2. THE EFFECT OF RELIGIOUS GROUPS’ DOMINANCE 47
show that by and large the effects of religious belief are minimal, and if
pupils’ socioeconomic background is taken into account as well, they dis-
appear altogether. Considerably more research has been done on the rela-
tion between school denomination and children’s educational outcomes
(Dijkstra, 1997). Such research focussed largely on cognitive outcomes at
the end of primary education; noncognitive outcomes have hardly been
researched at all. The results of this kind of research have not been con-
sistent: sometimes they indicate effects, sometimes not. To the extent that
effects are noted, the trend is for private schools, and more particularly
Roman Catholic ones, to do better than public and other private schools.
Of late the relation between religion and education has attracted renewed
attention (Dijkstra, 1997). This is largely attributable to a process of sec-
ularisation and declining church membership that started in the 1960s,
which has drastically affected the role of religion in society (Felling, Peters
& Schreuder, 1991; Hermans & Van Vugt, 1997). Noteworthy in this regard
is that, despite the radically altered situation in society at large, the Dutch
education system is still organised on the basis of religious classi cation
(Bax, 1988; Dekker & Ester, 1996). Although only 37% of Dutch people
regard themselves as members of a religious denomination, some 65% of
schools remain private, religious institutions (Becker & De Wit, 2000;
Dronkers, 1996). This paradoxical situation makes it interesting to deter-
mine what relations there are between religion, denomination and educa-
tional outcomes.
2. MAJORITY POSITION
This article investigates three aspects pertaining to religion. First I exam-
ine the relation between the religious af liation of parents of children in
primary education and the denomination of the schools attended by these
children on the one hand, and certain educational outcomes on the other.
In contrast to a lot of earlier research, the value of this study is that it is
relatively large-scale (about 550 schools and 10,000 pupils), is based on
very recent (1999) data, involves young children, analyses both cognitive
and noncognitive effect measures, and uses adequate analytical techniques
(multilevel analyses). A key factor in the study is the pupil’s position in
the school as regards religion: is he or she in a majority or a minority
position? Although this theme has been researched previously, those stud-
ies were conducted from the angle of ethnicity and social background
rather than from a religious perspective.
3. 48 GEERT W.J.M. DRIESSEN
In the Netherlands Dijkstra and Veenstra (2000) have recently initiated
research into the effects of pupils’ religious position. They conceive of a
school as a community functioning within the larger whole of a social
environment. Here they rely on the ideas of Coleman and Hoffer (1987)
about functional and value communities, as well as the ecological devel-
opment model of Bronfenbrenner (1979). A value community refers to the
congruence between the norms and values prevailing in the child’s imme-
diate social environment and those observed in the school. A functional
community exists when the school is assimilated into a network of mutual
and intergenerational relations. The expectation is that schools displaying
more attributes of functional and value communities provide a more
favourable context for achieving good educational outcomes.
In community theory the relation between school and environment is
focal. But one can also look into relations within a school. In this regard
Kassenberg, De Vos and Dijkstra (2000) refer to notions of belonging to
the school (school belonging and school membership) developed by
Goodenow (1993). School belonging is seen as a feeling that one is
accepted, valued, integrated and encouraged by others (teachers and fel-
low pupils) and that one is an important component of class life and activ-
ities. Research has shown that such feelings go hand in hand with stronger
motivation to achieve, greater self-con dence and well-being and ulti-
mately, as a result, better cognitive performance. Signi cantly, there is a
relation with pupils’ gender and ethnic background. Goodenow and Grady
(1993) and Voelkl (1996) found that girls have a stronger sense of school
belonging than boys. In addition children from ethnic minorities feel more
strongly af liated with the school. According to these researchers the lat-
ter nding is attributable to the value that the relevant minority cultures
assign to society and their motivation for schooling.2
Motivation to learn is strongly in uenced, not only by pupils’ home sit-
uation, but also by the extent to which one can speak of functional and
value communities. Within the school motivation is a product of inter-
action between teacher and pupils and among the pupils themselves
(Guldemond, 1994). Various socio-psychological comparison processes
also play a major role (Marsh, Köller & Baumert, 2001). Research in this
regard is often conducted within the framework of effects of school com-
position (Thrupp, 1995; Westerbeek, 1999). Thus such studies may exam-
ine the effect on performance of a large proportion of pupils from ethnic
minorities in a school. The percentage of pupils from ethnic minorities is
de ned as an attribute at school level.
As mentioned already, Dijkstra and Veenstra (2000) have pursued this
notion regarding ethnicity into the realm of religion. More particularly
4. THE EFFECT OF RELIGIOUS GROUPS’ DOMINANCE 49
they determined to what extent the religious philosophy of the school is
congruent with the pupils’ religious background, and whether the mea-
sure of congruence has any impact on educational performance. Their
study is an extension of the research by Coleman and Hoffer (1987) into
value communities, which also examined the congruence between the
dominant religion at school and pupils’ religious background. Both these
studies in fact demonstrated a relation between such congruence and edu-
cational performance.
With due regard to these theoretical notions, I have opted for a more
or less similar approach in my research. A key issue is the position occu-
pied by pupils in regard to religion, that is a minority or a majority posi-
tion. The child’s religious position is determined by comparing it with the
proportion of classmates whose parents share the same religious af lia-
tion. In contrast to studies of the effects of school composition, here it is
seen as an attribute at the level of the pupil rather than that of the school.
The assumption is that “dominant religion” refers to a value and func-
tional community. If the child is in such a majority position, there will be
more social networking between this pupil’s parents and other parents than
if the child is in a minority position. There will also be greater corre-
spondence between the norms and values observed at home and at school,
for instance regarding the importance of education and how children should
behave. Both factors will positively in uence the child’s school career,
mainly because there will be less need for discussion about ends and means
and greater ( joint) control and support. Not only will the relation between
school and home be more favourable; the same applies to the situation in
class. Children in a majority position have more characteristics in com-
mon, have a stronger sense of belonging and feel more at ease. They have
less dif culty asserting themselves and consequently they can devote more
energy to actual school tasks. This leads to better educational outcomes.
In this article majority/minority position in regard to religion is focal.
With the aid of recently collected data on grade 2 primary school pupils
I seek to answer the following question:
What effect does a pupil’s religious majority or minority position at pri-
mary school have on a number of cognitive and noncognitive educa-
tional outcomes?
As measures of cognitive effect I take pre-reading and pre-maths pro ciency,
and as noncognitive measures I use social position, sense of well-being
and self-con dence. It is anticipated that membership of a dominant reli-
gious group will have a positive effect on educational outcomes. Because
the children involved are so young, it is also expected that at this stage
5. 50 GEERT W.J.M. DRIESSEN
the effect on noncognitive effect measures will be greater than on cogni-
tive ones. This is because membership of a dominant religious group affects
primarily their social position, well-being and self-con dence and only
then – largely via these noncognitive effects – their cognitive skills.
From the research question it is evident that I am interested mainly in
the impact of religious position on educational outcomes. Since a num-
ber of attributes of pupils and schools are known to obscure the opera-
tion of this effect, they need to be taken into account in analyses. Hence
the analyses will be controlled for certain relevant background features,
such as parents’education and ethnicity and the denomination of the school.
The upshot is that we gain insight into the net effect of religious major-
ity/minority position.
3. METHOD
3.1 Sample
The data analysed here derive from a Dutch cohort study entitled “Primary
Education” (PRIMA). As part of this research project, test and question-
naire data on primary school pupils, their parents, teachers and the rele-
vant school administrators have been collected biennially since the 1994/95
school year. The project involves 700 primary schools, which is almost
10% of the total number of Dutch primary schools, and some 60,000 grade
2, 4, 6 and 8 pupils. The PRIMA project is characterised by an overrep-
resentation of schools with a relatively large number of minority group
students, which permits reliable estimates of the systematic effects of fac-
tors pertaining to ethnicity and social background. The present analyses
are based on the results of the third PRIMA measurement in the 1998/99
school year. The analyses involve 10,141 grade 2 pupils from 549 pri-
mary schools.
3.2 Dependent Variables
Five effect measures are used: two cognitive and three noncognitive. What
follows is merely a brief outline; for a more detailed explanation you are
referred to Driessen, Van Langen and Vierke (2000).
Language pro ciency. Language pro ciency was measured using the
Concepts test (Begrippen) developed by CITO (National Institute for
Educational Measurement). This test gives an indication of the student’s
achievement level in pre-reading skills. The test consists of 60 multiple-
choice items; reliability (K-R 20) is .96. By applying a calibration pro-
6. THE EFFECT OF RELIGIOUS GROUPS’ DOMINANCE 51
cedure the scores were rescaled so as to constitute a one-dimensional
metric pro ciency scale.
Maths pro ciency. Maths pro ciency was measured using the Ordering
test (Ordenen), also developed by CITO. This test gives an indication
of the student’s achievement level in pre-maths skills. The test contains
a total of 42 multiple-choice items; reliability (K-R 20) is .90. The scores
on this test were also rescaled to constitute a one-dimensional metric
pro ciency scale.
Social position. Social position was measured by means of what is
known as a pupil pro le. This pro le comprises a number of statements
presented to the grade teacher to indicate their applicability to each indi-
vidual pupil. Social position entails four items with a reliability (Cronbach’s
a) of .84. Two examples of items are “This pupil is popular with class-
mates” and “This pupil has few friends in his/her class”. Scores on the
items range from 1 (“de nitely false”) to 5 (“de nitely true”). A scale
score was computed by averaging the scores on the component items.
A low score indicates low social position, a high score points to high
social status.
Well-being. Well-being, too, was derived from the pupil pro le. It com-
prises three items and has a reliability (Cronbach’s a) of .79. Examples
of items include “The pupil feels perfectly at ease with me” and “The
pupil feels uncomfortable at school”. The scale score was calculated in
the same way as that for social position.
Self-con dence. Self-con dence was also measured by means of pupil
pro le. It entails four items and has a reliability (Cronbach’s a) of .75.
Examples of items include “The pupil is timid and anxious” and “The
pupil is easily upset”. The scale score was calculated in the same way
as that for social position.
3.3 Independent Variables
Predictors were identi ed at two levels: pupil level and school level. Key
variables were parental religious af liation, denomination of school and
religious minority/majority position. In addition some relevant background
data on children and their parents were collected; these served as control
variables in the analyses (cf. Scheerens & Bosker, 1997).
3.3.1 Pupil Level
At pupil level the following variables were used:
Religious af liation. The following question was asked with regard to
parents’ religious af liation: “Which church, religious community or
7. 52 GEERT W.J.M. DRIESSEN
ideological group do you and your partner subscribe to?” The follow-
ing categories were identi ed: (1) none, (2) Roman Catholic (RC), (3)
Protestant Christian (PC), (4) Islamic or Hindu, and (5) other.
Parental ethnicity. Parents’ birthplace was used to determine ethnicity.
On substantive grounds four categories were identi ed: (1) Dutch, (2)
Surinamese or Antillean (Sur/Ant.), (3) Turkish or Moroccan (Tur/Mor.),
and (4) other minority background. The fourth category includes a mix-
ture of Western and non-Western immigrants.
Parental education. Education was taken to be an indicator of social
background. The highest educational level in the family, that of the
father or mother, was used. Four categories were identi ed: (1) primary
education (PE), (2) junior secondary vocational education (JSVE), (3)
senior secondary vocational education (SSVE), and (4) higher voca-
tional and university education (HE).
Gender. There were two categories: (1) boys, and (2) girls.
Age. A relative measure standardised according to grade was taken to
provide a rough indicator of repeating a grade or delayed entrance by
immigrant children. There were three categories: (1) the “norm” or a
maximum of six months older than the norm, (2) more than six months
older, and (3) more than a year older.
Majority/minority position. To determine the pupil’s relative position in
regard to religion we rst established which religion had the highest
incidence in each class. Then we checked per pupil to determine whether
he or she belonged to the majority group. To distinguish between situ-
ations where there was a clear majority group (e.g. 90% RC, 5% PC,
5% other) and situations where the groups were more evenly distrib-
uted (e.g. 40% RC, 30% PC, 30% other) we also calculated the varia-
tion in the incidence of the different religions. On the basis of the median
of this distribution a distinction was made between relatively little
(diverse) and relatively great (dominant) variation. Thus majority/minor-
ity position had three categories: (1) the pupil belongs to a minority
group; (2) the pupil belongs to the biggest group, but this group is not
dominant; (3) the pupil belongs to the biggest group, which is dominant.
3.3.2 School Level
At school level one characteristic is identi ed:
Denomination. Information on schools’ denominations was obtained
from the Ministry of Education. The following categories were identi ed:
(1) public, (2) Roman Catholic (RC), (3) Protestant Christian (PC), and
(4) other private.
8. THE EFFECT OF RELIGIOUS GROUPS’ DOMINANCE 53
4. RESULTS
4.1 Descriptive Analyses
Table 1 shows the correlations between parental religious af liation and
the religious majority/minority position of pupils on the one hand, and the
control variables of ethnicity, education, gender and age on the other.
Correlations are indicated in the form of column percentages. For each
predictor the correlation coef cient Cramér’s V is given. Next table 2
shows the correlations between religious af liation and position and the
ve effect measures. These take the form of averages and the nominal
metric correlation coef cient Eta.
From table 1 we see that in the no religion, RC and PC categories the
Dutch are represented by between 89 and 96%. The Islamic/Hindu cate-
gory comprises 75% Turkish and Moroccan pupils. As for education, 40%
of Muslim and Hindu parents have at most primary school education. The
Islamic/Hindu category also has relatively many older pupils, probably
caused by late school entry. In the majority diverse category there are
comparatively many children whose parents have no religious af liation,
whereas the majority dominant category has relatively many Catholics.
As for the relation between religious af liation and effect measures,
table 2 shows signi cant differences in language and maths performance,
in which regard the Islamic/Hindu category – and to a lesser extent the
other religions category as well – deviate negatively from the other reli-
gious categories. These two categories also differ marginally from the other
pupils in the class in respect of social position. There are no signi cant
total correlations between majority/minority position when it comes to
religious af liation and the effect measures.
Table 3 gives the correlations between the school’s denomination and
the predictors at pupil level, while table 4 shows the correlations between
denomination and the effect measures.
Table 3 shows that there is in fact a signi cant correlation between
parental religious af liation and school denomination, although there is
manifestly no question of a one-to-one relation; clearly religious consid-
erations are not the only ones in uencing the choice of a school.
Table 4 indicates that the denominations do not differ in respect of the
cognitive and noncognitive effect measures.
10. THE EFFECT OF RELIGIOUS GROUPS’ DOMINANCE 55
Table2.Correlationsofreligionandreligiousmajority/minoritypositionwitheffectmeasures(averages)
ReligionMajority/minorityreligion
NoneRCPCIslam/OtherEtaMinorityMajorityMajorityEtaTotalsd
Hindudiversedominant
Language986986985953977.36979979981.0397937
Maths5657564755.28545555.045514
Socialposition3.93.83.83.73.7.123.83.83.8.053.8.6
Well-being4.14.14.14.04.0.074.14.14.1.024.1.5
Self-condence3.63.53.63.63.5.053.63.63.6.023.6.7
11. 56 GEERT W.J.M. DRIESSEN
4.2 Multilevel Analyses
Having given a descriptive summary of the distributions and correlations
of the attributes under investigation, we now offer an analysis in which
the data are correlated in a multivariate manner. To this end we make use
of multilevel analysis. This technique enables us to distinguish between
Table 3. Correlations of denomination with religion, ethnicity, education,
gender and age (in %)
Denomination
Public RC PC Other private V
Religion .37
None 47 18 25 14
RC 13 58 11 28
PC 10 7 52 22
Islam/Hindu 27 16 9 34
Other 2 1 3 2
Ethnicity .11
Dutch 68 77 85 60
Sur/Ant. 3 3 3 8
Tur/Mor. 21 13 7 22
Other 8 7 6 10
Education .06
PE 13 10 5 14
JSVE 32 30 31 27
SSVE 31 33 38 32
HE 23 27 26 26
Gender .02
Boys 53 53 52 49
Girls 47 47 48 51
Age .03
Norm 67 66 70 72
>1/2 year 32 33 29 28
>1 year 1 0 0 0
N=100% 2,905 4,053 2,381 802
Table 4. Correlations of denomination with effect measures (averages)
Denomination
Public RC PC Other private Eta
Language 977 981 981 974 .07
Maths 54 56 55 54 .06
Social position 3.8 3.8 3.8 3.7 .04
Well-being 4.1 4.1 4.1 4.1 .01
Self-con dence 3.6 3.6 3.6 3.6 .03
12. THE EFFECT OF RELIGIOUS GROUPS’ DOMINANCE 57
different levels in the explanation of variance. The underlying idea is that
variance in, say, pupil performance lies partly at pupil level and partly at
school level. Variance at school level is commonly known as systematic
variance. School variables can only account for variance in performance that
is associated with the school level. This means that a school variable which,
in an ordinary, one-level regression analysis including school and pupil
variables, explains only a little of the total variance, can still be an important
predictor if the non-systematic (i.e. pupil level) part of the variance is left
out of account (cf. Snijders & Bosker, 1999; Dijkstra & Veenstra, 2001).
In the multilevel analyses we test a series of models. These models are
constructed step by step, as follows:
First we compute what is known as a null model by introducing a con-
stant as predictor at both levels. From this we can infer which part of
the variance in effect measures is at pupil level and which part is at
school level.
Next parental religious af liation is introduced as four dummy variables
in model 1, with the no religion category serving as reference group.
In model 2 religious majority/minority position is added to the null
model as two dummy variables. Here the majority dominant category
is used as the reference group.
In model 3 both religion and majority/minority position are added to
the null model.
In model 4 the control variables ethnicity, education, gender and age
are added to model 3. Ethnicity entails three dummy variables with the
Dutch category serving as reference group; for gender boys are the ref-
erence group.
In model 5 school denomination is added to model 4. This entails three
dummy variables with public schooling serving as reference group.
Finally, in model 6 the cross-level interactions of religious af liation
and religious majority/minority position with denomination are added
to model 5. Here one needs to determine whether the effects of religious
af liation and position differ between schools of different denomina-
tions. Since no interaction proved signi cant, model 6 was not included
in the tables.
Tables 5 to 9 give the results of the multilevel analyses with the ve effect
measures. The tables re ect the unstandardised regression coef cients (B)
and the accompanying standard errors (SE). In addition they indicate the
degree to which the estimates deviate signi cantly from zero (under p).
Signi cant effects and highly signi cant effects are indicated with * and
** respectively.3
13. 58 GEERT W.J.M. DRIESSEN
The tables are constructed as follows. In the part labelled “variance
components” the null model (model 0) gives the percentage distribution
of the total variance in the effect measures according to pupil and school
level. Then, in the ensuing models, we calculate which part of the vari-
ance at each of these levels is “explained” by the predictors that were
introduced. For models 1, 2 and 3 these variance explanations were rst
calculated in relation to the null model. This indicated how much reli-
gious af liation, majority/minority position, and religious af liation plus
majority/minority position respectively help to explain variance. Then, for
model 4, the explanation provided by model 3 was deducted, so that under
this model the additional explanation of variance stands after the intro-
duction of ethnicity, education, gender and age. In model 5 the additional
variance stands after the introduction of denomination.
The values under x2
/df are used to test whether any one model devi-
ates signi cantly from another:
models 1, 2 and 3 are tested with reference to the null model;
model 4 is tested with reference to model 3;
model 5 is tested with reference to model 4.
Signi cant and highly signi cant effects are indicated with * and **
respectively.4
In the multilevel analyses the three noncognitive effect measures are all
multiplied by 10, since otherwise the effects and their standard errors are
often so slight that they cannot be presented in the same way as the
coef cients for language and mathematics, that is by means of one decimal.
Hence for purposes of data interpretation these effects have to be multi-
plied by 10. The estimate for the general average of the effect measures
has been omitted from all the tables, since, as a result of centring the pre-
dictors around their means (i.e. the individual score minus the grand mean),
these are equivalent to the averages presented under “total” in table 2.
Table 5 shows the results in regard to language pro ciency; tables 6
and 7 give the results for maths pro ciency and social position.
From the null model in table 5 we gather that as much as 80% of the
variance in language pro ciency pertains to the pupil level and nearly
20% to the school level. In model 1 parental religious af liation is intro-
duced as a predictor. From the regression coef cients it is evident that
children of Catholic parents score .2 marks lower than the reference cat-
egory, children of non-religious parents. The Protestant Christian category
scores .4 marks lower and the Islamic/Hindu category, no less than 27.6.
This last difference is highly signi cant, amounting to three quarters of
the standard deviation. The “other” category scores 6.1 marks lower. The
17. 62 GEERT W.J.M. DRIESSEN
variance components indicate that 4% of the variance in language scores
at pupil level and 48.7% of the variance at school level are “explained”
by religious af liation. These differences are highly signi cant.
In model 2 the effect of religious position is focal. None of the cate-
gories appear to be signi cant. In model 3 both religious af liation and
position are introduced simultaneously. In regard to religious af liation
there is hardly any change from model 1. The coef cients for position are
somewhat higher; pupils in a minority position score 2.8 marks lower and
those in a majority diverse position score 1.7 marks lower than pupils in
a dominant majority position. These coef cients are not signi cant, however.
In model 4 some background attributes are introduced as control vari-
ables. As a result we get some idea of the net effect of religious af lia-
tion and position, hence after allowing for pupils’backgrounds.The analysis
indicates that all four of the background features have pronounced effects.
For instance, Turkish and Moroccan children score 18.9 marks lower than
Dutch pupils (the reference category for ethnicity); in relation to parental
educational level pupils’ language scores increase by 6.9 marks per level;
thus between children of parents with at most primary education and those
whose parents have higher vocational or scienti c training this difference
comes to almost 21 marks; girls score 7.1 marks more than boys; older
pupils score 7.8 marks more than younger ones. In the case of these
coef cients it should be kept in mind that these are effects which take into
account the in uence of all other variables in the model.
If we look at the effects of religious af liation and position, we nd that
keeping the backgroundsconstant has the result that the language pro ciency
of the Islamic/Hindu category in particular “improves” markedly: they
have made up their backlog by some 20 marks and the effect now is only
slightly signi cant. The “other religions” category too has advanced some-
what. The effects of the other two religious categories and of religious
position remain much the same. When it comes to the variance compo-
nents we nd that introducing background attributes explains an additional
6.5% of the variance at pupil level and 7.8% of the variance at school
level, over and above the variance already explained in model 3. In model
5 denomination is added at school level. Here we conclude that the Roman
Catholic category fares somewhat better than the public school category,
and the Protestant Christian and other private schools category slightly
worse, although none of these effects are signi cant. Hence the variance
components indicate that the addition of denomination does not improve
the model signi cantly. As mentioned already, model 6, in which inter-
actions between religious af liation, position and denomination are speci ed,
18. THE EFFECT OF RELIGIOUS GROUPS’ DOMINANCE 63
was omitted because none of these interactions proved to be signi cant.
Hence the effect of religious af liation and position function identically
in each of the denominations. If we now calculate how much of the total
variance in language pro ciency is explained by the nal model, number
5, we get the following: at pupil level (4 + 6.5) ´ 80.7 = 8.5%, and at
school level (49.6 + 7.8 + .4) ´ 19.3 = 11.2%.
The results in table 6 relate to mathematical pro ciency. By and large
they are commensurate with those relating to language pro ciency. In
interpreting them it should be borne in mind that the coef cients also dif-
fer because the distribution attributes of the maths scores differ from those
of the language scores, more particularly the standard deviation. In model
4 we observe that in maths the effect of the minority position category is
slightly signi cant; however, this effect disappears in subsequent models
when pupils’ backgrounds are taken into account.
The results relating to social position in table 7 show a marked resem-
blance to those for maths. Noteworthy here is that ethnicity has no inde-
pendent effects. The results relating to the other two noncognitive effect
measures, well-being and self-con dence, are comparable with those for
social position. For this reason these tables are not included. The only real
difference is that in the case of well-being and self-con dence religious
af liation has no independent effects, whereas it does have some effect in
the case of social position: here we nd that Muslim and Hindu children
occupy a weaker social position in the class than other pupils.
5. CONCLUSIONS AND DISCUSSION
This article explored three aspects pertaining to religion. The cardinal issue
was the majority or minority position that children occupy in class by
virtue of their religion, and the effects this has on a number of cognitive
and noncognitive educational outcomes. It was anticipated that occupy-
ing a religious majority position would have a positive effect on educa-
tional outcomes. In addition we explored the relation between parental
religious af liation and the denomination of the school their children attend,
and educational outcomes. The ndings of the analyses can be summarised
as follows:
Although there is a slight trend for children who occupy a majority posi-
tion at school to score marginally higher in educational outcomes, this
effect is never signi cant once one has controlled for ethnicity, parents’
education, gender and age.
19. 64 GEERT W.J.M. DRIESSEN
Even when ethnicity, parents’ education, gender and age are kept con-
stant, the Islamic/Hindu category still shows a negative effect relative
to the nonreligious category in respect of language, mathematics and
social position.
Although there is also a slight trend for children at Roman Catholic
schools to score somewhat better than those attending other schools,
there is no question of any independent effect of denomination.
In terms of effects denomination does not interact with religious af lia-
tion and religious majority/minority position, indicating that the effects
of religious af liation and position are the same in all denominations.
There are independent effects of ethnicity on language and maths per-
formance, but not on social position, self-con dence and well-being. Edu-
cation, genderand age also have independenteffects on the effect measures.
The analyses show no independent effects of denomination, which accords
with earlier ndings (Dijkstra, 1997). Taking into account the other attrib-
utes, the category of school which pupils attend clearly does not do any-
thing for them. Undoubtedly this has to do with the steadily weakening
relation with parental religious af liation caused by the secularisation of
society, as emerged from the descriptive analyses as well.
As for the effect of the Islamic/Hindu religious category, this is not
unexpected (cf. Driessen & Valkenberg, 2000). Muslims in particular have
a consistently low educational level, are often unemployed or do lowly
skilled work. It is known that there is a strong correlation between this kind
of family-structural factors and educational prospects (Driessen, 2000). What
is alarming is that this negative effect of the Islamic/Hindu category per-
sists even after one has controlled for ethnicity, education, gender and age.
At all events, from the analyses one can conclude that aspects relating
to religion are much less important than attributes of family structure. Par-
ticularly parents’ education and – to a slightly lesser extent – ethnic origins
appear to be decisive for young children’s educational outcomes, both
cognitive and noncognitive.
The ndings also imply that the anticipation that membership of the
dominant religious group at school positively affects educational outcomes
cannot be con rmed. Here a few comments are called for. It should be
noted, for instance, that the analysed sample comprised young children
who on average have only attended school for about two years. It could
be that such effects only manifest themselves at a later stage. Hence it is
recommended that the analyses conducted in this study should be repeated
20. THE EFFECT OF RELIGIOUS GROUPS’ DOMINANCE 65
for the same children in a few years’ time. This is feasible, since the
PRIMA cohort, which we used, is longitudinal and measurements of the
same pupils are made every second year.
The absence of any effects of majority/minority position may also be
attributable to other, more conceptual factors. As mentioned already, there
has been virtually no research into the relation between position regard-
ing religious background and educational outcomes. Theorising in this
eld is also in its infancy. In our study we drew on Coleman’s concept of
value and functional communities, which was interpreted from a religious
perspective. It should be clear, however, that such communities are not
unidimensional entities. People consort with each other not only because
of their religious background, but also because of ethnic and socioeco-
nomic similarities; in addition these three aspects are closely interrelated.
In fact, it may be expected that in the present postmodern, highly indi-
vidualised and secularised society religious communities have lost much
of their importance. This is supported by the notions on the state of pre-
sent-day Western society recently proffered by Putnam (1995), to whom
social trust is a key concept. Putnam assumes that civic engagement and
social connectedness produce, for instance, better schools, lower crime
rates and faster economic development. He points out, however, that in
the United States (also in other Western societies) there is a high level of
individualisation. Thus religious sentiment seems to be becoming less tied
to institutions and more self-de ned. In addition, Putnam concludes that
American social capital in the form of civic associations has signi cantly
eroded over the last generation. This means that value communities as
envisaged by Coleman in his day are far less common nowadays. At most
they will probably still occur in more traditional, closed religious com-
munities such as the often locally organised, strictly Reformed Protestant
groups and Islamic communities in the Netherlands. But it is not only at
adult level that (religiously oriented) value communities have lost their
importance; this may also have ltered through to the level of children.
Future research should therefore be aimed at, rstly, development within
religious communities, and secondly, the signi cance of these develop-
ments for children.
Finally a few methodologically oriented comments. The rst concerns
the operationalisation of parental religious af liation. In this study it took
the form of a question about membership of a religious community. This
may be regarded as a crude operationalisation, which moreover tells us
little about the parents’ active involvement. As Putnam (1995) points out,
21. 66 GEERT W.J.M. DRIESSEN
religious experience is increasingly characterised by deinstitutionalisation
and individualisation. Hence instead of the institutional approach to reli-
gion adopted here it might be better to choose an operationalisation that
does more justice to active religious involvement, for instance by way of
prayer, church attendance and participation in activities organised by the
church community.
A nal point regarding the structure of the analyses. In regard to the
nested nature of the data (pupils within schools and both pupil and school
attributes) we used multilevel analysis. As a result, also through the intro-
duction of all sorts of control variables, we obtained insight into the net
effect of majority/minority position. In the process, however, we rather
lost sight of the interrelationship between these variables. In follow-up
analyses it may be advisable to disentangle the complexity of interactive
variables differently, perhaps by means of a Lisrel analysis.5
NOTES
1. The author would like to thank J. Doesborgh for his help with the multilevel analyses.
Grateful acknowledgment also to the Netherlands Organisation for Scienti c Research (NWO)
for funding the project on which this paper is based. The research was supported by grant #
411-20-005 from NWO’s Social Science Research Council.
2. It is questionable whether this applies in every respect. According to Ogbu (1988) there
is an oppositional and resistance culture growing among ethnic minorities, especially in inner
city schools. They set themselves up against the WASP ideal, which sets great store by a good
education. These minorities lack motivation to go to school and perform academically for the
very reason that it implies con rmation of the majority culture.
3. The degree of signi cance can be derived by calculating a z score, namely z = B/SE. The
exact meaning of signi cant and highly signi cant in terms of z scores depends on the num-
ber of units (here: schools) in the analysis (cf. Cohen, 1988). For N < 120 schools, an effect is
generally assumed to be just signi cant when the p value < .10 in keeping with a z value >
1.65. For N = 200 schools, just signi cant is p < .05 or z > 1.96. For N = 500 schools, just
signi cant is p < .001 or z > 3.29. In keeping with this, the following holds for N = 549 schools:
*: 3.5 > z < 4.7; **: z > 4.7.
4. The value obtained is an x2
value, calculated by subtracting the x2
value for the model to
be examined from the x2
for the reference model to see if they differ signi cantly. The differ-
ence in the x2
values is then divided by the difference in the degrees of freedom for the two
models. For N = 549 schools and df = 1, a x2
/df > 12 indicates a just signi cant difference.
5. In such a one-level analysis it would be better to omit the school variable “denomination”.
REFERENCES
Bax, E. (1988). Modernization and Cleavage in Dutch Society. A Study of Long Term Economic
and Social Change. Groningen: University of Groningen.
Becker, J. & De Wit, J. (2000). Secularisatie in de jaren negentig. Kerklidmaatschap, veran-
deringen in opvattingen en een prognose. Den Haag: SCP.
22. THE EFFECT OF RELIGIOUS GROUPS’ DOMINANCE 67
Bronfenbrenner, U. (1979). The Ecology of Human Development. Experiments by Nature and
Design. Cambridge, MA: Harvard University Press.
Cohen, J. (1988). Statistical Power Analysis for the Behavioral Sciences. Hillsdale, NJ: Erlbaum.
Coleman, J. & Hoffer, T. (1987). Public and Private High Schools: The Impact of Communities.
New York: Basic Books.
Dekker, P. & Ester, P. (1996). Depillarization, Deconfessionalization, and De-ideologization:
Empirical Trends in Dutch Society 1958-1992. Review of Religious Research, 37, 325-341.
Dijkstra, A. (1997). Onderwijskansen en richting van de school. In A. Dijkstra, J. Dronkers &
R. Hofman (Eds.). Verzuiling in het onderwijs. Actuele verklaringen en analyse (pp. 144-
184). Groningen: Wolters-Noordhoff.
—— & Peschar, J. (1996). Religious Determinants of Academic Attainment in the Netherlands.
Comparative Education Review, 40, 47-65.
—— & Veenstra, R. (2000). Functionele gemeenschappen, godsdienstigheid en prestaties in het
voortgezet onderwijs. Mens & Maatschappij, 75, 129-50.
—— & Veenstra, R. (2001). Do Religious Schools Matter? Beliefs and Life-Styles of Stu-
dents in Faith-based Secondary Schools. International Journal of Education and Religion,
2, 192-206.
Driessen, G. (2000). The Limits of Educational Policy and Practice? The Case of Ethnic Minor-
ity Pupils in the Netherlands. Comparative Education, 36, 55-72.
—— & Valkenberg, P. (2000). Islamic Schools in the Netherlands: Compromising Between
Identity and Quality? British Journal of Religious Education, 23, 15-26.
——, Van Langen, A. & Vierke, H. (2000). Basisonderwijs: Veldwerkverslag, leerlinggegevens
en oudervragenlijsten. Basisrapportage PRIMA-cohortonderzoek. Derde meting 1998-1999.
Nijmegen: ITS.
Dronkers, J. (1996). Dutch Public and Religious Schools between State and Market. A Balance
between Parental Choice and National Policy? In D. Benner, A. Kell & D. Lenzen (Eds.).
Bilding zwischen Staat und Markt (pp. 51-66). Weinheim und Basel: Beltz Verlag.
Felling, A., Peters, J. & Schreuder, O. (1991). Dutch Religion. The Religious Consciousness of
the Netherlands After the Cultural Revolution. Nijmegen: ITS.
Goodenow, C. (1993). Classroom Belonging Among Early Adolescent Students: Relationships
to Motivation and Achievement. Journal of Early Adolescence, 13, 21-43.
—— & Grady, K. (1993). The Relationship of School Belonging and Friends’ Values to Acad-
emic Motivation among Urban Adolescent Students. Journal of Experimental Education, 62,
60-71.
Guldemond, H. (1994). Van de kikker en de vijver: groepseffecten op individuele leerprestaties.
Leuven/Apeldoorn: Garant.
Hermans, C. & Van Vugt, J. (Eds.) (1997). Identiteit door de tijd. Re ecties op het confessionele
basisonderwijs in een geseculariseerde en multiculturele samenleving. Den Haag/Nijmegen:
ABKO/Katholiek Studiecentrum.
Kassenberg, A., De Vos, H. & Dijkstra, A. (2000). Gemeenschap binnen en rond scholen voor
voortgezet onderwijs en het gevoel ‘bij de school te horen’. Paper 27e
Onderwijs Research
Dagen, 24-26 mei 2000, Leiden.
Luyten, H., Cremers-Van Wees, L. & Bosker, R. (2001). Mattheus-effecten voor taal, rekenen
en non-verbaal IQ. Verschillen tussen scholen, lichtingen en leerlingen. Enschede: Twente
University Press.
Marsh, H., Köller, O. & Baumert, J. (2000). Reuni cation of East and West German School
Systems: Longitudinal Multilevel Modeling Study of the Big-Fish-Little-Pond Effect on
Academic Self-Concept. American Educational Research Journal, 38, 321-350.
Ogbu, J. (1988). Class Strati cation, Racial Strati cation, and Schooling. In L. Weis (Ed.).
Class, Race, and Gender in American Education (pp. 163-182). Albany: State University of
New York Press.
Putnam, R. (1995). Bowling Alone: America’s Declining Social Capital. Journal of Democracy,
6, 65-78.
Rasbash, J. & Woodhouse, G. (1996). MLn Command Reference. Version 1.0a. London: University
of London.
23. 68 GEERT W.J.M. DRIESSEN
Scheerens, J. & Bosker, R. (1997). The Foundations of Educational Effectiveness. Oxford:
Pergamon.
Sherkat, D. & Darnell, A. (1999). The Effect of Parents’ Fundamentalism on Children’s Educa-
tional Attainment: Examining Differences by Gender and Children’s Fundamentalism. Journal
for the Scienti c Study of Religion, 38, 23-35.
Snijders, T. & Bosker, R. (1999). Multilevel Analysis. An Introduction to Basic and Advanced
Multilevel Modeling. London: Sage.
Thrupp, M. (1995). The School Mix Effect: The History of an Enduring Problem in Educational
Research, Policy and Practice. British Journal of Sociology of Education, 16, 183-203.
Van Haaften, W. & Snik, G. (1999). Onderwijsvrijheid: voor en tegen. Pedagogisch Tijdschrift,
24, 1-15.
Voelkl, K. (1996). Measuring Students’ Identi cation with School. Educational and Psychological
Measurement, 56, 760-770.
Westerbeek, K. (1999). The Colours of my Classroom. A Study into the Effects of the Ethnic
Composition of Classrooms on the Achievement of Pupils from Different Ethnic Backgrounds.
Rotterdam: CED.
Address for correspondence:
Dr Geert W.J.M. Driessen
ITS – Institute for Applied Social Sciences
University of Nijmegen
PO Box 9048
6500 KJ Nijmegen
Netherlands
<G.Driessen@its.kun.nl>