This article was downloaded by: [Universite De Paris 1]On: 24 August 2013, At: 09:37Publisher: RoutledgeInforma Ltd Registered in England and Wales Registered Number: 1072954Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH,UK

Feminist EconomicsPublication details, including instructions for authorsand subscription information:http://www.tandfonline.com/loi/rfec20

Estimating the Impact of the2008–09 Economic Crisis onWork Time in TurkeySeçil A. Kaya Bahçe a & Emel Memiş ba Ankara University – Department of Economics Facultyof Political Sciences , Cemal Gürsel Caddesi Cebeci,Ankara , 06590 , Turkey E-mail:b Ankara University – Department of Economics Facultyof Political Sciences , Cemal Gürsel Caddesi Cebeci,Ankara , 06590 , TurkeyPublished online: 29 Apr 2013.

To cite this article: Seil A. Kaya Bahe & Emel Memi (2013) Estimating the Impact of the2008–09 Economic Crisis on Work Time in Turkey, Feminist Economics, 19:3, 181-207,DOI: 10.1080/13545701.2013.786182

To link to this article: http://dx.doi.org/10.1080/13545701.2013.786182

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all theinformation (the “Content”) contained in the publications on our platform.However, Taylor & Francis, our agents, and our licensors make norepresentations or warranties whatsoever as to the accuracy, completeness, orsuitability for any purpose of the Content. Any opinions and views expressedin this publication are the opinions and views of the authors, and are not theviews of or endorsed by Taylor & Francis. The accuracy of the Content shouldnot be relied upon and should be independently verified with primary sourcesof information. Taylor and Francis shall not be liable for any losses, actions,claims, proceedings, demands, costs, expenses, damages, and other liabilitieswhatsoever or howsoever caused arising directly or indirectly in connectionwith, in relation to or arising out of the use of the Content.

This article may be used for research, teaching, and private study purposes.Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expresslyforbidden. Terms & Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 09:

37 2

4 A

ugus

t 201

3

Feminist Economics, 2013Vol. 19, No. 3, 181–207, http://dx.doi.org/10.1080/13545701.2013.786182

ESTIMATING THE IMPACT OF T HE 2008–09EC ONOMIC CRISIS ON W OR K T IME IN

T URKEY

Seçil A. Kaya Bahçe and Emel Memis

ABSTRACT

Using the first nationwide Turkish Time-Use Survey of 2006, this contributionprovides estimates of the impact of the 2008–09 economic crisis on paid andunpaid work time in Turkey. Linking spouse’s unemployment risk with time-usepatterns of women and men, the authors find that a 1 percentage point risein spouse’s unemployment risk increases women’s total work time by 5 percent(22 minutes per day), while the rise is 1 percent (2.7 minutes per day) for men.The rise in unpaid work time for women is approximately four times more thanthat for men. These differences between women and men are much sharperin urban areas than in rural ones. Results support the argument that economiccrises reinforce the preexisting gender gap in work time. The method developedhere can be applied to other developing country cases, where there is a lack oflongitudinal data availability.

KEYWORDS

Economic crisis, gender inequality, time use, unemployment,unpaid work, Turkey

JEL Codes: B54, J22, J16

INTRODUCTION

The most salient effects of the 2008–09 economic crisis observed across theglobe, regardless of the level of economic development, have been higherunemployment and vulnerable employment. Rising joblessness, led by adramatic decline in aggregate demand with falling exports, particularly inthe developing world, has become a central issue. Much has been writtenabout the impact of the crisis on employment status (extensive margin)in labor markets; however, its consequences in terms of work time (theintensive margin), both market and nonmarket, have been neglected so far.

This version has been corrected. Please see Erratum (http://dx.doi.org/10.1080/13545701.2013.826893).

© 2013 IAFFE

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 09:

37 2

4 A

ugus

t 201

3

ARTICLES

The purpose of this contribution is to address the so-called hidden impactof this recent crisis on work time, focusing on the case of Turkey.1

In times of economic crisis, whether it is the individual’s job that is lostor the jobs of other household members, efforts are made to compensatefor the loss in household income through adjustments in work hours atboth margins. Depending on their reactions to loss of employment, thepattern of these adjustments shows distinct characteristics across householdmembers (Reuben Gronau 2006). Women are often the ones undertakingthe extra work burden due to the crises, compensating for the loss inhousehold income by working longer hours under informal conditionsin the market and/or doing more unpaid work, since a fall in householdincome necessitates home production of some goods and services previouslypurchased in the market: a common result observed following economiccrises since the Great Depression (for the United States, see Ruth Milkman[1976]).2

Although unpaid work is conducted outside the market, it is notnecessarily isolated from the impact of an economic crisis. On the contrary,the unpaid economy is more vulnerable and unprotected during crises thanthe paid economy; yet assessments of economic crises often neglect theimpact within the unpaid domain. In fact, economic shocks tend to increasethe dependence of the market economy on unpaid work.3 Considering bothunpaid and paid work time, we seek to answer whether and to what extentindividuals’ work time in Turkey has changed due to the 2008–09 crisis.Moreover, we investigate how these results vary between women and men,and who takes up the slack in the household when the crisis takes place:women, men, or both?

For our analysis, we use data from the Turkish Time-Use Survey of 2006,which provide information on time individuals spend in different activities,such as work (paid and unpaid), leisure, sleep, and personal care. Despitethe growing recognition of the importance of time-use data in economicanalysis, these data are not still on the lists of regularly collected data ofnational statistical institutes in many developing countries. Such is the casewith Turkey. The Turkish Statistical Institute compiled the first time-usesurvey data in 2006, and it is still the only database available with nationwidecoverage (TurkStat 2006b).4 This lack of data led us to propose a newand tractable method of estimation, which can also be applied to othercountries where infrequent collection of time-use data does not allow directcomparisons of pre- versus post-crisis time-use patterns.

Our method is based on a two-step estimation of paid and unpaidwork time. We first estimate the unemployment risk of women and menliving in couple households. Here, we define unemployment risk as theprobability of being unemployed. Second, we estimate influence of spouse’sunemployment risk, which is predicted in the first step, on paid andunpaid work time. The results show that a one percentage point increase

182

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 09:

37 2

4 A

ugus

t 201

3

WORK TIME IN TURKEY

in spouse’s unemployment risk substantially increases women’s work time,both paid and unpaid. Men’s paid work time, however, shows no majorchange, while their unpaid work time rises in response to their spouse’sunemployment risk.

Based on the estimation results, we conduct a simulation exerciseassuming the unemployment risk for married women and married menrises at a rate equal to the increase in relevant actual unemployment ratesfor both over the crisis period. Our calculations suggest that a 1 percentagepoint increase in spouse’s unemployment risk increases women’s total worktime by 5 percent, while the corresponding rise for men is only 0.7 percent.This increase widens the existing work-time gap between women and menby 25 percent (18 minutes per day). Differences between women and menare more pronounced in urban areas, where the gender gap in total worktime increases by 50 percent (27 minutes per day). When these effectsare compared in absolute terms, nationwide averages show that in Turkeywomen’s total work time rises approximately eight times more than thatof men.

The contribution of this research is twofold. First, from an empiricalperspective, we provide a quantitative evaluation of the possible impact ofthe recent economic crisis on paid and unpaid work time in Turkey and thushelp complete the picture of the costs of the economic crisis by focusing ona usually neglected dimension: the hidden costs observed in the unpaidsphere of the economy. Second, in terms of methodology, we introducea straightforward method applicable when the time-series dimension ofavailable time-use data is incomplete. The proposed method could be usedto decompose adjustments in employment into an intensive and extensivecomponent, the differences between which have been highlighted recentlyin analyses of the responses of labor supply to economic and policy changes.

THEORETICAL ISSUES AND EMPIRICAL EVIDENCE

Earlier literature on the time-use patterns of women and men with respect tounpaid and paid work time assumed that intrahousehold time allocation isdetermined exogenously (Gary S. Becker 1991). By revealing the limitationsof modeling household decisions based on this assumption, researchershave now established that there are several intrahousehold interactionsendogenously determined along with the allocation of time. This studybuilds its methodology for estimating the impact of the crisis on time-usepatterns on the basis of the endogenous relationship between employmentstatus and time allocation. Behind our analysis is the hypothesis that, likeany economic change, economic crises affect the intrahousehold allocationof paid and unpaid work time.

On the one hand, as studies have pointed out, being unemployed in themarket releases more time that can be devoted to unpaid work activities

183

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 09:

37 2

4 A

ugus

t 201

3

ARTICLES

(Namkee Ahn, Juan F. Jimeno, and Arantza Ugidos 2003).5 Thus, one mightexpect that an unemployed spouse, by sharing the unpaid workload, maycause a fall in their spouse’s unpaid work burden. On the other hand, if notshared, extra unpaid workload due to loss in household income increasesspouses’ work burden. Between these two effects, the one that dominatesduring the crisis depends particularly on the characteristics of the existingdivision of labor within the household. There is ample global evidence thatwhile men’s work is highly associated with paid market work, women devotea higher proportion of their work time to doing unpaid work (Jonathan I.Gershuny 2000; Lourdes Benería 2003).

In theoretical literature, there is no agreement on which model betterpredicts this traditional time allocation. While some studies build uponBecker’s (1991) unitary household model and consider gender-basedspecialization in paid and unpaid work as an efficient outcome of rationalchoice (for example, Reuben Gronau [1973]), others introduce game-theoretic, cooperative, and non-cooperative household models showing thathouseholds can be both cooperative units and institutions where conflictualrelations are involved (Marilyn Manser and Murray Brown 1980; ShellyLundberg and Robert A. Pollak 1993).6

Several feminist scholars have critiqued both unitary household modelsas well as game-theoretic models, arguing that household allocation of timeis much more complicated than these models describe, as decision-makingrules may coexist (Janet A. Seiz 1999). Other critiques maintain that thesemodels primarily pay attention to the outcomes but ignore the processes bywhich the bargaining power of each member is endogenously determinedwithin the household. Household allocations are not exclusively determinedby outside options, but also by ethical principles, social biases in perception,patriarchal relations, gender biases, policies, and state actors. Feministscholars emphasize the linkages between the two domains, arguing thatindividual decision-making processes cannot be isolated from the complexsocial context including social norms, regulations, laws, and policies (see,for example, Heidi Hartmann [1979]; Michael Bittman, Paula England,Liana Sayer Nancy Folbre, and George Matheson [2003]). At the same time,institutions indirectly influence the normative context in which the decisionis embedded.

The sparse empirical evidence on the impact of economic crises indeveloping countries highlights major disparities in the allocation of extrawork burden for women. In the Indonesian financial crisis of 1997, the ratioof women working for pay in urban areas increased by 2.8 percent; however,the rise in the ratio of women doing both market and unpaid work wasconsiderably larger (10.1 percent), indicating a substantial increase in theratio of women doing unpaid work. The ratio of men doing unpaid workincreased as well, although the impact was quite modest compared to thatof women, whereas the ratio of men working for pay declined (Elizabeth

184

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 09:

37 2

4 A

ugus

t 201

3

WORK TIME IN TURKEY

Frankenberg, Duncan Thomas, and Kathleen Beegle 1999). Similarly, inthe Philippines, following the 2008–09 food, fuel, and financial crisis, thelikelihood of employment declined for both men and women while increasesin employment as unpaid family workers were biased toward women (Yanavan der Meulen Rodgers and Nidhiya Menon 2012).

Earlier studies also show that women’s labor market participationincreases during times of crises in order to compensate for the loss ofhousehold income, as observed in the 2001 financial crisis in Turkey (A.Burça Kızılırmak 2008). With increased labor market participation, onemight then expect a decline in women’s unpaid work time. Empiricalevidence, however, shows that increases in women’s labor force participation(LFP) are not necessarily substituted by unpaid work time. Women’sparticipation in the labor market cumulates demands on themselves ratherthan increasing men’s unpaid work time.

DATA

This study uses data from the first and only national Turkish Time-UseSurvey, which provides data for 10,893 individuals who are 15 years old orolder living in 4,345 households (TurkStat 2006b).7 Through interviews anddaily dairies, respondents provide information about their time allocationfor two specified days (one a weekday, the other a weekend day), recordingtheir daily activities in 10-minute intervals over a 24-hour period.8 If therespondent performs more than one activity simultaneously, he or shedetermines one of these activities as the main activity, and the data showthe distribution of the time spent in the main activity during 24 hours.Daily activities are classified according to the EuroStat (2000) activity codinglist. For our purposes, we focus on women and men in a spousal/partnerrelationship and limit our sample to working-age individuals – that is, age15 or older but younger than age 65. Once we also exclude the individualswith missing values in the variables of interest, we are left with an overallsample of 2,491 married couples living in nuclear families for which usabledata are available (see Table 1).9

Sample statistics show that the average age is 40 years for women and43 years for men, slightly higher than that of the entire dataset.10 Given thatthe final sample consists of individuals in a spousal relationship, householdcharacteristics do not vary between women and men: 68 percent of thehouseholds live in urban areas. The average number of children youngerthan 16 years is 1.3, and the average household size is 3.9 people. Thirty-twopercent of the couples in our sample have no children.

Statistics on individual characteristics point to major differences betweenwomen and men in Turkey. The ratio of women with higher levels ofeducation substantially lags behind that of men (Table 1). There are alsostriking gender-based disparities in the labor market: 20 percent of women

185

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 09:

37 2

4 A

ugus

t 201

3

ARTICLES

Table 1 Descriptive statistics

Data Sample

Women Men Women Men

Number of observations 5,193 4,642 2,491 2,491Age group 2.64 2.77 2.91 3.22Married (%) 44 44 100 100HouseholdUrban (%) 62 63 68 68Ratio of couples with no children (%) 37 37 32 32Number of children under 16 years old 1.37 1.31 1.30 1.30Number of people in household 4.45 4.45 3.86 3.86Household income (%)

Less than 300 TL 10 9 8 8301–450 14 14 16 16451–600 18 17 18 18601–750 11 11 12 12751–1,000 18 19 19 191,001–250 8 8 7 71,251–750 9 10 9 91,751–2,500 7 7 7 72,501–4,000 3 3 3 3More than 4,000 TL 1 1 1 1Educational status (%)

No primary school 24 7 19 6Primary school 39 39 53 49Secondary school 15 22 8 14High school 16 22 14 20University or above 6 10 6 12Labor market status (%)Employed 25 72 20 82Unemployed 2 5 1 3Homemaker 61 0 75 0Student 6 8 0 0Retired 3 11 3 13Economically inactive (elderly/unable to work) 2 0 1 1Other 2 3 0 1Sector (%)

Agriculture 47 20 45 15Manufacturing 16 31 13 31Services 37 49 42 54Types of employment (%)

Employed (regular) 41 49 40 54Employed (irregular) 7 12 9 11Unpaid family worker 39 6 38 0Employee 0 7 1 9Self employed 11 26 12 26

Source : Authors’ calculations based on TurkStat (2006b).

186

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 09:

37 2

4 A

ugus

t 201

3

WORK TIME IN TURKEY

Table 2 Household composition according tothe number of earners

Frequency %

No earner 466 18.71One earner (woman) 44 1.77One earner (man) 1,509 60.58Dual earner 472 18.95Total 2,491 100

Source : Authors’ calculations based on data for couplehouseholds (TurkStat 2006b).

are employed in the market, while, for men, the ratio is 82 percent.Unemployment rates are 3 and 1 percent for men and women, respectively.11

These figures, when compared to the averages of the entire dataset, suggestthat labor market participation is much lower among the women in oursample: 75 percent of the women designate themselves as homemakers,compared to 61 percent for the whole dataset.

The LFP rate for our sample is consistent with the official figure reportedby TurkStat (2006a): only 24.9 percent of women participate in the Turkishlabor market.12 The sectoral distribution of employed women and men,as well as the types of employment, show a high degree of gender-basedsegregation in Turkey. The ratio of women employed in agriculture remainsas high as 45 percent, and 60 percent of employed women are in vulnerableforms of employment, including irregular employed, self-employed, andunpaid family workers (Table 1). While 19 percent of couples in our samplehave no income earners (that is, neither spouse is employed in the labormarket), 62 percent of them have only one earner: in 60 percent the husbandis the earner, the remaining 2 percent correspond to women-earner couples.The rest (19 percent) are dual-earner couples (Table 2).

Table 3 presents mean duration of time (hours per day) devoted to unpaidand paid work activities by women and men with respect to their labormarket status and the type of employment.13 We group the daily activitiesbased on the following categories: (1) paid work consists of all employmentand employment-related activities;14 (2) unpaid work includes householdmaintenance (food preparation, dishwashing, cleaning, laundry, ironing,gardening, repairing, and shopping) and caring for other householdmembers (childcare, caring for a dependent adult household member, andso on). Total work time is the sum of paid and unpaid work time.

Employed women spend approximately 9 hours per day in paid andunpaid work (Table 3). They have a much larger workload compared totheir male counterparts, who devote 7 hours per day to total work. Ofwomen’s total work time, 55 percent is allocated to unpaid work while

187

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 09:

37 2

4 A

ugus

t 201

3

ARTICLES

Table 3 Paid, unpaid, and total work (hours/day), weighted averages for the sample

Women Men

Unpaid Paid Total Unpaid Paid Totalwork work work work work work

Labor market statusEmployed 4.74 4.06 8.80 0.82 6.19 7.01Unemployed 6.46 0.03a 6.49 1.54 0.67 2.22Retired 6.07 0.00 6.07 1.40 0.47 1.87Economically inactive

(elderly/unable to work)4.90 0.00 4.90 1.11 0.00 1.11

Homemaker 6.94 0.09 7.03 – – –Employment statusEmployed (regular) 4.48 4.13 8.62 0.89 6.39 7.28Employed (irregular) 4.95 3.45 8.40 0.72 5.51 6.23Employee 3.57 5.00 8.57 0.75 6.35 7.10Self employed 5.36 3.23 8.58 0.76 5.87 6.63Unpaid family worker 5.08 3.85 8.93 0.75 5.07 5.82

Notes: a Positive paid work time for being unemployed is due to the survey’s categorization of timespent looking for jobs as paid work time. Mean duration of paid work hours is slightly higher thanzero for homemakers. This might reflect the fact that women living in rural Turkey tend to describethemselves as housewives even when they participate in paid work activities; for such women, beinghousewives confers a higher social status and is therefore seen as desirable and prestigious (FerhundeÖzbay 1990).Source : Authors’ calculations based on data for women and men living in couple households (TurkStat2006b).

only 12 percent of men’s total work time corresponds to unpaid work. Thedifference between unemployed and employed women with respect to themean duration of unpaid work time is much lower when compared to thedifference in their paid work time. This finding shows that there is no one-to-one substitution between paid and unpaid work time, confirming thatemployed women put in a double shift. It is also interesting to observe thatemployed women’s total work time is more than 8 hours per day, regardlessof their employment status. Again, this result signifies an uneven distributionof unpaid work between spouses. Thus, married women in Turkey who havean opportunity to work in the market seem to “choose” between workingvery long hours, including unpaid work hours, or not participating in thelabor market.

METHODS

In previous empirical literature, scholars have taken employment statusas exogenous when they investigate the relationship between spouses’employment status and their time allocation. Estimates were obtained by

188

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 09:

37 2

4 A

ugus

t 201

3

WORK TIME IN TURKEY

controlling several other socioeconomic factors, for example, individual,household, and life-course characteristics as the major determinants ofallocation of time that might potentially influence both time-use patternsand employment status (F. Thomas Juster and Frank P. Stafford 1991). Weformalize the generic reduced form equation as follows:

yji = αj Pi + β ′j xi + εji (1)

where represents time allocated to activity j by individual i, which is censoredat zero: as commonly observed in time-use data, a large number of therespondents appear to spend zero time in work activities. The notation Pi

represents a binary variable that indicates whether the spouse is employed(=1) or unemployed (=0), xi is a vector of explanatory variables otherthan the employment status including the constant term. The notationincludes individual and household demographic characteristics such as age,education, and household composition variables. The notations αj and βj

are vectors of parameters, and εji is the error term.Despite this single-equation specification where Pi is considered

exogenous, there is a great deal of emphasis in the literature on the roleof unobservable social norms and institutions as the potential source ofendogeneity between total work time and spouse’s employment. Recentempirical research has also raised the joint nature of time allocation andspouses’ employment decisions (Rachel Connelly and Jean Kimmel 2009;José I. Giménez-Nadal and Jose A. Molina 2012). These findings pointto the endogeneity issue in estimation. In order to address this potentialendogeneity, we apply a two-step estimation procedure. By using thistechnique, we accomplish two goals: we control for endogeneity, and weestimate the impact of the crisis on work time where comparable pre- andpost-crisis time-use data are not available.

There are four main approaches to censored data estimation withendogeneity in the literature: substitution, control function, system-reducedform, and artificial instrumental regressor. All these approaches include atwo-step procedure with an instrumental variable. However, they vary withrespect to the type of estimation used in the second stage, as well as interms of asymptotic properties of the estimators they provide. Comparisonsof these four estimators through a simulation show that the substitutionmethod performs quite well (Changhui Kang and Myoung-jae Lee 2010).The proposed substitution method here involves two steps: (1) estimatingindividual unemployment risk and (2) estimating influence of spouse’sunemployment risk on paid and unpaid work time.

To estimate unemployment risk, we first derive the dependent variableusing the standard labor market statuses as: (1) employed, (2) unemployed,and (3) out of labor force.15 Unemployment risk is conditional onparticipating in the labor market. In estimating the probability of being

189

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 09:

37 2

4 A

ugus

t 201

3

ARTICLES

unemployed, we use a binary logit model,16 which is a widely usedmodeling technique to estimate binary dependent variables (unemployed/

employed = 1/0). More formally, unemployment risk is determined by thelatent variable for each individual:

P li = γ Zi + ϑ

γ

i (2)

where represents the vector of explanatory variables, ϑγ

i is symmetricallydistributed with zero mean random term denoting unobservabledeterminants of unemployment risk, and F(ϑ

γ

i ) is the cumulativedistribution function. Given that F(ϑ

γ

i ) is a logistic distribution, Equation 2provides a binary logit model, which we estimate separately for women andmen in order to capture distinct effects for each.

In the second step, we estimate determinants of unpaid and paid worktime for each spouse assuming time spent in both work times is determinedsimultaneously. Spouse’s influence here is captured through a singlevariable, the predicted unemployment risk obtained in the first step. We alsoincorporate several controls that may confound the relationship betweenspouses’ unemployment risk and work time. A large number of respondentsreport zero value for unpaid and/or paid work time. Data sets with truncationof this sort require specific methods (Jeffrey Wooldridge 2009).17 UsingTobit empirical specification given censored outcome variables, we estimate:

ylji = α′

j Pls + β ′

j xi + εji (3)

where ylji is the latent variable representing time allocated to activity j by

individual i; xi is a vector of explanatory variables including individual andhousehold characteristics; P l

s is the spouse’s predicted unemployment risk;βj and αj are vectors of parameters; and ∈ji is the error term. The observedtime allocation (yji) variables are related to the corresponding latent timeallocation variables by:

yj i = ylji if yl

ji > 0, yji = 0 otherwise (4)

An ordinary least squares (OLS) estimation of Equations 1 and 3 couldbe consistent if cov(εji , ϑ

γ

i ) = 0 and cov(εji , ∈ji) = 0. However, given thepotential endogeneity problem, necessary conditions for OLS estimation arehighly likely to fail. We address this endogeneity issue using the maximumlikelihood estimation with logistic estimation in the first step and Tobitestimation in the second step. However, we note that maximum likelihoodestimation in two stages does not give consistent estimators of parameters for(βj , αj ) unless instrumental variables are used in the first stage. Consistentestimators may be obtained by nonlinear instrumental variables where anatural instrument for P − i is F (γ Zi) provided that Zi includes at least onevariable, which is not contained in xi (Begoña Álvarez and Daniel Miles2003). Here we use rural and urban unemployment rates by sex, education,

190

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 09:

37 2

4 A

ugus

t 201

3

WORK TIME IN TURKEY

and age groups as the instrumental variable. Given data limitations in time-use surveys, we constructed this variable using a different data source: theLabor Force Survey (LFS) conducted in 2006 (TurkStat 2006a). Age andeducational categories in the LFS data are different from the categories usedin time-use data. In order to make them compatible with Time Use Surveys’(TUS) data categories (TurkStat 2006b), we first converted the classificationand calculated the unemployment rates by age and education for marriedmen and women accordingly.18

We examine the validity and relevance of our instrumental variable bytesting the first-stage results. First, the instrument proves to be valid, asit is not serially correlated with the outcome variables, paid and unpaidwork time.19 Likelihood Ratio (LR) specification test results also indicatethat the weak instrument problem is not an issue here (see SupplementaryTable 1 online).20 In order to see the extent of potential bias fromendogeneity, we run our first step regression excluding the instrument.Biased results present higher coefficient estimates of the impact on women’stotal work time and lower estimates of the impact on men’s work time (seeSupplementary Table 2 online). The results show that ignoring endogeneityleads to overestimating spouse’s influence on women’s total work time whileunderestimating the influence on men’s work time.21

Depending on the assumption that unobserved factors that influence timespent in unpaid and paid work activities might be correlated, the empiricalspecification we use in the second step is a multivariate Tobit that entailssimultaneous determination of time spent in unpaid and paid work activities.This method provides statistical efficiency gains by using the full informationabout the error correlation and allows one to analyze the correlationsbetween error terms of the equations, which reflect the correlations inallocation of time among different activities not accounted for explanatoryvariables. The maximum simulated likelihood (MSL) method, based on theGeweke–Hajivassiliou–Keane (GHK) simulator, is conducted in estimatingthe model in Equations 3 and 4.22 Then, we calculate the marginal effects bymultiplying the coefficients obtained with the proportion of non-censoredobservations in the sample in order to interpret estimation results (WilliamH. Greene 1999, 2008).

Turning to the explanatory variables controlled in our estimations, Zi

includes individual characteristics (including age and age squared, whichenables capturing possible nonlinear effects of age; years of educationcompleted; the age–education variable for the interaction effect of age andeducation; and the instrumental variable). We also include the ownershipof home appliances in Zi . We incorporate age- and education-relatedvariables, as these variables have the potential to indicate the factors thatinfluence employers’ hiring and firing decisions. We expect to obtaina negative relationship both for the education and age variables withunemployment risk.

191

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 09:

37 2

4 A

ugus

t 201

3

ARTICLES

The instrumental variable reflects the factors that influence demand forlabor. We expect individual unemployment risk to be higher in accordancewith the higher average unemployment rates for the specific categoryindividuals belong to, where categories are constructed by the individuals’cohort, sex, level of education, and residential location.

Among household characteristics, we include ownership of domesticappliances as a dummy variable in the estimation.23 Ownership of appliancesindicates a household’s well-being and living standards; thus, we expect toget a negative effect on an individual’s unemployment risk. In addition,given the close association between ownership of domestic appliancesand unpaid work time, such ownership might also play a critical rolein determining time spent in job-seeking activities. Higher intensity ofjob search has a potential to decrease unemployment rates. Thus, weincorporate ownership of domestic appliances as a control to capture bothof these effects.

The notation xi stands for the vector of explanatory variables wecontrol in the second step estimation. Except for the instrumental variableincorporated in Zi , xi includes all the independent variables used in thefirst step (age and age squared, education years, interaction variable foreducation and age, and ownership of domestic appliances) as well asthe variables that may affect both unemployment risk and total worktime. Urban/rural (0/1) dummy and the number of children (by sexand age group: older than 15 years, and younger than 16 years) areour additional controls here. Based on experiences of other countries,we expect to get a positive relationship between unpaid work time andurban/rural dummy, and we expect the reverse for paid work time (RaniaAntonopoulos and Indira Hirway 2010). We also expect to get a positiveinfluence of number of children living in the household on unpaid worktime. Yet, we might obtain this result only for women, as care work hasbeen stressed widely as being divided sharply between women and men.However, young children might also spend a considerable amount of timehelping their parents in unpaid work activities (Nadeem Ilahi 2000), whichmight also lead to an inverse relationship. Unlike unpaid work time, paidwork time is expected to decrease with the number of children living inhousehold.

We compute spouse’s influence on unpaid and paid work time bymultiplying the coefficient estimates obtained in the second step withcorresponding scale factors used to transform the outcomes to uncensoredones (Greene 1999). Marginal effects obtained show the change in anindividual’s work time in minutes per day, given a 1 percentage pointincrease in spouse’s unemployment risk. Then, to get an average estimatefor the impact of the crisis, we assume that unemployment risk for womenand men in couple households rises at a rate equal to the increase in relevantactual unemployment rates for both (Table 4).

192

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 09:

37 2

4 A

ugus

t 201

3

WORK TIME IN TURKEY

Table 4 Average unemployment rates by sex, age 15 years and over (%)

Pre-crisis (beginning of 2008) 2009

Married Women Men Women Men

National 6.0 7.0 9.1 10.2Urban 10.8 7.6 14.9 11.3Rural 1.6 5.7 3.1 7.6

Unemployment rates by sector Women Men Women MenAgriculture 1.9 5.7 3.0 7.0Industry 16.7 13.9 23.0 22.0Services 19.0 8.0 19.0 11.0

Notes: First-time job seekers are distributed among sectors proportional to thepercentage distribution of unemployed among the three sectors. Industry includes theconstruction sector.Sources: Unemployment rates by sector are authors’ calculations based on TurkStat(2007, 2009).

EMPIRICAL RESULTS

Data suggest that women from older age groups seem to face a higherunemployment risk in Turkey, which increases at a decreasing rate (seeSupplementary Table 1 online). Having relatively less work experience inthe labor market, elderly women are more likely to be unemployed. Thepositive and statistically significant influence of education for both womenand men indicates that populations with higher levels of education face ahigher risk of unemployment. Official statistics also support this observation:except for the group with a university degree, the average unemploymentrate increases with the level of education both for women and men (TurkStat2006a). The joint effect of education and age in our results indicates similartrends, as well. The statistically significant and negative coefficient of theinteraction term implies a lower unemployment risk for individuals whoare in higher age groups and have relatively higher education level, whileyounger individuals with high education face a higher unemployment risk.24

As expected, the coefficients for the instrumental variable present apositive relationship between individuals’ unemployment risk and actualaverage unemployment rates by gender, education, age group, andrural/urban location. This finding simply reflects the influence of generallabor market conditions specific to individuals’ cohort and the group theybelong to on their likelihood of being unemployed.

Turning to our main results provided by the second-step estimation results,we see that figures in Tables 5 and 6 present the key relationships weseek, that is, influence of spouse’s unemployment risk on unpaid and paidwork time. Estimates are presented both at the national level as well as at

193

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 09:

37 2

4 A

ugus

t 201

3

AR

TIC

LE

S

Table 5 Multivariate Tobit estimates of women’s paid and unpaid work time: Marginal effects are conditional on the outcome beinguncensored

National (2,491 couples) Urban (1,694 couples) Rural (797 couples)

Row Dep. var.: Daily minutes (1) (2) (3) (4) (5) (6)Variables Paid work Unpaid work Paid work Unpaid work Paid work Unpaid work

Individual characteristics(1) Age (group median) 11.63∗∗∗ −8.11∗∗∗ 8.53∗∗∗ −4.63∗ 17.97∗∗∗ −18.51∗∗∗

(9.72) (1.99) (17.42) (2.35) (11.84) (3.74)(2) Ageˆ2 −0.12∗∗∗ 0.07∗∗∗ −0.11∗∗∗ 0.04 −0.18∗∗∗ 0.17∗∗∗

(0.11) (0.02) (0.22) (0.03) (0.13) (0.04)(3) Education years 8.41∗∗∗ −6.31∗∗∗ 4.99∗∗ −3.13 4.81 −18.72∗∗∗

(12.70) (2.75) (20.10) (3.13) (17.63) (6.13)(4) Education∗age −0.11∗∗ 0.10∗∗ −0.02 0.00 −0.08 0.50∗∗∗

(0.29) (0.06) (0.48) (0.07) (0.41) (0.14)(5) Spouse’s unemployment risk 349.61∗∗∗ 311.35∗∗∗ 315.62∗∗∗ 496.46∗∗∗ 746.92∗∗∗ −104.26

(predicted) (473.13) (134.22) (782.51) (178.581) (611.87) (210.14)Household characteristics

(6) Rural/urban (1=Rural) 61.29∗∗∗ −3.78 – – – –(24.87) (6.27)

(7) Number of daughters (older than 15 years) 1.48 −28.97∗∗∗ 10.90∗∗ −33.61∗∗∗ −11.62 −20.26∗(23.27) (5.09) (41.08) (5.91) (27.15) (10.19)

(8) Number of sons (older than 15 years) −4.87 −3.42 −1.47 −6.29 −4.99 0.99(24.86) (5.59) (47.78) (6.79) (26.65) (10.17)

194

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 09:

37 2

4 A

ugus

t 201

3

WO

RK

TIM

EIN

TU

RK

EY

(9) Number of sons (younger than 16 years) −11.22∗∗∗ 31.38∗∗∗ −10.90∗∗ 36.46∗∗∗ −12.48∗∗ 25.64∗∗∗(17.18) (4.05) (32.13) (5.33) (17.70) (6.52)

(10) Number of daughters (younger than 16 years) 5.80 −8.39∗ 1.19 −5.62 13.136 −14.32(23.90) (5.10) (43.08) (5.94) (26.67) (9.78)

(11) Ownership of domestic appliances −0.21 0.77 50.61∗∗ 9.08 15.80 −32.74∗(1=Own) (58.98) (16.43) (195.98) (29.74) (57.20) (18.64)N 4,971 3,380 1,591Lnsigma1 6.264∗∗∗ 6.503∗∗∗ 5.983∗∗∗

(0.025) (0.034) (0.038)Lnsigma2 5.096∗∗∗ 5.086∗∗∗ 5.098∗∗∗

(0.012) (0.015) (0.020)atrh0 −0.730∗∗∗ −0.762∗∗∗ −0.709∗∗∗

(0.030) (0.045) (0.040)

Notes: ∗∗∗, ∗∗, and ∗ indicate statistical significance at the 1, 5, and 10 percent levels, respectively. Scale factor used to calculate marginal effects at the national levelfor women is 0.99 and 0.18 for unpaid work and paid work equations, respectively. Corresponding figures for unpaid and paid work equations are 0.99 and 0.12for women living in urban areas and 0.99 and 0.31 for women in rural areas. Scale factors are calculated as the proportion of non-zero respondents; that is, 0.99indicates that 99 percent of women participate in unpaid work and spend positive number of minutes > 0. Observation numbers (N) are higher than number ofrespondents (2,491). There are two observations for 99 percent of the respondents who report their time-use data in both diaries. Half of the diaries were collectedon a weekday and half on a weekend day. Diary weights used are provided by TurkStat (2006b). Coefficients show marginal effects. For the statistical significancevalue, these marginal effects need to be converted to their original values then divided by the standard errors provided in parentheses. Statistically significantcorrelation found between paid and unpaid work time as shown by the test parameter (atrh0) provides supporting evidence for the simultaneous determinationof the paid and unpaid work time.

195

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 09:

37 2

4 A

ugus

t 201

3

AR

TIC

LE

S

Table 6 Multivariate Tobit estimates of men’s paid and unpaid work time: Marginal effects are conditional on the outcome beinguncensored

National (2,491 couples) Urban (1,694 couples) Rural (797 couples)

Row Dep. var.: Daily minutes (1) (2) (3) (4) (5) (6)Variables Paid work Unpaid work Paid work Unpaid work Paid work Unpaid work

Individual characteristics(1) Age (group median) 21.60∗∗∗ −3.17∗∗∗ 26.89∗∗∗ −3.50∗∗∗ 7.46 −3.98

(5.88) (2.06) (7.34) (2.50) (10.15) (3.66)(2) Ageˆ2 −0.34∗∗∗ 0.04∗∗∗ −0.44∗∗∗ 0.04∗∗∗ −0.15∗∗ 0.05∗∗

(0.07) (0.02) (0.08) (0.03) (0.11) (0.04)(3) Education years −0.30 3.11∗∗ −4.46 2.75∗∗ −9.03 0.67

(6.69) (2.36) (7.95) (2.76) (14.12) (4.91)(4) Education∗age −0.08 −0.02 0.03 −0.02 0.16 0.07

(0.15) (0.05) (0.19) (0.06) (0.31) (0.11)(5) Spouse’s unemployment risk −143.23 87.42∗∗ −141.89 92.62∗∗ −78.99 −39.06

(Predicted) (195.59) (68.00) (205.15) (72.68) (646.14) (218.92)Household characteristics

(6) Rural/urban (1 = Rural) 8.05 3.20 – – – –(13.99) (4.90)

(7) Number of daughters (older than 15 years) 22.98∗∗∗ −11.08∗∗∗ 31.77∗∗∗ −14.31∗∗∗ 7.69 −2.92(12.32) (4.27) (14.34) (5.08) (23.34) (7.89)

(8) Number of sons (older than 15 years) −28.52∗∗∗ −6.41∗∗∗ −23.44∗∗ −4.38∗ −19.31a −13.71∗∗(12.78) (4.45) (16.42) (5.27) (19.66) (8.15)

196

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 09:

37 2

4 A

ugus

t 201

3

WO

RK

TIM

EIN

TU

RK

EY

(9) Number of sons (younger than 16 years) 1.47 4.21∗∗ 9.17 4.95∗∗ −8.59 3.68(7.79) (2.96) (9.56) (3.72) (14.18) (5.04)

(10) Number of daughters (younger than 16 years) 0.16 −2.42 −4.78 −2.87 9.33 −1.77(10.57) (3.76) (12.07) (4.32) (21.11) (7.44)

(11) Ownership of domestic appliances 100.29∗∗∗ −16.61∗∗∗ 110.22∗∗∗ −5.71 73.05∗∗∗ −28.67∗∗∗(1=Own) (27.28) (10.18) (49.59) (15.88) (32.70) (13.43)N 4,967 3,376 1,591Lnsigma1 5.86∗∗∗ 5.86∗∗∗ 5.85∗∗∗

(0.02) (0.02) (0.03)Lnsigma2 4.82∗∗∗ 4.80∗∗∗ 4.87∗∗∗

(0.03) (0.04) (0.04)atrh0 −0.52∗∗∗ −0.52∗∗∗ −0.54∗∗∗

(0.02) (0.03) (0.04)

Notes: ∗∗∗, ∗∗, ∗ denote statistical significance at the 1, 5, and 10 percent levels, respectively. aStatistical significance at 15 percent. Scale factor used to calculatemarginal effects at the national level for men is 0.74 and 0.56 for unpaid work and paid work equations, respectively. Corresponding figures for unpaid and paidwork equations are 0.74 and 0.55 for men living in urban areas and 0.73 and 0.60 for men in rural areas. Scale factors are calculated as the proportion of non-zerorespondents; that is, 0.74 indicates 74 percent of men participate in unpaid work and spend positive number of minutes > 0.

197

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 09:

37 2

4 A

ugus

t 201

3

ARTICLES

the urban/rural distinction separately. Note that coefficients reported inTables 5 and 6 show the marginal effects of the independent variables onunpaid and paid work time conditional on the outcome being uncensored.To give an example, a 1 percentage point increase in men’s unemploymentrisk raises women’s unpaid work time by 3 minutes per day (311.4 multipliedby 0.01) at the national level. In urban areas, this increase is 5 minutes perday (496.5 multiplied by 0.01).

Our findings show substantive differences between women and men.As Table 5 shows, there is a positive relationship between spouse’sunemployment risk and time women devote to both paid and unpaid workactivities. While in rural areas this impact is only observed for women’s paidwork time, our findings reveal that, on average, women in Turkey worklonger hours with higher unemployment risk of their spouse. However,men’s work time does not show a strong spousal influence as such. We onlyobserve a statistically significant impact on men’s unpaid work time, but theeffect is smaller when compared to that on women.

Nationwide averages show that a 1 percentage point rise in spouse’sunemployment risk raises women’s total work time approximately eighttimes more than that of men. In rural settings, the impact on unpaid workdoes not seem to be substantial for both men and women, as the type ofunpaid work activities differ between rural and urban areas. In rural areas,gender-based divisions of labor and the distinction between paid and unpaidwork are not clear cut (Ferhunde Özbay 1990; Deniz Kandiyoti 1997).

Regarding the effects of other individual characteristics, we observe that,for both women and men, paid work time increases at a decreasing rate withage. On the contrary, unpaid work time increases at a decreasing rate. Higherlevels of education increase women’s paid work time while decreasing theirunpaid work time (Table 5). This result might reflect the fact that womenwith higher education have higher bargaining power both at home and inthe market.

Consistent with earlier evidence based on US households (Joan Huberand Glenna Spitze 1983; Julie Brines 1994), the figures in Table 6 indicatethat educated men do more unpaid work in Turkey. On the other hand,the education and age interaction term shows that women’s paid work timedecreases, while unpaid work time increases, which might point particularlyto the status of middle-class women in Turkey. Despite their high educationlevels, these women do not participate in the labor market.

Regarding the relation between the number of children and unpaid worktime, our results show that both women and men in Turkey appear to sharetheir unpaid work burden with their daughters. In addition, compared tomen, the effects on women’s unpaid work time are much larger. We alsoobserve a positive relationship between number of daughters and the meanduration of women’s paid work time for those living in urban areas. Couplesliving in urban areas participate in paid work when there are girls to bear the

198

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 09:

37 2

4 A

ugus

t 201

3

WORK TIME IN TURKEY

unpaid work requirements within the household, such as caring for siblings,a common practice in Turkey.

It is also interesting to observe a negative relationship between the numberof daughters younger than 16 years old and women’s unpaid work time.Even if daughters are not yet adults, they share the unpaid work burdenwithin the household like other women members. However, we do notobserve this relationship in urban areas, which may reflect the higher schoolenrollment rates for girls in these areas: by not staying at home, they arefreed from domestic chores.25 Unlike girls, the number of boys over age15 does not appear to have any major impact on mothers’ total work time.Yet, the presence and number of sons influences fathers’ total work timenegatively. Larger household size might bring about an allocation amongmale members with respect to paid work.

Contrary to the effect of girls, the greater the number of boys in thehousehold, the lower the amount of paid work time and the higher theunpaid work time for women, since caring for children constitutes a majorportion of women’s unpaid work. The results obtained may also reflectpatriarchal values, as boys may be more valued than girls (Carol Delaney1991). Since caring for boys is perceived to be more important than forgirls, mothers of boys are released from paid work. In addition, we observea positive effect of number of boys in the household on men’s unpaidwork time, except for men living in rural areas, a finding that supportsthe argument in the literature that men who have boys spend more time incaring activities (W. Jean Yeung, John F. Sandberg, Pamela E. Davis-Kean,and Sandra L. Hofferth 2001).

Finally, as expected, estimation results show that domestic appliances inthe household increase men’s paid work time while decreasing their unpaidwork time. However, except for women living in urban areas, we do notsee any major influence of owning domestic appliances on women’s unpaidand paid work time. In fact, these findings are consistent with earlier studiesraising the argument that so-called labor-saving domestic technology doesnot save time for women (Michael Bittman, James Mahmud Rice, and JudyWajcman 2004).

SIMULATION

Using the marginal effects of spouse’s influence on paid and unpaid worktime and the actual change in corresponding unemployment rates officiallyreported in Turkey, we conduct a simulation analysis to estimate the impactof the recent economic crisis on work time. Based on our assumption thatunemployment risk rises at a rate equal to the increase in relevant actualunemployment rates for both, we calculate the change in women’s andmen’s paid and unpaid work time simply by multiplying the percentage pointchanges in actual unemployment rates by their marginal effects. To illustrate,

199

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 09:

37 2

4 A

ugus

t 201

3

ARTICLES

Table 7 Summary results (time in minutes/day)

Actual Simulation

Row National Paid Unpaid Total Paid Unpaid Total(1) (2) (3) (4) (5) (6)

(1) Women 53 387 440 65 397 462(2) Men 313 55 369 313 58 371(3) Gender gap −260 332 72 −249 339 90

% Change in work time(4) Women 21.0 2.6 4.8(5) Men 0.0 4.9 0.7(6) Gender gap −4.3 2.2 25.6

Actual Simulation

Urban Paid Unpaid Total Paid Unpaid Total(7) Women 42 387 428 54 405 459(8) Men 320 54 374 320 58 378(9) Gender gap −279 333 54 −267 347 81

% Change in work time(10) Women 28.7 4.8 7.1(11) Men 0.0 7.0 1.0(12)Gender gap −4.3 4.4 49.1

Actual Simulation

Rural Paid Unpaid Total Paid Unpaid Total(13) Women 78 389 467 93 389 482(14) Men 298 58 356 298 58 356(15)Gender gap −220 330 111 −205 330 126

% Change in work time(16) Women 19.4 0.0 3.2(17) Men 0.0 0.0 0.0(18)Gender gap −6.9 0.0 13.6

Notes: We calculate the gender gap and percentage change in work time in the followingway:Gender gap at the national level is computed by subtracting the figures in row 1 fromthe figures in row 2. Percentage change in paid work time for women is calculated bysubtracting the figure in row 1 and column 1, from the figure in row 1 column 4 anddividing by the former.Source : Authors’ calculations based on TurkStat (2006a, 2006b, 2007, 2009).

as can be observed from the actual nationwide average unemploymentfigures in Table 4, the unemployment rate for married men increased by3.2 percentage points during the crisis period, rising from 7 percent atthe beginning of 2008 to 10.2 percent by the end of 2009. Multiplying thischange by the estimates obtained for spouse’s influence on women’s paid

200

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 09:

37 2

4 A

ugus

t 201

3

WORK TIME IN TURKEY

and unpaid work time, respectively, we get an average estimate that showsthe change in women’s total work time: accordingly, married women livingin nuclear couple households work approximately 20 minutes longer perday, which adds up to more than 2 hours per week.

Overall simulation results show that the percentage increase in women’spaid work time is higher than the increase in their unpaid work time, andthat the reverse is true for men. This is because the original mean durationof time spent in unpaid and paid work activities is much higher/lowerfor married women than married men in Turkey (see actual figures forpaid and unpaid work time in Table 7). Second, the gap between womenand men in unpaid and paid work time widens and declines, respectively.While women’s total work time on average increases by 22 minutes per day(5 percent), the corresponding rise for their spouses is only 2.7 minutes perday (0.7 percent). This widens the existing gap of total work time betweenwomen and men by 25 percent. In urban areas, the percentage change inwomen’s total work time is 7 percent (30.4 minutes/day); much higher thanmen’s, which is 1 percent (3.7 minutes/day). Thus, the gap between womenand men grows more dramatically in urban areas (by 50 percent).

CONCLUSION

This contribution provided estimates of the impact of the 2008–09 economiccrisis on Turkey by focusing on a usually neglected dimension: bothpaid and unpaid work time. Using the first national Turkish Time-UseSurvey conducted in 2006, we model time spent in paid and unpaid workusing a two-step estimation specification in which we first estimate theunemployment risk of coupled women and men and, second, estimatespouse’s influence on the duration of unpaid and paid work time.

We find that the change in women’s unpaid work burden is quitesubstantial: women have considerably higher initial levels compared to menin terms of time devoted to unpaid work, yet still, women’s unpaid worktime rises approximately four times more than that of men. The results aremuch more striking as far as the total workload is considered: the increasein women’s total work burden is approximately eight times more than thatof men. The differences are sharper in urban settings.

Previous empirical research on the economic crisis’ gendered impact onwork time in developing countries is very limited. In the case of the 1997Indonesian crisis, while the number of weekly paid work hours declined forboth women and men in rural areas, in urban areas, women’s weekly workhours increased by 5.4 percent (2.1 hours per week), whereas for men thischange was only 1.1 percent (0.5 hours per week; Frankenberg, Thomas,and Beegle 1999). Although it is not possible to make an exact comparisonof results between the Indonesian crisis and our findings here, we observea smaller change in women’s paid work hours (1.4 hours/week) in urban

201

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 09:

37 2

4 A

ugus

t 201

3

ARTICLES

Turkey, whereas men’s paid work hours show no change. Since our resultspertain to the intensive component of employment adjustment (that is,changes in work hours given current employment status), this difference isexpected. Our findings are not inconsistent with existing evidence; however,it is important to note that a simulation exercise based on the data collectedin a non-crisis period would not allow us to take into account any behavioraldifferences of the households in time allocation at the time of a crisis period.Depending on the transitions in labor force states, estimated results mightdiffer from the actual changes occurred. For instance, Günseli Berik andEbru Kongar (2013), conducting a trend analysis based on the actual dataavailable for pre- and post-recession periods, find that the 2007–09 recessionnarrowed the disparity in both and unpaid work hours of married mothersand fathers in the US.

This study provided empirical evidence from Turkey supporting theargument that pre-existing gender inequalities in work time are deepened byeconomic crises and that the impact of economic crises takes a gender-biasedform, putting most of the work burden upon women. Any policy or actionundertaken for gender equality as well as social and economic recovery inpost-crisis should take into account these invisible costs. Increase in publicinvestment and reallocation of public resources across different sectors aremore conducive for economic recovery than contractionary fiscal policy.However, attempts to this end should not just serve for financial bailouts.Public investment in social sectors such as education, health, community,and care services could create new jobs while releasing households fromadditional unpaid workload due to crises.

Seçil A. Kaya BahçeAnkara University – Department of Economics

Faculty of Political Sciences, Cemal Gürsel Caddesi Cebeci, Ankara, 06590, Turkeye-mail: secil.bahce@politics.ankara.edu.tr

Emel MemisAnkara University – Department of Economics

Faculty of Political Sciences, Cemal Gürsel Caddesi Cebeci, Ankara, 06590, Turkeye-mail: emel.memis@politics.ankara.edu.tr

ACKNOWLEDGMENTS

The authors thank the anonymous reviewers and Murat Kırdar for theirvaluable input and suggestions on an earlier version of this study. Theauthors are also grateful to the participants of the Feminist Economics andUN Women Symposium, Critical Perspectives on Financial and EconomicCrises: Why Gender Matters, for their comments, which helped to improvethe manuscript. Finally, the authors acknowledge the generous support

202

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 09:

37 2

4 A

ugus

t 201

3

WORK TIME IN TURKEY

provided by the Scientific and Technological Research Council of Turkeyfor this research (TUBITAK-109K127 coded research project). The usualdisclaimer regarding errors and omissions applies.

NOTES1 While Turkey was a fast-growing developing country during the 2000s, it was hard hit

by the 2008–09 crisis. The average unemployment rate reached the unprecedentedlevel of 14 percent at the end of 2009, one of the highest rates among OECD countries(TurkStat 2010).

2 See also Rania Antonopoulos (2009), Maria S. Floro, Annika Tornqvist, and EmcetOktay Tas (2009), and Stephanie Seguino (2010) for the discussions on this issue inthe context of the 2008–09 crisis.

3 See Diane Elson (1993) and Nilüfer Çagatay and Korkut Ertürk (2004) for an extensivediscussion on the relationship between economic shocks and their impact on theunpaid sphere of the economy.

4 The survey was planned for every five years; but the second one, scheduled in 2011, waspostponed due to budgetary concerns.

5 Ahn, Jimeno, and Ugidos (2003) find that unemployed people in Spain allocate 3.5more hours per weekday than employed people to money-saving activities involvinghousework and childcare.

6 For a more detailed discussion of this theoretical literature, please see the workingpaper version of the current study: Emel Memis and Seçil A. Kaya Bahçe (2011).

7 A total of 5,070 (3,380 urban and 1,690 rural) households were randomly contactedfor the survey. The response rate was quite high: 85.7 percent (4,345 households).

8 Data was collected over a thirteen-month period between December 2005–06 and wascontinuous on a weekly basis.

9 Only five couples among 2,491 report that they are unmarried partners living together.We also exclude four households where there is more than one woman who reportsherself as the wife. These appear to be polygamous households, although not recordedas such in the data.

10 Instead of the actual age information, only the age groups into which the respondents’ages fall are available in the data. We calculated average age here based on the medianvalue of each age group.

11 Sociological research shows that many women who are considered economicallyinactive wish to work, although they are not active job seekers (Ferhunde Özbay 1990).

12 See Emel Memis, Umut Önes, and A. Burça Kızılırmak (2011) and Ipek Ilkkaracan(2012) for a discussion on this so-called Turkish puzzle.

13 Mean duration of time is the weighted average calculated using the weight variablenamed “factor” provided in the dataset, which differs by day (weekday/weekend) foreach respondent.

14 Economic activities and occupations are classified according to the StatisticalClassification of Economic Activities in the European Community, NACE Rev.1.1 andInternational Standard Classification of Occupations (ISCO-88), respectively. Travel towork is not included. Travel for all activities is classified as a separate and single category.

15 Some of the respondents outside the labor force report positive amounts of paid worktime without a sectoral code of economic activity. We redefined unemployed to includethese respondents, assuming their paid work time as the time spent searching for ajob. The number of unemployed increased by fourteen respondents: four students,two retired, one sick and elderly, three homemakers, and four others.

16 An alternative specification to logit estimation is probit estimation, which requiresa restrictive normality assumption. Compared to probit log likelihood, logit log

203

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 09:

37 2

4 A

ugus

t 201

3

ARTICLES

likelihood obtained in our estimation indicates that logit model provides a betterspecification (see the note to Supplementary Table 1 online).

17 Heckman and double-hurdle models, which are alternative estimation methods forsolving large number of respondents reporting zero time, consider the decision toparticipate in doing unpaid work as an independent process from the decision onthe duration of market work. However, given that modeling the participation decisionprocess of doing unpaid work is not as straightforward as in the case of the laborsupply model, a misspecified participation equation introduced in double-hurdleor Heckman’s model can produce worse results than implementing a Tobit model(Lennart Flood and Urban Grasjö 1998).

18 First-step results are presented in Supplementary Tables 1 and 2, which are available inthe online version of this article under the “Supplementary Information” tab.

19 Men’s unpaid and paid work time is not correlated with the instrumental variable (withp-values equal to 0.2 and 0.6, respectively). Women’s unpaid and paid work time is alsonot correlated (p = 0.53); however, women’s paid work time is found to be correlated(p = 0.08).

20 In order to check instrument validity, we use the LR χ2 test, since the partial F test thatis used in OLS regression cannot be conducted in logistic regression.

21 When calculated with biased marginal effects, results show that women’s total worktime would increase by 6 percent while men’s would rise by 0.5 percent at the nationallevel, given a 1 percentage point rise in spouse’s unemployment risk.

22 See William H. Greene (2008) and Vassilis Hajivassiliou and Paul Ruud (1994), amongothers, for an explanation of the GHK simulator.

23 Given the categorical household income variable in the data, other householdmembers’ income cannot be isolated from individual income. Instead of using incomeas a proxy, we constructed a variable: owning domestic appliances (based on whethera household owns a washing machine and/or dishwasher, which reflects the quality ofliving conditions specific to Turkish data).

24 Research establishes that members of the highly educated young population in Turkeyface a much higher risk of unemployment with the rise in unemployment rates ofwhite-collar workers (Aksu Bora, Ilknur Üstün, Necmi Erdogan, and Tanil Bora 2011).

25 Ilahi (2000) points to the two opposing effects to explain children’s effect on women’shousework time: income effect (as mother’s income increases, her demand forchildren’s education increases) and substitution effect (children have to step in formother’s forgone housework). Accordingly, the income effect might dominate here.

REFERENCES

Ahn, Namkee, Juan F. Jimeno, and Arantza Ugidos. 2003. “Mondays at the Sun:Unemployment, Time Use and Consumption Pattern in Spain.” FEDEA Working Paper2003–18.

Álvarez, Begoña and Daniel Miles. 2003. “Gender Effect on Housework Allocation:Evidence from Spanish Two-Earner Couples.” Journal of Population Economics 16(2):227–42.

Antonopoulos, Rania. 2009. “The Current Economic and Financial Crisis: A GenderPerspective.” Working Paper 562, Levy Economics Institute of Bard College.

Antonopoulos, Rania and Indira Hirway, eds. 2010. Unpaid Work and the Economy: Gender,Time Use and Poverty in Developing Countries. New York: Palgrave Macmillan.

Becker, Gary S. 1991. A Treatise on the Family. Cambridge: Harvard University Press.Benería, Lourdes. 2003. Gender, Development, and Globalization: Economics as if All People

Mattered. New York: Routledge.

204

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 09:

37 2

4 A

ugus

t 201

3

WORK TIME IN TURKEY

Berik, Günseli and Ebru Kongar. 2013. “Time Allocation of Married Mothers and Fathersin Hard Times: The 2007–09 US Recession.” Feminist Economics, http://dx.doi.org/10.1080/13545701.2013.798425.

Bittman, Michael, Paula England, Liana Sayer, Nancy Folbre, and George Matheson.2003. “When Does Gender Trump Money? Bargaining and Time in Household Work.”American Journal of Sociology 109(1): 186–214.

Bittman, Michael, James Mahmud Rice, and Judy Wajcman. 2004. “Appliances and TheirImpact: The Ownership of Domestic Technology and Time Spent on HouseholdWork.” British Journal of Sociology 55(3): 401–23.

Bora, Aksu, Ilknur Üstün, Necmi Erdogan, and Tanil Bora. 2011. “Bosuna mi Okuduk?”Turkiye’de Beyaz Yakali Issizligi [So, we studied for nothing? White-collar unemployed inTurkey]. Ankara: Iletisim Yayinevi.

Brines, Julie. 1994. “Economic Dependency, Gender, and the Division of Labor at Home.”American Journal of Sociology 100(3): 652–88.

Çagatay, Nilüfer and Korkut Ertürk. 2004. “Gender and Globalization: A MacroeconomicPerspective.” Working Paper 19, International Labour Organization.

Connelly, Rachel and Jean Kimmel. 2009. “Spousal Influences on Parents’ Non-MarketTime Choices.” Review of Economics of the Household 7(4): 361–94.

Delaney, Carol. 1991. The Seed and the Soil: Gender and Cosmology in Turkish Village Society.Comparative Studies on Muslim Societies. Berkeley: University of California Press.

Elson, Diane. 1993. “Gender-Aware Analysis and Development Economics.” Journal ofInternational Development 5(2): 237–47.

Eurostat. 2000. “Survey on Time Use, Activity Coding List, Final Draft.” DOCE2/TUS/5/00, United Nations Statistics Division.

Flood, Lennart and Urban Grasjö. 1998. “Regression Analysis and Time Use Data AComparison of Microeconometric Approaches with Data from the Swedish Time UseSurvey (HUS).” Working Papers in Economics 5, Department of Economics, Schoolof Economics and Commercial Law, Göteborg University, Sweden.

Floro, Maria S., Annika Tornqvist, and Emcet Oktay Tas. 2009. “The Impact ofthe Economic Crisis on Women’s Economic Empowerment.” Swedish InternationalDevelopment Agency Working Paper Series. http://www.american.edu/cas/economics/pdf/upload/2009-26.pdf.

Frankenberg, Elizabeth, Duncan Thomas, and Kathleen Beegle. 1999. “The Real Costsof Indonesia’s Economic Crisis: Preliminary Findings from the Indonesia Family LifeSurveys.” Labor and Population Program Working Paper Series 99–04, RAND.

Gershuny, Jonathan I. 2000. Changing Times: Work and Leisure in Postindustrial Society.Oxford: Oxford University Press.

Giménez-Nadal, José I. and Jose A. Molina. 2012. “Regional Unemployment, Gender andTime Allocation of the Unemployed.” IZA Working Paper 7043.

Greene, William H. 1999. “Marginal Effects in the Censored Regression Model.” EconomicsLetters 64(1): 43–9.

———. 2008. Econometric Analysis, 6th ed. Cranbury, NJ: Pearson Education.Gronau, Reuben. 1973. “The Intrafamily Allocation of Time: The Value of the Housewives’

Time.” American Economic Review 63(4): 634–51.———. 2006. “Home Production and the Macro Economy – Some Lessons from Pollak

and Wachter and from Transition Russia.” Working Paper 12287, National Bureau ofEconomic Research.

Hajivassiliou, Vassilis A. and Paul Ruud. 1994. “Classical Estimation Methods for LDVModels Using Simulation.” In Handbook of Econometrics, vol. IV, edited by Robert Engleand Daniel McFadden, 2383–441. New York: Elsevier.

Hartmann, Heidi. 1979. “The Unhappy Marriage of Marxism and. Feminism: Towards aMore Progressive Union.” Capital and Class 3(2): 1–33.

205

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 09:

37 2

4 A

ugus

t 201

3

ARTICLES

Huber, Joan and Glenna Spitze. 1983. Sex Stratification: Children, Housework, and Jobs. NewYork: Academic Press.

Ilahi, Nadeem. 2000. “The Intra-household Allocation of Time and Tasks: What HaveWe Learnt from the Empirical Literature?” Policy Research Report on Genderand Development, Working Paper Series 13, World Bank Development ResearchGroup/Poverty Reduction and Economic Management Network.

Ilkkaracan, Ipek. 2012. “Why So Few Women in the Labor Market in Turkey?” FeministEconomics 18(1): 1–37.

Juster, F. Thomas and Frank P. Stafford. 1991. “The Allocation of Time: EmpiricalFindings, Behavioral Models, and Problems of Measurement.” Journal of EconomicLiterature 29(2): 471–522.

Kandiyoti, Deniz. 1997. “Kadınlar ve Hane Içi Üretim Türkiye’de Kırsal DönüsümünEtkileri” [Women and household production, the impact of rural transformation inTurkey]. In Cariyeler, Bacılar, Yurttaslar: Kimlikler ve Toplumsal Dönüsümler [Concubines,sisters, citizens: Identities and social transformation], edited by Deniz Kandiyoti, 54–68.Istanbul: Metis Yayınları.

Kang, Changhui and Myoung-jae Lee. 2010. “Performance of Various Estimators forCensored Response Models with Endogenous Regressors.” Pacific Economic Review15(4): 447–63.

Kızılırmak, A. Burça. 2008. “Labor Market Participation Decisions of Married Women:Evidence from Turkey.” In Globalization and Development: A Handbook of New Perspectives,edited by Ashwini Deshpande, 176–84. Oxford: Oxford University Press.

Lundberg, Shelly and Robert A. Pollak. 1993. “Separate Spheres Bargaining and theMarriage Market.” Journal of Political Economy 101(6): 988–1010.

Manser, Marilyn and Murray Brown. 1980. “Marriage and Household Decision-Making:A Bargaining Analysis.” International Economic Review 21(1): 31–44.

Memis, Emel and Seçil A. Kaya Bahçe. 2011. “Estimating the Impact of the RecentEconomic Crisis on Work Time in Turkey.” Working Paper Series 686, Levy EconomicsInstitute of Bard College.

Memis, Emel, Umut Önes, and A. Burça Kızılırmak. 2011. “Housewifization of Women:Contextualising Gendered Patterns of Paid and Unpaid Work.” In Gender and Society inTurkey: The Impact of Neoliberal Policies, Political Islam and EU Accession, edited by SaniyeDedeoglu and Adem Y. Elveren, 87–102. London: I. B. Tauris.

Milkman, Ruth. 1976. “Women’s Work and Economic Crisis: Some Lessons of the GreatDepression.”Review of Radical Political Economics 8(1): 71–97.

Özbay, Ferhunde. 1990. “Kadınların Ev Içi ve Ev Dısı Ugraslarındaki Degisme” [Change inwomen’s domestic and outdoor chores]. In 1980’ler Türkiyesi”nde Kadın Bakıs AçısındanKadınlar [Women from the viewpoint of women in Turkey in the 1980s], edited by S.Tekeli, 115–40. Istanbul: Iletisim Yayınları.

Rodgers, Yana van der Meulen and Nidhiya Menon. 2012. “Impact of the 2008–2009 TwinEconomic Crises on the Philippine Labor Market.” World Development 40(11): 2318–28.

Seguino, Stephanie. 2010. “The Global Economic Crisis, its Gender and EthnicImplications, and Policy Responses.” Gender and Development 18(2): 179–99.

Seiz, Janet A. 1999. “Game Theory and Bargaining Models.” In The Elgar Companion toFeminist Economics, edited by Janice Peterson and Margaret Lewis, 379–90. Cheltenham:Edward Elgar.

TurkStat. 2006a. Household Labour Force Statistics. http://www.turkstat.gov.tr/Kitap.do?metod=KitapDetay&KT_ID=8&KITAP_ID=27.

———. 2006b. Time-Use Survey. www.turkstat.gov.tr/metaveri/15_m14.doc.———. 2007. Household Labour Force Statistics. www.turkstat.gov.tr/metaveri/15_m14.doc.———. 2009. Household Labour Force Statistics. www.turkstat.gov.tr/metaveri/15_m14.doc.———. 2010. Household Labour Force Statistics. www.turkstat.gov.tr/metaveri/15_m14.doc.

206

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 09:

37 2

4 A

ugus

t 201

3

WORK TIME IN TURKEY

Wooldridge, Jeffrey. 2009. Introductory Econometrics: A Modern Approach. Florence,KY: South Western Cengage Learning.

Yeung, W. Jean, John F. Sandberg, Pamela E. Davis-Kean, and Sandra L. Hofferth. 2001.“Children’s Time with Fathers in Intact Families.” Journal of Marriage and Family 63(1):136–54.

NOTES ON CONTRIBUTORS

Seçil A. Kaya Bahçe is Assistant Professor of Economics at Ankara University,Turkey. She received her PhD at the Middle East Technical University,Turkey. She also holds BS and MSc degrees in economics from the MiddleEast Technical University. She has teaching experience in macroeconomicsand microeconomics. Her research interests include macroeconomics andeconomic development.

Emel Memis is Assistant Professor of Economics at Ankara University, Turkeyand currently works as Research Associate in the Gender Equality and theEconomy program at the Levy Economics Institute of Bard College. Shereceived her PhD at the University of Utah. She holds BS and MSc degrees ineconomics from the Middle East Technical University, Turkey. Her researchinterests include macroeconomics, gender, and economic development.

207

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 09:

37 2

4 A

ugus

t 201

3