Laurent Snoeckx

Jaap Dronkers

Dimitri Mortelmans

Peter Raeymaeckers

Last update 12 September 2007

Paper to be presented at the fifth Meeting of the European Network for the

sociological and demographic Study of Divorce, 17-18 September 2007, the London

School of Economics in London.

2

Laurent Snoeckx*, Jaap Dronkers**, Dimitri Mortelmans* & Peter Raeymaeckers*

* Research Centre for Longitudinal and Life Course Studies (CELLO), University of Antwerp,

Belgium

** European University Institute, Florence, Italy

This article uses a cross-national research perspective on divorce risks within a single country.

We will argue that Belgium as such is an interesting test case in international perspective

since it combines - in a quasi experimental setting - a mutual national divorce legislation with

different regional cultural traditions. Belgium is divided into a rather catholic northern part

(Flanders) and a secular southern part (Wallonia), respectively referred to as a southern

(Spain, Italy) and a northern (Scandinavian) cultural pattern. In this contribution we analyse

the effects of different micro-, macro- and interaction-determinants to examine to what extent

they can account for the difference in divorce proneness between Flanders and Wallonia. Our

results show that the different regional divorce risks can partly be attributed to different

regional characteristics concerning cultural and religious traditions.

The spectacular increase of the divorce rates of the latest decennia in Western Europe has

been scrutinized by many social researchers. In understanding these patterns of family

dissolution and other demographic behaviour, both structural and cultural macro determinants

have been central and appearing side by side. In this article, we want to focus on the influence

of cultural and religious attitudes on the occurrence of partnership dissolution. While previous

research on divorce proneness focuses on individual (Jalovaara, 2002; Kalmijn, 2004) and

couple-level (Raeymaeckers, 2006) socio-economic or couple-related relational determinants

(Wagner, 2004; Cohan, 2002; Rogers, 2004; Snoeckx, 2006), we want to look at the influence

of macro cultural characteristics and micro determinants. More specifically, we focus on the

differences in divorce proneness between Flanders and Wallonia (Belgium).

The French speaking and Dutch speaking parts of Belgium are an interesting case to look at

the relative influence of cultural and religious determinants on divorce behaviour. The two

regions can be considered a micro setting of Europe’s geographical pattern of cross-national

differences in cultural and religious values. As the well-known Yin Yang symbol, part of

Europe’s Latin culture is present in the Northern Flemish part of the country while the

Southern Walloon region is culturally more related to the protestant and more progressive

Northern part of Europe. The uniqueness in this setting comes from the unitary Belgian state

structure holding both cultures together with a unified welfare system and a central divorce

law. This one-country-two-regions-setting gives the unique opportunity to analyse the neteffect

of cultural macro-determinants on divorce behaviour.

3

In this article, we join a demographic research tradition in which the impact of values on

behaviour is studied (Kalmijn, 2004; Thornton, 1983; Lesthaeghe, 1986; Moors, 1996;

Clarkberg, 1995). We agree with Surkyn and Lesthaeghe that a cultural shift is by no means

people’s life course decisions and behaviour (Inglehart, 2003). Earlier research already

showed the importance of cultural values as an explanatory factor for the differences in

divorce proneness between the Scandinavian countries (being Europe’s forerunners on

divorce) and the Southern European countries and Ireland (Ester, 1993). Even though - from

1965 onward - the European countries are reaching more convergence and coming closer to

the Scandinavian demographic patterns (Lesthaeghe, 1986; Ester, 1993), there still remains a

clear geographical pattern in divorce figures (Europe, 2002). Southern European countries

like Italy, Spain and Greece show significantly lower divorce rates than countries in the

northern region like Lithuania, Estonia and Denmark (Eurostat, 2004). The Eurostat

researchers note that, taking the divorce rates of Ireland, Slovenia and the Czech Republic –

respectively Roman Catholic and atheist in general - into account, religion also has an

important influence, interrelated with this geographical north-south pattern (Eurostat, 2004).

This is an indication of the findings of Inglehart and Baker (Inglehart, 2000) who argue that

cross-cultural differences linked with religion are being transmitted through cultural and

educational institutions, and through the shared experience of the people of a nation. In

explaining the national differences in value patterns across Europe, Gundelach (Gundelach,

1994) agrees with them by emphasizing that values are influenced by the specific institutional

characteristics which exist in different nations.

In an article on the persistence of traditional values despite modernization, Inglehart and

Baker (Inglehart, 2000) describe how modernization processes across the world and the

European continent cause cultural and religious value changes and show how path-dependent

these cultural values are. By mapping 65 societies into two dimensions (traditional/secularrational

and survival/self-expression), and combining those with economic zones (based on

annual per capita gross national products), the structure of the labour force (percentage of

labour force in industrial and service sector) and the historical cultural heritage, they show

how a society’s cultural and religious tradition remains to affect the cultural values, even

when taking the influence of economic determinants into account. Industrialisation and postindustrialisation

cause value changes in accordance with the traditional/secular-rational and

survival/self-expression dimensions, but historical traditions and especially religious heritages

remain to influence the rank of societies on those dimensions. Historically Protestant societies

for example, mostly situated in the northern part of Europe, obtain a higher score on the

survival/self-expression scale than all of the Roman Catholic societies, situated in central and

southern Europe, even when controlling for economic determinants. Ex-communist societies

rank higher on the traditional/secular-rational dimension while historically Roman-Catholic

societies show relatively traditional values, giving proof again of the persisting influence of

path-dependent cultural zones (Inglehart, 2000; Inglehart, 2003). Like Inglehart and Baker

(Inglehart, 2000) state, this indicates that economic development brings convergence in

nations value systems, while long-established cultural zones persist two centuries after the

industrial revolution began. Modernization theories are thus partly right, but belief systems

and traditions seem to be highly path-dependent.

4

Differences and similarities in behavioural outcomes between regions can be looked upon in a

European context, but also on the level of a single country. For Belgium, France and

Switzerland, Lesthaeghe and Neels (Lestheaghe, 2002) analysed the cultural path-dependency

and similar spatial pattern of the First and Second Demographic Transition, respectively

producing different behavioural outcomes: a strong population growth, a decline of the age at

first marriage and birth control on the one hand, and a dramatic decline of marriage rates, a

sharp increase of divorces and a fall in (marital) fertility rates on the other (Lesthaeghe, 2002;

influential until the 1960’s. As a consequence, the spatial pattern of the determinants of the

first value shift (First Demographic Transition) strongly influences the spatial pattern of the

second value shift (Second Demographic Transition) which can be linked to the behavioural

outcomes of the latter. In this perspective, the Second Demographic Transition is a

continuation of the First, in spite of their different and sometimes contrasting behavioural

outcomes (Lestheaghe, 2002).

When looking at Belgium, we expect to see regional differences in divorce behaviour that can

be attributed to cultural differences and barriers underpinned by different regional historicocultural

continuities. Given the different early regional economic developments in the

northern and southern part of Belgium, and the regional division of competencies with regard

behaviour could be traced to the different economic and cultural history of Flanders and

Wallonia. Before taking a look at the divergences in divorce proneness and other relational

characteristics between the northern and southern part of Belgium, we present a short

overview of the historical and political situation in Belgium. In doing so, we provide a general

framework in order to fully understand the regional disparities and their impact on the divorce

behaviour.

Belgium’s history is characterised by remarkable different regional developments, consisting

of a Catholic Northern (Flanders) and an atheist Southern part (Wallonia and Brussels) with

different spheres of influence. Flanders belonged to the German part of Europe. Wallonia

resided in the Latin part. Belgium has always been under the influence of foreign powers like

Spain (1555-1713), Austria (1713-1794), France (1794-1815) and The Netherlands (1815-

1830). Especially during those final two periods, a strong economic development took place.

When Napoleon came to power (1795), Belgium became part of the French empire, and the

industry took off, especially in the southern part of the country. British immigrants smuggled

machines into Belgium, erected factories, and Wallonia became one of the most industrialised

regions in the world. In Flanders, only the city of Ghent could profit from these economic

developments (Baetens, 1984; Gaus, 1996; De Brabander, 1984). Later, under the rule of the

Dutch King William I (1815-1830), the industry was supported by infrastructural

developments and financial assistance to entrepreneurs. However, these advantages were

especially favourable for industrialists in Ghent and Wallonia, thereby causing dissatisfaction

and opposition in the rest of Flanders, where the non-mechanised industry could hardly meet

the new competition. This rural opposition was supported by the Catholic Church, as their

power was to be limited by the protestant king. Politically, Catholics formed a union with

young Liberals, thereby inspiring an anti-Dutch, anti-Protestant movement that ultimately led

to the collapse of the United Netherlands and to the new independent state Belgium. In the

next decades, the Belgian economy grew fast and the first Industrial Revolution took place,

5

1984). However, within the boundaries of the unitary Belgian state, the linguistic borders

always remained a cultural demarcation line, resulting in important disparities with regard to

the measure of secularisation and birth control (Lesthaeghe, 2002). The enlightened ideas

stemming from the French Revolution were more widely spread in the southern part of the

country than in the Roman Catholic Flanders.

After World War II, the Belgian state structure evolved from a unitary state to a federal state

where the central federal authority weakened and important competences on employment,

education and welfare were transferred to two different types of regional authorities:

Communities and Regions. First, Communities (three in total: the Flemish Community, the

French Community and the German-speaking Community) have competences on personrelated

matters. At this level, policy measures are taken on cultural matters (theatre, libraries,

etc.), education, the use of languages and matters relating to the individual which concern on

the one hand health policy (curative and preventive medicine) and on the other hand

assistance to individuals (protection of youth, social welfare, aid to families, immigrant

assistance services, etc.) (Federal Government, 2006). Next, the Regions (also three, but

geographically different from the Communities: the Flemish Region, the Brussels-Capital

Region and the Walloon Region) execute responsibilities on territorial competences. Their

power extends to economic matters, employment, agriculture, water policy, housing, public

works and others (Federal Government, 2006).

An exception crucial for this article is the domain of family law. The divorce legislation,

which is rooted in the French “Code Napoléon” from 1804 (Senaeve, 2004) remained on the

central level (together with tax policy, national defence, foreign affairs and the social security

legislation). In this way, the Belgian situation forms a unique natural experiment since it

combines a joined divorce legislation and different macro-characteristics, in the heart of the

border area between the Latin and German part of Europe. Our central hypothesis is that –

despite the central (and liberal) divorce legislation - the historical regional disparities in

Belgium have a long-term effect on the opinions and culture in Wallonia and Flanders,

resulting in different demographic behaviour and more specific, in distinct divorce figures.

The fact that the Belgian state structure was highly influenced by this historical regional

dichotomy, thereby shifting competencies to the regions and producing Walloon and Flemish

institutions, endorses this hypothesis.

The Belgian divorce rates are among the highest in Western Europe, almost drawing level

with the northern forerunners like Denmark (Eurostat, 2004). The Belgian divorce rates have

quadrupled over the last thirty years. However, if we look closer at the divorce figures, we

notice there is quite a difference between the regions of this highly federalised state. Earlier

research (Snoeckx, 2006) already indicated that the mere fact of being Flemish, compared to

being inhabitant of Brussels or Wallonia, significantly reduces the risk on divorce, and keeps

reducing it when socio-demographic, relational and fertility characteristics are taken into

account. The divorce figures of 2002 of the Belgian National Institute for Statistics also

confirm this regional division: per 100000 inhabitants, 263 divorces took place in Flanders.

For Wallonia and Brussels, this number respectively rises to 283 and 550. The number of

divorces is twice as high in the Capital region compared to the other regions (Nationaal

Instituut voor de Statistiek, 2005). However, it should be noted that a distortion appears

because marriages of Belgian citizens who marry abroad are registered in the Brussels Capital

Region. Looking at the Belgian divorce figures of the latest decennia (Nationaal Instituut voor

6

de Statistiek, 2005), one can notice that the proportional divorce rates in Flanders, in

comparison with the other Belgian regions, have been the lowest since 1970. Jacobs (Jacobs,

2000) comes to the same finding: the Belgian high divorce rates can primarily be attributed to

the divorce figures of Brussels and Wallonia.

The dichotomy between the northern and southern part of Belgium also reveals itself in the

differences with regard to the relational value system. Flemings adhere more to a relational

homogamous orientation with regard to religious and political opinions, and social

background. This religious orientation is not surprising, given the fact that 41% of the

Flemings find religion (quite) important whereas this applies for only 29% of the Walloons.

With regard to church attendance, only one out of six Flemings never goes to church; for

Wallonia this is one out of four. When looking at the actual relational design, Flemish

inhabitants seem to stick more to the traditional family culture, according to Van den Troost

(Van den Troost, 2000). More than half of the Flemings (54%) has a lot of contact with their

family, but hardly meet their personal friends apart from their partner, compared to 39% of

the Walloon inhabitants.

In this article, we examine whether the regional differences in divorce proneness can be

explained by specific characteristics that can be attributed to Flanders and Wallonia.

Likewise, we examine to what extent the specific composition of the two regions accounts for

the regional divorce effect. We estimate two different models where micro determinants,

macro-variables and their interaction-effects play together. First, we interpret the different

ways in which micro determinants affect the divorce risk across the regions or cohorts.

Second, we examine whether the regional effects disappear by introducing macro-variables

that can account for the differences in divorce proneness across the regions. On the one hand,

the dichotomy in divorce rates and other relational characteristics should reflect itself in the

regional interaction effects of the micro determinants. On the other hand, the macro clustering

effect should be reflected in the effects of the different macro-variables. In the following, we

will briefly describe some relevant micro determinants and how we expect their effects to be

different across the regions. Afterwards, we will elaborate on the macro-determinants.

The micro-level predictors are expected to affect the divorce risk according to the wider

socio-cultural context (Wagner, 2004). Therefore, we present some explanations on how these

micro effects are influenced by characteristics typical of the composition of the different

regions.

The first richly illustrated risk-factor that can vary due to the different composition of the

Wagner, 2004; Wolfinger, 2005). Different mechanisms have been put forward to account for

this intergenerational divorce-effect, from which the most recent and popular approach

concentrates on the socialisation or role model explanation (Engelhardt, 2002; Greenberg,

1982; Traag, 2000). Like Engelhardt, Trappe and Dronkers mention (Engelhardt, 2002), there

are interactions between these different mechanisms (e.g. economic and stress arguments),

and they could succeed each other over time. If the socio-cultural barriers to divorce are low,

the divorce cycle will weaken. Thereby, we believe that the effect of parental divorce will be

weaker in less traditional societies where the barriers to divorce are lower.

7

In a traditional society where marriage is highly institutionalised and the barriers to divorce

a more selective group in such a society (selection-hypothesis). Partners who choose to be in a

cohabitational relationship, have a distinct set of characteristics. They have less conventional

and more progressive values and their unions are characterised by a more heterogeneous

composition concerning religion and age (Amato, 2003; Axinn, 1992; Brown, 1996;

Bumpass, 1991; Cohan, 2002; Kalmijn, 2004). In addition, premarital cohabitation will be

more strongly correlated with the divorce risk in traditional societies, causing different effects

across Flanders and Wallonia.

Married couples with young children have considerably more chance of survival than

childless couples (Parsons, 1956; Härkönen, 2004; White, 1990; Kalmijn, 2004; Kitson, 1985;

Wagner, 2004; Tzeng, 1995). We believe this marital stabilising effect should be stronger

when the divorce barriers are high because in less traditional societies - where divorces occur

more frequently - a culture and policy will emerge that makes it easier for one-parent or

newly composed households to raise their children.

(e.g. educational level, income level) becomes greater in affecting the divorce risk. More

resources are necessary to overcome the high costs of divorce. In a traditional society,

partners who divorce must have relatively high educational attainment, high incomes, and a

low chance on unemployment to overcome the high divorce barriers. Moreover, in case of a

marriage with children, a single parent needs a lot of resources when living in a traditional

society with a policy that does not facilitate divorce.

The disparities in divorce proneness across Flanders and Wallonia can be explained by the

different effects of the micro level determinants that vary in accordance with the distinct

regional characteristics. We expect the effects of micro variables like parental divorce,

premarital cohabitation, presence of children, income, employment, and education to be

stronger in the more traditional Flanders than in Wallonia.

Bearing in mind Durkheim’s maxim (Durkheim, 1951) that a society is more than the sum of

the parts, we also want to examine whether macro-characteristics, on the level of the two

regions, show a straightforward effect on the risk of divorce. People live together, interact, are

part of larger communities, and act in different institutions. In that way, it would not be

surprising that a life decision like divorce - just like Durkheim’s anomic suicide - is

influenced by the characteristics of the social environment. The cultural or religious

orientations of a country for example have an influence on the divorce decision of individuals,

regardless of their own orientation. Therefore, we expect that the divorce disparities between

Flanders and Wallonia can be explained by varying regional macro-characteristics. As stated

earlier, the divorce legislation is still equal across both regions forming a unique case to test

whether other structural (e.g. income, employment) and cultural (e.g. religion, postmaterialism)

regional characteristics influence the divorce risk.

8

In our models, we will introduce cultural and structural macro variables examining to what

extent they explain the divorce differences across the regions. Building on the historicocultural

and religious background of the two regions, we expect religion to play an important

variables, the lower we expect the divorce proneness to be in that region. Other cultural

Based on the cultural determinants of Lesthaeghe’s ‘Second Demographic Transition’

(Lesthaeghe, 1995; Lesthaeghe, 2002) and Inglehart’s ‘Silent Revolution’ (Ester, 1993;

in a region to have a positive effect on the regional divorce differences.

We also include structural macro-determinants in our models. In line with Surkyn and

Lesthaeghe (Surkyn, 2004), we believe that culture is not the only factor responsible for

changes in demographic behaviour. Therefore, we also incorporate two structural

regional level to have a positive impact on the divorce risk, given the fact that richer societies

with well-educated and working inhabitants are more likely to have a policy that makes it

financially easier for individuals to break up a relationship.

decisive elements in individual life decisions. Tolerance towards divorce decisions has

changed over time, the economic independence of women (female employment) has

increased, and policy measures that facilitate divorce decisions have emerged (Rogers, 1997;

Rogers, 1997). These changing time circumstances are empirically funded with the well-know

steep rise of divorce rates since the early 1960’s, which continued during the 1970’s and

slowed down a bit during the 1980’s (Teachman, 2002). In order to control for these changing

attitudes towards divorce, we incorporate four marriage cohorts (< 1960, during the 60’s, the

70’s and after 1980). The division into four groups results in an almost egalitarian number of

individuals in each cohort, as can be seen in table 1a (cf. Appendix).

Based on the theoretical framework concerning the importance of cultural and religious

factors in explaining demographic behaviour, and on the empirical studies on the difference in

divorce figures and other relevant characteristics between the Belgian regions, we want to

unravel the regional effect in our models by introducing micro and macro determinants, as

well as some of their interactions. The structure of this analysis is double. First, a basic model

with the regional variable it is estimated, controlling for individual characteristics. Using this

model as a baseline, we will introduce interaction effects of these individual determinants

with the marriage cohort variable and the regional variable. Next, we add the macro variables

to examine whether they can account for the regional divorce differences.

9

Second, we turn the sequence of the analysis upside-down: the basic model with individual

determinants will then be elaborated by first introducing the macro variables. Afterwards,

interactions between significant macro variables and individual determinants are added.

Thereafter, we test whether an additional regional or cohort effect can still be recognized and

whether the interaction effects between the individual characteristics and these two variables

become significant. By doing so, we introduce a double control for our hypothesis: if adding

the regional variable does not improve the model after introducing micro- and macro-level

variables, this indicates that the macro variables can account for the regional differences.

The data for the analysis come from the Panel Study of Belgian Households (PSBH). This

panel study followed both individuals and households from 1992 until 2001. The total

database provides eleven waves on a representative sample of the Belgian population,

covering a broad range of socio-economic topics, as well as themes relating to family

sociology. The prospective structure of the data set surpasses the main share of biases in

retrospective and in particular cross-sectional studies.

The most appropriate method to use is a variant of survival analysis, more in particular Cox’

proportional hazards technique. Cox regression models can handle both time-constant and

time-varying explanatory variables (Allison, 2000; Kleinbaum, 1996). The technique assesses

the effect of a set of explanatory variables on survival or event times. For this article, we

estimate the effect of a set of explanatory variables - described more detailed in the next

section - on the occurrence and timing of divorce. Unfortunately, information about mere

union dissolution is not available in the PSBH-data, so we can only measure the official

divorce proneness. By using Cox-regressions we take into account the differences in duration

of marriage (our time-dependent covariate) and the right censoring of the dependent variable

(spouses could still experience marriage dissolution after the panel-study has taken place).

The starting point for our analysis is the year of the first survey wave (1992), being equal for

all subjects. The basic unit of the file is the individual, more specifically the individual that

was married at the time of entering the panel study in wave 1. Second or higher-order

marriage are excluded from the analysis, taking into account the specific divorce dynamics for

this subgroup (Cherlin, 1996). For our time-dependent covariates, we use the lagged

covariates to account for the correct causal ordering. For these covariates, we take the value of

The basic data set contains 3911 individuals, who were married in the first wave (1992). Each

subject at risk was followed throughout the full length of the panel study, where possible, and

in case of attrition, the timing variable got the score of the wave of attrition, and the event was

coded zero. The problem of attrition is a downside to panel studies, especially when this is

selective and certain subgroups are more inclined to drop out. We are aware of the extreme

amount of censoring (n = 3756). Therefore an additional set of respondents was integrated in

the panel study from wave 7 (1998) until the end. However, it is crucial for our research

design to follow the same subset. A total of 165 individuals did break up their marriage

through the course of the follow up, meaning that only 4 % was not censored. The small

10

amount of divorcees in the sample size is also observed for other European panel studies

(Poortman, 2000; Jarvis, 1999; Burkhauser, 1991).

in the sample until the event occurred. The event is a legal divorce or a separation, and takes

on value ‘1’ if it occurs and ‘0’ for the censored individuals who remained married or left the

panel study prematurely. The timing variable refers to the wave in which the divorce occurred

or the censoring wave, and ranges from 1 to 11.

percentages, frequencies and means are shown in tables 1a and 1b (cf. Appendix). The

individual variables consist of demographic background, relational and fertility

categories, “Flanders” (n= 1947) and “Wallonia + Brussels” (n= 1964), whereby the latter is

the reference category. This two folded division had both theoretical and pragmatical reasons.

Theoretically, as earlier stated, the linguistic border (under which Wallonia and Brussels co

reside) forms a cultural demarcation line, marked by important disparities with regard to

fertility control and secularisation. Pragmatically, the number of divorces for Brussels in our

data set (n= 20) is too low in order to perform separate analyses. Furthermore we include two

dummy (reference category = employed). The frequency table shows that there is almost an

equal number of employed (n = 1949) and unemployed (n = 1886) people in our sample. The

1499 Euro” and the highest one being “more than 5000 Euro” (“less than 1000 Euro” is the

individual experienced a parental divorce and 0 if it did not. Apparently 529 people out of

3911 witnessed a parental divorce.

category. Table 1a (cf. appendix) shows that 404 respondents did live together before getting

that was directly obtained from the respondents. The final two variables refer to the fertility

years”.

1990. The EVS covers nearly all European countries and is one of the most extensive data

collections on opinions and ideas concerning work, religion, morals, politics and society,

relationships and parenting. The used methodology has become an important standard in the

international comparative research on norms and values (Ester, 1994). By combining the 1990

EVS macro variables with the 1992-2001 PSBH time-dependent individual covariates, we

maintain the correct causal order between values and behaviour.

To obtain the macro-level variables, we created eight clusters by introducing four marriage

cohorts for Flanders and Wallonia: one group of spouses that got married before 1960,

another group during the 60’s, a third group during the 70’s and a final group that got married

11

during the 80’s or later. By combining these groups with the corresponding macro cultural

value indicators from the EVS-study, we can control for the clustering of individuals in these

higher hierarchical entities and how these can hold responsible for the differences in divorce

proneness. The values of these variables are obtained by taking the percentages of the extreme

categories of the corresponding variables in the EVS-study. The categories, frequencies and

values can be found in table 1b (Cf. appendix). We are aware of the small number of clusters.

In order to obtain some statistical power, we prefer less clusters containing a larger number of

observations. A division into provinces for example leads to a very small number of valid

observations per cluster and is less appropriate to test whether our theoretical assumptions

concerning regional effects are correct.

The following macro variables, which are different for the eight combinations of two regions

The latter variable originates from Gaus (2001), not from the EVS-data.

Figure 1 depicts the survival function of marriages for Flanders and Wallonia. The Flemish

marriages tend to last longer in Flanders than the Walloon ones. By introducing micro, macroand

interaction effects, we want to explain the disparities between these two lines.

Figure 1 Marriage survival per region

0,00 10,00 20,00 30,00 40,00 50,00

0,93

0,94

0,95

0,96

0,97

0,98

0,99

1,00

Region

Wallonia

Flanders

The result of the first set of analyses is shown in table 2. We used stepwise models, first

introducing only our regional variable, followed by individual determinants, interaction

decline in -2 log likelihood as the model gets more extended. As explained earlier, the

12

purpose of this stepwise structure is to explain the regional effect by introducing parameters

that take account for the regional divorce differences.

mere fact of being Flemish reduces the chance on getting divorce, compared to being

Walloon. Model 2 shows that this effect remains the same when background, relational and

elaborately shown in international literature, older marriage cohorts tend to divorce

significantly less than the younger ones. However, it should be noted that the cohort-effects

are probably overestimated due to the fact that marriages of older cohorts are initially more

likely to survive within the data set. The other effects are also generally in the line of the

will be higher as the marriage lasts. It also confirms findings of Hiedemann et al.

(Hiedemann, 1998; Snoeckx, 2006) concerning the decreasing pressure of upbringing on the

relation between spouses as the children get older.

As the cohort-effects in our model are very strong, we introduce cohort-interaction-effects in

a third step to examine whether they can account for the divorce differences. The results show

that getting married in an old marriage cohort and having lived together before marriage

(4.35; p < 0.05) increases the divorce proneness, which is not very surprising when bearing in

mind the selection-hypothesis (Bumpass, 1991; Axinn, 1992; Cohan, 2002; Amato, 2003;

Kalmijn, 2004; Brown, 1996). Those few couples in the oldest cohort who lived together

before marriage must have had more liberal (and thus more pro-divorce) attitudes than the

majority of the other couples of their generation who started living together after marriage.

their inclusion in the equation.

Model 4 shows the introduction of the regional interaction effects. If the effects of the

individual determinants differ across the regions, it indicates regional differences in divorce

processes. Apparently, some determinants seem to affect the divorce risk in different ways

across the two regions. Spouses in Flanders who lived together before marriage (3.40; p <

0.01), or have no children (3.49; p < 0.01) or have children older than 18 years (4.75; p <

0.05) have a higher divorce proneness than comparable couples in Wallonia. The first result

fits also into a selection-hypothesis: Flemish couples who lived together before marriage must

have more liberal attitudes about marriage and divorce than the comparable Walloon couples,

for whom living together before marriage is more normal. In the same way Flemish couples

without children will have more liberal attitudes than the comparable Walloon couples,

because childlessness is less common among the former than among the later. The stronger

cultural aversion against divorce in Flanders makes it less probable that couples with young

children divorce. As a consequence Flemish couples with adult children will divorce more

often than comparable Walloon couples, who divorce more often when they have young

children. However, the regional main effect continues to significantly affect the chance on

getting divorced; the individual and cohort- and region-interaction effects are only able to

explain a part (from 0.71 to 0.37) of the effect of region.

13

Region 0.70* 0.69* 0.71* 0.37*** 16.77

Tertiary education 1.66* 1.73* 1.79* 1.78*

Employment 0.44*** 0.43*** 0.42*** 0.43***

Age at marriage 0.95* 0.95* 0.95* 0.95

Premarital cohabitation 3.43*** 0.01 0.01* 0.008*

Age youngest child

Number of children >3 0.39*** 0.38*** 0.35*** 0.36***

Parental divorce 2.57*** 2.62*** 2.62*** 2.59***

Marriage Cohort ‘61-‘70 25.455*** 25.10** 23.65** 6.90

Marriage Cohort ’71-‘80 155.236*** 152.18*** 140.30*** 29.83**

Marriage Cohort > 1980 343.300*** 255.72*** 244.99*** 34.71**

Cohort*premarital cohab. 4.35* 4.99** 4.47*

Region*premarital cohab. 3.40** 2.69*

Number of religious pers. 0.84*

* p <.05; ** p< .01; *** p<.001

In a final step, all macro-variables are added to examine whether they can fully explain the

decreasing effects of the marriage cohorts in the final column; the effect of the retained

insignificance of the Wald statistic of their parameters. Yet the most important conclusion of

this final model is the disappearance of the significant regional effect due to the addition of a

cultural macro variable. This means that the divorce differences between Flanders and

Wallonia can be explained by the regional and cohort differences in religious beliefs.

Moreover, the -2 log likelihood ratio of this model shows the best fit.

As announced earlier (cf. 6.1 Procedure), we reverse the story in our second set of analyses.

Considering the power of the macro variables to account for the regional effect, we start with

introducing them to see whether a regional effect can still be observed. Before doing so, our

their explanatory power towards the dependent variable.

14

standard basis model is constructed. The first column (model 1) in table 3 shows that the

effects are similar to the ones in the first model of our first set of analyses (cf. table 2).

Tertiary education 1.71* 1.66*

Employment 0.45*** 0.44***

Age at marriage 0.95 0.95*

Premarital cohabitation 3.52*** 3.44***

Age youngest child

Number of children >3 0.40*** 0.39***

Parental divorce 2.38*** 2.57***

Marriage Cohort ‘61-‘70 25.46** 21.93**

Marriage Cohort ’71-‘80 152.19*** 130.45***

Marriage Cohort > 1980 323.97 275.73***

Number of religious pers. 0.98*

* p <.05; ** p< .01; *** p<.001

the lower the divorce chances of couples, living in that region. This forms an indication of the

influence of cultural and religious determinants on people’s divorce behaviour.

In a third and fourth step, we added regional and cohort effects and interactions between these

two parameters and micro determinants. However, in accordance with our expectations, the

stepwise model selection removes the regional effect, thereby clearly indicating that this

effect does not (no longer) make any difference in predicting the risk on divorce. Neither

cohort effects, nor interaction effects of micro determinants with region or cohort are retained

in the Cox-regression. When looking at the variables not included in the regression, we find

supportive evidence for the disappearance of the regional effect when macro characteristics

are taken into account: the significance of the regional variable amounts to 0.44. The main

conclusion is that the differences in divorce proneness across the two Belgian regions are not

caused by the different composition of the inhabitants, but can be attributed to different

regional cultural characteristics.

The Belgian case lends itself to use a cross-national perspective within one single country,

since the unique opportunity is given to analyse the net-effect of different regional

characteristics on divorce behaviour, centrally regulated by one national divorce legislation.

In this contribution, we join a demographic research tradition in which the impact of cultural

and religious values on behaviour is emphasized, without neglecting the influence of

15

structural conditions. Following earlier research on the explanatory power of macro-cultural

values on diverging divorce risks in European perspective (Ester, 1993) and the cultural pathdependency

of the First and Second Demographic Transition in Belgium, France and

Switzerland (Lesthaeghe, 2002), we give a culturally coloured interpretation on the differing

Walloon and Flemish divorce risks in Belgium.

To explain these regional disparities, we estimate the impact of both micro and macro

determinants and their interaction-effects by using a twofold stepwise strategy. Our first

analysis takes the model with the regional variable as starting point and examines whether and

to what extent this regional effect can be explained by micro-, interaction- and macrodeterminants.

In line with previous research (Snoeckx, 2006), the regional effect remains to

influence the divorce risk when controlling for micro background, relational and fertility

characteristics. As expected, some micro-determinants affect the divorce risk differently

according to the regional context (Wagner, 2004); the interaction-effects show the varying

divorce processes in Flanders and Wallonia. These interaction-effects can be explained by the

stronger selection effect of living together in Flanders, where it is less normal and thus

reflecting a more liberal attitude toward marriage and divorce, or where Flemish couples

without children will also have more liberal attitudes than the comparable Walloon couples.

The stronger aversion against divorce in Flanders explains the other significant interaction

effects of children older than 18 years. Flemish couples with children will postpone their

divorce more often than comparable Walloon couples, who divorce more often when they

have still young children. The final step in our analysis introduces macro-determinants to

explain the remaining divorce differences between the two regions. The non-redundant

influence of cultural traditions on people’s life-decisions, as shown by Ester, Lesthaeghe et al.

(Ester, 1993; Lesthaeghe, 2002) also applies for the Belgian case since the religious tradition

in the Flemish region is an important factor for the different regional divorce numbers.

On the one hand, the cultural Yin Yang pattern of this country, with a South-European Latin

culture in the northern Flemish region and a North-European progressive culture in the

southern Wallonia, remains very specific. On the other hand, the Belgian test-case is very

relevant in international perspective since it shows the net-effect of cultural and religious

traditions on divorce behaviour. We believe the Belgian case ads value to the existing

research tradition in divorce behaviour. Further research should take advantage of this

peculiar Belgian case, instead of treating it as an anomaly.

16

Cary, North Carolina: SAS Press.

Amato, P., R., Johnson, D.R., Booth, A., and Rogers, S.J. (2003). Continuity and Change in

Amato, P.R.and Booth, A. (1991). Consequences of Parental Divorce and Marital

Axinn, W.G.and Thornton, A. (1992). The relationship between cohabitation and divorce:

Antwerpen: Standaard.

Brown, S.L.and Booth, A. (1996). Cohabitation versus marriage: A comparison of

Bumpass, L., Sweetser, D.A., and Cherlin, A.J. (1991). The Role of Cohabitation in the

Burkhauser, R., V., Duncan, G., J., Hauser, R., and Berntsen, R. (1991). Wife or Frau, women

do worse: a comparison of men and women in the United States and Germany after marital

Clarkberg, M., Stolzenberg, R.M., and Waite, L.J. (1995). Attitudes, Values, and Entrance

Cohan, C.L.and Kleinbaum, S. (2002). Toward a Greater Understanding of the Cohabitation

Dronkers, J.and Harkönen, J. 2006. "Is the divorce cycle really related with the societal

context? A cross-national test." Presented at the Fourth Conference of the European Network

for the Sociological and Demographic Study of Divorce ‘Family Dynamics and Family

structures in a comparative perspective’, 22-24 June 2006, Italy, Florence.

www.iue.it/Personal/Dronkers/English/divorcecycle.pdf.

17

Engelhardt, H., Trappe, H., and Dronkers, J. (2002). Differences in family policy and the

intergenerational transmission of divorce: A comparison between the former East- and West-

Ester, P., Halman, L., and Moor, R.d. 1993. "The individualizing society. Value change in

Europe and North America." Pp. 252. Tilburg: Tilburg University Press.

Tilburg University Press.

Europe, C.o. 2002. "Recent demographic developments in Europe." Pp. 121: Council of

Europe.

Eurostat. 2004. "Population and Social Conditions." Pp. 150, edited by P. Statistics: Eurostat.

Federal Government, 2006, http://www.belgium.be/eportal/index.jsp.

Greenberg, E.F.and Nay, R.W. (1982). The Intergenerational Transmission of Marital

Gundelach, P. (1994). National value differences. Modernization or institutionalization?

Härkönen, J. and Dronkers, J. (2004). Stability and Change in the Educational Gradient of

Hiedemann, B., Suhomlinova, O., and O'Rand, A.M. (1998). Economic Independence,

Inglehart, R.and Baker, W.E. (2000). Modernization, cultural change, and the persistence of

Antwerpen: Universiteit Antwerpen.

Jalovaara, M. (2002). Socioeconomic Differentials in Divorce Risk by Duration of Marriage,

18

Jarvis, S.and Jenkins, S.P. (1999). Marital splits and income changes: Evidence from the

Kalmijn, M., de Graaf, P.M., and Poortman, A.-R. (2004). Interactions Between Cultural and

Lesthaeghe, R. 1995. "The second demographic transition in Western countries: an

K.O. Mason and A.-M. Jensen. Oxford: Clarendon Press.

Lesthaeghe, R.and Meekers, D. (1986). Value Changes and the Dimensions of Familism in

Lesthaeghe, R.and Neels, K. (2002). From the First to the Second Demographic Transition: an

interpretation of the spatial continuity of demographic innovation in France, Belgium and

Lestheaghe, R.and Neels, K. (2002). From the First to the Second Demographic Transition: an

interpretation of the spatial continuity of demographic innovation in France, Belgium and

Algemene Directie Statistiek en Economische Informatie (2005), FOD Economie,

Bevolkingsstatistieken.

Kegan Paul Ltd.

Poortman, A.-R. (2000). Sex differences in the economic consequences of separation. A panel

Raeymaeckers, P., Snoeckx, L., and Mortelmans, D. (2006). Marriage and divorce in

Belgium. The influence of professional, educational and financial resources on the risk for

19

Rogers, S.J. (2004). Dollars, Dependency, and Divorce: Four Perspectives on the Role of

Rogers, S.J.and Amato, P.R. (1997). Is marital quality declining? The evidence from two

Snoeckx, L., Raeymaeckers, P., and Mortelmans, D. (2006). Relationele kenmerken en

echtscheiding in België. Een analyse op basis van de Panel Studie van Belgische

Surkyn, J.and Lesthaeghe, R. (2004). Value orientations and the Second Demographic

331–352.

Thornton, A., Alwin, D.F., and Camburn, D. (1983). Causes and consequences of sex-role

Traag, T., Dronkers, J., and Vallet, L.-A. 2000. "The intergenerational transmission of divorce

risks in France." Presented at ISA Research Committee 28 (Social Stratification) Spring

Conference, May 11th-14th, France.

Tzeng, J. and Mare, R.D. (1995). Labor Market and Socioeconomic Effects on Marital

Van den Troost, A. (2000). De relationele markt anno 2000. Een exploratie van

Wagner, M. and Weiss, B. (2006). On the Variation of Divorce Risks in Europe: Findings

483-500.

Zwaan, T. (1993). Recente transities in huwelijk, gezin en levenscyclus. In Zwaan, T. (Red.),

Universiteit.

20

Table 1a: Descriptive information on the micro-level determinants for Belgium, Flanders and

Wallonia

Duration of marriage 28.5 28.7 28.4

Age at marriage 23.9 23.8 24.2

21

Table 1b: Descriptive information on the macro-level determinants for the different clusters

(n = 538)

(n = 382)

(n = 455)

(n = 572)

(n = 543)

(n = 363)

(n = 432)

(n = 626)

Importance of religion

(very important) 34.2 15.2 16.3 9.6 24.75 22.25 13.65 11.35

Belonging religious

organisations (belong) 23.9 15.2 16.3 9.6 14.25 16.95 8.8 6.1

Attend religious services

(>once a week) 21.9 6.9 3.4 2.3 16.1 5.65 2.6 1.2

Number of religious persons

(religious person) 89.1 86.4 81.7 76.6 69.9 57.2 57.6 56.5

Number of catholic votes

(high) 47.8 35.5 23.1 18.5 18.75 15.95 16 13.8

Justification of divorce

(never justified) 36.4 24.2 19.4 13 27.85 15.35 8.5 11.9

Post-materialism (12 items)

(post-materialist) 5.1 13.9 17 20.4 10.9 14.05 11.8 21.75

Post-materialism (4 items)

(post-materialist) 15.3 20.2 23.9 36.7 3.6 17.25 16.65 26.5

Socio-economic status (high) 91.7 58.2 24.0 15.0 88.5 55.8 26.45 17.15

Income (10 categories) (high) 3.3 4.8 15.5 9.5 4 16.65 12.95 8.8

Income (4 categories) (high) 10.8 29.3 47.5 47.3 19.3 33.65 46.9 34.6

Unemployment rate

(percentage unemployment) 3.6 1.8 9.6 8.0 2.65 2.75 11.1 15.15