0.18 (0.10) sample (Fig. 1). Where this questionnaire is used as a research
tool to assess the ability of interventions to reduce anxiety, failure
SE, standard error. to differentiate the level of “extreme” anxiety is not problematic.
Interventions which can significantly reduce anxiety are those
which cause a significant decrease in mean anxiety. Therefore,
Construct Validity while it is important that the questionnaire can reliably measure
State anxiety scores were significantly higher for patients attend- the majority of people close to the mean, it is less important that
ing because of an emergency compared with those attending for it measures those who are at the extremes, i.e., who are in the
a routine appointment (t (114) = -3.92; P < 0.001). tails of the normal distribution. Furthermore, Figure 2 and
Table 2 show that each item has a different item measure, i.e.,
each item measures a different level of anxiety. The majority of
Discussion studies using the shortened scale calculate anxiety by adding raw
scores. This approach assumes that each item has equal difficulty
The results of this analysis show that the 6-item Spielberger State
and therefore contributes equally to the final measurement.
Anxiety Scale [21] is a valid measurement tool with which to
However, the results from this analysis suggest that each item
quantify anxiety in general medical practice. The questions on
should be weighted in the final measure. In other words, use of
this scale work well together to form a valid unidimensional
raw anxiety scores could degrade measurement precision.
interval scale, i.e., it measures a single underlying latent trait,
Rasch analysis also provides “separation reliabilities” that
“state anxiety.”
describe the reliability of the item and person estimates. The
Rasch analysis is a powerful tool allowing identification of
separation reliabilities were high (0.99 for items and 0.78 for
items which are not sensitive to the underlying trait. “Misfitting”
persons), indicating that the estimated measures can discriminate
items increase the level of noise within the measurement and
items and persons well along the anxiety scale. For individual
therefore should be removed [41]. Tenenbaum, 1985, identified
patient use, the person separation reliability indicates that the
nine items in the full version of the STAI state scale with poor fit
questionnaire will reliably discriminate patients into at least three
statistics [31]. Whereas, analysis of the 6-item scale showed that,
levels of anxiety [33]. In other words, practitioners will be able to
although not all items perfectly fit the Rasch model, all items are
recognize patients with high levels of anxiety. Category analysis
productive for measurement i.e., infit and outfit mean square
also confirmed the use of a four response option. Item and
values were below 1.5 [33]. These results suggest that the short-
category threshold estimates were used to create a scoring key,
ened scale measures one underlying construct. The unidimen-
allowing conversion of raw scores to a continuous measure of
anxiety (Table 4).
One limitation of the study was that receptionists were not
Table 3 DIF in the 6-item Spielberger State Anxiety Scale item calibra- consistent in recording the number of people who refused to
tions for gender (male, female) and age (<50 years as younger, >50 years
complete the questionnaire. Although every practice commented
as older)
that the majority of people accepted a questionnaire, we do not
Gender (male, female) Age (<50, ⱖ50 years) know the response rate. Nonresponse can introduce bias into the
Item DIF t P DIF t P sample [42]. Comparison to recent population estimates for
general practice attendance in Wales identifies that the current
1 0.12 0.58 0.56 0.09 0.43 0.67 study may have slightly underrepresented older patients (Welsh
2 0.34 1.57 0.12 0.03 0.16 0.88
estimates; median age range 55–64 years) [43]. It is possible that
3 0.34 1.26 0.21 0.47 1.79 0.08
4 0.25 1.22 0.23 0.25 1.27 0.21 older people were slower at questionnaire completion and did
5 0.14 0.66 0.51 0.24 1.19 0.23 not return the questionnaire. However, previous studies within
6 0.26 1.18 0.24 0.12 0.58 0.56 health care suggest that there is no association between age and
state anxiety [16,44]. This indicates that the age bias may not
DIF, differential item functioning.
have effected the distribution of questionnaire scores. The
current study also included a slightly higher proportion of
women compared with population estimates (65% vs. 55%)
Table 4 Scoring key for the 6-item State Anxiety Scale
[43]. Health-care studies suggest that women tend to report
Item Not at all Somewhat Moderately Very much higher anxiety levels compared with men [16,44,45]. Therefore,
it is possible that there was a slight overrepresentation of higher
1 2.66 0.60 -1.41 -3.55 anxiety scores in our sample.
2 -2.90 -0.76 1.25 3.31
3 -1.27 0.87 2.88 4.94 Although some slight bias may exist in the sample because of
4 4.09 0.01 -2.00 -4.14 an overrepresentation of women, this should not significantly
5 2.25 0.19 -1.82 -3.96 compromise the calibration of the questionnaire. Unlike the cali-
6 -2.95 -0.81 1.20 3.26 bration of questionnaires in traditional test design which are
This key may be implemented by assigning the appropriate score for each response category
dependant upon the sample, Rasch analysis allows relatively
selected; adding up the scores and dividing by the number of questions answered. sample-free test calibration [46,47]. The Rasch model simply
818
seeks to describe what happens when any person encounters any
item, therefore removing the interaction between person ability
and item difficulty [46]. In other words, the calibration of the test
is not bound by the ability (anxiety) distribution of the sample.
Furthermore, the absence of DIF for both age and gender further
suggests that our assessment of the questionnaire was not com-
promised by age or gender bias.
In conclusion, the 6-item Spielberger State Anxiety Scale
allows measurement and identification of anxious patients within
general practice. The short length of the scale lends itself to busy
clinical practice, ensuring low respondent burden and increasing
the likelihood that the questionnaire will be completed fully and
truthfully. The unidimensionality of the scale has also been estab-
lished by Rasch analysis. Furthermore, calculation of item and
category estimates allows conversion of raw scores into continu-
ous level data, allowing improved measurement precision with
the questionnaire. This can be easily done using the scoring key
(Table 4). Therefore, the results of this study would indicate that
the shortened scale is an acceptable alternative to full STAI for
use in health-care research.
Source of financial support: This study was supported by Cardiff Univer-
sity and a grant from Bausch and Lomb, Ltd.
References
1 Matthews A, MacLeod C. Cognitive Vulnerability to emotional
disorders. Annu Rev Clin Psychol 2005;1:167–95.
2 Court H, Greenland K, Margrain TH. Evaluating the association
between anxiety and satisfaction. Optom Vis Sci 2009;86:216–
21.
3 Lang F, Floyd MR, Beine KL. Clues to patients’ explanations and
concerns about their illnesses—a call for active listening. Arch
Fam Med 2000;9:222–7.
4 Corah NL. Dental anxiety—assessment, reduction and increasing
patient satisfaction. Dent Clin North Am 1988;32:779–90.
5 Skaret E, Kvale G, Raadal M. General self-efficacy, dental anxiety
and multiple fears among 20-year-olds in Norway. Scand J
Psychol 2003;44:331–7.
6 Taani D. Dental anxiety and regularity of dental attendance in
younger adults. J Oral Rehabil 2002;29:604–8.
7 Connelly JE, Philbrick JT, Smith GR, et al. Perceptions of primary
care patients and the influence on health-care utilization. Med
Care 1989;27(Suppl.):S99–109.
8 Frostholm L, Fink P, Christensen KS, et al. The patients’ illness
perceptions and the use of primary health care. Psychosom Med
2005;67:997–1005.
9 House A, Stark D. ABC of psychological medicine—anxiety in
medical patients. Br Med J 2002;325:207–9.
10 Muglali M, Komerik N. Factors related to patients’ anxiety
before and after oral surgery. Br J Oral Maxillofac Surg
2008;66:870–7.
11 Griffiths HS, Wilson MA, Kincey JA. Anxiety levels, patient sat-
isfaction and failed appointment rate in anxious patients referred
by general practitioners to a dental hospital unit. Br Dent J
1998;185:134–6.
12 Dailey YM, Humphris GM, Lennon MA. Reducing patients’ state
anxiety in general dental practice: a randomized controlled trial.
J Dent Res 2002;81:319–22.
13 Sepulcri R, do Amaral VF. Depressive symptoms, anxiety, and
quality of life in women with pelvic endometriosis. Eur J Obstet
Gynecol Reprod Biol 2009;142:53–6.
14 Coudeyre E, Poiraudeau S, Revel M, et al. Beneficial effects of
information leaflets before spinal steroid injection. Joint Bone
Spine 2002;69:597–603.
15 Skinner TC, Davies MJ, Farooqi AM, et al. Diabetes screening
anxiety and beliefs. Diabet Med 2005;22:1497–502.
16 Nijkamp M, Kenens C, Dijker A, et al. Determinants of surgery
related anxiety in cataract patients. Br J Ophthalmol 2004;
88:1310–14.
Court et al.
17 Habib NE, Mandour NM, Balmer HGR. Effect of midazolam on
anxiety level and pain perception in cataract surgery with topical
anesthesia. J Cataract Refract Surg 2004;30:437–43.
18 Mokashi A, Leatherbarrow B, Kincey J, et al. Patient communi-
cation during cataract surgery. Eye 2004;18:147–51.
19 Spielberger C. Theory and research on anxiety. In: Spielberger
C, ed. Anxiety and Behavior (1st ed.). New York: Academic Press,
1966.
20 Barlow DH. Anxiety and Its Disorders: The Nature and Treat-
ment of Anxiety and Panic (2nd ed.). New York: Guilford Press,
2002.
21 Marteau TM, Bekker H. The development of a 6-item short-form
of the state scale of the Speilberger State Trait Anxiety Inventory
(Stai). Br J Clin Psychol 1992;31:301–6.
22 Humphris GM, Clarke HMM, Freeman R. Does completing a
dental anxiety questionnaire increase anxiety? A randomised con-
trolled trial with adults in general dental practice. Br Dent J
2006;201:33–5.
23 Rice G, Ingram J, Mizan J. Enhancing a primary care environ-
ment: a case study of effects on patients and staff in a single
general practice. Br J Gen Pract 2008;58:465–70.
24 Park P, Simmons RK, Prevost T, Griffin SJ. Screening for type 2
diabetes is feasible, acceptable, but associated with increased
short-term anxiety: a randomised controlled trial in British
general practice. BMC Public Health 2008;8:350.
25 Bond TG, Fox CM. Applying The Rasch Model: Fundamental
Measurement in the Human Sciences. Mahwah, NJ: Lawrence
Erlbaum Associates, Inc., 2001.
26 Wright BD, Linacre JM. Observations are always ordinal; mea-
surements, however, must be interval. Arch Phys Med Rehabil
1989;70:857–60.
27 Tennant A, McKenna SP, Hagell P. Application of Rasch analysis
in the development and application of quality of life instruments.
Value Health 2004;7(Suppl.):S22–6.
28 McHorney CA, Monahan PO. Postscript: applications of Rasch
analysis in health care. Med Care 2004;42:1–73.
29 Bezruczko N. Rasch Measurement in Health Sciences. Maple
Grove, MN: JAM Press, 2005.
30 Tenenbaum G, Furst D, Weingarten G. A statistical reevaluation
of the STAI anxiety questionnaire. J Clin Psychol 1985;41:239–
44.
31 Tenenbaum G, Furst DM. Similarities between retrospective and
actual anxiety-states. J Psychol 1985;119:185–91.
32 Andrich D. Rating formulation for ordered response categories.
Psychometrika 1978;43:561–73.
33 Linacre JM. A User’s Guide to Winsteps: Rasch-Model Computer
Programs. Chicago: Winsteps.com, 2005.
34 Smith RM, Schumacker RE, Bush JM. Using item mean squares
to evaluate fit to the Rasch model. J Outcome Meas 1998;2:66–
78.
35 De Ayala RJ. The Theory and Practice of Item Response Theory.
New York: Guilford Press, 2008.
36 Pickard AS, Dalal MR, Bushnell DM. A comparison of depressive
symptoms in stroke and primary care: applying Rasch models to
evaluate the center for epidemiologic studies-depression scale.
Value Health 2006;9:59–64.
37 Linacre JM. Sample size and item calibration stability. Rasch
Meas Trans 1994;7:328.
38 Bland JM, Altman DG. Statistics notes—validating scales and
indexes. Br Med J 2002;324:606–7.
39 Linacre JM. MESA Research Note #2. Midwest Objective Mea-
surement Seminar; June; Chicago, 1997.
40 Pesudovs K, Garamendi E, Keeves JP, Elliott DB. The activities of
daily vision scale for cataract surgery outcomes: re-evaluating
validity with Rasch analysis. Invest Ophthalmol Vis Sci
2003;44:2892–9.
41 Stelmack JA, Szlyk JP, Stelmack BR, et al. Psychometric proper-
ties of the Veterans Affairs Low-Vision Visual Functioning Ques-
tionnaire. Invest Ophthalmol Vis Sci 2004;45:3919–28.
42 Evans SJW. Good surveys guide. Br Med J 1991;302:302–3.
43 WAG. Welsh Health Survey. Cardiff: Welsh Assembly Govern-
ment, 2009.
 Rasch Analysis of the 6-item STAI
44 Stauder A, Kovacs M. Anxiety symptoms in allergic patients:
identification and risk factors. Psychosom Med 2003;65:816–23.
45 Badner NH, Nielson WR, Munk S, et al. Preoperative anxiety:
detection and contributing factors. Can J Anaesth 1990;37:
444–7.
View publication stats
819
46 Wright BD, Stone MH. Best Test Design: Rasch Measurement.
Chicago: MESA Press, 1979.
47 Tinsley HEA, Dawis RV. An investigation of the Rasch simple
logistic model: sample free item and test calibration. Educ Psychol
Meas 1975;35:325–39.