Content uploaded by Philipp Yorck Herzberg
Author content
All content in this area was uploaded by Philipp Yorck Herzberg on Nov 14, 2017
Content may be subject to copyright.
Multistudy Report
Psychometric Properties and
Validation of a German High
Sensitive Person Scale (HSPS-G)
Sandra Konrad and Philipp Yorck Herzberg
Department of Personality Psychology and Psychological Assessment, Helmut-Schmidt-University, University of the German Federal
Armed Forces Hamburg, Germany
Abstract: High sensitivity is an individual disposition to perceive and process external and internal stimuli more intensely than the average
population. For measuring high sensitivity, Aron and Aron (1997) developed a unidimensional self-report questionnaire. However, Smolewska,
McCabe, and Woody (2006) fitted a model with three correlated factors: ease of excitation, aesthetic sensitivity, and low sensory threshold.
Both models were questioned by Evans and Rothbart (2008) who postulated a two-factor structure: negative affect and orienting sensitivity.
Nonetheless, the studies presented so far are based on small samples and did not address the issues of the ordinal data and measurement
invariance. We adopted the High Sensitive Person (HSP)-Scale for German-speaking populations and found that a three-factor model provided
the best fit. However, we excluded 13 items because of their low factor loadings or high intercorrelations. The revised HSP-Scale fit a three-
factor model. Furthermore, we could establish a high level of measurement invariance (strict invariance), indicating equality of loadings,
thresholds, and residual variances across sex. The scale showed good psychometric properties and high test-retest reliability. Finally,
relationships with psychological symptoms were presented.
Keywords: sensory-processing-sensitivity, highly sensitive person scale, validation, personality, BSCL
High sensitivity is a phenomenon postulated by Aron and
Aron (1997), in which the affected person perceives and
processes external and internal stimuli more intensely.
Highly sensitive individuals (HSP) report stronger
emotional reactivity and behavioral inhibition. Aron and
Aron call the underlying basis of temperamental property
sensory-processing-sensitivity (SPS; Aron & Aron, 1997,
p. 345). According to Kagan, Snidman, Arcus, and
Reznick (1994) approximately 15–25% of the population
can be considered to be highly sensitive. High sensitivity
is often confused with neuroticism, introversion, and
shyness, but Aron and Aron found that SPS is separable
from social introversion and emotionality. To measure
high sensitivity Aron and Aron (1997) developed the
Highly Sensitive Person Scale (HSPS), conceptualized as a
unidimensional measure. However, empirical findings do
not convincingly support the authors’original assumption
that the postulated factor structure of the HSPS is unidi-
mensional. In the next section we will give an overview
of the divergent results according to the dimensionality
of the HSPS. Since the HSPS is the only, and the most
frequently, used measure of sensory-processing-sensitivity,
the aim of the present study is to address a number of
critical issues which have been raised about the scale.
One of them, the focus of this article, is whether the test
measures one dimension or more dimensions. Further-
more, none of the studies using the HSPS have investi-
gated the measurement invariance of the scale, neither
for females and males nor for high versus low sensitive
individuals. Finally, our objective was to give first evi-
dence for the psychometric properties and validity of
the HSPS in German-speaking populations with Aron
and Aron scale as a starting point.
Theoretical Background
One-Factor-Structure
The HSPS (Aron & Aron, 1997) measures high sensitivity
with 27 items,theanswersareprovidedona7-point-
Likert-Scale which ranges from 1(= not at all)to
7(= extremely). Hofmann and Bitran (2007)confirmed
the one-factor structure through principal component
analysis (PCA) based on the scree-plot decision. Neal,
Edelmann, and Glachan (2002) also accepted a one-factor
European Journal of Psychological Assessment (2017) Ó2017 Hogrefe Publishing
DOI: 10.1027/1015-5759/a000411
http://econtent.hogrefe.com/doi/pdf/10.1027/1015-5759/a000411 - Tuesday, April 11, 2017 2:14:51 AM - Helmut-Schmidt-Universität IP Address:139.11.10.124
solution “as the most parsimonious explanation of the
data”(Neal et al., 2002,p.367). The reliabilities ranged
from α=.85 (Aron & Aron, 1997)toα=.87 (Aron & Aron,
1997; Hofmann & Bitran, 2007;Nealetal.,2002).
However, other studies could not confirm the unidimen-
sional structure of the HSPS. While Evans and Rothbart
(2008) and Cheek, Bourgeois, Theran, Grimes, and Norem
(2009) suggest a two-factor structure, Smolewska et al.
(2006), Liss, Mailloux, and Erchull (2008), Evers, Rasche,
and Schabracq (2008) reported a three-factor solution.
Other multi-factor solutions originate from Meyer, Ajchen-
brenner, and Bowles (2005; four-factor structure) and
Blach and Egger (2014; six-factor structure). In the follow-
ing sections of this paper, previous studies reporting more
than one factor solution for the HSPS will be discussed
briefly. All studies cited here, excluding Evers et al.
(2008), used the original version of the HSPS by Aron
andAron(1997).
Two-Factor-Structure
Evans and Rothbart (2008) developed a theoretical concept
of SPS containing the two factors “Negative Affect”and
“Orienting Sensitivity.”The concept was only partially
confirmed because the HSPS-items did not fit sufficiently
within their conception. Nevertheless, it was shown that
most of the items load either on “negative affect”or on
“orienting sensitivity.”An additional analysis of the one-
and three-factor solution showed that three factors fit better
than one-factor, but Evans and Rothbart deemed the two-
factor solution as most suitable. Cheek et al. (2009)tested
the factor structure in a large random sample of college
women (N=826) and extracted the factors “Temperamen-
tal Sensitivity”(Cronbach’sα=.86)and“Rich Inner Life”
(Cronbach’sα=.66). Despite the different labels of the
factors, Cheek et al. (2009) reported a pattern of loadings
which was similar to results reported by Evans and
Rothbart (2008).
Three-Factor-Structure
In an exploratory factor analysis (EFA) based on sample-
split-data from a large random sample of undergraduates
(N=851) and in a subsequent confirmatory factor analysis
(CFA), Smolewska et al. (2006) extracted the three
factors: Ease of Excitation (EOE), Aesthetic Sensitivity
(AES), and Low Sensory Threshold (LST). Both, the one-
and the three-factor-model fitted well. Further comparison
of the models shows that the three-factor model was
significantly better than the one-factor model (Smolewska
et al., 2006).
Liss et al. (2008) compare a two-factor structure with a
three-factor structure. Both solutions showed an adequate
fit, but the three-factor model fit significantly better. Evers
et al. (2008) tried to replicate the three-factor structure in a
Dutch sample and confirm the conclusion of Smolewska
et al. (2006) that the HSPS does not measure a unidimen-
sional construct.
Multi-Factor-Structure
Meyer et al. (2005) proposed more than three factors.
The PCA, through varimax rotation, resulted in a four-
structure solution: (1) General Sensitivity/Overstimulation,
(2) Adverse Reaction to Strong Sensations, (3)Psychological
Fine Discrimination, (4) Controlled Harm Avoidance.
All four factors were moderately correlated (.23–.51,
p<.01) and accounted for 48.4% of the variance, which
equals the explained variance reported by Aron and Aron
(1997) for the one-factor solution.
Central Issues
The studies reported so far do not conclusively answer the
question of whether high sensitivity is a homogeneous
construct or if it consists of several factors as the methodi-
cal procedure is criticized. Most of the cited studies rely on
PCA (sometimes in combination with CFA with the same
samples, except Smolewska et al., 2006). The small sample
sizes in several of the cited studies must also be noted as a
limiting aspect. Additionally, the studies with larger sample
sizes (e.g., Cheek et al., 2009;Smolewskaetal.,2006)
were limited because they used very homogeneous
samples, which consisted of either students or exclusively
women with limited age variation.
The aim of the present study is to examine the factor
structure of the German version of the HSPS (HSPS-G)
using a large heterogeneous sample. The study is aimed
to clarify the factor structure of the HSPS, to test measure-
ment invariance across sex and high and low sensitivity
individuals, and to provide the psychometric properties of
a German version of the HSPS. Finally, initial evidence is
provided for the validity of the HSPS-G.
Study 1
Method
Protocol
We combined active (snowball-technique) and passive
(invitation text with link to a website or web forum, offline
2 S. Konrad & P. Y. Herzberg, Psychometric Properties and Validation of a German High Sensitive Person Scale (HSPS-G)
Ó2017 Hogrefe Publishing European Journal of Psychological Assessment (2017)
http://econtent.hogrefe.com/doi/pdf/10.1027/1015-5759/a000411 - Tuesday, April 11, 2017 2:14:51 AM - Helmut-Schmidt-Universität IP Address:139.11.10.124
speech) selection by recruiting subjects through advertise-
ments placed in magazines, radio, television, and Internet
serving in the German-speaking area in order to reach a
broad spectrum of the general public. To ensure that a
sufficient number of highly sensitive people take part in
the study, approximately 30% of the participants were
recruited through the association “hochsensibel.org.”
Participation was voluntary and without payment, however,
a feedback of the questionnaire results was offered.
Anonymity was ensured.
Sample
Each participant received a code, which made it possible to
exclude duplicate records by cleansing of demographic data,
like age, sex, level of education, and mail address. Also,
incomplete or records with suspicious response patterns
were excluded. No other exclusion criteria were used and
atotalof3,588 individuals participated in the study.
On average, respondents were 39.17 years (range 18–80)
of age (SD =11.47); 3,015 (84%) out of 3,588 were female.
Sociodemographic information for the full sample is
describedindetailinTable1. For analytical reasons, the
sample was split randomly into two halves and consisted
of 1,794 cases per sample. Participants from subsample 1
(n=1,794) were on average 39.19 years (range 18–77)of
age (SD =11.41)and288 (16%) were male. In subsample 2
(n=1,794), the average age was 39.14 (range 18–80;
SD =11.53)and285 were male (16%).
Instruments
High Sensitive Person Scale (HSPS; Aron & Aron, 1997),
German Translation
To begin with, the HSPS was translated independently
into German by two qualified psychologists. Afterwards,
the two translations were compared with each other and
with the German version from Blach and Egger (2011).
Table 1.Demographics validation sample
Study 1 (CFA) Study 2 (HSP-BSCL)
Frequency Percent Frequency Percent
Gender Male 573 16.0 91 22.9
Female 3,015 84.0 306 77.1
Marital status Single 1,190 33.2 130 32.7
Stable relationship 885 24.7 99 24.9
Married 1,106 30.8 119 30.0
Divorced 382 10.6 45 11.3
Widowed 25 0.7 4 1.0
Household Living alone 1,356 37.8 145 36.5
Partnership 1,681 46.9 172 43.3
Parental home 141 3.9 17 4.3
Shared accommodation with others 410 11.4 63 15.9
Maximum educational attainment Left school without a certificate 11 0.3 1 0.3
Diploma from main school 63 1.8 10 2.5
Diploma from middle school 321 8.9 32 8.1
Advanced technical school certificate 293 8.2 29 7.3
Abitur (general qualification for university entrance) 706 19.7 98 24.7
Completed apprenticeship 436 12.2 34 8.6
Diploma from a college of applied sciences 601 16.8 49 12.3
University degree 931 25.9 119 30.0
Graduated and habilitated 98 2.7 14 3.5
Other degrees 128 3.6 11 2.8
Employment Pupil 53 1.5 8 2.0
Student 579 16.1 79 19.9
Civil servants 1,540 42.9 141 35.5
Salaried employees 193 5.4 42 10.6
Housewife/househusband 197 5.5 21 5.3
Self-employed 582 16.2 54 13.6
Unemployed 273 7.6 33 8.3
Pensioners 171 4.8 19 4.8
S. Konrad & P. Y. Herzberg, Psychometric Properties and Validation of a German High Sensitive Person Scale (HSPS-G) 3
European Journal of Psychological Assessment (2017) Ó2017 Hogrefe Publishing
http://econtent.hogrefe.com/doi/pdf/10.1027/1015-5759/a000411 - Tuesday, April 11, 2017 2:14:51 AM - Helmut-Schmidt-Universität IP Address:139.11.10.124
The translation was then proofed for correctness by con-
ducting a retranslation into English by a native speaker.
This procedure maximized conceptual uniformity (Brislin,
Lonner, & Thorndike, 1973). Slight variations from the
translations were resolved through discussion. Some items
(1,7,22,26,seeTable2) from Aron’s original version were
modified. Some included complex enumerations and items
consisting of more than one aspect were split. The overall
content was not changed. The resulting German version
contains 12 additional items than the original version,
resulting in 39 items in total. In accordance with Aron
andAron(1997; study 4) and Blach and Egger (2011), a
5-point-Likert-Scale (0=hardly at all to 4=extremely)
was adopted. The original German version can be provided
by the authors upon request.
Statistical Analysis
Given the large sample size, a CFA was conducted on half
of the sample and a further confirmatory factor analysis
was conducted on the second half based on the results of
the first sample.
The three models were subjected to a CFA using LISREL
9.1. Since the variables were categorical and the sample size
was sufficiently large, analyses were based on diagonally
weighted least squares estimations (WLSMV) with robust
standard errors. This approach takes the categorical nature
of the variables into account and also considers that
categorical variables violate the assumption of multivariate
normality in CFA. Furthermore, it ensures unbiased
parameter estimations and standard errors (Beauducel &
Herzberg, 2006).
Several fit indices were examined to evaluate the overall
fit of each model. The chi-square goodness-of-fit statistic
was statistically significant for both models, suggesting that
neither fit the data. However, the chi-square statistic is
sensitive to sample size, so it is rarely used as a sole index
of model fit. Therefore, three incremental indices of fit
were examined: the normed fit index (NFI), the compara-
tive fit index (CFI), and the Tucker-Lewis index (TLI).
Incremental indices reflected the improvement in fit
gained by a given factor model relative to the most
restrictive (null or independence) model. All three incre-
mental indices are scaled from 0(= no fit)to1(= perfect
fit). Hu and Bentler (1999) advised that values close to
.95 are indicative of a good fit. Finally, the root mean square
error of approximation (RMSEA) is a population discrep-
ancy function that compensates for the effects of model
complexity. The closer the RMSEA coefficient is to 0,the
better the fit of the model. According to Browne and
Cudeck (1993), a RMSEA value of .05 or less indicates a
close fit of the model in relation to the degrees of freedom,
whereas a value of .08 or less indicates a reasonable error
of approximation.
Results
Factorial Structure of the HSPS
The statistics used to evaluate the three CFA models are
showninTable3. By this standard, the one-factor model
did not meet the evaluation criteria. Although the two-
factor model is an improvement compared over the one-
factor model, the RMSEA exceeds the critical value of .08.
The three-factor model also exceeds the critical RSMEA
value of .08,butisnotaspronouncedasthetwo-factor
model. Therefore, we concluded that the three-factor model
was the best fitting model and then we subjected it to further
investigation by consecutive modification; thus, 11 items
with a low loading (< .40)orsmallR
2
(< .25) were deleted.
Furthermore, two items (24,34)withhighintercorrelations
to other items were removed. Finally, modification indices
suggested that it was better to assign two items to another
factor (items 8and 27 were suggested to load on EOE).
This modified model was tested on the other half of the
sample (n=1,794). The modified model met the evaluation
criteria for a good fit with regard to all fit-statistics consid-
ered (see Table 3). The factor intercorrelations for the three
factors are relatively high, with .36 between EOE and AES,
.65 between EOE and LST, and .49 between AES und LST.
Correlations are significant at a 1% level. The factor load-
ings ranged from .58 to .86 and are reported in Table 2.
The item-factor assignment, also in comparison with Aron
and Aron (1997) and Smolewska et al. (2006), is shown
in Table 2as well.
We also tested a second-order factor model to justify the
computation of a general score of the HSPS (HSPS-GS) with
the full sample. Again, we based our analyses on diagonally
weighted least squares estimation (WLSMV) with robust
standard errors. The second-order model, with the three
scales as first-order factors, showed a statistically significant
chi-square goodness-of-fit statistic (w
2
=7,027.99,df =296,
p<.001), suggesting a poor fit to the data. However, the
chi-square statistic is sensitive to sample size, so it is rarely
used as a sole index of model fit (Hu & Bentler, 1999).
In terms of the absolute model fit indices, the second-order
model provided a reasonable fit to the data, RMSEA = .080,
90%CIofRMSEA[.079,.082]. However, in terms of
relative model fit, indices provided a very good fit to the
data, CFI = .974,TLI=.972. Factor loadings between items
and factors ranged from .56 to .86. Second-order factor
loadings were .80,.68,and.94 for EOE, AS, and LST,
respectively. At the end of the process the German version
of HSPS resulted in a 26-item version with the three sub-
scales EOE (10 items), AES (5items), and LST (11 items).
Measurement Invariance
The final model was tested for factorial invariance across
sex and high versus low SPS individuals. The analysis of
4 S. Konrad & P. Y. Herzberg, Psychometric Properties and Validation of a German High Sensitive Person Scale (HSPS-G)
Ó2017 Hogrefe Publishing European Journal of Psychological Assessment (2017)
http://econtent.hogrefe.com/doi/pdf/10.1027/1015-5759/a000411 - Tuesday, April 11, 2017 2:14:51 AM - Helmut-Schmidt-Universität IP Address:139.11.10.124
Table 2.Results of the CFA for Sample 2, Model 3: factor loadings and coefficient of determination for HSPS-G
Equivalent to item Assignment according to Scales
from Aron and Aron Smolewska et al. (2006; EOE AES LST Removed
Item (1997) with loadings) λR
2
λR
2
λR
2
λR
2
27. I become unpleasantly aroused when a
lot is going on around me.
19 LST (.53) .81 .66
08. I find myself needing to withdraw
during busy days into bed or into a
darkened room or any place where I
can have some privacy and relief from
stimulation.
5 AES (.39) .78 .60
22. I find it unpleasant to have a lot going
on at once.
23 EOE (.68) .75 .56
32. I get rattled when I have a lot to do in a
short amount of time.
14 EOE (.68) .74 .54
04. When I must compete, I become so
nervous or shaky that I do much worse
than I would otherwise.
26 EOE (.58) .68 .46
33. I make it a high priority to arrange my
life to avoid upsetting or overwhelming
situations.
24 EOE (.36) .68 .46
05. Other people's moods affect me. 3 EOE (.36) .67 .45
36. When I´m observed while performing a
task, I become so nervous or shaky
that I do much worse than I would
otherwise.
26 EOE (.58) .65 .43
30. Changes in my life shake me up. 21 EOE (.65) .65 .42
21. I startle easily. 13 EOE (.42) .59 .35
03. I seem to be aware of subtleties in my
environment.
2 AES (.65) .75 .56
12. I have a rich, complex inner life. 8 AES (.76) .72 .52
31. I notice and enjoy delicate or fine
scents.
22 AES (.68) .64 .41
39. I notice and enjoy works of art. 22 AES (.68) .58 .34
15. I am deeply moved by music. 10 AES (.69) .58 .33
35. I am bothered by intense stimuli, like
loud noises or chaotic scenes.
25 LST (.74) .86 .75
16. I am easily overwhelmed by things like
sirens close by.
7 LST (.70) .85 .72
38. I am easily overwhelmed by strong
sensory input like hearing.
1 .80 .64
14. I am made uncomfortable by loud
noises.
9 LST (.70) .78 .62
01. I am easily overwhelmed by strong
sensory input like smell.
1 .78 .61
06. I am easily overwhelmed by things like
strong smells.
7 LST (.70) .76 .57
10. I am easily overwhelmed by things like
bright lights.
7 LST (.70) .75 .56
20. I am easily overwhelmed by strong
sensory input like see.
1 .75 .56
28. I am easily overwhelmed by strong
sensory input like feel.
1 .69 .48
13. I am easily overwhelmed by strong
sensory input like taste.
1 .68 .46
18. I am easily overwhelmed by things like
coarse fabrics.
7 LST (.70) .62 .39
24. I am annoyed when people try to get
me to do too many things at once.
16 EOE (.62) .63 .39
34. I notice and enjoy sounds. 22 AES (.68) .61 .38
(Continued on next page)
S. Konrad & P. Y. Herzberg, Psychometric Properties and Validation of a German High Sensitive Person Scale (HSPS-G) 5
European Journal of Psychological Assessment (2017) Ó2017 Hogrefe Publishing
http://econtent.hogrefe.com/doi/pdf/10.1027/1015-5759/a000411 - Tuesday, April 11, 2017 2:14:51 AM - Helmut-Schmidt-Universität IP Address:139.11.10.124
the factorial invariance across sex was based on the full
sample of 2,977 female and 542 male participants.
We applied a sequential constraint imposition approach to
test for factorial invariance by means of multiple-group
CFA (Byrne, 1989). Acknowledging that the variables were
categorical and our sample size was sufficiently large, we
again used the WLSMV estimator with robust standard
errors. Five models were tested. Configural invariance
implies that the number of latent variables and the pattern
of loadings of latent variables on indicators are similar
across sex. Weak invariance (i.e., metric invariance) implies
that the magnitude of the loadings is similar across sex.
Strong invariance for ordinal indicators implies that, not
only the item loadings, but also the item thresholds are
similar across females and males. Strict invariance implies
that the factor loadings, thresholds, and residual variances
are constrained to be equal across groups. Finally, the strict
invariance can be completed with equal means across both
sexes. Thus, the final invariance model constrains factor
loadings, thresholds, residual variances, and means are
equal across groups.
Fit indices for the five invariance models are reported in
the upper part of Table 4. The goodness-of-fit indices indi-
cated a reasonable fit for the configural invariance model,
thus configural invariance is established. Restricting
loadings to be equal for both sexes yielded a significant
w
2
-difference value
1
of 139.54 with df = 29,(p<.001).
Although the w
2
-difference value is statistically significant,
we assume invariance of the factor loadings across the
two groups. This reasoning is based on the fact that the
w
2
-difference value is sensitive to sample size. Given our
extensive sample size, it is very likely that even a small
w
2
-difference becomes statistically significant. Furthermore,
the relative (CFI) and the absolute model fit (RMSEA) are
only marginally affected, differences for both fit indices
are very small (.0008 and .002, respectively), lending
support for equal factor loadings across the two groups.
This is underpinned by simulation studies, where ΔCFI
proved the empirically best supported criterion to define
invariance; ΔCFI < .01 was chosen to decide whether a
substantial decrease in model fit occurs (Chen, 2007).
In addition, inspection of factor loadings reveals only slight
differences between females and males. The next model,
holding not only the item loadings, but also the item thresh-
olds similar for both sexes, results in a good fitting model.
Fit differences are evaluated with the same reasoning as
before, that is, although the w
2
-difference value is statisti-
cally significant; differences in CFI and RMSEA are smaller
Table 2. (Continued)
Equivalent to item Assignment according to Scales
from Aron and Aron Smolewska et al. (2006; EOE AES LST Removed
Item (1997) with loadings) λR
2
λR
2
λR
2
λR
2
26. I make a point to avoid violent movies
and TV shows.
18 LST (.57) .50 .25
02. I notice and enjoy tastes. 22 AES (.68) .49 .24
07. I tend to be more sensitive to pain. 4 EOE (.36) .47 .22
29. Being hungry triggers a strong reaction
in me and disrupts my concentration
or mood.
20 EOE (.56) .45 .20
17. My nervous system feels so frazzled
sometimes that I just have to by
myself.
11 .44 .20
37. When I was a child, my parents or
teachers saw me as sensitive or shy.
27 EOE (.47) .40 .16
23. When people are uncomfortable in a
physical environment I tend to know
what needs to be done to make it more
comfortable (like changing the lighting
or the seating).
15 AES (.53) .38 .14
25. I try hard to avoid making mistakes or
forgetting things.
17 EOE (.36) .36 .13
11. I am conscientious. 12 AES (.53) .35 .12
09. I am particularly sensitive to the
effects of caffeine. 6LST (.70) .33 .11
19. I am tidy. 12 .31 .10
Notes. EOE = Ease of Excitation; AES = Aesthetic Sensitivity; LST = Low Sensory Threshold (N= 1,755).
1
Please note that w
2
-difference test with the WLSMV estimator is not tested by computing simple difference between the baseline model and the
more constrained model. The w
2
-difference test applied here used the Satorra-Bentler scaled w
2
-square approach (Satorra & Bentler, 2001).
6 S. Konrad & P. Y. Herzberg, Psychometric Properties and Validation of a German High Sensitive Person Scale (HSPS-G)
Ó2017 Hogrefe Publishing European Journal of Psychological Assessment (2017)
http://econtent.hogrefe.com/doi/pdf/10.1027/1015-5759/a000411 - Tuesday, April 11, 2017 2:14:51 AM - Helmut-Schmidt-Universität IP Address:139.11.10.124
than the < .01 criterion. This reasoning also holds true for
the strong invariance model (see Table 4). The best fitting
model was the strict invariance model, with highest values
for TLI and lowest for RMSEA (.966 and .076,respec-
tively). This was supported by a nonsignificant w
2
-difference
and differences in CFI and RMSEA smaller than .01.
Progressing to the final model with strict invariance
supplemented with equal means resulted in poor absolute
model fit with RMSEA = .092 and poor relative fit
(TLI = .950,CFI=.946). Adding the mean equality across
sex resulted in a marked ΔCFI (.0169) which exceeded the
<.01 criterion. Thus, imposing the mean constraint led to a
poor model fit. Thus, the strict invariance model provided
the best fit in absolute terms compared to the preceding
models, and –given the imposed restrictions –was not
significantly inferior to these models, indicating equality
of loadings, thresholds, and residual variances across sex.
The issue of mean differences is investigated in the next
section.
As above, we used the whole sample of Study 1to inves-
tigate the measurement invariance between high versus
low HSP individuals and report the fit of the measurement
model for the whole sample. For the analysis of the factorial
invariance between high versus low SPS individuals, we set
the 80th percentile of the total score as cut-off point, which
is the mean of the estimate (15–25%) provided by Kagan
et al. (1994) of the population that can be considered as
highly sensitive. For females, the cut-off point for HSPS-
GS was a sum score > 88 and for males > 82.Thiscut-off
point divided the sample into n=718 high sensitive and
n=2,801 low sensitive individuals. We applied the same
sequential constraint imposition approach as for sex, but
we were not able to treat the variables as categorical
because two variables had one empty category in the high
sensitivity sample, which does not allow for utilization of
the WLSMV. Therefore, we treated the variables as contin-
uous variables and used robust Satorra-Bentler maximum
likelihood estimation. Fit indices for the five invariance
Table 4.Testing for measurement invariance across sex and between high versus low sensory processing sensitivity (SPS) individuals
Modell
a
w
2
df Δw
2
Δdf p TLI CFI ΔCFI RMSEA ΔRMSEA
Females versus Males
Configural 10,881.35 922 –––.964 .966 –.078 –
Weak (+ loadings) 11,156.82 951 139.54 29 < .001 .964 .965 .0008 .078 .002
Strong (+ thresholds) 11,801.35 1044 379.23 93 < .001 .965 .963 .0019 .077 .004
Strict (+ residual variances) 12,074.00 1076 27.80 32 0.68 .966 .963 .0008 .076 .008
All (+ means) 17,060.80 1079 83.72 3 < .001 .950 .946 .0169 .092 .009
High versus Low HSPS
Configural 3,461.80 592 –––.916 .860 –.051 –
Weak (+ loadings) 3,973.70 615 158.40 23 < .001 –.849 .0110 .052 .001
Strong (+ intercepts) 4,952.70 638 665.70 23 < .001 –.810 .0390 .057 .005
Strict (+ residual variances) 6,910.00 664 1,319.30 26 < .001 –.730 .0800 .067 .010
All (+ means) 28,807.30 667 5,853.10 3 < .001 –.000 .7300 .139 .072
Notes. Number of observations per group: female = 2,977, male = 542, high sensory processing sensitivity individuals = 718, low sensory processing
sensitivity individuals = 2,801.
a
Models following a sequential constraint imposition. The analysis begins with the least constrained model (configural),
subsequent restrictions comprise the restriction that are imposed in the preceding model (e.g., the weak model is the configural model with equal loadings).
All models are nested. Chi square difference test according to Satorra-Bentler approach.
Table 3.Fit Indices for the three models for Sample 1 and for the three-factor model for Sample 2
Model w
2
df AGFI NFI TLI CFI RMSEA 90% CI RMSEA
Sample 1
1 Factor 1,2783.19 702 .940 .931 .931 .935 .099 [.097, .100]
2 Factor 9,351.85 593 .953 .948 .948 .951 .092 [.090, .093]
3 Factor 6,709.21 492 .955 .946 .946 .950 .085 [.083, .086]
Sample 2
3 Factor 3,445.97 296 .974 .975 .975 .977 .078 [.076, .080]
Notes. NFI = normed fit index; TLI = Tucker-Lewis index; CFI = comparative fit index; RMSEA = root mean square error of approximation. Sample 1:
n= 1,794; Sample 2: n= 1,794.
S. Konrad & P. Y. Herzberg, Psychometric Properties and Validation of a German High Sensitive Person Scale (HSPS-G) 7
European Journal of Psychological Assessment (2017) Ó2017 Hogrefe Publishing
http://econtent.hogrefe.com/doi/pdf/10.1027/1015-5759/a000411 - Tuesday, April 11, 2017 2:14:51 AM - Helmut-Schmidt-Universität IP Address:139.11.10.124
models are reported in the lower part of Table 4. The good-
ness-of-fit indices indicated a reasonable fit for the
configural invariance model, thus configural invariance is
established. Restricting loadings to be equal for both groups
yielded a significant w
2
-difference value, but this difference
is attributed to the large sample size (see explanation
above). In contrast, the absolute model fit (RMSEA) is only
marginally lower compared to the previous model; the
difference is small (ΔRMSEA = .001). The CFI difference
is .011, which slightly exceeded the ΔCFI < .01 criterion.
The next model, holding not only the item loadings, but
also the residuals similar for both groups’results clearly
exceeds the ΔCFI < .01 criterion. The subsequent models
also exceed the ΔCFI < .01 criterion. Therefore, we find
merely support for equal factor loadings across the two
groups, which indicate weak measurement equivalence
only between high and low sensitivity individuals.
Mean Differences Between Male and Female
To examine the effect of gender on high sensory-
processing-sensitivity, a multivariate variance analysis
(MANOVA) was conducted. The factor was gender; the
dependent variables were the three factors, EOE, AES,
and LST, which were gained by means of CFA. Age was
included as a control variable. The analysis showed signifi-
cant group differences between men and women on all
three HSPS-subscales. The multivariate test (Pillai’sTrace)
was statistically significant for main effect, “sex”:
F(3,3,514)=108.62,p<.001,η
p
2
=.085. Differences in
mean values of men and women are significant on all three
HSPS-subscales. Results are reported in Table 5.
Summarizing mean differences, men reported lower
values on all subscales and had larger variances than
women with effect-sizes: Cohen’sd=.51 (AES)–.81 (LST).
Age had a positive relationship to EOE (r=.20), AES
(r=.20), LST (r=.26), and HSPS-GS (r=.26). All correla-
tions reported were significant at a 1% level.
Psychometric Properties of the HSP-Scale
Reliabilities for the final model were calculated using
subsample 2.Cronbach’sαfor the subscales was:
EOE = .87,AES=.70,LST=.91 and for HSPS-Global
Score (HSPS-GS) = .93. It was established that all
scales can be considered to be internally consistent in the
self-report while taking into account the number of items.
Mean values and standard deviations for the subscales
and HSPS-GS were: EOE: M=29.34,SD =7.08;AES:
M=16.16,SD =3.09;LST:M=28.71,SD =9.17;andfor
HSPS-GS: M=74.21,SD =16.85.
Test-Retest Reliability
Participants were recruited for Study 2in the same way as
in Study 1. Participants in the HSPS-G once again as part of
other studies. After data cleansing (see above), a total of
296 subjects were analyzed to determine the test-retest
reliability. On average, the respondents were 39.73 years
(18–69)ofage(SD =11.65), and 49 (15%) were male.
Two hundred ninety-six subjects had participated 1–3-
months before.
After 1–3months, the following test-retest reliability was
noted: EOE = .85,AES=.81,LST=.86, and for HSPS-
GS = .88. All results reported are significant at a 1% level.
Study 2
Construct Validity
Several authors related high sensory-processing-sensitivity
with psychological symptomatology and mental health
(e.g., Ahadi & Basharpoor, 2010; Benham, 2006;Hofmann
&Bitran,2007; Meyer et al., 2005; Neal et al., 2002).
Therefore, in Study 2we investigated the relationship
between high sensory-processing-sensitivity and psycholog-
ical symptomatology.
Method
Statistical Analysis
Initially, the normal distribution had been tested by the
Kolmogorov-Smirnov test. It was disapproved and a non-
parametric procedure was calculated. Nevertheless,
grouping was conducted in order to examine differences
in the psychic symptom stress between highly-sensitive
(HSP) and non-highly-sensitive persons (NON-HSP).
Table 5.Mean differences between male and female
Men (N= 542) Women (N= 2,977)
M SD M SD F-value (1, 3,516) pCohen’sd95% CI
EOE 25.45 8.41 30.05 6.41 208.65 < .001 0.68 [.588, .774]
AES 14.92 3.44 16.42 2.87 111.72 < .001 0.51 [.414, .598]
LST 22.74 10.15 29.75 8.35 303.12 < .001 0.81 [.717, .904]
Notes. EOE = Ease of Excitation; AES = Aesthetic Sensitivity; LST = Low Sensory Threshold; M= mean; SD = standard deviation; F-value: MANOVA;
d= Cohen’s value (effect size); CI = confidence interval.
8 S. Konrad & P. Y. Herzberg, Psychometric Properties and Validation of a German High Sensitive Person Scale (HSPS-G)
Ó2017 Hogrefe Publishing European Journal of Psychological Assessment (2017)
http://econtent.hogrefe.com/doi/pdf/10.1027/1015-5759/a000411 - Tuesday, April 11, 2017 2:14:51 AM - Helmut-Schmidt-Universität IP Address:139.11.10.124
In this, men and women have been looked at separately
because they differ from each other in respect to the mean
values in SPS. As above, the 80th percentile of the HSPS-GS
has been used to determine the group of highly-sensitive
individuals. This resulted in a cut-off for sum score of
81 in the HSPS-GS of the men and in a cut-off of 88
of the women. The subjects of the high-sensitivity group
were chosen in that way. An arbitrary, non-highly-sensitive
person has been contrasted to, matched by age, gender,
and profession resp. highest educational attainment.
For the ongoing analyses, using the Mann-Whitney-
U-Test, mean values of two very homogeneous groups were
compared in the scales of the Brief Symptom Checklist
(BSCL). The problem of the α-error-accumulation by
multiple testing was countered by adjusting αto p.004
(division of p=.05 bythenumberoftests12, Bonferroni
correction).
Sample
We collected data as part of a bigger study on the website of
Professorship of Personality Psychology and Psychological
Assessment at the Helmut Schmidt University/University
of the Federal Armed Forces, Hamburg. Incomplete and
duplicate data and data with conspicuous answer patterns
were removed. In the end, a total of 397 persons partici-
pated in the study. On average, respondents were
39.01 years old (range 18–68,SD =12.26)and306 (77%)
were female. Detailed sociodemographic information is
describedinTable1. After group classification, 184 subjects
were analyzed, from those 140 (76%) were women.
The age ranged from 21 to 68 years (M=43.71,SD =11.02).
Instruments
High Sensitive Person Scale (HSPS; Aron & Aron,
1997), German Translation
The psychometrically tested HSPS from study 1was used in
study 2. The internal consistency (Cronbach’sα)instudy2
was .95 for HSPS-GS. For subscales, the following internal
consistencies are applicable: EOE = .90,AES=.75,and
.94 for LST. The factor intercorrelations for the three
factors were higher than in the first study with .62 between
EOE and AES, .76 between EOE and LST, and .63 between
AES und LST. Correlations were significant at the 1% level.
Brief Symptom Inventory (BSI; Derogatis &
Melisaratos, 1983; German version: Brief Symptom
Checklist (BSCL; Franke, 2017; formerly Brief
Symptom Inventory, BSI; Franke, 2000)
BSCL is a 53-item short version of Symptom Checklist-
90-Revised-Standard (SCL-90®-S; Derogatis & Melisaratos,
1983;Franke,Stenzel,Rank,Herbold,&Küch,2015), which
only included the 4–7items with highest loadings per scale.
The checklist included the same items as of the BSI, but the
order of a few items was slightly modified. Psychological
symptomatology can be rated on a 5-point-Likert-Scale
from notatallto extremely in the last 7days and evaluated
for nine scales (hostility, anxiety, depression, paranoid
ideation, phobic anxiety, psychoticism, somatization, inter-
personal sensitivity, obsessive-compulsive) and three global
values. The Global Severity Index (GSI) records a funda-
mental burden. The GSI is presumed to be the best
indicator for the current extent of mental load in total
because it correlates with the intensity of mental load with
all 90 items. The Positive Symptom Distress Index (PSDI)
provides information about the intensity of the answers.
It also provides information about the extent of mental
load of those items, where a mental load was reported.
The Positive Symptom Total (PST) is also regarded as a
simple measure for the number of symptoms, which are
subject to a mental load, regardless of the extent. Thereby,
the scale comprehends a multidimensional view of the
psychological strain of a person. Furthermore, with use of
the checklist, it can be determined if the respondent
fulfilled the conditions of case definition. The requirements
of considerable psychological strain are considered to be
fulfilled if the respondent achieved a T-value 63 on at
least two BSCL-scales and/or the GSI-value. Reliabilities
in this study reached from Cronbach’sα.63 (interpersonal
sensitivity) to .96 for GSI. Moreover, the scales’“
obsessive-
compulsive,”“psychoticism,”and “phobic anxiety”were
lower than .70. For the remaining scales, internal consisten-
cies were higher than .72.
Results
The analysis of the case definition shows that 297 (74%)
participants met the criteria of conspicuous load, that is,
the respondent achieved a T-value 63 on at least two
BSCL-scales and/or the GSI-value. Particularly, 60%of
the respondents between the ages of 25 and 55 are
encumbered conspicuously.
Results of the correlation analysis are shown in Table 6.
The results of the correlation coefficient show that EOE,
LST, and the HSPS-GS exhibited low to moderate positive
correlations to all subscales and global values from BSCL.
The HSPS-subscale AES showed low to middle positive
correlations to all subscales and global values from BSCL
and seems to be distinct from the other subscales of HSPS.
Since the factors of the HSPS were found to be highly
intercorrelated, the other two factors were partialed out,
respectively. Results are represented as well in Table 6
(second row). After partialing out very weak significant
correlation is still showing between AES and “anxiety”
S. Konrad & P. Y. Herzberg, Psychometric Properties and Validation of a German High Sensitive Person Scale (HSPS-G) 9
European Journal of Psychological Assessment (2017) Ó2017 Hogrefe Publishing
http://econtent.hogrefe.com/doi/pdf/10.1027/1015-5759/a000411 - Tuesday, April 11, 2017 2:14:51 AM - Helmut-Schmidt-Universität IP Address:139.11.10.124
(r=.15)andPSDI(r=.11). All the other scales of the BSCL-53
were not significant and are negatively associated with
psychological symptomatology partly. The scale, “psychoti-
cism,”was only not significant for factor LST. For all the
other scales of BSCL-53, only low correlations in the range
of r=.11 (depression) to .24 (hostility) can be determined.
For the factor EOE, low significant associations were found
in the range of r=.20 (phobic anxiety) to r=.36 (GSI).
Group differences based on the Mann-Whitney-U-Test
were significant for all subscales of the BSCL and showed
medium or large effect-sizes for all scales (Table 7).
Therefore, HSP differs on all subscales and global scores
from NON-HSP in their psychological strain. The averages
of the T-values are presented for both groups in Figure 1.
On the scales “hostility,”“anxiety,”and “psychoticism,”
NON-HSP shows T-values 63 as well. On the other
scales, T-values from 56.86 (depression) up to 62.38
(phobic anxiety) were reported. The HSP shows the lowest
values on the “depression”scale as well. For all other
scales, T-values of 63 are reported. The values were
between 64.35 for “interpersonal sensitivity”and 71.88
for “hostility.”As a further association, the contingency
coefficient and the odds ratios were determined separately
for both sexes. For men the coefficient of contingencies was
C=.37 and the OR =12, whereas for women, a coefficient
of contingencies of C=.26 and an OR =8.5were deter-
mined. Also, medium correlations between SPS and
psychological strain do exist. For men, the odds ratio of
developing a mental disorder is as high as 12,iftheyare
counted among the HSP-group. The odds ratio of
developing a mental disorder for women being part of the
HSP-group is at 8.5, which is a bit less than for men.
Discussion
The aim of the presented studies was to investigate the
factorial structure of the construct of high sensory-
processing-sensitivity using a large and broad sample as
all previous studies were based on small and homogeneous
samples (age, students, female). The results show Aron’s
one-dimensional model (Aron & Aron, 1997) is not suitable.
A two-factor structure (Evans & Rothbart, 2008)couldnot
be confirmed either. Based on the analyses, the presented
translation of Aron’s HSP-Scale (Aron & Aron, 1997)
supported a three-factor structure with the factors EOE,
AES, and LST. The four indicators identified by Aron, Aron,
and Jagiellowicz (2012)forSPS(1. inhibition of behavior,
2. deeper processing of sensory information, 3. sensitivity
to stimuli, and 4. emotional/physiological reactivity) will
be described with slight modifications as follows. EOE
records emotional reactivity to physiological stimuli,
whereas AES describes reflection and awareness as
Table 6.Correlations of sensitivity and symptom severity-related scales of BSCL-53 (N= 397)
Hostility Anxiety Depression
Paranoid
ideation
Phobic
anxiety Pychoticism Somatization
Interpersonal
senitivity
Obsessive-
compulsive GSI PST PSDI EOE AES LST
EOE .55*** .56*** .41*** .45*** .40*** .45*** .50*** .49*** .52*** .56*** .59*** .51***
.32*** .28*** .29*** .25*** .20*** .35*** .28*** .26*** .32*** .36*** .34*** .29***
AES .33*** .43*** .23*** .29*** .27*** .25*** .36*** .30*** .31*** .36*** .36*** .39*** .62***
.04 .15** .01 .02 .04 .03 .05 .01 .05 .04 .06 .11*
LST .50*** .52*** .35*** .41*** .42*** .34*** .47*** .45*** .47*** .51*** .54*** .45*** .76*** .63***
.24*** .16** .11* .15** .19*** .06 .12* .17*** .22*** .20*** .20*** .15**
HSPS-GS .54*** .57*** .39*** .45*** .43*** .40*** .52*** .49*** .52*** .56*** .58*** .51*** .92*** .76*** .95***
Notes. First row display raw correlation, second row display partial correlation (corrected for the remaining factors). EOE = Ease of Excitation; AES = Aesthetic Sensitivity; LST = Low Sensory Threshold;
HSPS-GS = High Sensitive Person Scale-Global Score. ***p.001; **p.01; *p.05.
10 S. Konrad & P. Y. Herzberg, Psychometric Properties and Validation of a German High Sensitive Person Scale (HSPS-G)
Ó2017 Hogrefe Publishing European Journal of Psychological Assessment (2017)
http://econtent.hogrefe.com/doi/pdf/10.1027/1015-5759/a000411 - Tuesday, April 11, 2017 2:14:51 AM - Helmut-Schmidt-Universität IP Address:139.11.10.124
Figure 1. Group differences between NON-HSP and HSP on the BSCL-53. Error bars represent the standard deviations.
Table 7.Comparisons of NON-HSP and HSP regarding their T-scores on the BSCL-53 analysed by the Mann-Whitney non-parametric test of
variance
NON-HSP (N= 92) HSP (N= 92)
MSDMean rank
Sum of
ranks Mean rank
Sum of
ranks Mann-Whitney-U-test UpCohen’sd
Hostility 68.08 9.69 71.29 6,558.50 113.71 10,461.50 2,280.50 5.43 < .001 0.85
Anxiety 68.07 8.08 73.55 6,767.00 111.45 10,253.00 2,489.00 4.84 < .001 0.74
Depression 59.36 7.79 74.72 6,874.50 110.28 10,145.50 2,596.50 4.53 < .001 0.68
Paranoid ideation 62.72 9.16 71.91 6,616.00 113.09 10,404.00 2,338.00 5.25 < .001 0.84
Phobic anxiety 65.01 7.67 73.87 6,796.00 111.13 10,224.00 2,518.00 4.76 < .001 0.73
Psychoticism 65.54 7.73 78.95 7,263.50 106.05 9,756.50 2,985.50 3.46 .001 0.50
Somatization 62.49 9.93 74.47 6,851.00 110.53 10,169.00 2,573.00 4.60 < .001 0.70
Interpersonal sensitivity 60.77 9.20 70.05 6,445.00 114.95 10,575.00 2,167.00 5.72 < .001 0.84
Obsessive-compulsive 62.39 8.51 71.49 6,577.50 113.51 10,442.50 2,299.50 5.36 < .001 0.83
GSI 65.04 8.23 70.53 6,489.00 114.47 10,531.00 2,211.00 5.60 < .001 0.85
PST 64.55 8.22 71.90 6,614.50 113.10 10,405.50 2,336.50 5.26 < .001 0.80
PSDI 63.30 8.66 73.33 6,746.00 111.67 10,274.00 2,468.00 4.89 < .001 0.80
Notes. GSI = Global Severity Index; PSDI = Positive Symptom Distress Index; PST = Positive Symptom Total; M= mean; SD = standard deviation;
d= Cohen’sd.
S. Konrad & P. Y. Herzberg, Psychometric Properties and Validation of a German High Sensitive Person Scale (HSPS-G) 11
European Journal of Psychological Assessment (2017) Ó2017 Hogrefe Publishing
http://econtent.hogrefe.com/doi/pdf/10.1027/1015-5759/a000411 - Tuesday, April 11, 2017 2:14:51 AM - Helmut-Schmidt-Universität IP Address:139.11.10.124
indicator for depth of processing. The factor LST describes
the over-excitability by external sensory input. The three
scales might be presented as defined by the content of
the respective items better than in Smolewska et al.
(2006). In addition, the general construct fits well into
the context of the above-defined three indicators, with
one exception being the AES scale, which is quite short
and of low reliability. Therefore, specific items containing
“withdrawal”and avoidance had to be eliminated due to
insufficient loads being met and thus insufficient depiction
of the indicator “behavioral inhibition.”
Some items have been removed because of their low
loadings. Furthermore, most of the contained items
improved their loadings in comparison with previous
studies (Aron & Aron, 1997;Smolewskaetal.,2006).
Accordingly, SPS is a multidimensional construct. This
result essentially agrees with the analysis of Smolewska
et al. (2006). However, the model had to be modified
and adjusted slightly. In addition, the original version of
the HSP-Scale contained smaller psychometric shortages,
which were to be eliminated in the new version. This
approach has been proven because it has appeared that
some of the items which had been split had in most cases
higher loadings. It can be assumed that they may involve
different aspects and so they are different indicators for
forecasting SPS. The same applies for the differentiations
between sensory modalities at EOE and LST. This indicates
that not all individual sensory modalities are equally
affected, and thus must be differentiated. Items 38,1,20,
28,and13 could not be compared with Smolewska’smodel
as item 1from Aron’soriginalversionwasdeleted.Thisisa
shortcoming of the model from Smolewska, since this item
is an essential aspect of LST, namely the stronger percep-
tion and intense processing of sensory stimuli and the
related lower threshold. If such an item is not taken into
account, LST and thus SPS cannot be adequately measured.
Loadings of more than .85 show that it is an essential aspect
of SPS. However, there are differences between the sensory
modalities. Item 1load of .56 in Aron’sstudy(Aron&Aron,
1997) and so a further improvement could be achieved in
the new model. Some other items, measuring empathy
and conscientiousness, do not thematically match the
postulated factors. Furthermore, it has not been ascertained
to what extent empathy and conscientiousness can be
considered to be aspects of SPS.
Originally, the HSP-Scale has been designed by Aron and
Aron as a one-dimensional scale. Therefore, different
subscales have not been destined and the scale included
a different number of items for each topic, consequently
varying strong factors from the beginning. As a conse-
quence, it is possible that these adaptations may contribute
to the finding of a three-factor version more than a one- or
two-factor structure.
Psychometric Properties
The presented translation and adaption of Aron and Aron’s
HSP-Scale shows good measurement properties. The sub-
scales are factorially valid and allow a reliable measure-
ment of the corresponding items. Previous reliabilities
(Cronbach’sα)forHSPS-GSrangedfrom.84 (Cheek
et al., 2009; Evers et al., 2008), .85–.87 (Aron & Aron,
1997), .87 (Hofmann & Bitran, 2007; Neal et al., 2002),
and .89 (Smolewska et al., 2006). The new, three-factor
modelresultedinCronbach’sαfrom .93 to .95 for
HSPS-GS. Better reliabilities can be reported for the three
factorsaswell:EOE=.87–.90,AES=.70–.75,LST=.91–.94.
All previous studies reported lower Cronbach’sα,indicat-
ing that our modifications of the items improved the
measurement properties of the questionnaire. By splitting
the items, less ambiguous factors which possess higher
reliabilities were generated. Furthermore, the investigation
of test-retest reliability after 1–3months indicates satisfac-
tory stability of the scales.
The relatively high correlations between factors suggest
restricted discriminant validity between these related
aspects of high sensitivity. In other words, as specified by
theory, the scales measure related but still separate aspects
of high sensitivity. The high correlations also explain the
reasonable (but not better) fit of the second-order model.
In contrast, the strong correlation between EOE and
LST suggests a lack of discriminant validity. However,
Smolewska et al. (2006) argued that the positive intercorre-
lation among these factors reflects the common underlying
mechanisms of high sensitivity. Hence, from a theoretical
point of view, this correlation is comprehensible.
Measurement Invariance
The measurement invariance of the German translation
shows that self-evaluation is equivalent for men and
women. Thus, the content-related interpretation of similar-
ities or differences is permissible between sexes. Further
analysis of sex differences revealed significant differences
between men and women across all three factors and
HSP-GS. Women indicated significantly higher mean
values and lesser dispersions for all factors and higher
global scores than men. These effects are dependent upon
the HSPS-G subscales and are characterized by medium to
large effects. These findings are consistent with those of
other studies, for example, Aron and Aron (1997), Blach
and Egger (2014).
With regard to measurement invariance across HSP
versus NON-HSP, only weak measurement equivalence
between high and low sensitivity individuals was estab-
lished. It can be expected that the latent constructs have
the same content-related meaning in all groups examined.
12 S. Konrad & P. Y. Herzberg, Psychometric Properties and Validation of a German High Sensitive Person Scale (HSPS-G)
Ó2017 Hogrefe Publishing European Journal of Psychological Assessment (2017)
http://econtent.hogrefe.com/doi/pdf/10.1027/1015-5759/a000411 - Tuesday, April 11, 2017 2:14:51 AM - Helmut-Schmidt-Universität IP Address:139.11.10.124
The intercepts of the manifested variable are not identical
over the groups and it is conceivable that there are
differences in item difficulties.
Construct Validity of the HSPS-G
In general, low to moderate correlations can be stated
between the SPS-factors and psychological symptomatol-
ogy, which is also reflected in case definition. AES corre-
lates only a bit lower with BSCL-scales. Due to high
intercorrelations of HSPS-G subscales, impacts from both
of the other factors were partialed out and show differenti-
ated interrelations with the BSCL-scales. EOE correlates
only low, LST shows few or even no correlations with
psychological symptoms. AES is merely lowly associated
with anxiety. Results are consistent with results from other
studies. Neal et al. (2002) reported a higher level of anxiety
but not of depression in conjunction with a higher level
inSPS.TheHofmannandBitran(2007) results revealed
that SPS is correlated with agoraphobia but not with social
anxiety. According to Hofmann and Bitran (2007), individ-
uals with a generalized subtype of social anxiety disorder
also reported higher levels of SPS than individuals with
non-generalized subtype, which implied that SPS is specifi-
cally associated with generalized subtype of social anxiety
disorder.
Correlations between higher SPS-values with higher
perceived stress and common illness symptoms were
reported, however the perceived stress level does not
moderate health (Benham, 2006). The relationship
between SPS and physical symptoms can only be attributed
to EOE and LST (Ahadi & Basharpoor, 2010). Furthermore,
Ahadi and Basharpoor (2010) report a significant relation-
ship between the three HSP-factors and anxiety as well as
between EOE and depression. They also reported a high
correlation in regard to neuroticism and physical
symptoms,anxiety,depression,lowmentalhealth,and
EOE (Ahadi & Basharpoor, 2010). Other contextually
related studies report a significant correlation between the
factors EOE and LST with anxiety and depression. AES
shows no connection to depression, however, to anxiety.
The researchers then assume that AES is conceptually
distinct from the other two factors (Liss et al., 2008).
This finding coincides with the results of the present study.
The relationship between AES and depression revealed the
least significant connection while the relationship between
AES and anxiety is relatively high, as is the relationship
between EOE and LST to anxiety and depression.
Meyer et al. (2005) report a relationship between SPS and
depression, anxiety, avoidant personality disorder, border-
line personality disorder, and anger. The correlation discov-
ered between SPS and hostility parallels the correlation
between SPS and anger reported by Meyer et al. (2005).
With regard to psychological symptomatology and SPS, a
group comparison was performed for the first time.
The results of this study showed a very large difference
between groups. The HSP-group indicates higher values on
all BSCL-53-Scales. The T-values of the NON-HSP group
were also exceptionally high, which indicates that high
psychological symptomatology is not equated with high
sensitivity, but they can occur together. When counted
among the HS-group, there is a higher susceptibility
(men = 12,women=8.5) of experiencing mental illness. This
is in accordance with results of Aron and Aron (1997)from
two different groups of high sensitive individuals. One group
reported an unhappy childhood and related variables and
the second group which was similar to nonsensitive individ-
uals, except for their sensitivity. It is assumed that sensitivity
moderates the relationship between parental environment
and the report of a negative childhood. The first group seems
to have a higher vulnerability especially for depression and
anxiety (Aron & Aron, 1997).
Since about 75% of the samples meet the criteria of the
case definition, it is possible that results could be biased
due to sample characteristics and so the relation between
SPS and the psychological symptomatology is overesti-
mated. The results of the partial correlation support this
assumption.
In total, the findings of the previous studies coincide with
the results of the present study. The present study shows
that all three factors of HSP-Scale correlate from low to
moderate with all scales of BSCL. The values in hostility
and anxiety are prominent and suggest that they are
mutually dependent. The low sensory threshold and the
related perceptual overload can lead to aggressiveness,
whichinturntriggersfearbecausethepersonhasnot
reacted adequately. Another possibility is that the
perceptual overload leads to anxiety, which then leads to
aggressiveness due to the feelings of helplessness.
Most of the highest correlations were established
between the factor EOE and all scales of BSCL-53 followed
by HSPS-GS. The lowest correlations were exhibited on the
factor AES. The factor intercorrelations are very high
and not independent, which could be a result of higher
correlations with the factor AES.
InsummaryitcanbesaidthatbothHSPandNON-HSP
experience relatively high psychological symptomatology.
But the difference is that HSP reports consistently higher
scores in all scales and global scores. With the increase of
SPS, the psychological symptomatology also increases.
If the influence of the other HSPS factors is partialed out,
these high correlations cannot be found. In particular, the
factor AES showed only a minor significant relationship
with anxiety and the PSDI of the BSCL-53. These differ-
ences between HSP and NON-HSP are not related to
differences in age, sex, or education level.
S. Konrad & P. Y. Herzberg, Psychometric Properties and Validation of a German High Sensitive Person Scale (HSPS-G) 13
European Journal of Psychological Assessment (2017) Ó2017 Hogrefe Publishing
http://econtent.hogrefe.com/doi/pdf/10.1027/1015-5759/a000411 - Tuesday, April 11, 2017 2:14:51 AM - Helmut-Schmidt-Universität IP Address:139.11.10.124
Limitation
The lack of differentiation between the items in previous
studies and the reliability for AES have already been
discussed and modified. It has already been discussed that
the modifications of HSPS could contribute to the finding of
a three-factor solution rather than a one- or two-factor solu-
tion. This cannot be excluded completely because more
items were generated for single factors by items being split.
The increase of items could contribute to an increase of
reliability and to a better data fit because the factors are
similar in terms of content. Regarding this point, it must
be stated that due to the modification of the answer format,
the results are not able to be easily compared to those from
previous studies. Since the scale shows similar results to
previous studies, the change of answer format should not
be a problem. However, adaptations of the scale through
item splits and changes to the answer format simultane-
ously indicate that further studies are necessary in order
to check if the scale endures or it is useful to repatriate
the answer format to 7-point-Likert-Scale. The adaptation
seems to have been proven because not all sensory
modalities were rated equally and some items are better
or poor predictors for high sensory-processing-sensitivity.
The samples used are another point of criticism. They
were not representative as they were not randomized. But
generating a genuine random sample is almost impossible.
In order to reduce the results to one big sample by spread-
ing a variety of invitations, diverse selection techniques
(active and passive) have been used, particularly with
Study 1(Thielsch, 2008). The samples selected online show
a much bigger diversity opposed to the offline ones
(Gosling, Vazire, Srivastava, & John, 2004).
A possible role could be the self-selection bias. This influ-
ence seems less relevant because not each person, who
identified with high sensitivity, achieved values which are
classified as a high sensitive person. A confounding of
variables (psychological symptomatology and sensitivity)
then comes into question. The data of the second study
suggests that along with psychological symptomatology,
sensitivity increases. Another question that arises is
whether or not people, due to their negative feedback from
their environment, are affected by more psychological
symptomatology. Both perspectives are conceivable and
must be differentiated in future studies.
Item formats are only formulated positively and there are
no items which have to be inverted. This makes the scale
particularly comfortable to use because the item scores only
need to be summed up for the respective factor. However,
in regard to the test and questionnaire construction, it is
recommended that scales and questionnaires meet
psychometric minimum standards and refer to acquies-
cence bias. Various diverging measures to control response
tendencies are recommended, that is, an adequate item
polarity with a good balance between positive and negative
items may be performed. This proceeding is controversial,
but it should be considered, because many purported HSP
could draw secondary advantages due to the transparency
of the process. Alternatively, items could be chosen, which
definitely exclude SPS in order to separate desired results
from undesired results.
References
Ahadi, B., & Basharpoor, S. (2010). Relationship between sen-
sory processing sensitivity, personality dimensions and mental
health. Journal of Applied Sciences, 10, 570–574. doi: 10.1093/
scan/nsq028
Aron, E. N., & Aron, A. (1997). Sensory-processing sensitivity and
its relation to introversion and emotionality. Journal of Person-
ality and Social Psychology, 73, 345–368. doi: 10.1037/0022-
3514.73.2.345
Aron, E., Aron, A., & Jagiellowicz, J. (2012). Sensory processing
sensitivity: A review in the light of the evolution of biological
responsivity. Personality and Social Psychology Review, 16,
262–282. doi: 10.1177/1088868311434213
Beauducel, A., & Herzberg, P. Y. (2006). On the performance of
maximum likelihood versus means and variance adjusted
weighted least squares estimation in CFA. Structural Equation
Modeling, 13, 186–203. doi: 10.1207/s15328007sem1302_2
Benham, G. (2006). The highly sensitive person: Stress and
physical symptom reports. Personality and Individual
Differences, 40, 1433–1440. doi: 10.1016/j.paid.2005.11.021
Blach, C., & Egger, J. W. (2011). “Hochsensible Persönlichkeit”–
Bericht zum Forschungsprojekt Hochsensibilität [“The highly
sensitive personality”–Report to research project high
sensitivity]. Psychological Medicine, 22,59–63.
Blach, C., & Egger, J. W. (2014). Highly sensitive persons –An
empirical investigation to a complex phenomenon. Psycholog-
ical Medicine, 25,4–16.
Brislin, R. W., Lonner, W. J., & Thorndike, R. M. (1973). Cross-
cultural research methods. New York, NY: Wiley.
Browne, M. W., & Cudeck, R. (1993). Alternative ways of assessing
model fit. In K. A. Bollen & J. S. Long (Eds.), Testing structural
equation models (pp. 136–162). Newbury Park, CA: Sage.
Byrne, B. M. (1989). A primer of LISREL. Basic applications and
programming for confirmatory factor analytic models. New York,
NY: Springer.
Cheek, J. M., Bourgeois, M. L., Theran, S. A., Grimes, J. O., &
Norem, J. K. (2009, February). Interpreting the factors of the
Highly Sensitive Person scale. Poster session presented at the
annual meeting of the Society for Personality and Social
Psychology, Tampa, FL.
Chen, F. F. (2007). Sensitivity of goodness of fit indexes to lack of
measurement invariance. Structural Equation Modeling, 14,
464–504. doi: 10.1080/10705510701301834
Derogatis, L. R., & Melisaratos, N. (1983). The brief symptom
inventory: An introductory report. Psychological Medicine, 13,
595–605. doi: 10.1017/S0033291700048017
Evans, D. E., & Rothbart, M. K. (2008). Temperamental sensitivity:
Two constructs or one? Personality and Individual Differences,
44, 108–118. doi: 10.1016/j.paid.2007.07.016
Evers, A., Rasche, J., & Schabracq, M. J. (2008). High sensory-
processing sensitivity at work. International Journal of Stress
Management, 15, 189–198. doi: 10.1037/1072-5245.15.2.189
14 S. Konrad & P. Y. Herzberg, Psychometric Properties and Validation of a German High Sensitive Person Scale (HSPS-G)
Ó2017 Hogrefe Publishing European Journal of Psychological Assessment (2017)
http://econtent.hogrefe.com/doi/pdf/10.1027/1015-5759/a000411 - Tuesday, April 11, 2017 2:14:51 AM - Helmut-Schmidt-Universität IP Address:139.11.10.124
Franke, G. H. (2000). BSI. Brief Symptom Inventory –Deutsche
version. Manual [Brief Symptom Inventory –German version.
Manual]. Göttingen, Germany: Beltz.
Franke, G. H. (2017). BSCL. Brief Symptom Checklist, manual.
Göttingen, Germany: Hogrefe.
Franke, G. H., Stenzel, S., Rank, C., Herbold, D., & Küch, D. (2015).
Die Brief Symptom Checklist (BSCL) im Einsatz bei Patientinnen
und Patienten der orthopädischen Rehabilitation [The Brief
Symptom Checklist (BSCL) in use with female and male patients
in the orthopedic rehabilitation] Berlin, Germany: Jahrestagung
des AK Klinische Psychologie. Retrieved from http://www.
researchgate.net/publication/281833605_Die_Brief_Symptom_
Checklist_%28BSCL%29_im_Einsatz_bei_Patientinnen_und_
Patienten_der_orthopaedischen_Rehabilitation
Gosling, S. D., Vazire, S., Srivastava, S., & John, O. P. (2004).
Should we trust web-based studies? A comparative analysis of
six preconceptions about internet questionnaires. The Ameri-
can Psychologist, 59,93–104. doi: 10.1037/0003-066X.59.2.93
Hofmann, S. G., & Bitran, S. (2007). Sensory-processing sensitivity
in social anxiety disorder: Relationship to harm avoidance and
diagnostic subtypes. Journal of Anxiety Disorders, 21, 944–954.
doi: 10.1016/j.janxdis.2006.12.003
Hu, L., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in
covariance structure analysis: Conventional criteria versus
new alternatives. Structural Equation Modeling, 6,1–55.
doi: 10.1080/10705519909540118
Kagan, J., Snidman, N., Arcus, D., & Reznick, J. S. (1994). Galen’s
prophecy: Temperament in human nature. New York, NY: Basic
Books.
Liss, M., Mailloux, J., & Erchull, M. J. (2008). The relationship
between sensory processing sensitivity, alexithymia, autism,
depression, and anxiety. Personality and Individual Differences,
45, 255–259. doi: 10.1016/j.paid.2008.04.009
Meyer, B., Ajchenbrenner, M., & Bowles, D. P. (2005). Sensory
sensitivity, attachment experiences, and rejection responses
among adults with borderline and avoidant features. Journal of
Personality Disorders, 19, 641–658. doi: 10.1521/pedi.2005.
19.6.641
Neal, J. A., Edelmann, R. J., & Glachan, M. (2002). Behavioural
inhibition and symptoms of anxiety and depression: Is
there a specific relationship with social phobia? The British
Journal of Clinical Psychology, 41, 361–374. doi: 10.1348/
014466502760387489
Satorra, A., & Bentler, P. M. (2001). A scaled difference chi-square
test statistic for moment structure analysis. Psychometrika, 66,
507–514. doi: 10.1007/BF02296192
Smolewska, K. A., McCabe, S. B., & Woody, E. Z. (2006). A
psychometric evaluation of the Highly Sensitive Person Scale:
The components of sensory-processing sensitivity and their
relation to the BIS/BAS and “Big Five”.Personality and
Individual Differences, 40, 1269–1279. doi: 10.1016/j.paid.
2005.09.022
Thielsch, M. T. (2008). Ästhetik von Websites. Wahrnehmung von
Ästhetik und deren Beziehung zu Inhalt, Usability und
Persönlichkeitsmerkmalen [Aesthetics of websites. Perception
of aesthetics and their relation to content, usability and
personality traits]. Münster, Germany: MV Wissenschaft.
Received January 28, 2016
Revision received November 8, 2016
Accepted November 14, 2016
Published online April 7, 2017
Sandra Konrad
Department of Personality Psychology and Psychological Assessment
Helmut-Schmidt-University
University of the German Federal Armed Forces Hamburg
Holstenhofweg 85
22043 Hamburg
Germany
konrads@hsu-hh.de
S. Konrad & P. Y. Herzberg, Psychometric Properties and Validation of a German High Sensitive Person Scale (HSPS-G) 15
European Journal of Psychological Assessment (2017) Ó2017 Hogrefe Publishing
http://econtent.hogrefe.com/doi/pdf/10.1027/1015-5759/a000411 - Tuesday, April 11, 2017 2:14:51 AM - Helmut-Schmidt-Universität IP Address:139.11.10.124
