Women’s employment and the decline of home cooking: Evidence from France, 1985–2010

Abstract

We here investigate the extent to which labour-market changes explain the decline in the time spent home cooking by married women in France between 1985 and 2010. Using time use data and Oaxaca-Blinder decompositions, we find that rising women’s employment and observed wages together account for about 60% of the fall in the time married women spent cooking. We then use a semi-parametric matching technique to construct an implicit wage rate, which better reflects the change in labour-market incentives that individuals face. The rise in women’s implicit wages explains no more than 20% of the decline in their cooking time, while the wage of their partner has no effect. Changing labour-market incentives are thus far from being the main driver of the decline in home-cooking. We also find evidence that home cooking continues to be structured by the gendered social norm of the “proper family meal”.

1 Introduction

In most human societies, home cooking was and still is traditionally carried out by women. The rise in women’s labour-force participation is thus likely to have had an impact on food preparation at home. Aggregate data does indeed reveal a negative correlation between trends in women’s labour supply and non-market work such as home-cooking. For instance, American women spent 3.8 h more per week at work in 2003 than in 1965, but 10.3 h less in non-market work (Aguiar and Hurst 2007). Similarly, the rise in French urban women’s labour-force participation between 1974 and 2010 is associated with a four-hour fall in their supply of non-market work (Brousse 2015). The leading economic explanation of this negative correlation is price substitution. As women obtain more opportunities to participate in the job market and to earn decent wages, the opportunity cost of time spent in household production rises, as does the relative full price of cooking from raw ingredients at home. This rise in the relative price of home cooking, combined with time-saving innovations in food processing, may thus explain why the purchase of processed or semi-prepared food and the consumption of food away has progressively become more attractive than “cooking from scratch”.

We here use French time-use data collected in 1985–1986 (INSEE 1986) and 2009–2010 (INSEE 2010) to investigate the extent to which labour-market changes explain the fall in home-cooking time in recent decades. France is of particular interest here for two reasons. First, similar negative trends in cooking time are observed for the US, France and the UK. However, French women still spend more time on cooking, despite the popular view of a loss of food culture and cooking skills (Warde et al. 2007).¹ Second, the labour-force participation of French women rose continuously over the period. According to official statistics, this reached 65.8% in 2010 for women aged between 15 and 64, as against 56.7% in 1985.²

We focus specifically on partnered women under the legal retirement age for four reasons. First, our data show comparatively little change in home-cooking time for single women (with or without children) or for men. Second, the rise in women’s labour-force participation came almost exclusively from partnered women. Third, labour-market incentives disappear after retirement.³ Last, we want to examine the question of gender balance in home cooking.

We use Oaxaca-Blinder decomposition techniques to identify the contribution of labour-market changes to the change in home-cooking time. We first find that rising women’s employment and observed wages together account for about 64% of the decline in married women’s cooking time (57% when we also control for the increase in education levels and ownership in kitchen equipments). However, observed labour-force statuses and wages are the outcomes of individual choices. They thus mix individual selection on preferences and on labour-market incentives, as well as other environmental factors (from child-care subsidies to the institutional determinants of spouses’ bargaining power). We then use semi-parametric matching techniques to construct implicit wage rates, i.e. the expected wage rates of individuals were they to work full time. This implicit wage represents the labour-market incentives faced by individuals at the moment of deciding their labour supply, and directly determines the opportunity cost of time spent in household production (Becker 1965). We find that 28% of the decline in married women’s cooking time comes from higher implicit wages, and only 18% when we also control for the confounding effects of education and kitchen equipments. This is much lower than the above figure of 64%. The difference (46 percentage points) is explained by changes in preferences, technologies and environmental factors that are positively correlated with labour-market incentives and negatively correlated with cooking. By way of comparison, educational expansion, i.e. the increase in the schooling level of both men and women, sharply reduced home cooking, and represents about half of the observed fall in time spent in cooking.

These results confirm that labour-market changes have a significant impact on women’s home cooking, even in a country like France, where home cooking is a strong cultural value. This is in line with empirical results from different countries and cultures, such as the US and Japan (Davis 2014; Davis and You 2010; Hamermesh 2007; Kohara and Kamiya 2016). Beyond cooking, our semi-parametric matching treatment of the endogeneity of observed wages and the measurement of time costs is also a methodological contribution to the large literature on the effect of partners’ wages on non-market work (see Bloemen and Stancanelli 2014, for a recent study).

We also find that changing labour-market incentives are far from being the main driver of the decline in home cooking. Our regression results reveal the time-saving role of technological innovations in home cooking (kitchen appliances). Male wages have little effect on the distribution of cooking tasks and home cooking remains strongly gendered, even among the more educated. These results confirm previous evidence that men and women are not substitutes for most household chores (Bertrand et al. 2015; Sullivan 2011; Sofer and Thibout 2015; Bloemen and Stancanelli 2014).

Our results have direct consequences for food and nutritional policies. The decline in women’s home cooking has been related to the rise in obesity and diabetes (Anderson et al. 2003; Cawley and Liu 2012; Fertig et al. 2009; Hamermesh 2010; Liu et al. 2009; von Hinke Kessler Scholder 2008).⁴ A popular view among nutritionists and health-policy makers is that public-health programs should promote home cooking (Smith et al. 2013). Our results suggest that such programs may fail, because of the trade-offs faced by women between paid work and home cooking, women’s preferences over these choices, and the lack of substituability between men and women in meal preparation. Given the impact of technologies on cooking trends, encouraging innovations designed to ease the preparation of healthy meals might be a better way of ensuring the healthiness of food-at-home whatever the time inputs that women can and want to devote to home cooking. This point is not specific to France, as Davis and You (2011) reached a similar conclusion after an empirical analysis of the consistency of household decisions of food production and food assistance programs in the U.S.

The remainder of this paper is organized as follows. Section 2 uses insights from the economics of household production and the sociology of food to discuss the effect of labour-market incentives, technology and social norms on home cooking. We derive three testable empirical hypotheses. Section 3 describes the French Time Use Surveys used to test these hypotheses, and outlines the major changes in cooking time, labour-market participation and socio-demographic variables between 1985–1986 and 2009–2010. Section 4 presents the statistical models. The main results are presented in Section 5, and further discussed in Section 6. Last, Section 7 concludes.

2 The economics of home cooking

This section presents the main economic and sociological mechanisms that may lie behind the trends in home cooking. We highlight the key role of changing labour-market incentives and technological progress. We also emphasise the interactions between economic forces and social norms regarding the gender division of household production and what is considered to be a proper meal. We derive three hypotheses that will be tested in the empirical work.

2.1 The economics of home-cooking

The economic approach to cooking revolves around individuals substituting cooking time for food preparation partly or entirely done in restaurants and factories, depending on consumer preferences and on the money and time costs of home cooking relative to the market price of food preparation away. These relative prices depend on food processing technologies and time-saving kitchen appliances (e.g. the invention of deep freezing combined with microwaves) and on the characteristics of non-food markets, especially labour markets.

Consumer responses to changing market prices, time costs and technologies can be formalized in a simple static model that stems directly from household-production theory (Becker 1965; Pollak and Wachter 1975). We consider a simple utility structure that corresponds to a consumer or a unitary household allocating time and money resources between the home production of meals and purchase of food-away for one-period. Utility is defined as:

$$U\left( {q_a,F_h(T_h,q_h;z_1),L,c;z_2} \right)$$

(1)

where q_a is food entirely prepared away from home, F_h is food prepared at home using time T_h and food products q_h, L is leisure commodity, c is the numeraire good, z₁ is cooking technology and z₂ is a vector of preference variables.⁵ We let N be the time spent working, w the hourly wage rate and ν non-labour income. The prices of food-away and food products for cooking are denoted respectively as p_a and p_h.

Cooking, as well as paid work, may produce utility in their own right. Following Kooreman and Kapteyn (1987), this “joint production” can be formalized as L = l + g_T(T_h) + g_N(N), with l being available leisure time. Assuming that g_T(T_h) = (1 − k_T)T_h and g_N(N) = (1 − k_N)N with k_T, k_N < 1,  one hour spent cooking (working) costs the individual only a fraction k_T (k_N) of 1 h of leisure. The household optimization program is then:

$$\begin{array}{l}Max_{q_a,T_h,q_h,l,c,N}\,U\left( {q_a,F_h(T_h,q_h;z_1),L,c;z_2} \right)\cr p_aq_a + p_hq_h + c = wN + v\cr T_h + l + N = T\cr L = l + g_T(T_h) + g_N(N)\end{array}$$

(2)

Assuming away corner solutions, the first-order conditions yield:

$$\partial F_h{\mathrm{/}}\partial T_h = \eta \frac{{k_T}}{{k_N}}\frac{w}{{p_a}}$$

(3)

where η = (∂U/∂x_a) / (∂U/∂F_h) is the marginal rate of substitution (MRS) between food consumed away and meals prepared at home.⁶ Then, at a very general level, optimal cooking time is given by the following equation:

$$T_h = f(w,p_h,p_a,\nu ;z)$$

(4)

where z = (z₁,z₂) is a vector of factors related to preferences and technologies, including k_T and k_N.

When the returns in household production are decreasing, F_h is increasing and concave in its arguments. Eq. (3) then tells us that cooking will decline with a fall in the price p_a of food-away and an increase in the wage rate w. Note also that non-labour income has a pure income effect through the consumer’s full budget constraint; this income effect is likely positive. The opportunity cost of time spent cooking is a fraction η k_T/k_N of wages. As labour-market opportunities for women have changed, wages have risen and so has the opportunity cost of time. All else equal, it became more advantageous to work more and purchase meals prepared away rather than buying food and spending time preparing it. This prediction has been confirmed by a number of empirical contributions: see, for instance, the review of US research in Davis (2014). The impact of increasing wage rates is all the more important that the direct utility of paid work is high (i.e. when k_N decreases), which is the case if, for instance, paid work yields non-monetary benefits (social status, feeling of social inclusion, etc.).

Changes in technology and women’s labour-market participation are related. Technological innovations have increased the productive efficiency of time spent home cooking.⁷ A number of innovations in food processing, preservation and preparation have occurred since the 1960s. Frozen ready-meals require only a few minutes of preparation. Frozen ingredients avoid spending time chopping, dredging and sautéing. Most home-cooked meal include ultra-processed food that is ready-to-eat (e.g. yogurts). Sauces or cakes are easier to prepare thanks to prepared mixes. Washed and sliced fresh vegetables have now become available in all French supermarkets. For the household meal planner, outsourcing cooking operations is a way to save time while benefitting from considerable economies of scale, with lower fixed and variable costs per meal. The mass processing and preparation of food away from home is more capital-intensive, with larger and more efficient equipment and devices. It exploits the division of labour and the specialization of food professionals along the production chain. Technological advances in the food sector (and more generally in the household sector) have contributed to render labour-market participation more attractive for women, the rise in which has stimulated in turn the demand for innovations.⁸

Standard household-production theory then yields two empirical hypotheses that we will test:

Hypothesis 1—Labour-market incentives. Rising wages for women on the labour market have had a negative impact on cooking time.

Hypothesis 2—Household technology. The diffusion of kitchen appliances has had a negative impact on cooking time.

We now consider how within-household social interactions can add to these predictions.

2.2 Social norms and spousal interactions

As noted above, the impact of changing wages will be affected by contemporaneous variations in the marginal rate of substitution between home cooking and food away, and in the enjoyment from cooking. These factors depend notably on social norms and interactions between household members.

Sociologists have examined trends in eating patterns in relation to the possible decline of the “proper family meal” (Charles and Kerr 1988; Murcott 1982). The latter refers to a social norm regarding the time pattern of meals, their structure (food and dishes), the labour input in terms of cooking or table dressing, the ways of cooking, and the participants. Its precise definition and normative strength varies across western cultures. For instance, regularity in time patterns and eating commensality are more important for the French than for Anglo-Americans. In France, the family meal retains both strong normative and descriptive content for everyday living (McIntosh et al. 2009; Grignon and Grignon 2009; Fjellström 2009).⁹ Eating home-cooked food is a key element of the “proper family meal”, with a crucial gender connotation. Family members, including mothers, tend to see cooking for the family as an “act of love” (Moisio et al. 2004). French mothers tend further to endorse a “caring ideology”, whereby concerns over their children’s and partner’s health play a significant role in meal composition (Le Bigot Macaux 2001). Cooking remains a strong moral imperative for women. The impact of a weakening of this norm can be formalized in Section 2.1's model as either a shock to the joint production function g_T(T_h) leading to a fall in the marginal enjoyment from home-cooking (a rise in k_T), or alternatively by a rise in the MRS η, which may tend towards one (perfect substitution).

A more formal treatment of the concept of the “family meal” in a choice model would consider collective household decisions over cooking and eating. The main difficulty in going down this path is that households produce both family and individual meals. Family meals should be treated as public commodities, whereas individual meals are private commodities. Cooking a family meal is likely to produce more enjoyment than cooking just for oneself. Although we do not formally write down the complex model that follows from these observations, it is worth considering the theoretical results from the economic literature on household production in non-unitary models. This literature analyses the allocation of goods and time between household members, either by assuming that each household member produces, supplies and demands commodities on intrahousehold shadow markets (Grossbard-Shechtman 1984), or under the assumption that intrahousehold decisions are Pareto-efficient (Apps and Rees 1997; Browning et al. 2014; Chiappori and Lewbel 2015; Pollak 2005; Rapoport et al. 2011).¹⁰ Whatever the approach, the main prediction is that each spouse’s time spent in cooking will depend on the wage rates of spouses and the market price of food away, but also on preferences and on factors affecting spouses’ relative bargaining power.

Using the framework proposed by Grossbard-Shechtman (1984), cooking can be seen as a form of Work-in-Household (WIH) that is essentially supplied by women, as they have historically been in charge of food preparation and spouses’ cooking times are technical substitutes in meal production. The shadow cost of women’s cooking, and thus its supply price, will depend positively on their own labour market wages and the marginal utility of labour, and negatively on the marginal utility of cooking. Higher shadow costs of production result in a lower men’s equilibrium demand for women’s cooking. Men’s willingness to pay for women’s cooking increases with men’s wage rates, due to an income effect and a substitution effect in home production.¹¹ The equilibrium cooking time of women will eventually depend negatively on their own wage and positively on their partner’s wage. Higher wages for women should also result in a higher consumption of food away from home,¹² and in higher cooking times by men, but only for male partners who do have cooking skills. Absent these skills and the willingness to acquire them, there is no substitutability in household production, so that wives’ wages should not affect husbands’ cooking.

In addition, the “proper family meal” is a not-perfectly marketable good. It has one specific characteristic—being home-cooked—that food prepared away does not have. Hence, the wage effects will be smaller if the “proper family meal” norm is strong and gendered, in the sense that partners prefer not substituting women’s cooking for food-away or for men’s cooking.

Following these insights, we have one more testable empirical hypothesis:

Hypothesis 3—Household bargaining process. For couples, an increase in the husband’s wage has a positive impact on the wife’s cooking time.

Failure to validate Hypothesis 3 would suggest that the impact of changing economic incentives is largely attenuated by social norms regarding gender roles and family meals. We now present the data that will be used to test Hypotheses 1–3.

3 Data

We exploit the 1985–1986 and 2009–2010 French Time Use Surveys (FTUS—Enquêtes Emploi du Temps), which are conducted about every ten years by the French National Statistics Office (INSEE), and are included in the Multinational Time Use Study dataset.¹³ Time-use surveys (TUS) have two key advantages. They provide more accurate accounts of the time devoted to domestic chores than traditional questionnaires; refusal to complete the survey generates very little bias in the estimated durations. Additionally, because TUS collect information on every activity, they avoid the selection or declaration bias that a survey focusing on food or eating might generate. Their main limitation is that there is no information on the foods consumed and their prices; there is, however, information on earned and unearned incomes (Gershuny 2003).

The survey samples are nationally representative of households and individuals not living in institutions, with calibration by labour-force status, age by gender, education, household structure, place of residence, days of week, months and school holidays. Various types of household- and individual-level data are collected. We will in particular use information on the kitchen equipment in the household, i.e. freezer and microwave, labour-market outcomes (labour-force status, earned income, and usual working hours), education and non-labour income. When individuals are in a couple, their partner is also interviewed and also provides time-use data for the same day. The latter consist of self-completed 24-hour paper diaries. The 1985 FTUS data include (i) a base of 10,373 households with complete household-level data and at least one time diary for a randomly drawn household member (the main respondent);¹⁴ (ii) information on 29,723 household members, of which 77.6% are aged over 15. For the 1985 survey, one day was randomly drawn for the main respondent, and the activities were coded into slots of 5 minutes (min) each. When the main respondent had a partner (6582 households), the latter was invited to complete a time diary. The refusal rate was 13.8% only. We have 16,047 completed time diaries, among which 11,348 are same-day diaries completed by 5674 couples. The 2010 data includes: (i) a base of 12,069 households, with missing time diaries for 1394 households; (ii) a base of 18,521 individuals, of which 2279 have missing time diaries and 98.8% are aged over 15. One week day and one week-end day were randomly drawn for one randomly drawn household member individual, with activities being coded into 10-min slots.¹⁵ When this respondent had a partner, the latter was asked to provide time diaries on the same days. The refusal rate was 5.24%. We have 27,903 completed time diaries for 16,242 individuals, with 8966 pairs of same-day diaries from married couples.

3.1 Construction of the estimation sample and definition of the main variables

The empirical analysis focuses on households that fully completed the time diaries, where all adults are aged between 18 and 64, at least one adult is active on the labour market and without self-employed. We keep households with at least one adult woman, and without other cohabiting adults (grand-parents, brothers, sisters etc.). We drop households with more than two adults, as well as same-sex couples, as social norms and decision processes may be structurally different for these. We drop diaries completed on a sickness day. Starting with a dataset of 20,994 households observed either in 1985 or in 2010, this leaves us with an initial sample of 5579 households and 9227 individual-days observed in 1985, and 5345 households and 13,658 individual-days in 2010.

We use a strict definition of time spent cooking, which does not include meal-related chores such as setting and clearing the table, washing dishes, grocery shopping, etc.¹⁶ In some analyses, we will use household cooking time, which is defined as the sum of partners’ cooking times. We ignore time overlaps, because there is little spousal synchronization in cooking: around 2 min in 1985 and 4 min in 2010. The direct substitute for home cooking, food away, will be captured by the frequency of restaurant eating, which includes meals at worksite restaurants and commercial restaurants. We do not distinguish between meals at non-company restaurants with work colleagues and meals with family or friends. To test the robustness of our results, we have also looked more broadly at the frequency of eating away from home, which encompasses all eating-away occasions (e.g. eating at friends).

The labour-market outcomes of interest are labour-force status and wages. We code the former into four categories: inactive or unemployed (including a few students), part-time workers, full-time workers with missing wage information, and full-time workers with observed wages. The latter are calculated as total self-reported annual earned income from all activities divided by self-reported usual weekly working hours. They are adjusted for annual changes in the Consumer Price Index, so that all monetary variables are expressed in 2010 Euros.

A key variable in any economic analysis of time-use decisions is the opportunity cost of time. As outlined in Section 2.1, this is a function of the wage that the individual earns or may expect to earn on the labour market. This wage is not observed for those who do not work, and imperfectly measured for those who work part-time. Section 4 proposes a semi-parametric matching method to construct implicit wages for these individuals. This method further reduces the sample size, as we drop individuals for whom the matching is unreliable.¹⁷ Considering only the female partner, the full estimation sample finally includes 3949 households from the 1985 FTUS (3949 days), and 3566 households from the 2010 FTUS (5727 days, as 2 days are available for about half of the sample).

Importantly, we control for changes in non-labour income, which is an important determinant of choices over work, leisure and household production. The FTUS allow us to measure this.¹⁸ In both years, information is missing for about 15% of the sample. Instead of dropping these observations, and given the likely lack of accuracy of our measure, we construct a categorical variable for non-labour income, with five interval categories ([0,50], [50, 250], [250,500], [500, 1000], over 1000 Euros/month), and a sixth category for missing values.

Last, technological innovations in home cooking will be captured by variables for whether the household has a freezer and/or a microwave, and can therefore prepare meals from food products that have been partly processed away from home (ready-meals and frozen products). In some regressions, we will also use a proxy measure of household time-saving kitchen technology by summing up ownership of freezer, microwave and dishwasher.

3.2 Changes in cooking times and labour-market choices between 1985 and 2010

Table 1 illustrates the decline in time spent cooking. All durations are in minutes per day. The upper panel shows the descriptive statistics for household cooking time for the whole sample, whether partnered or not, while the middle and lower panels consider the trends for women and men respectively. The first line of Table 1 shows that households cooked about 10 min less in 2010 than in 1985 (61.0 vs. 71.7 min). A large part of this fall is due to a rise in the proportion of zeros, i.e. days where households do not cook at all, from 4.1 to 16.7%. The conditional mean cooking time is actually quite stable at around 74 min per day. These trends also apply for married couples, which represent 75.1% of household-day observations in 1985, as against 65.5% in 2010. This shift in sample composition reflects a rise in divorce and later union formation, with more single women and single mothers in 2010. Interestingly, there has been little change in unconditional mean cooking times for these last two categories. Although they have more “no-cooking” days in 2010 than in 1985 (33.2% for single women), they nevertheless spend more time cooking when they do so (50.5 min in 2010 vs. 42.9 min in 1985 for single women).

Table 1 Trends in household cooking times

The trends in individual cooking time reveal the same pattern for married women (see the middle panel of Table 1), with more “no-cooking” days and little variation in conditional mean cooking times. Splitting the sample into working days and week days (Monday–Friday) and Saturdays or Sundays off reveals little effect from the potential constraint of work.¹⁹ This can be interpreted as evidence that the time input into cooking is on average the outcome of long-term decisions, rather than short-term changes in constraints.

As shown in the lower panel of Table 1, partnered men cook only 4 min more per day in 2010 than they did in 1985, with an average cooking time of 18.0 min that remains far below that of their partners. The same trend as for women is observed at the extensive margin, with a 15% point rise in the proportion of “no-cooking” days (59.5%). However, when they do cook, they spend more time doing so (44.4 vs. 26.3 min). This is not just because cooking has become a fancy means of amazing friends at week-end dinners, as they also spend more time in cooking (when they do cook) during weekdays.

Table 2 compares these trends in cooking times with trends in meal times, meal chores and working time. A number of notable figures emerge. First, the French spent more time in 2010 than in 1985 eating. For women in couples, this figure rose from 94.6 min per day in 1985 to 136.5 min in 2010. Interestingly, this is not because having a meal is declared as a secondary activity, undertaken while watching TV for instance. Eating remains a primary activity, unlike in the US (Hamermesh 2010). The French take more time to eat at home (+18 min for women in couples) and away-from-home (+22 min). The time devoted to meal chores has fallen, thanks to the diffusion of dishwashers. Similar trends are observed for single women, single mothers and men in couples. The only difference between these categories regards paid work. Married women have longer working hours in 2010, unlike single women, single mothers or married men. This is due to a large increase in the proportion of days not worked for these last three categories, as all individuals report longer work hours on working days. The stability in the proportion of days not worked for married women reflects two opposing trends: more days off for everyone on the one hand, and greater female participation in the formal labour market on the other.²⁰

Table 2 Trends in individual meal times

This major change in labour-force participation, together with the stability of cooking time for single women and single mothers, explains why we will particularly focus on married women. In addition, marital status is an important determinant of cooking and eating practices, with likely consequences on family health. We also want to examine trends in the sharing of home cooking between spouses, to see whether labour-market incentives are moderated by norms regarding the gender division of household chores.

Table 3 describes the changes in married women’s characteristics in the estimation sample of individual-days between 1985 and 2010.²¹ Labour-force participation increased by 13.3 percentage points (pp), from 61.8 to 75.0%. Part-time employment represents about 40% of this rise (+5.6 pp). Working women’s real wages rose from 7.3 to 9.6 Euros/h (in 2010 Euros), while those of men grew slightly more (+2.6 Euros over the period, vs. +2.3 Euros for women). The distribution of non-labour income has changed (except for the proportion of missing values), with higher unearned incomes in 2010. For instance, 14.3% of the sample reported unearned incomes of over 500 Euros/month in 2010, as against 0.4% in 1985. Although these trends in unearned income are in line with observations from administrative and fiscal data (Piketty 2006; OECD 2007), we remain cautious about the accuracy of our measures here.

Table 3 Descriptive statistics—estimation sample of partnered women

The lower part of Table 3 shows how kitchen equipment and education have changed. As almost all households have a freezer in 2010, we construct a dummy variable for the joint ownership of a freezer and a microwave in 2010. In 1985 less than half of households had a freezer and the questionnaire did not include the microwave which was very rare in France. Our index for household kitchen equipment, which adds up the ownership of a freezer, a microwave and a dishwasher, rose from 0.72 in 1985 to 2.60 in 2010.

The impact of educational expansion is massive, with 33.3% of married women now having a higher-education diploma (as against 29.1% for their partner). Educational expansion is likely to be an important confounder of the impact of changing labour-market incentives, as education is both a determinant of wages and has a direct impact on food and foodways through social norms and nutritional knowledge.

4 Empirical framework

We test hypotheses 1–3 by applying an Oaxaca-Blinder decomposition approach to the changes in unconditional mean cooking times between 1985 and 2010 in the FTUS. This section briefly presents this econometric method. Our empirical strategy largely relies on the comparison of decomposition results from two specifications. In the first, women’s labour-market changes are given by their observed labour-market status and wages (which are set to zero for non-workers). In the second, we replace these choice variables by implicit wages, which pick up the change in the labour-market incentives faced by individuals, whatever their preferences. Section 4.2 proposes a non-technical presentation of the semi-parametric matching technique that we use to construct implicit wages. The technical details appear as online Supplementary Appendix.

4.1 Econometric model

Let T_i,t be the time spent cooking by individual i on a given day in survey year t = 1985, 2010. We model T_i,t as a simple linear function of a set of regressors X_i,t, which include labour-market variables, non-labour income, two variables for time-saving kitchen equipment (a freezer and a microwave) and a large set of sociodemographic controls: age (in seven categories); the age difference between spouses and its square (continuous variables); family structure (four categories: at least one child aged under 4, at least one child aged strictly under 6 and no child aged under 4, all children aged 6 or over, no children); dummies for whether the individual is off on the diary day, and whether the diary day is a Saturday or a Sunday; five residential-area dummies (rural area, small town, middle-sized town, city, and Paris and its suburbs), and administrative region (8 dummies).²²

We estimate the following equation by OLS for each survey year t:

$${T}_{{i},{t}} = {\beta }_{t}{X}_{{i},{t}} + c_{t} + {\it{\epsilon }}_{i,t}$$

(5)

where β_t measures the association between X_i,t and T_i,t in survey year t, c_t is a constant and ϵ_i,t is an error term with mean zero.

The Oaxaca–Blinder method decomposes the changes in T_i,t between two surveys and identifies the contribution of labour-market changes to changes in cooking time, holding all other characteristics constant on average (Blinder 1973; Oaxaca 1973). We construct the decomposition by first taking the unconditional mean of equation (5) in each survey year t = 1985, 2010:

$$\left\{ {\begin{array}{*{20}{c}} {{\Bbb E}\left( {{\mathrm{T}}_{{\mathrm{i}},1985}} \right) = {\mathrm{\beta }}_{1985}{\mathrm{X}}_{{\mathrm{i}},1985} + c_{1985}} \cr {{\Bbb E}\left( {{\mathrm{T}}_{{\mathrm{i}},2010}} \right) = {\mathrm{\beta }}_{2010}{\mathrm{X}}_{{\mathrm{i}},2010} + c_{2010}} \end{array}} \right.$$

(6)

We can then write the change in the unconditional mean over time as:

$$\begin{array}{l}{\Bbb E}\left( {{\mathrm{T}}_{{\mathrm{i}},2010}} \right) - {\Bbb E}\left( {{\mathrm{T}}_{{\mathrm{i}},1985}} \right) = \underbrace {{\mathrm{\beta }}_{2010}\left[ {{\Bbb E}\left( {{\mathrm{X}}_{{\mathrm{i}},2010}} \right)-{\Bbb E}\left( {{\mathrm{X}}_{{\mathrm{i}},1985}} \right)} \right]}_{composition\,effect}\cr + \underbrace {\left[ {{\mathrm{\beta }}_{2010} - {\mathrm{\beta }}_{1985}} \right]{\Bbb E}\left( {{\mathrm{X}}_{{\mathrm{i}},1985}} \right)}_{structure\,effect} + \underbrace {c_{2010} - c_{1985}}_{unexplained\,variation}\end{array}$$

(7)

The first term on the right-hand side of equation (7) refers to the composition effect. This will provide the answer to our main research question: To what extent do labour-market changes explain changes in cooking time? The second term is the structure effect: this measures the contribution of changes in the associations between the dependent variable and the covariates. The last term, i.e. the difference in the constants, is the residual change, which is explained by neither the changes in observed covariates nor the changes in the impact of these covariates. There may be some unobserved factors that influence cooking decisions beyond their indirect effect via the X variables. For instance, women may on average have fewer cooking skills in 2010 than in 1985, which would generate an unobserved composition effect. It may also be the case that cooking skills matter less now for producing a meal, which would yield an unobserved structure effect. The unexplained variation thus captures both unobserved composition and structure effects.

We here focus on composition effects. An important question, then, is whether we should evaluate the composition effect at the coefficients of 2010 or 1985.²³ The β_t coefficients reflect to a large extent the impact of period-specific preferences on choices. For instance, if individuals have a strong taste for home cooking, then we may expect only a small impact of wages on cooking time. Hence, as we want to analyse past changes from the perspective of current individuals, we choose 2010 as the reference year.

Another important question is whether modelling unconditional mean cooking time is appropriate. The statistics in Table 1 show that a large fraction of individuals report no cooking time on the diary day, especially in 2010. The changes in the unconditional mean observed between 1985 and 2010 seem to be largely driven by these zeros. However, we here take a long-run perspective on women’s time use. In this context, we are not interested in day-to-day variations in time use, and the zeros reflect the infrequency of choices rather than long-term equilibrium outcomes of individual time-allocation decisions. These zeros do not affect our estimates of average long-run time use via OLS (Frazis and Stewart 2012; Stewart 2013).²⁴ However, we will test the robustness of our results by decomposing changes in the frequency of cooking, which is defined as the probability of having cooked at least 3 min on the diary day, i.e. $T_{i,t}^ \ast = Prob\left( {{\mathrm{T}}_{{\mathrm{i}},{\mathrm{t}}} \ge 3{\mathrm{|}}X_{i,t}} \right) = F({\mathrm{\beta }}_{\mathrm{t}}{\mathrm{X}}_{{\mathrm{i}},{\mathrm{t}}} + c_{\mathrm{t}})$. We use the extension of the Oaxaca-Blinder method in (Yun 2004), which uses a first-order linear approximation of the function F(.). This approach will also be applied to eating away occasions.

4.2 Measuring labour-market changes

Labour-market changes can obviously be reflected in women’s labour-force participation and observed wages. However, as we take a long-run perspective on our time-use data, labour-market decisions are endogenous. This is in line with the theoretical framework in Section 2. Individuals self-select into the labour market as a function of expected wages, which are unobserved for non-workers. As is usual in household-production analysis, we have to construct implicit wages. One common solution is to estimate a wage equation with correction for self-selection into employment. Implicit wages can then be predicted for non-workers (see for instance Hamermesh 2007).

A key challenge with this approach is that observed wages may depend on the number of hours worked, i.e. the budget constraint of the leisure-work choice problem is not linear. In France, many part-time workers benefit from specific wage regulations. To avoid this problem, we measure implicit wages by the wage that the individual would have earned in a full-time job. We define three labour-market statuses: non worker, part-time worker, and full-time worker. The latter includes all individuals with working time greater than or equal to the OECD threshold of 32 h. Note that this threshold is lower than the reference value of legal weekly working hours (39 h in 1985; 35 h in 2010), as in our dataset there are a considerable number of observations just under the legal reference level, suggesting the presence of individual- or firm-level agreements whereby full-time workers do not work a complete week. We assume that individuals can freely choose hours above these norm levels and/or jobs (e.g. managerial) that necessarily imply longer hours (and higher earnings). Those in part-time jobs or who do not work are considered to have potentially self-selected into these labour-market statuses. We then use a semi-parametric matching method to match each non-worker with a full-time worker, and each part-time worker with a full-time worker, by gender and survey year. We also apply this method to construct wages for the few full-time workers with missing wages. Section S.1 in the Supplementary Appendix provides additional details and statistics on the econometrics, implementation and statistical quality of the matching procedure.

Full-time wages are modelled as a linear function of a set of variables W that are commonly used in wage equations: age and age-squared, seven education categories (Nothing, primary school, incomplete lower secondary school—general, lower secondary school—technical, upper secondary school—general, upper secondary school—technical, two-year university degrees, three or more year university degrees).²⁵ An individual’s implicit wage can then be written as the sum of an expected wage conditional on W and an individual-specific error term reflecting unobserved ability. Propensity-score matching is applied to predict this error term for each non-worker and part-time worker. The propensity scores depend on a set of variables Z that includes the W variables, but many variables in Z are excluded from W. We thus use additional exclusion restrictions to obtain more precise estimates (Heckman et al. 1997).

The exclusion restrictions potentially include the following variables that commonly appear in work on labour-market participation or time use (Bloemen and Stancanelli 2014; Hamermesh 2007; Kimmel and Connelly 2007; Duguet and Simonnet 2007): number of children aged under three (children go to school at three in France) and aged under six, total number of children; whether there is free help for child care; whether the household has to pay for child care; whether the individual’s mother was active or not, four socio-occupational dummies (private-sector managers, private-sector intermediate professions, private-sector workers, and public-sector employees); and interactions between age and the socio-occupational dummies. These interactions capture cohort differences in the relative unemployment risk faced by individuals with different professional skills. We carefully select an optimal subset of exclusion restrictions to avoid “weak instrument” bias and impose a strict common support condition. Finally, we remain cautious about the causal interpretation of our results, as we do not exploit quasi-natural shocks to identify wages. Table 3 shows mean implicit wages in 1985 and 2010, and the change between the two surveys. The implicit wages have converged to observed wages, which means that selection into employment now depends less on unobserved characteristics affecting wage offers.

5 Results

Tables 4 and 5 present the main regression and decomposition results for married women. Three specifications are tested: specification 1 controls for labour-market choices (observed wages and labour-market status) and unearned income; specification 2 replaces labour-market choices by implicit wages; and specification 3 adds the kitchen-equipment and education controls. All specifications include controls for non-labour income being missing, household structure, partner’s age, type of diary day, type of residential area, and region. All regressions use individual-day survey sample weights and, for 2010, cluster standard errors at the individual level (as there are two days per individual).²⁶

Table 4 Cooking time of partnered women—regression results by year

Table 5 Decomposition of the change in mean cooking times

Table 4 presents the main regression results by survey year. In columns 1–6, the dependent variable is unconditional cooking time, and three specifications are estimated for each year. Columns 7 and 8 show the results of logit regressions for the frequency of cooking (defined as cooking ≥ 3 min on the diary day) for specification 3 only. Table 5 presents the decomposition results for unconditional mean cooking time (left panel) and the frequency of cooking (middle panel), for all specifications. The right panel of Table 5 shows additional decomposition results for restaurant eating (time spent eating at a restaurant ≥ 3 min on the diary day).²⁷

5.1 Labour-market choices vs. labour-market incentives

Columns 1 and 2 of Table 4 show the estimation results for specification 1. The coefficients on women’s wages and working status are both negative and significant at the 1% level. The impact of labour-force participation rose from −22.5 min in 1985 to −40.2 min in 2010. In 1985, this effect was attenuated for women in part-time jobs, who cooked 13.4 min longer than women in full-time jobs. This is not the case in 2010. The estimated coefficient on wages is significant and similar over time: −1.5 and −1.2 min per additional Euro of hourly earnings in 1985 and 2010 respectively. Columns 3 and 4 of Table 4 report the results for specification 2. The coefficient on women’s implicit wages is negative and significant in 1985 and 2010 and of similar size, with point estimates of −2.8 min/Euro in 1985 and −2.6 min/Euro in 2010. The regression results for specification 2 are in line with the main predictions of household economics regarding time allocation. This is in line with Hypothesis 1.

The increase in wages shown in Table 3, combined with the negative marginal effects in Table 4, produces the negative composition effects in Table 5. The upper-left panel of Table 5 reports the values of unconditional mean cooking time in 1985 and 2010 and the difference between them. The middle-left panel displays the composition effects for the covariates of interest.²⁸

The first column of Table 5 shows that the contribution of changing labour-market outcomes to the reduction in married women’s time spent cooking is −8.4 min (−3.175–5.328 + 0.074). This is the sum of the composition effects from wages, working and working part-time. Labour-market changes then represent 63.9% (CI 95: [34.1%; 93.7%]) of the overall decline observed over the period (−13.2 min). The second column of Table 5 applies the decomposition to specification 2. The contribution from women’s rising implicit wages is −3.7 min, i.e. 28.2% of the overall decline (CI 95: [13.5%; 42.8%]). It is also worth noting that the increase in unearned income yields a positive contribution (+3.6 min), which was not the case in specification 1.

5.2 Impact of technology and partner’s labour-market variables

The regression results for specification 3 appear in columns 5 and 6 of Table 4. The impact of kitchen equipment (freezer) is zero in 1985, but strongly negative in 2010. Having a freezer and a microwave is associated with almost 10 min less per day home cooking. However, this effect is only imprecisely estimated, as they are very few unequipped households. In column 3 of Table 5, the diffusion of microwave and freezers is associated with 7.0 min less mean cooking time, which is however insignificant. Our second empirical hypothesis seems not to hold, but this is likely due to a lack of cross-sectional variation in our equipment variables in 2010. We have tested this explanation by replacing the two equipment dummies by our index for household kitchen technology, which is treated as a continuous variable and has more cross-sectional variation. Column 4 of the Supplementary Appendix Table S6 reports the results. The decomposition effect of technology is now large (−10 min) and significant at the level of 5%.

Specifications 1 and 2 produce negative coefficients on partner’s wages and working status (Table 4). As shown in the second column of Table 5, a rise in partner’s implicit wages reduces women’s time spent cooking. These results imply the rejection of our third empirical hypothesis, as within-household bargaining would produce positive effect from partner’s wage.

5.3 Education effects

As shown by columns 3–6 of Table 4, controlling for education clearly attenuates the coefficients on wages, the marginal effect of which is now −1.4 min/Euro in 1985 and −1.6 min/Euro 2010. These coefficients again do not change significantly over time. The fall in the marginal effect of wages is explained by the joint impact of education on wages and cooking time.²⁹ As a result, the decomposition results in the fourth column of Table 5 show a fall in the contribution of women’s rising implicit wages. This latter is now estimated to be −2.3 min over the period, representing 17.7% of the overall decline in time spent cooking (CI 95:[4.5%; 30.9%]). The increase in female education has had a larger effect, at about −3.7 min.

When we control for education, the marginal effect of partner’s wages becomes slightly positive in 2010 (see column 6 of Table 4), albeit not significant. Our third empirical hypothesis about spousal bargaining is then definitely rejected. The negative estimated coefficient on partner’s wages in specifications 1 and 2 likely reflected the income effect of partner’s wages, as well as the omission of education. The rise in partner education has also reduced women’s cooking time (−3.1 min).

5.4 Frequency of cooking and restaurant eating

The middle panel of Table 5 replicates the results for cooking frequency. The results are globally similar to those for unconditional mean cooking times, although with some differences. The composition effects estimated from specification 1 show that observed labour-market choices have large, negative, but not significant effects on the fall in cooking frequency. The sum of these composition effects is −6.2 pp, for a total reduction of −14.7 pp. The results reported in columns 5 and 6 show that the estimated contribution of women’s implicit wages fluctuates between −1.5 pp and −1.9 pp. This contribution increases slightly with controls for technological innovation and education. The rise in women’s education has not significantly affected the probability of cooking, while the rise in partner education has had a large negative effect. In additional regressions, we have decomposed conditional mean cooking time, i.e. the time spent cooking for those women who declare at least 3 min cooking. The estimates of specification 3 show that the composition effects of implicit wages on conditional mean cooking time is small and not significant.³⁰ Changing labour-market incentives then affected unconditional mean cooking time more through lower cooking frequency, i.e. via the day-to-day possibility of substituting with food-away or batch-cooking, than through large reductions in conditional mean cooking time.

As implicit wages rise and technological innovations lower the price of food-away, households may substitute food-away for cooking time. We test this prediction by decomposing the increase in the frequency of eating at restaurants.³¹ The right panel of Table 5 shows the decomposition results for specifications 2 and 3. The frequency of restaurant eating is 8.7 pp higher in 2010 than in 1985. As expected, implicit wages play a role with an estimated contribution of +1.8 pp over the period in specification 3.³² Educational expansion still appears as a key driver, with a contribution of +3.7 pp to the change in restaurant frequency, rising to +5.6 pp when we add partner education.

6 Discussion

There are three conclusions from our estimation results. First, women’s implicit wages and non-labour income have respectively negative and positive effects on their time spent home cooking. This is in line with the predictions of household-production theory, so that Hypothesis 1 is confirmed. Second, technology has reduced cooking time, supporting Hypothesis 2, although the estimates are not precise. This underlines the importance of labour-saving technological innovations for home production. Third, men’s implicit wages do not affect women’s cooking. Intra-household bargaining over home cooking is then only weakly related to spousal wages, and Hypothesis 3 is rejected. We have done a similar analysis for partnered males’ cooking and eating away, which shows that women’s implicit wages are positively correlated with their partners’ cooking (+0.679 min/Euro), although this correlation is not statistically significant. Women’s implicit wages only significantly affect men’s cooking frequency.³³ Our results thus confirm that cooking does not seem to be an activity that can easily be transferred from one spouse to another. This confirms previous qualitative findings in sociology on female cooking as an input to the “proper family meal”.

The estimated wage effects are qualitatively similar to findings in Bittmann (2015) and Sofer and Thibout (2015). They find negative correlations between women’s observed wages and their domestic work, and no significant effect of their partner’s domestic work. We add to these studies with evidence that the impact of labour market participation and wages on the decline of cooking time is largely explained by women’s self-selection into employment based on observable and unobservable preference shifters. Once we account for self-selection, the rise in wage offers explains 28% of the decline in cooking time, i.e. less than half of the total effect of changing labour market participation. When we additionally control for education and kitchen equipments, the composition effect of implicit wage explains no more than 18% of the decline, vs. 57% in a specification with the observed labour-market variables and the same set of control variables (not reported here). The observed increase in implicit wages may a priori result from both supply and demand shocks on labour. With perfect labour markets, an increase in women’s employment driven by a shift in preferences may have a depressive effect on wages. This would imply that the estimated composition effect of changing labour market incentives is lower than what would have been expected in the absence of preference changes. However, we do not think that we are under-estimating the composition effect, because the French labor market is characterized by important rigidities that largely limit downward adjustments of wages (high minimum wage, high unemployment benefits, strict employment protection, powerful labour unions).³⁴ In addition, wage increases have largely been driven by innovations and technological changes, which have played an important role in the demand for skills. This has in turn stimulated women’s investment in education and human capital, with spillovers in terms of improved productivity of the labour force and technological development.³⁵ Hence, a non-negligible share of the estimated composition effect of implicit wages might also well have been caused by women’s increasing preference for work, implying that we would still overestimate the direct composition effect of changing labour-market incentives.

Our results also suggest that educational expansion has been an important direct determinant of changes in cooking time, beyond its indirect effects through wages. A last question, then, is whether our conclusions still hold across education or social groups. We investigate this question by replicating the analysis separately for women under the “Baccalaureat” (A-level), and women with the ‘Baccalaureat’ or over. Since there are few less educated women in 2010 (N = 800), we cannot robustly evaluate the composition effects at the coefficients of 2010. We thus compute these effects using the coefficients obtained from pooled regressions on the 2010 and 1985 education subsamples.

Table 6 presents the decomposition results for unconditional cooking time and restaurant eating (specification 3 only). The change in cooking time is about the same for the two education groups (−9.3 and −7.2 min).³⁶ The more educated cook less than the less educated, both in 1985 and 2010. The more educated eat at restaurants more often, and their frequency of restaurant eating increased by 8.3 pp, as against +4.1 pp for the less educated.³⁷

Table 6 Decomposition of changes in cooking times by education

For cooking, the composition effect of implicit wages is significant only for the more educated (−1.1 min for cooking). As the average rise in implicit wages between the two groups is similar (about +1 Euro over the period), the difference between education subgroups is explained by the estimated coefficients. The estimated marginal effect of wage is −0.96 min/Euro for the less educated (significant at the 10% level), as against −1.42 min/Euro for the more educated (significant at the 5% level). For restaurant eating, the composition effect of wages is significant and does not differ by education level. Kitchen technology is associated with large significant negative composition effects on cooking, whatever the education level, but it has no impact on restaurant eating. The composition effect of non-labour income is always significant and of the expected sign.³⁸ There is a residual composition effect of education for the less-educated, associated to the decline in the proportion of women with very low education.

The picture that emerges from Table 6 is that better-educated women are slightly more prone to trade cooking time for market work when faced with changing labour-market incentives. We may then ask whether the differences in responses to incentives by education are related to a weakening of the gender-related norm of the “proper family meal” in more educated couples. For instance, the latter might be more likely to move away from having women specializing in cooking in order to increase their joint consumption of eating-away occasions, and more generally leisure. Table 6 shows that the more educated have more restaurant occasions, and that the gap with the less educated increased between 1985 and 2010. A closer analysis of spouses’ synchronized times reveal however that going to the restaurant with the partner seems exceptional: this represents only about 10% of restaurant occasions in 2010, whatever the education level. This is in line with results from Barnet-Verzat et al. (2011), who find that education has no significant impact on leisure synchronization in dual-earners French couples. Most restaurant occasions are indeed lunch meals during work days. Since 1967, employers have the legal requirement to propose to their employees either a subsidized access to a worksite restaurant, or vouchers to eat at commercial restaurants or to buy takeaways. Increasing labour market participation (the extensive margin of employment) thus explains part of the rise in restaurant eating.

We can also explore the question of changes in the proper family-meal norm via the family meals and via the share of total household cooking time provided by women. The average number of family meals has decreased only slightly between 1985 and 2010, from 1.92 to 1.85 meals/day. There are no significant differences in trends by week days/week-ends, or by education level. We then create a first dummy for the woman’s share being over 95% (no sharing) and a second for the share being over 50% (signaling whether sharing is more or less favorable to the woman). Table 7 shows that, in the full sample, the proportion of couples who do not share cooking rose by 7.0 pp between 1985 and 2010, from 37.8 to 44.8%. On the other hand, the proportion of households where women cook more than their partner fell by 11.7 pp, from 87.1 to 75.4%. Is this polarization of the gender balance related to education? The statistics in the middle and right panels of Table 7 show that this is not the case. There is rather a convergence between education groups, with 47.4% of the less educated women fully in charge of home-cooking in 2010, vs. 42.6% of the more educated women. The difference was larger in 1985 (41.5% for the less educated, 26.3% for the more educated). While men’s cooking time represented 32% of their partner’s cooking time in 2010, as against 20% in 1985, this move to more balanced task-sharing hides a polarization between more equal households and households where women remain in charge of almost all the cooking. The rise in women’s implicit wages has improved the sharing of cooking within households, as predicted by models of household decision-making, but not in all households or not on a daily basis.

Table 7 Decomposition of the changes in the sharing of cooking time between partners within households

In line with our discussion in Section 2.2, we can thus conclude that social norms (the “proper family meal”) mitigate the explanatory power of standard economic theories of the household, even amongst the more educated. Another potential and complementary explanation is that many men feel, rightly or wrongly, that they do not have the required skills to cook.

7 Conclusion

Our work here has revealed that, in France, the estimated contribution of changing labour-market incentives to the decline in home cooking between 1985 and 2010 is much smaller than that of rising female labour-market participation. This has two implications. First, as implicit wages should continue to rise with productivity, people will certainly cook less and less. They will increasingly rely on processed food, kitchen technologies and food-away, even in countries with a strong culinary culture such as France. Second, although labour-market choices have a considerable impact on time use, these choices largely reflects individual preferences over household production vs. market work, rather than just changing relative prices. The impact of policies promoting home cooking may thus depend on whether they can provide advices to help households to conciliate labour market participation with meal preparation at home, for instance by increasing the productivity of cooking time through the use of appropriate and health-preserving technologies. In a public health perspective, the improvement of the quality of food prepared away also appears as a complementary objective.

Our results also confirm the lack of substitution between men and women in home cooking. Our analysis of the changes in the sharing of cooking reveals rising polarization. The proportion of ‘absolutely unequal’ households as well as that of “more equal” households has grown. However, more equality does not imply that the rise in men’s cooking frequency has offset the decline in women’s cooking time. While the pervasiveness of the gendered norm of the “proper family meal” partially explains this result, it is also likely that men lack the required cooking skills to prepare everyday meals.

We last note that a large part of the fall in home cooking remains unexplained by the composition effects we observed, and specifically by monetary time costs. Future research may aim to quantify the contribution of other determinants of home cooking, such as cooking skills and the changes that have occurred in leisure markets.