Beruflich Dokumente
Kultur Dokumente
11
1. Introduction argue those challenging the model, the
paradigmatic use has not been reflected
In the 1960s, a new sub discipline of in a sufficiently critical way. The econo-
general psychology became institu- mist Brendan McSweeney, one of Hof-
tionalized, called cross-cultural psy- stede’s most famous critics, observes an
chology ( Jahoda and Krewer 1997: “on-going unquestioning acceptance of
3, 24). Until today, researchers of this Hofstede’s national culture research by
sub discipline have been following the his evangelized entourage” (McSweeney
aim of “comparing data from several 2002b:112).
cultures” (Triandis 1997:ix) in order to
detect intercultural differences, usually The present study retraces the discrep-
by means of standardized question- ancy of the extreme reactions to Hof-
naires (Smith and Schwartz 1997:81). stede’s model. We suspect that it is the
One of the most famous examples is debate about the validity of the cultural
the work of Geert Hofstede (Barmeyer dimensions which essentially contrib-
and Genkova 2010:122, Harrison / utes to the discrepancy. Can the Hofst-
McKinnon 1999:485). His comparison edean dimensions really serve as interna-
of national cultures2 is based on cultural tional standards of comparison? On
dimensions, which serve as standards which methodic foundation are they
of comparison. For a high number of based? To contribute to this debate, the
countries (Hofstede 2009:xix), Hof- remainder of the present article pro-
stede calculated scores in different ceeds as follows: First of all, the origins
dimensions, i. e. numeric values, which of the dimensions will be retraced. In
allow establishing international rank- chronological order, the different steps
ings and country clusters (Hofstede will be outlined, on the basis of which
2009:150). His approach has been the dimensions were composed and
evoking extreme reactions, in a positive filled with content. Thereafter, we will
as well as in a critical sense. One body present follow-up studies to the Model
of research uses it as a framework for a of Cultural Dimensions. Secondly,
high number of cross-cultural research Hofstede’s statement about the dimen-
projects (Blodgett et al. 2008:762, sions’ validity will be recapitulated. His
Barmeyer 2010:87). The list of areas in statement will then be confronted with
which the model is employed, contains the criticisms it evoked. This confronta-
a great variety3, ranging from informa- tion leads to the research questions of
tion technology (Myers / Tan 2002:25), the present study, if Hofstede’s cultural
management controlling (Harrison / dimensions can be considered valid con-
McKinnon 1999:485), innovation, structs or not. We test this hypothesis
leadership styles ( Jones 2007:6), over with regard to one of the dimensions,
intercultural relations, decision-making, the Uncertainty Avoidance dimension.
selection, training, job design, motiva- Therefore, we replicate Hofstede’s origi-
tion and human resource management nal questionnaire in a survey interrogat-
to marketing (Søndergaard 1994:453f.) ing German and French public school
and market research. Hofstede himself teachers and factory workers, and test
realizes: “Since the later 1980s the idea for measurement equivalence. In the
of dimensions of national cultures has final section we present the results and
become part of […] ‘normal science’” comment on their implications.
(Hofstede 2002:2). Accordingly, he
calls his book Culture’s Consequences.
Comparing Values, Behaviors, Institu- 2. Hofstede’s cultural
tions, and Organizations Across Nations dimensions theory
(1. edition 1980, 2. edition 2009, cop.
2001) a classic (Hofstede 2009:xvii). The starting point for the develop-
Thus, his dimension model has been ment of Hofstede’s cultural dimen-
cited, reviewed, replicated (Sønder- sions theory is best summarized by the
gaard 1994:447, Nakata 2009:3, Taras / following quotation of Geert Hofstede
Steel 2009:41) and used as a paradigm and his son, Gert Jan Hofstede: “In the
(Taras / Steel 2009:53). However, so late 1960s Geert accidentally became
13
For this purpose, he points out correla- Area of Live Low score on the High score on the
tions between the national indices and Uncertainty Avoi- Uncertainty Avoi-
external data (Hofstede 2009:41). As dance dimension dance dimension
external data, he includes the findings
of empirical studies, anecdotes and per-
sonal experiences (Hofstede 2009:27) Expression of Expressions have to be Expression of emotions
as well as socio-economic indices Emotions controlled. normal
(Hofstede 2009:68f.). The number of
data considered is high: “The count of
significant and independent correlations
Subjective Well-Being More subjective well-being Less subjective well-being
15
within the area of psychology (Merritt (Søndergaard 1994:447). Hofstede
2000), whereas Power Distance and Un- slightly modifies this categorization.
certainty Avoidance were rather focused He divides the applications of his
on by sociologists and business scientists model into the following categories: 1)
(Hoppe 1990, Hastings / Hastings reviews and criticism, 2) extensions to
1981). Later on, Hofstede revises his new nations and regions, 3) replication
specification of disciplines that should studies and 4) paradigmatic uses (Hof-
apply his model: “Reviews and criti- stede 2009:461). Some examples of 1)
cisms are most interesting when they Reviews and criticism have already been
come from unexpected areas” (Hofstede mentioned. As far as 2) extensions to
2009:463). As unexpected areas, he new nations and regions is concerned,
mentions, among others, information two examples shall be mentioned: first
technology, archive management, and the master thesis by Nanhekhan (1990)
nuclear power regulation. Indeed, the and second the study by Nasierowski
variety of disciplines that apply the and Mikula (1998). The former applied
model is great. But, so Hofstede asserts, the IBM questionnaire in Surinam and
the follow-up studies also show a great the latter in Poland. 3) The replica-
variety of quality (Hofstede 2009:66). tions are the most frequent category
Therefore, he dedicates a whole chap- of follow-up studies. “Replications are
ter of Culture’s Consequences, entitled studies hat administer questions used
“Using Culture Dimension Scores in in the IBM research to new samples
Theory and Research” to his require- from two or more of the same coun-
ments for continued research (Hof- tries” (Hofstede 2009:463). Hofstede
stede 2009:461ff.). Here, he warns of appreciates those kinds of studies, as
several pitfalls. Researchers should, for
he uses them to confirm the validity of
instance, not misunderstand the IBM
his dimension constructs. Within the
questionnaire as a personality test for
validation of his constructs by interpret-
individuals. Furthermore, Hofstede
ing the correlations between his and
criticises an ethnocentristic approach
external data, he dedicates one sub-
of those researchers who leave out the
chapter per dimension to “[s]traight
Masculinity/Femininity dimension
replications of the IBM survey” (Hof-
because they consider it politically
stede 2009:91f., 154f., 219f., 295f.).
incorrect.
Within straight replications, the IBM
Now we will further analyze the dif- questionnaire is either applied in his
ferent kinds of follow-up studies. It is original version, or the formulation of
difficult to give a representative over- certain items is adapted to the context
view covering the past 35 years, as many of the sample (Søndergaard 1994:448,
follow-up studies are master thesises Hofstede 2009:67). In any case, most
and dissertations which either have not replication studies follow the aim of
been translated into English or which testing Hofstede’s international rank-
have not been published at all (Søn- ings and the majority of them confirm
dergaard 1994:450). Therefore, we will them (Søndergaard 1994:451). Those
concentrate on, first, the works analyzed not confirming are usually blamed for
by the Danish business scientist Mi- not having adapted the items adequately
kael Søndergaard, and, second, studies to the context of the samples (Sønder-
mentioned by Hofstede in his texts and gaard 1994:452). As far as the samples
third some additional ones. In 1994, are concerned, most replication studies
Søndergaard analyzed approximately do not include more than one sam-
550 applications of the Hofstedean ple per nation. They aim at detecting
model. He had access to Hofstede’s international differences rather than
private library which allowed him to in- differences between subcultures. Only
clude unpublished works that had been very few studies compare units that are
sent to Hofstede. Søndergaard divides smaller than nations. One study entitled
the applications into four categories: “Exploring subcultural differences in
Citations, reviews, empirical replica- Hofstede: The case of the Chinese”
tions and applications as a paradigm (Huo / Randall 1991:160) compares
17
are confirmed by those found by others Item Position Title Request for Response options
in other matched samples, our matching of item of battery answer
was adequate” (Hofstede 2009:23). within
Whereas Hofstede is convinced of the the
validity of his dimensions, his critics battery
remain skeptical: “The use of Hofstede’s
dimensions […] raises more problems A37 18 MORE How often do you 1. I always feel this way
than it solves” (Baskerville 2003:10). ABOUT feel nervous or tense 2. Usually
What they find problematic is the YOUR at work? 3. Sometimes
JOB 4. Seldom
dimensions’ face validity, the wording 5. I never feel this way.
of the items in the IBM questionnaire
and the specificity of the IBM sample. A43 18 MORE How long do you 1. Two years at the most
ABOUT think you will conti- 2. From two to five years
The face validity has been tested in the THE nue working for this 3. More than five years
course of a study conducted in 2008. COMPANY company? (but I will probably
Here, two groups of altogether 157 leave before I retire)
students with experience in behavioral 4. Until I retire
science were asked to match the items of B60 9 ABOUT Company rules 1. Strongly agree
the IBM questionnaire to the dimen- GENERAL should not be broken 2. Agree
sions. For instance, the items belonging BELIEFS – even when the 3. Undecided
to the Uncertainty Avoidance dimen- employee thinks it is 4. Disagree
in the company’s best 5. Strongly disagree
sion were correctly matched by only
interests.
30,4% of the one group and by only
26% of the other one. This means that
only a third respectively a fourth of the Table 2: The original items on the basis of which Hofstede calculated the Uncertainty Avoidance
respondents considered the three items indices. Source: Hofstede et al. 1976.
as reflectors of the Uncertainty Avoid-
ance dimension as elaborated by Hof-
stede. This rate is evaluated as problem- Steiner / Benesch 2010:51; Lienert /
atically low (Blodgett et al. 2008:762). Raatz 1998:52). Moreover, the response
These kinds of findings or simply the options of this item do not constitute
reconsideration of the wording of the fixed references points, so they are not
items, as we will do in the following necessarily understood in the same
section, leads scholars like Behrens way by every respondent (Saris 1988).
(2007:96) to conclude that the Uncer- Whereas, for instance once per week
tainty Avoidance turns out as fiction. could by perceived as sometimes by one
respondent, it could be considered
A possible lack of validity could be due
seldom by another. In turn, the catego-
to the wording of the items. The items
ries always and never are fixed reference
which the Uncertainty Avoidance di-
points, but according to recent findings
mension is based on are presented Table
they might be perceived unrealistic
2. Using the means of these items, Hof-
by the respondents (Raab-Steiner /
stede created a formula8 to calculate the
Benesch 2010:51). Furthermore, as
Uncertainty Avoidance indices for 50
far as item A43 is concerned, respond-
nations and three regions, namely West
ents could consider more than one
Africa, East Africa and several Arab
response option adequate. They could,
countries. In the international ranking,
for instance, plan to continue working
Greece has the highest score of 112 and
for the same company two years at the
Singapore has the lowest of 8. The mean
most, and at the same time, until they
is 65 and the standard deviation is 24
retire. Finally, regarding item B60, the
(Hofstede 2009:151).
wording is long-winded and, therefore,
From these items, a set of problems incomprehensible. Moreover, it con-
evolves. To begin with, Item A37 is tains a negation. Negations increase the
double-barreled because it contains two complexity of the item and, therefore,
different states of feelings: “nervous” decrease the capability of respondents
and “tense”. Therefore, it does not meet to answer in a spontaneous and intuitive
the claim of unidimensionality (Raab- way. In this case, not only the item but
19
scope of IBM study, we test for measure-
ment invariance. Measurement invari-
ance means that individuals’ answers are
not dependent on their group charac-
teristics (Mellenbergh 1989, Meredith /
Millsap 1992, Meredith 1993), and,
thereby, that the concept is valid beyond
specific groups. Figure 4 illustrates the
measurement model of the Uncertainty
Avoidance dimension.
In Figure 1 Uncertainty Avoidance is
the unobserved latent concept which is
reflected by the three indicators A37,
A43 and B60. λi is the loading, τi is the
intercept and ei is the disturbance terms
for the ith item. It is assumed that the
disturbance terms have a mean of zero,
and are uncorrelated with each other
and with the latent variable. Moreover,
in order to assign a scale to the latent Figure 1: Measurement model of Uncertainty Avoidance. Source: Authors own figure.
variable the loading λ2 is fixed to 1 and
the intercept τ2 to 0.
There are three different levels of invari- and testing we rely on Jrule software
ances testing, in order: configural, met- (Van der Veld et al. 2008) based on the
ric, and scalar invariance. Configural in- procedure developed by Saris, Satorra
variance is achieved if the measurement and van der Veld (2009). Saris et al.
model fits the data well in the different show that the commonly used evalua-
groups and all item loadings are signifi- tion procedures for structural equation
cant (Davidov 2008:37). We will test models cannot be trusted as the test
for configural invariance in each group statistics and Fit indices are unequally
separately. Only if this is achieved, sensitive for different misspecifications.
metric invariance can be tested. Metric They propose using the modification in-
invariance requires that the loadings λi dex (MI) as test statistic for detection of
are the same across groups and is a nec- misspecifications (expressed as expected
essary condition for comparing relation- parameter change; EPC) in combina-
ships of the latent variable Uncertainty tion with the power of the MI test. This
Avoidance with other variables. If metric means that Jrule tests the model on the
invariance is established, in a final step parameter level rather than the model as
scalar invariance can be tested. Scalar a whole.
invariance requires that the intercepts
τi are also equal across groups and if
established allows comparing means 5. Results
across groups (Horn 1983, Meredith
1993, Steenkamp / Baumgartner 1998). The results can be summarized rather
Recalling quickly because in none of the groups
Hofstede’s approach, this means that configural invariance was found. In
only if this final condition is met, Hof- other words, in none of the groups the
stede’s formula can be used and groups items load on the latent concept Uncer-
can be compared. tainty Avoidance. Therefore, testing for
metric and scalar invariance has become
In order to conduct these tests we redundant, as both require configural
employ multigroup confirmatory factor equivalence as a precedent condition.
analysis (MGCFA) ( Jöreskog 1993, Table 4 summarizes the non-significant
Billiet 2002) the maximum likelihood effects conducting factor analysis in
estimator of LISREL 8.72 ( Jöreskog / each group separately; it presents the
Sörbom 2005). For model evaluation standardized solution which allows
Table 4: Findings of the test for configural invariance in each group. Source: Authors own table.
comparing the loadings within a group within these nations. From our results
as well as the unstandardized solution we conclude, that criticisms suspect-
which allows comparing the load- ing that the Uncertainty Avoidance
ing across groups. We report this for dimension is specific to the Hofstedean
the sake of completeness, although IBM sample are justified. Consequently,
the non-significant loadings make all this dimension is neither to be used as a
comparisons of loadings redundant. The standard of cross-national comparisons,
items which were supposed to reflect nor as the basis for general descriptions
the Uncertainty Avoidance dimension about countries as wholes.
according to Hofstede, do not act as
expected. The lack of validity we find leads to the
question what it is due to. One theo-
retical explanation would be cultural
6. Discussion change over time. Uncertainty Avoid-
In this study, we aimed to test the ance might have been a valid construct
validiy of the Uncertainty Avoidance in the 1970s and might have lost
dimension of the Hofstedean model of relevance over the years. This might be
Cultural dimensions. We traced back the reason for our sample from 2011 to
the origins of this model, highlighted not replicate it. This explanation though
the importance and use of it, mentioned does not convince in two respects. First,
the criticisms to it, and tested with Hofstede himself excludes the idea of
original data from France and Germany cultural change. “Cultures, especially
(2011), if the Uncertainty Avoidance national cultures, are extremely stable
dimension can be replicated and hence, over time” (Hofstede 2009:34). More
proves to be a valid construct beyond concretely speaking, national cultures
the original IBM study. We find that are supposed to be stable “across many
this is not the case. In none of our four generations” (Hofstede 2009:10), at
samples, the items have proven to be least until the year 2100 (Hofstede
indicators of the construct Uncertainty 2009:36). Whenever his data is blamed
Avoidance. This contradicts Hofstede to be out of date, Hofstede points to
who claims that his dimensions are the century-old roots of his dimensions
applicable not only in a high number of (Hofstede 2009:73). Here, he mentions
nations, but also among all subsamples the second objection to the explana-
21
tion of cultural change over time. If the Uncertainty Avoidance dimension has
dimensions really do reflect universal not proven valid, those studies show a
problems of every human society and if lack of sustainable foundation. Con-
country scores shall be published and struct validity should have been assured
discussed until today, then it is indis- by testing for measurement invariance
pensable that the dimensions have not across groups. As a conclusion, this does
lost their validity. not mean that a concept such as the
Uncertainty Avoidance does not exist,
In the beginning of the present study, but it means that it cannot be measured
we recapitulated in detail the founda- employing (variations of ) the items
tion and methodological approach of from the IBM study.
the Hofstedean model. Thereby, we de-
tected several reasons which could lead
to the lack of validity of dimension we 7. Bibliography
studied here. First of all, the aim of IBM Barmeyer, C. / Genkova, P. / Scheffer, J.
questionnaire was not to compare cul- (2010): Interkulturelle Kommunikation und
tures across countries and the idea of the Kulturwissenschaft. Passau: Stutz.
Uncertainty Avoidance dimension was
not present at all when the question- Barmeyer, C. (2010): Kulturdimensionen
naire was designed. Secondly, once the und Kulturstandards. In: Barmeyer, C. /
data from the IBM study was obtained, Gankova, P. / Scheffer, J. (Eds.): Interkul-
Hofstede split the three factors which turelle Kommunikation und Kulturwissen-
he found by means of factor analysis schaft. Passau: Stutz, pp. 87-117.
into four dimensions. The theoretical
Baskerville-Morley, R. (2005): A research
reason was to bring the dimension in
note: the unfinished business of culture.
line with the standard analytic issues by Accounting, Organizations and Society 30,
Inkeles and Levinson (1954). However, pp. 389-391.
practically this means that the dimen-
sions are not statistically independent Baumgartel, H. / Thomas, H. (1982):
from each other. Thirdly, given that the Review of Geert Hofstede: `Culture’s
aim of the questionnaire design was not Consequences: international differences in
to detect the Uncertainty Avoidance work-related values´. Personnel Psychology
dimension, the wordings of the items 35(1), pp. 192-196.
are doubtful. Hofstede himself consid-
ers this problematic and recommends Behrens, L. (2007): Konservierung von
adjustments in the formulation of the Stereotypen mit Hilfe der Statistik. Köln:
Institut für Linguistik, Allgemeine Sprach-
items. However, he still believes that
wissenschaft.
these three items are the indicators of
the Uncertainty Avoidance dimension.
Billiet, J. (2002): Cross-cultural equiva-
This reasons seems to explain why it was lence with structural equation modeling.
impossible to detect the Uncertainty In: Harkness, J. A. / Van de Vijver, F. J. R. /
Avoidance dimension with our data. Mohler, P. P.(Eds.): Cross-Cultural Survey
Methods. Hoboken: John Wiley, pp. 247-
All in all, our study made clear that the 264.
Uncertainty Avoidance dimension is
not a valid construct and, therefore, Blodgett, J. / Bakir, A. / Rose, G. (2008): A
does not serve as base for follow-up Test of the Validity of Hofstede’s Cultural
studies. The majority of these studies Framework. Advances in Consumer Research
are replications of the original study, 35, pp. 762 f.
mainly, in order to update the original
Bond, M. H. / Smith, P. B. (1993): Social
study with more recent data or to derive
psychology across cultures. London: Har-
action guidelines that are practically ap- vester Wheatsheaf.
plicable. In contrast to our replication,
the majority of the studies does not Briley, D. A. (2009): Cultural Influence on
question the validity of the constructs Consumer Motivations: A Dynamic View.
and hence, take the dimensions as given. In: Cheryl, N. (Eds.): Beyond Hofstede. New
Following from our finding that the York: Palgrave Macmillan, pp. 181-200.
23
S. J. (Eds.): Testing structural equation Saris, W. E. (1988): Variation in Response
models. Newbury Park: Sage Publications, Functions: A Source of Measurement Error in
pp. 294-316. Attitude Research. Amsterdam: Sociometric
Research Foundation.
Korman, A. K. (1985): Review of `Culture’s
Consequences: International differences Saris, W. E. / Satorra, A. / Van der Veld, W.
in work-related values´. Journal of Occupa- (2009): Testing structural equation models
tional Behaviour 6(3), pp. 243 f. or detection of misspecifications? Structural
Equation Modeling: A Multidisciplinary
Lienert, G. A. / Raatz, U. (1998): Testauf- Journal 16(4), pp. 561-582.
bau und Testanalyse. Weinheim: Beltz,
Psychologie Verl.-Union. Smith, P. B. / Schwartz, S. H. (1997):
Values. In: Berry, J. W. / Segall, M. H. /
McSweeney, B. (2002a): The Essentials of Kagitcibasi, C. (Eds.): Handbook of Cross-
Scholarship: A Reply to Geert Hofstede. Cultural Psychology. Boston: Allyn and
Human Relations 55(11), pp. 1363-1372. Bacon.
25
26 interculture j ourna l 13/22 ( 2 0 1 4 )