ABSTRACT
Despite substantial country variation in gender-role attitudes and female labor supply and theoretical arguments stressing the consequences of contextual attitudes for individual behavior, prior research did not find evidence for an effect of a country's gender-role attitudes on female labor supply. In this study I reassess this finding using a powerful multilevel design on the 2008 wave of the European Values Study on 33 countries. I find a substantial positive and independent effect of a country's egalitarian gender-role attitudes on individual women's odds of labor market attachment. The original, gross effect can for one-fourth be attributed to an effect of individual gender-role attitudes and one-tenth to an institutional effect. These findings indicate that the cultural (attitude) context matters for female labor supply.
1. Introduction
Cross-comparative studies have demonstrated that gender-role attitudes vary between the citizens of European countries (Kalmijn 2003; André et al.2013). People from the Scandinavian countries are, for example, much more in favor of an active position of women on the labor market than people elsewhere in Europe. This paper's question is whether – and if, how – these differences in the cultural climate of countries are consequential for individual behavior, in specific the labor supply of women. Empirical research has demonstrated that individual gender-role attitudes affect individual women's labor supply (Jansen and Kalmijn 2002; Corrigal and Konrad 2007) – whereby women with more modern, egalitarian gender-role attitudes have a higher labor supply than women with less egalitarian attitudes – yet do a country's gender-role attitudes also affect individual women's labor supply?
Some scholars analyzing the sizeable variation in female labor supply across countries in Europe suggest that individual women's labor supply is affected by the cultural context in which they are embedded, among others the prevailing gender norms regarding the role of men and women in the home and in work (Pfau-Effinger 1998; Albrecht et al. 2000; Geist 2005; Heineck 2007). This cultural context effect would exist independent of effects of institutional factors (such as childcare provision) and country-level economic factors (such as economic affluence). The scholars have not been very clear about the mechanisms of such an effect of gender norms in society on female labor supply, yet it is not difficult to come up with arguments why such an effect would exist. Gender norms and attitudes in society could affect individual women's female labor supply indirectly via socialization of individual-level gender attitudes, or directly via normative sanctioning (women who do not live up to the expectations may face sanctions).
In the light of the empirical findings (cross-country variation in gender-role attitudes and female labor supply) and theoretical arguments (contextual attitudes would matter), it is surprising that cross-national studies applying a multilevel design do not find evidence for an effect of national gender-role attitude on individual women's labor supply (Pettit and Hook 2005; Uunk et al. 2005; Steiber and Haas 2009; for a review see Steiber and Haas 2012). Net of institutional determinants a contextual effect of gender-role attitudes does not seem to exist.1 Yet, these studies have their shortcomings since they analyze few countries (Pettit and Hook 2005; Uunk et al.2005), model the odds of employment rather than being in the labor force or not (all three studies), use indirect measures for the prevailing gender norms (Pettit and Hook 2005), are restricted to young mothers' labor supply (Uunk et al.2005; Steiber and Haas 2012), or miss a control for individual-level attitudes (Pettit and Hook 2005; Uunk et al. 2005). Using a multilevel design on 33 countries from the 2008 wave of the European Values Study (EVS), the current study reassesses the effect of national gender-role attitudes on female labor supply; that is, on the decision to be on the labor market – employed or unemployed – or not. Although there exist stronger norms on work and home for women with children at home than women without children, I include both groups. I also include an institutional measure for female employment that more generally applies to all women (gender empowerment; see below) instead of institutions targeted on women with young children (maternity leave, public childcare provision). An accompanying advantage of the more general institutional measure is that is not collinear with gender norms (Uunk et al. 2005), yet only shows a moderately high association. This leaves room for independent effects.
In comparison to the previous studies testing the effect of a country's gender-role attitudes on female labor supply, I also want to go one step further. Although the multilevel study of Steiber and Haas (2009) did assess whether country-level gender-role attitudes affect female labor supply net of individual-level gender attitudes, they did not assess the extent to which the country-level gender-role attitude effect can be attributed to individual-level gender attitudes. This may sound trivial, but it is not. The control for individual-level gender attitudes was done to control for population composition. Yet, if individual-level attitudes are themselves influenced by the contextual-level attitudes, the part of the contextual effect attributable to individual-level attitudes does not represent a simple compositional effect but an indirect contextual effect. While I cannot sort out with the cross-section data at hand whether the relationship between contextual-level attitudes and female labor supply occurs through affecting individual-level attitudes (that is, indirect causation) or simply reflects the influence of individual-level attitudes (that is, composition), it seems worthwhile to find out to what extent the contextual attitude effect can be attributed to an individual-level attitude effect because it can answer the question why we would find a relationship between contextual-level attitudes and female labor supply. An immediate, ‘standard’ control for individual-level attitudes in the models may be too strong a control, and could hide an indirect contextual effect.
From a societal and policy standpoint, the question whether gender-role attitudes on the country level affect female labor supply and whether its effect is attributable to institutional effects or individual-level attitude effects is obviously relevant. Despite rapid changes in female labor market participation, women still participate less often on the labor market than men and there are marked differences in female labor supply across countries in Europe (Stier et al.2001; Anxo et al.2007). Knowing the structural and cultural factors impeding female labor supply is of paramount interest.
2. Theoretical background
My baseline hypothesis is there will be an initial, gross effect (when institutions and individual-level attitudes are still not taken into account) of a country's prevailing gender-role attitude on individual women's labor supply. I expect that the more favorable the attitudes are toward women's role on the labor market vis-à-vis household work, the more likely it is that women are attached to the labor market (Hypothesis 1). Below, I will elaborate on three explanations for an effect of contextual-level gender-role attitudes: direct social influencing, composition or indirect social influencing, and institutional variation.
2.1. Direct social influencing
A first explanation for a contextual gender-role effect can be found in classical sociological literature. A more general idea in sociology – building upon Durkheim's (1897) theory of social integration – is that people's behavior is not only led by their own attitudes, but also led by the attitudes, values, and norms of people in the context in which they are integrated. The people in the context that may influence a person are numerous and diverse, and may include one's close interaction persons, but also national media.
But why do the attitudes in the country affect individual women's labor supply? A specific sociological explanation for a direct contextual attitude effect on individual behavior is normative sanctioning. Attitudes and values of surrounding others may operate as social norms for the individual. Such norms may not solely be typical for the community in which one resides, but may reflect the social norms of a broader geographical context such as the country. The normative expectations in the context are likely to influence female labor supply because women may face costs when they do not behave according to these expectations. Others may apply sanctions when the woman does not comply with their expectations. They may limit instrumental help or social contacts. For example, a nonemployed woman living in a traditional-oriented country who is preferring paid work over household work (that is, who has modern-egalitarian gender-role attitudes), may decide not to enter the labor market because she receives no help in her search for work or even disapproval from persons in her social setting. Social norms, therefore, may counteract the effects of individual attitudes and as such have a direct effect on female labor supply independent of individual attitudes. That is: more egalitarian gender-role attitudes on the country-level will increase individual women's labor supply independent of individual gender-role attitudes (Hypothesis 2).
I expect that this net, positive effect of a country's gender-role attitudes is stronger for women with children at home than for women without children at home, implying a positive interaction effect of national (egalitarian) gender-role attitudes by the presence of children (Hypothesis 3). The reason for this expectation is that gender norms are stronger for women with than women without children at home. From women it is generally less often expected that they work than men, and this is especially so when women have young children.2 Sanctions for not living to the norm may therefore also be higher for women with children than women without children at home. An additional argument why it is important to test whether the effect of a country's gender-role attitudes on female labor supply is distinct for women with than women without children is that prior multilevel studies found a null effect of a country's gender attitude for women with young children (Uunk et al. 2005; Steiber and Haas 2009). Maybe the context effect will only show up for one of the two groups.
2.2. Composition or indirect social influencing
A second explanation for an initial, gross effect of contextual gender-role attitudes on female labor supply denies that there is a direct, independent effect and states that the contextual effect is due to individual-level gender-role attitudes, either through a compositional effect or through an indirect effect.
To understand the reasoning, it is first necessary to know why individual-level gender attitudes may affect female labor supply at all. An effect of individual gender attitudes on labor supply can be argued by the work of social psychologists Fishbein and Ajzen (1975) and Cooper and Croyle (1984), and is also central to the sociological ‘Preference Theory’ of Hakim (1991). Under certain conditions, attitudes determine behavior because what people do is partly guided by their preferences and these preferences are anchored in their attitudes. Thus, individual gender-role attitudes may affect women's employment decisions because these attitudes reflect women's preferences to work: if women hold more traditional attitudes on the gender roles, whereby men specialize in paid work and women specialize in household and caring for the children, women's preferences to work are assumed to be lower than if women hold more egalitarian attitudes on the gender division of paid work and care (Jansen and Kalmijn 2002). Although the reverse argument is also upheld, postulating an effect of behavior on attitudes, prospective panel studies have shown that the effect of gender-role attitudes on employment is about as strong as the effect of employment on gender-role attitudes (Corrigal and Konrad 2007).
The compositional interpretation via individual attitudes, then, is that higher female labor supply in more modern-oriented countries is simply due to a higher share of women who hold modern gender-role attitudes. There will exist an effect of individual gender attitudes, but no additive effect of contextual attitudes. The indirect effect interpretation is more complicated. It assumes that individual-level gender attitudes are in itself influenced by contextual-level attitudes, for example via socialization (country-level gender attitudes during the socialization period influence individual attitudes) and adaptation of individual attitudes to the context (current country-level gender attitudes affect individual attitudes). Country-level gender-role attitudes may in this view affect female labor supply indirectly via individual-level attitudes.
Both the composition and indirect effect interpretation have the same hypothesis, namely that the initial, gross effect of national gender-role attitudes on female labor supply will disappear when controlling for individual-level gender attitudes (Hypothesis 4). However, the two interpretations have different implications. While the compositional interpretation dismisses the view that a country's gender-role attitudes matter, the indirect effect interpretation nuances it: there is no direct effect of the context independent of individual attitudes, but an indirect effect via individual attitudes. With the cross-section design employed I can unfortunately not disentangle the two interpretations, but if the truth lies somewhere in the middle (that is, there will be a weak indirect effect) the message is that a simple, standard control for individual gender attitudes may be too strong and may take away some of the interesting music.
2.3. Institutional variation
A third and final explanation for an initial gross effect of national gender-role attitudes on individual women's labor supply focuses on the role of institutions. Countries with more egalitarian gender-role attitudes may have higher female labor supply because these countries are characterized by more generous state support for female employment than more traditional countries (Uunk et al. 2005), and because institutional support for female employment increases women's labor supply. Multilevel, cross-country and within-country, cross-regional studies provide support for the claim that female employment supportive institutions increases women's labor supply, in specific public childcare provision (Pettit and Hook 2005; Uunk et al. 2005; Van Ham and Mulder 2005; Van Ham and Buchel 2006; Steiber and Haas 2009; Del Boca et al. 2009). It is possible, therefore, that the (positive) effect of a country's egalitarian gender-role attitudes on individual women's supply can be accounted for (will be mediated) by female employment supportive institutions.
The above reasoning assumes that attitudes causally precede institutions. This may be the case because institutional arrangements are partly formed on the basis of prevailing attitudes and norms, and because institutional arrangements need to be backed up by public acceptance. Governments and their employment-supportive policies respond to how men and women think about the combination of work and employment (Pfau-Effinger 1999). However, a rival view on the causal relationship between attitudes and institutions holds that institutions precede attitudes, and not vice versa. New institutional arrangements may arise for reasons not much related to the public opinion, but these new institutions may alter the public opinion (Gelissen 2002). This rival view on the causal relationship between attitudes and institutions implies that the effect of contextual gender-role attitudes on female labor supply is spurious upon institutions. Institutional arrangements affect both the gender-role attitudes of the public at large and the labor supply of women, and it is therefore that national gender-role attitudes and female labor supply will show an association.
Since I use cross-sectional data I cannot empirically distinguish the two rival versions of the institutional argument, yet both versions have the same implication: if existing, the effect of contextual gender-role attitudes on female labor supply should disappear once institutional arrangements are taken into account (Hypothesis 5). In their multilevel, cross-country study of young mothers' labor supply Uunk et al. (2005) found support in favor of this expectation. My study will test whether this applies also to female labor supply in general, using an institutional measure that is not merely targeted at young mothers but all women.
3. Data, measures, and method
3.1. Data
To test the effect of individual and contextual gender-role attitudes on female employment I use data from the 2008 wave of the EVS. The EVS is a large-scale, cross-national, and longitudinal survey research program on ideas, beliefs, preferences, attitudes, values, and opinions of citizens all over Europe. The EVS started in 1981, when a thousand citizens in the European Member States of that time were interviewed using standardized questionnaires. Every nine years, the survey is repeated in an increasing number of countries. The fourth wave in 2008 covers 47 European countries/regions, from Iceland to Azerbaijan and from Portugal to Norway (for more information see http://www.europeanvaluesstudy.eu/). I restrict my analyses to countries for which a measure of institutional support for female employment is available (six countries were dropped: Albania, Bosnia and Herzegovina, Belarus, Luxemburg, Serbia, Kosovo), for which reliable estimates of the gender attitude scale could be observed (seven countries were dropped: Azerbaijan, Armenia, Czech Republic, Georgia, Moldova, Romania, Turkey), and that show sufficient within-country variation in female labor force supply (Sweden dropped). This leaves me with 33 countries of which 20 are Western European and 13 Eastern European countries (for names, see Figure 1). Northern Cyprus, although only recognized by Turkey as an independent state, is also counted as a contextual unit in the analyses as the EVS 2008 wave also sampled in some regions separately.
Female labor force participation rate by national gender-role attitudes (solid line is regression line)
Source: EVS 2008 (own calculations).
Source: EVS 2008 (own calculations).
In the EVS 2008, representative multistage or stratified random samples of the adult population of 18 years old and older were drawn, and about 1500 respondents per country were interviewed at home using face-to-face interviews in the period 2008–2010 (except for Finland that used an Internet panel on a sample of people aged 18–74). In my study, the analytical sample consists of 13,924 women aged 18–55 (upper limit because of retirement and health-related reduction in labor supply). Women with missing values on the individual-level measurement of gender-role attitude items were excluded, as well as women with missing values on the central main activity status question, and women with missing values on educational level (in total, 1% of the original sample). Women in the selected age range who are restricted in their labor market participation because their main activity status is student, retiree, permanently disabled, in military service, or unknown were also excluded (12% of the sample).
3.2. Measures
The dependent variable in the analyses is women's labor force supply. Labor force supply is defined as being on the labor market or not and is measured by a self-report question on the main activity status. Women who report to be employed (more than 30 hours, less than 30 hours, or self-employed) or unemployed are coded as being in the labor force (score = 1). Women who report to be a housewife or otherwise not in the labor force are coded as not in the labor force (score = 0). As noted above, I exclude retirees, students, persons in military service, and (permanently) disabled persons, because these women do not offer themselves for employment. I restrict my analyses to labor force supply because this labor decision is still a central cause of gender inequality, because there are wide differences in female labor supply across countries (see below), and because the gender-role attitudes in the EVS data center on this issue (being a housewife or working for pay). I do not analyze the odds of being employed for those who are on the labor market since that depends more on labor market opportunities. I also do not opt for analyzing the odds of full-time versus part-time employment because part-time work is often institutionally arranged, which implies that an analysis of part-time work suffers from endogeneity as to the institutional determinant.
Independent, control variables at the individual level are age, the presence of children living in the household, living with a partner – married or unmarried – and education. The age of children is not known which is unfortunate for an analysis of female employment, yet note that not only mothers but also women without children are analyzed. Education is measured as the highest level of full-time education accomplished and is coded according to the International Standard Classification of Education 1997 using its first digit code: (1) preprimary education or none education; (2) primary education or first stage of basic education; (3) lower secondary or second stage of basic education; (4) upper secondary education; (5) postsecondary nontertiary education; (6) first stage of tertiary education; (7) second stage of tertiary education. Education is modeled as an interval variable since female labor force supply proved to increase almost linearly with educational level. It is important to control for variables as age, children, partner status, and education because these variables are known to influence women's labor force supply as well as gender-role attitudes. For example, earlier cross-comparative research has shown that countries with a larger share of higher educated are characterized by more egalitarian gender-role attitudes (Kalmijn 2003; André et al.2013). It may therefore not be the national gender-role attitudes that affect individual women's labor supply, but the educational composition of the population (cf. Uunk et al. 2005).
The key independent variable on the individual level is the gender-role attitude. Individual gender-role attitudes are measured by a block of eight items that center on the role of women (and partly of men) in paid work, care, and household tasks. They are stated as follows:
People talk about the changing roles of men and women today. For each of the following statements I read out, can you tell me how much you agree with each.
A working mother can establish just as warm and secure a relationship with her children as a mother who does not work;
A pre-school child is likely to suffer if his or her mother works;
A job is all right, but what most women really want is a home and children;
Being a housewife is just as fulfilling as working for pay;
Having a job is the best way for a woman to be an independent person;
Both the husband and wife should contribute to household income;
In general, fathers are as well suited to look after their children as mothers;
Men should take as much responsibility as women for the home and children.
Although factor analyses identified two underlying dimensions (distinguishing the first four from the latter four items), I decided to construct one scale. A first, pragmatic reason is that reliability dropped when distinguishing two underlying scales. The alpha for the entire scale is 0.65, and for the subscales, respectively, 0.58 and 0.63. In addition, reliability scores for single countries were substantially higher when using the one-scale measure. A second reason is more methodological. Taking two (or more) measures for gender-roles (at the country level) would lower the chances of finding an effect of national gender-role attitudes since the degrees of freedom on the contextual level are fairly low. The final gender-role attitude measure was constructed by computing the mean over the eight items, under the condition that for each respondent at least five items were nonmissing (only 17 respondents did not meet this criterion and were dropped from the analyses). Alpha reliability scores were higher in Western European countries (0.69) than Eastern European countries (0.57), a finding that has been noted earlier (André et al.2013). Note that countries with scores lower than 0.50 were already dropped from the analytical sample.
Country-level gender-role attitudes are aggregated from the individual EVS data. Using a similar measurement procedure as for the individual-level gender-role attitudes (distinguishing one scale and using the same items and coding), I computed the mean level of egalitarian gender-role attitudes in respondent's country of residence. In contrast to the individual-level attitude measures, the country-level measures are obtained from all survey respondents in the country of residence, not only from women (N = 54,743 respondents).4 The reason for this is that I assume that all sorts of people can influence individual women's labor supply, not only the women who belong to the sample of analysis. The country scale therefore measures the prevailing gender norm.
To measure the institutionally support for female employment, I use the Gender Empowerment Measure (GEM). GEM measures the gender equality in ‘women's participation in political decision-making, their access to professional opportunities, and their earning power’ (United Nations Development Program [UNDP] 1995: 72; in Fuwa 2004: 756). GEM is obtained from the Human Development Report 2007/2008 (UNDP 2007; for Montenegro from the 2009 report, UNDP 2009) and is constructed from the combination of the percentage of parliamentary seats held by women, the percentage of administrators and managers who are women, the percentage of professional and technical workers who are women, and women's share of earned income compared to that of men (Fuwa 2004: 756). It ranges from 0 to 1 with higher values representing greater gender equality. In contrast to institutional measures as public childcare provision, GEM is not a direct measurement of state-provided institutional support for female employment, yet it does measure how the state has shaped gender equality and promoted female employment and therefore mirrors the degree of female friendliness of a state's institutions. The advantage is that it applies to all women and is not specifically targeted at young mothers' employment such as public childcare provision. In addition, GEM does not correlate too highly with national gender-role attitudes (in the sample of 33 countries, r = 0.53, p = 0.002) so that both national gender attitudes and institutions may have independent effects.5
The level of economic affluence is a control variable at the country level. It is also taken from the Human Development Report 2007/2008 (UNDP 2007), and measures the gross domestic product (GDP) per capita for the year 2007 at the price levels and exchange rates (Purchasing Power Parity) of 2005 US dollars. It is important to control for GDP in the analyses because not doing so may suppress the effect of contextual gender-role attitudes (Uunk et al.2005). Women in countries with traditional gender-role attitudes still work quite often due to economic necessity. Once economic affluence would be netted out, differences in labor force supply between countries with traditional and modern gender-role attitudes would be larger.
Table 1 lists descriptive statistics of the independent and dependent variables in the analyses.
| Mean | Standard deviation | Min. | Max. | |
|---|---|---|---|---|
| On the labor market | 0.85 | 0.00 | 1.00 | |
| Age | 38.55 | 10.05 | 18.00 | 55.00 |
| Child(ren) | 0.64 | 0.00 | 1.00 | |
| Partner | 0.68 | 0.00 | 1.00 | |
| Education | 4.39 | 1.25 | 1.00 | 7.00 |
| Gender attitude | 3.67 | 0.59 | 1.13 | 5.00 |
| Country: gender attitudea | 3.55 | 0.21 | 3.19 | 4.13 |
| Country: GDP (/10,000)a | 3.03 | 2.04 | 0.30 | 8.24 |
| Country: gender empowermenta | 0.68 | 0.13 | 0.46 | 0.91 |
| Mean | Standard deviation | Min. | Max. | |
|---|---|---|---|---|
| On the labor market | 0.85 | 0.00 | 1.00 | |
| Age | 38.55 | 10.05 | 18.00 | 55.00 |
| Child(ren) | 0.64 | 0.00 | 1.00 | |
| Partner | 0.68 | 0.00 | 1.00 | |
| Education | 4.39 | 1.25 | 1.00 | 7.00 |
| Gender attitude | 3.67 | 0.59 | 1.13 | 5.00 |
| Country: gender attitudea | 3.55 | 0.21 | 3.19 | 4.13 |
| Country: GDP (/10,000)a | 3.03 | 2.04 | 0.30 | 8.24 |
| Country: gender empowermenta | 0.68 | 0.13 | 0.46 | 0.91 |
aCountry-aggregate descriptive statistics.
Source: EVS 2008 (own calculations).
3.3. Method
I use multilevel logistic regression models to test whether individual and country-level gender-role attitudes influence individual women's labor supply. The multilevel regression models distinguish two levels: individuals and countries. All models are estimated in Stata 13.
4. Results
In line with findings by Kalmijn (2003) for the 1999 EVS wave and André et al. (2013) for the 2002 wave of the International Social Survey Programme, I find variation in gender-role attitudes between the 33 countries I selected for analysis from the 2008 EVS wave. Unreported multilevel analysis on all survey respondents (both men and women) document that the country variance is statistically significant (p < 0.01), but the proportion of variance explained by countries (the intraclass correlation) is small with 12%. Gender-role attitudes vary much more between individuals than between countries (cf. Kalmijn 2003; André et al. 2013).
How do the gender-role attitudes at the country level associate with female labor supply? Figure 1 plots women's labor force participation rate by the mean level of (egalitarian) national gender-role attitudes. Although this plot suffers from the lack of individual-level and country-level control variables, it provides insight in the country-specific scores on gender-role attitudes and female labor supply, it provides insight in the strength of the relationship between these two characteristics, and it provides insight in countries deviating from the pattern. The figure displays a positive association between national, egalitarian gender-role attitudes, and female labor supply. The more egalitarian the gender-role attitudes of a country, the higher female labor supply. The association is strong and statistically significant (Pearson r = 0.64, N = 33, p < 0.01). This indicates that although countries differ modestly in gender-role attitudes, these country differences in attitudes might be consequential for female employment. Female labor market participation in the most modern-oriented countries Denmark and Norway (respectively, 99% and 94%) is, for example, at least 30 percentage points higher than in the most traditional-oriented countries Malta and Northern Cyprus (respectively, 50% and 58%). Noticeable is that relatively many Eastern European countries have rather traditional gender-role attitudes but high female labor force participation (Lithuania, Russia, Poland, Estonia have participation rates of nearly 90%). The high female labor supply in these countries may be explained by an economic necessity for women to work.
The multilevel models in Table 2 estimate whether the country-level gender-role attitudes affect individual women's labor supply independent of other country- and individual-level characteristics. Model 1 is the ‘empty’ (random-intercept) model, estimating variation in the intercept only. As suggested by Figure 1, the model parameters display significant country variation in women's odds of labor force supply. The intraclass coefficient is (1.96/(1.96 + 3.29) =) 0.37 which suggests that 37% of the total variation in labor supply is due to differences between countries. Model 2 estimates the total or ‘gross’ effect of the country-level gender-role attitudes on women's odds of being in the labor force. As hypothesized and in line with Figure 1, the effect is positive and statistically significant. The more egalitarian gender-role attitudes are in a country, the higher individual women's odds to be in the labor force. The effect (b = 3.15) is substantial: the odds ratio of being in the labor force versus not being in the labor force between the most traditional country (Malta; mean gender attitude is 3.19) and the most modern country (Norway; mean gender attitude is 4.13) is (exp [(4.13–3.19) × 3.15]) a factor 19.
| Model 1 | Model 2 | Model 3 | Model 4 | Model 5 | Model 6 | Model 7 | |
|---|---|---|---|---|---|---|---|
| Constant | 1.955** (0.158) | −9.230** (2.082) | −9.416** (2.197) | −10.748** (2.187) | −11.075** (2.232) | −11.139** (2.158) | −10.787** (2.441) |
| Individual-level | |||||||
| Age | 0.019** (0.003) | 0.019** (0.003) | 0.019** (0.003) | 0.019** (0.003) | 0.019** (0.003) | ||
| Child | −1.133** (0.073) | −1.134** (0.073) | −1.109** (0.075) | −1.108** (0.074) | −1.284 (1.833) | ||
| Partner | −1.154** (0.077) | −1.155** (0.768) | −1.113** (0.078) | −1.113** (0.078) | −1.117** (0.079) | ||
| Education | 0.507** (0.025) | 0.507** (0.025) | 0.439** (0.025) | 0.439** (0.025) | 0.441** (0.025) | ||
| Gender attitude | 1.028** (0.051) | 1.028** (0.051) | 1.029** (0.051) | ||||
| Country-level | |||||||
| Gender attitude | 3.147** (0.587) | 2.897** (0.618) | 3.375** (0.634) | 2.522** (0.648) | 2.178** (0.666) | 2.085** (0.750) | |
| GDP (/10,000) | −0.122~ (0.062) | −0.140* (0.064) | −0.239** (0.091) | −0.212* (0.093) | |||
| Gender empowerment | 2.333 (1.576) | 2.140 (1.606) | |||||
| Gender attitude (country) × child (ind.) | 0.056 (0.522) | ||||||
| Individual intercept variancea | 3.29 | 3.29 | 3.29 | 3.29 | 3.29 | 3.29 | 3.29 |
| Country intercept variance | 1.955** (0.158) | 0.633** (0.087) | 0.668** (0.092) | 0.626** (0.087) | 0.638** (0.089) | 0.614** (0.086) | 0.591** (0.094) |
| Country child-effect variance | 0.311** (0.100) | ||||||
| Log likelihood | −5381.6 | −5370.7 | −4709.4 | −4707.6 | −4489.6 | −4488.6 | −4486.1 |
| Model 1 | Model 2 | Model 3 | Model 4 | Model 5 | Model 6 | Model 7 | |
|---|---|---|---|---|---|---|---|
| Constant | 1.955** (0.158) | −9.230** (2.082) | −9.416** (2.197) | −10.748** (2.187) | −11.075** (2.232) | −11.139** (2.158) | −10.787** (2.441) |
| Individual-level | |||||||
| Age | 0.019** (0.003) | 0.019** (0.003) | 0.019** (0.003) | 0.019** (0.003) | 0.019** (0.003) | ||
| Child | −1.133** (0.073) | −1.134** (0.073) | −1.109** (0.075) | −1.108** (0.074) | −1.284 (1.833) | ||
| Partner | −1.154** (0.077) | −1.155** (0.768) | −1.113** (0.078) | −1.113** (0.078) | −1.117** (0.079) | ||
| Education | 0.507** (0.025) | 0.507** (0.025) | 0.439** (0.025) | 0.439** (0.025) | 0.441** (0.025) | ||
| Gender attitude | 1.028** (0.051) | 1.028** (0.051) | 1.029** (0.051) | ||||
| Country-level | |||||||
| Gender attitude | 3.147** (0.587) | 2.897** (0.618) | 3.375** (0.634) | 2.522** (0.648) | 2.178** (0.666) | 2.085** (0.750) | |
| GDP (/10,000) | −0.122~ (0.062) | −0.140* (0.064) | −0.239** (0.091) | −0.212* (0.093) | |||
| Gender empowerment | 2.333 (1.576) | 2.140 (1.606) | |||||
| Gender attitude (country) × child (ind.) | 0.056 (0.522) | ||||||
| Individual intercept variancea | 3.29 | 3.29 | 3.29 | 3.29 | 3.29 | 3.29 | 3.29 |
| Country intercept variance | 1.955** (0.158) | 0.633** (0.087) | 0.668** (0.092) | 0.626** (0.087) | 0.638** (0.089) | 0.614** (0.086) | 0.591** (0.094) |
| Country child-effect variance | 0.311** (0.100) | ||||||
| Log likelihood | −5381.6 | −5370.7 | −4709.4 | −4707.6 | −4489.6 | −4488.6 | −4486.1 |
aIndividual (intercept) variance in a multilevel logistic regression is π²/3 =3.29 (Snijders and Bosker 1999).
**p < 0.01; *p < 0.05; ~p < 0.10.
Source: EVS 2008 (own calculations).
Model 3 adds the effects of individual-level confounders age, children, partner status, and education in order to test whether the above-noted gross effect of national gender-role attitudes is due to population composition on these characteristics. The model shows that age and education increase the odds of labor force attachment and the presence of children and of a partner decrease these odds. Adding these individual-level covariates to the model reduces the country-level gender-role attitude effect from 3.15 (Model 2) to 2.90 (Model 3). That is, 8% of the gross effect of country-level gender-role attitudes on female labor force supply is due to compositional heterogeneity in the above characteristics. Additional unreported analyses have shown that education is the one responsible covariate: why in countries with more egalitarian gender norms women have a higher odds to be in the labor force than elsewhere can to some extent be explained by the educational composition of the population (modern-oriented countries have a larger share of higher educated persons). Yet, educational composition cannot ‘explain away’ the effect of country-level gender-role attitudes since the contextual effect remains significant net of education.
Model 4 adds the effect of GDP. The effect is negative and significant when using the more tolerant 10% p-level (p = 0.050). It means that the higher GDP, the lower the odds of women's labor force attachment. As expected, GDP suppresses the relationship between national gender-role attitudes and female labor supply (Uunk et al2005): inclusion of GDP in the model increases the effect of national gender-role attitudes from 2.90 (Model 3) to 3.38 (Model 4). That is, women in countries with more egalitarian gender-role attitudes have even a greater labor supply once one reckons that the economic affluence suppresses their labor supply. Vice versa, women in traditional-oriented countries are relatively often in the labor market due to their low GDP (work as economic necessity); once reckoned with GDP, the difference in labor supply with women from modern-oriented countries becomes stronger. This was earlier suggested for some Eastern European countries that display high female labor supply yet traditional gender-role attitudes.
Model 5 adds the effect of individual-level gender-role attitudes and tests to what extent the effect of national gender-role attitudes is due to covariance with individual-level attitudes. The model displays a positive effect of individual gender-role attitudes on labor force attachment. This individual-level attitude effect is the strongest predictor of women's labor supply (relative effects not shown in table). Including individual gender attitudes reduces the contextual attitude effect from 3.38 (Model 4) to 2.52 (Model 5), a reduction in effect size by 25%. That is, one-fourth of the total effect of national gender-role attitudes is due to covariance with individual-level attitudes. This part, the one-fourth, may reflect either an indirect effect of national gender-role attitudes on female labor supply via individual gender-role attitudes, or a compositional effect whereby countries with more modern gender-role attitudes are characterized by higher shares of women with modern gender-role attitudes, without national gender-role attitudes being the cause of these higher shares. Important to note is that net of individual-level gender-role attitudes a significant effect of national gender-role attitudes remains in Model 5. Women in countries with more egalitarian gender-role attitudes offer themselves more often for employment than women in countries with less egalitarian attitudes, and this holds also when women have traditional or modern attitudes themselves.
Model 6 tests whether the cultural context still matters when reckoned with the country's institutional support for female employment. It adds the effect of a country's score on GEM. In contrast to studies on institutional effects on mother's labor supply (Pettit and Hook 2005; Uunk et al. 2005; Steiber and Haas 2009), the effect of GEM appears non-significant. Yet, the p-value (p = 0.13) is close to reaching significance, and the direction of the effect is in the predicted positive direction. Furthermore, when the same model was run excluding national gender-role attitudes, the effect of GEM was significant (b = 4.14, p = 0.015). That the GEM effect disappears when introducing national gender-role attitudes indicates that the effect of GEM on female labor supply is in itself spurious on national gender-role attitudes, and not the other way around (effect national gender-role attitudes on female labor supply not spurious on GEM). Introducing GEM in the model, therefore, does not alter the national gender-role attitude effect much. Model 6 displays this. The national gender-role attitude effect only slightly reduces from 2.52 (Model 5) to 2.18 (Model 6), a reduction in effect size of 14%. To compare the amount of the national attitude effect explained by GEM with the amount explained by individual-level gender-role attitudes, I reran Model 4 adding GEM. The effect of national gender-role attitudes of this model is b = 3.04 which is a reduction of the total effect of [1 – (3.04/3.38) × 100% =] 10%. Thus, GEM explains substantially less of the national gender-role attitude effect than individual-level gender-role attitudes (part explained by GEM is 10% and by individual attitudes 25%).
Model 6 further shows that a significant effect of national gender-role attitudes on female labor supply remains. This means that – at least for the current sample of women and countries – the national gender culture influences women's labor supply rate, and it does so independent of institutions and individual-level gender attitudes. The effect is substantial. Its relative effect is on par with that of having children and having a partner, yet lower than that of education, individual-level gender attitudes, and GDP (relative effects not shown).
Model 7 of Table 2, finally, tests whether there exists an interactive effect of national gender attitudes and the presence of children on the odds of female labor supply. Judged by the effect parameter the interaction effect is not significant, despite significant country variation in the child-effect. That the effect of national gender-role attitudes is as strong for women with children as women without children at home may come as a surprise given the more pronounced gender norms toward the home and work for women with children at home, yet it is quite feasible that the labor supply of women with children at home is – due to the restrictions the care for (young) children imposes – less sensitive to external factors than the labor supply of women without children at home.
5. Conclusions and discussion
In this study I investigated the effect of country-level gender-role attitudes on individual women's labor supply. Although this has been examined earlier in the literature with respect to mothers' labor supply (Pettit and Hook 2005; Uunk et al. 2005; Steiber and Haas 2009), this has not been done with a powerful multilevel design with many countries and not for all women (women with and without children at home). In addition, I wanted to contribute to the literature by estimating the extent to which an effect of national gender-role attitudes on female employment – if existing – can be attributed to individual-level gender-role attitudes and institutions.
My analyses have shown that there is a positive effect of national gender-role attitudes on women's individual odds of labor force participation. Women in countries with more modern-oriented, egalitarian gender-role attitudes are more often attached to the labor market (employed or unemployed) than women in countries with less modern-oriented, egalitarian attitudes. This national gender-role attitude effect can for one-fourth be attributed to an effect of individual-level attitudes and for one-tenth by institutional support for women, yet a large independent effect of national gender-role attitudes remains. The analyses have further shown that the national gender-role attitude effect is equal for women with and without children at home.
The findings have some implications. First, the independent effect of a country's gender-role attitudes on female labor supply implies that the substantial country variation in gender-role attitudes that has been observed for Europe (Kalmijn 2003; André et al.2013) is not without consequences for individual behavior, and it supports the claim of some scholars that contextual gender norms and attitudes matter for female labor supply (Pfau-Effinger 1998; Albrecht et al. 2000; Geist 2005; Heineck 2007). The contextual attitude effect that exists independent of institutions and individual gender attitudes, may be explained by normative sanctioning. Even the traditional woman in a modern-oriented country – who not adapted to the attitudes around her – may be pushed to the labor market due to the threat of informal sanctions from the community. Earlier cross-national studies applying a multilevel design did not find evidence for an independent effect of national gender-role attitude on individual women's labor supply (Pettit and Hook 2005; Uunk et al. 2005; Steiber and Haas 2009), yet these studies are restricted because they analyzed fewer countries, modeled the odds of employment instead of the odds of labor force attachment, and used institutional measures specifically targeted at young mothers' employment (for example, public childcare provision) that were collinear with national gender-role attitudes. In the current study I used a more powerful multilevel design with 33 countries, modeled the odds of labor force attachment, and used an institutional measure (GEM) that applies to all women and correlates moderately high with gender norms. In this setup, the contextual attitude effect displayed a substantial and significant effect. The restriction in previous studies to analyze young mothers' labor supply only cannot explain the discrepancy in findings between the current and previous studies, as my analyses showed that the national gender attitude effect was equal for women with and without children at home.
A second implication of the findings is that it is important to control for individual-level (gender-role) attitudes when estimating the national (gender-role) attitude effect. One-fourth of the national gender-role attitude effect could be attributed to the effect of individual-level gender-role attitudes. At first sight this may simply reflect a compositional effect: women in countries with more modern-oriented gender attitudes are (in part) more attached to the labor market just because these countries have higher shares of modern-oriented women, and because a more modern gender-role attitude promotes labor supply. In this view, the methodological lesson to be learned would be to control for the individual namesake variables in making claims on effects of (derived) contextual factors (Moore and Vanneman 2003; Diez Roux 2004). On second thoughts, it may not be as simple as that. The question arises why women in the first place are more modern-oriented in more modern-oriented countries. The two may not be unrelated. Country-level gender-role attitudes may have influenced individual-level gender-role attitudes, and these individual attitudes in turn influence women's labor supply decision. Thus, the one-fourth part of the gross effect of national gender-role attitudes that is due to individual-level attitudes may next to composition also be understood as an indirect effect of country attitudes on individual behavior via individual attitudes. This indirect influence of the cultural context may have come about by socialization or adaption of individual attitudes to the context, and as such does not weaken the claim that the cultural context matters but nuances it. If the truth lies somewhere in the middle – that is, both composition and indirect influencing play a role – a further implication of my analyses is that a simple, ‘standard control’ for individual (gender) attitudes in analyses of contextual (gender) attitude effects, is too strong.6
My study is not without shortcomings, the prime being the cross-sectional research design. This design may bias the contextual gender-role attitude effect on female labor supply. However, overestimation of the effect due to reverse causation is not very likely. Although individual women's labor supply may change her gender attitudes, it is not probable that she changes national gender-role attitudes. Residence selection (women who are employed migrate to countries with more modern gender attitudes) is neither a likely mechanism accounting for the reported contextual gender attitude-individual labor supply relationship. In fact, underestimation of the contextual attitude effect is more likely. Cross-sectional analyses, namely, overestimates due to reverse causation the individual attitude effect on female labor supply, and because of this the statistical control for the individual-level attitude effect is too strong, which undermines the strength of a contextual attitude effect. Additional problems with the cross-sectional design is that one cannot assess whether individual-level attitudes are the outcome of national gender-role attitudes and thus represent an indirect or compositional effect, and whether contextual gender attitudes precede institutions or vice versa. The conclusions of my study on contextual attitude effects therefore have to be treated with caution and future research should investigate the causality of effects by using longitudinal designs (Steiber and Haas 2012). Other ways to proceed research on contextual (gender) attitude effects are to test effects of the more immediate social context and of one's social-cultural reference groups.
Acknowledgments
I would like to thank Matthijs Kalmijn and anonymous reviewers from European Societies for their helpful comments on an earlier draft of this paper.
Footnotes
Using a macro–macro design with more power to disentangle institutional effects from cultural effects, Fortin (2005) did observe an independent effect of the national gender culture. Yet, this macro–macro design suffers from a lack of (important) individual controls, including the individual-level gender attitudes (see arguments below).
My analyses on the gender-role items in the EVS display that people (all survey respondents, including men and women) on average are most traditional on the items involving the consequences of women's work for children (items 2 and 3; see below).
I also reran analyses using the original four response categories (excluding the ‘don't knows’), yet did not find distinct effects.
The attitude of the respondent is included in the country means. This does not introduce any bias since the number of respondents used to compute the country mean is large (at minimum 600 per country). An alternative country-level mean measure excluding the respondent correlates 0.997 with the country-level measure including the anchor respondent.
For 21 countries in the sample a measure of public childcare expenditure was available (see André et al.2013). GEM correlates strongly with the public childcare measure (r = 0.65, p = 0.001), which raises confidence that GEM measures institutional support for female employment.
Even effects of ‘standard compositional’ variables such as education may reflect indirect causation, provided they are (partly) caused by the cultural context.
References
Wilfred Uunk is an Assistant Professor at the Department of Sociology, Tilburg University, The Netherlands. In 1996 he took his Doctoral Degree on the dissertation ‘Who marries whom? The role of social origin, education and high culture in mate selection of industrial societies during the twentieth century’. His research interests include cross-comparative research, ethnic minorities, and sociology of the family.