PARTICIPATION OF GREEK MARRIED WOMEN IN FULL-TIME PAID EMPLOYMENT

Transcription

1 Sout Eastern Europe Journal of Economics 2 (2004) PARTICIPATION OF GREEK MARRIED WOMEN IN FULL-TIME PAID EMPLOYMENT J. DAOULI, M. DEMOUSSIS, * N. GIANNAKOPOULOS Department of Economics, University of Patras, Greece Abstract In ts paper we analyze te static labor supply beavior of married women in Greece using micro data from te National Houseold Budget Survey of 1998/ 99. To tat end, a conventional Probit model is employed for te estimation of te participation decision; a typical OLS regression corrected for selectivity bias is used for te estimation of te wage equation; and, a Tobit procedure is utilized for te estimation of te labor supply function. Te econometric results confirm te presence of a sample selection criterion tat separates working from non-working married women. Te labor force participation decision is affected by te presence of cldren, uman capital caracteristics, region of residence, usbands earnings and non-work income. Te same factors appear to influence, in a plausible manner, te ours of work decision. Moreover, te labor supply elasticity wit respect to te ourly wage is about 0.84 (calculated at te sample means). Finally, uman capital variables suc as education and age appear to explain adequately te observed wage structure of Greek married women. JEL Classification: J22, J48, C24 Keywords: Labor supply, Married women, Greece 1. Introduction As a rule, female LFP rates are substantially lower tat tose for men and ave attracted te attention of many labor economists worldwide. It is well known tat female economic beavior wit regard to labor market decisions depends on several factors tat result in complex labor supply scedules over te life cycle. Ts bea- * Corresponding autor: University Campus, Rio 26500, Greece micdem@econ.upatras.gr Te autors wis to acknowledge two anonymous referees for teir elpful comments and suggestions. Financial support from te Researc Committee of te University of Patras under grant KARATHEOTHORI is greatly appreciated.

2 20 J. DAOULI, et al., Sout-Eastern Europe Journal of Economics 2 (2004) vior as been at te core of labor supply researc since te early seventies and it as focused on te investigation of te factors tat influence te participation decision and its extent, te explanation of non-participation, te role of preferences and te interdependencies wit decisions concerning marriage, cld-bearing, divorce and retirement, (Pollak, 2003). Te Greek LFP rate of women increased from 29.8 per cent in 1980 to 35.1 per cent in 1989 and to almost 39 per cent in In te same period and in te pursuit of teir labor market objectives, Greek women acquired ger levels of formal education, delayed marriage and reduced cldbearing, (Kanellopoulos and Mavromaras, 2002). Despite recent increases owever, Greek women display low levels of LFP wen tey are compared wit teir counterparts in oter European countries and te USA. Table 1 presents information regarding te LFP of women between te ages of 15 and 64 in te USA, te EU and Greece. Wit regard now to married women, Greece again exbits low rates of LFP. According to Megr et. al. (1989) te results of te Greek Labor Force Survey for 1981 indicate tat te LFP of married women is ger tan tat for women in general, as expected, muc lower tan te corresponding rate for Greek men and it does not exceed te proportion of 36 per cent of te female active population. Evidence from publised data also reveals tat for te most productive ages, years, te LFP of married women in 1983 was about 41% wle in 1991 it exceeded te 50% mark, (Kanellopoulou 1994). Finally, in te period Greece ranks 8 t among EU-11 member-states in te LFP rate of married women (Pietro-Rondriquez J., Rodriquez-Gutierrez C. 2003). Te average LFP rate of married women in te 11 member states was about 53 per cent, wle Greece exbited a rate of about 44 per cent. Table 1. Labor Force Participation Rates of Women, aged years United States European Union Greece Source: OECD Employment Outlook 2002

3 J. DAOULI, et al., Sout-Eastern Europe Journal of Economics 2 (2004) Te LFP of Greek women and te interrelated issues ave not been studied to any substantial extent. Ts is mainly due to te lack of informative data sets available to researcers. Data sets amenable to labor market studies ave become available only recently. Te NHBS is a case in point and contains te minimum data requirements for carrying out researc on LFP issues. Te survey contains information on earnings and income, ours of work as well as personal, family and employment caracteristics of women. Te major objective of ts paper is to investigate te labor supply scedule of married women in Greece. Te investigation concerns te Labor Force Participation (LFP) of married women, te offered market wage and te amount of ours supplied in te market. For ts purpose, we will employ cross-sectional data from te National Houseold Budget Survey (NHBS) and contemporary econometric tools. Te paper is structured as follows. Section 2 briefly reviews te teoretical and empirical issues concerning labor force participation. Section 3 presents information in te form of descriptive statistics of LFP of married women as well as of te related economic and demograpc variables. Section 4 reports te results of estimation of te participation, wage and labor supply equations. A discussion of te results is also contained in ts section. Major findings are summarized in Section 5, wc also includes suggestions for furter researc. 2. Teoretical and Econometric considerations One of te major objectives of most labor supply studies is te identification and investigation of te factors tat influence te labor force participation decision. As a rule, te teoretical approac is based on te utility-maximizing framework of consumer teory and time allocation. More specifically, te labor supply beavior of an individual is determined primarily by te relationsp between te market wage offered and te sadow wage, i.e., te value of time in non market activities. Te first wage pertains to te market demand function wle te second to te individual s supply function. 1 Te empirical findings in te relevant literature indicate tat women s labor supply beavior, i.e., decision to participate and labor supply in term of ours allocated to work, is responsive to canges in wage rates, uman capital caracteristics, non-labor income, usband s wage, if te woman is married, and to te presence of cldren, particularly of pre-scool age 2 in te ouseold. 1. For a detailed presentation of tese issues see (Becker, 1965; Heckman, 1974; Smit, 1980; Killingswort 1983; Mroz, 1987). 2. Killingswort and Heckman (1986) provide a compreensive review of te related literature.

4 22 J. DAOULI, et al., Sout-Eastern Europe Journal of Economics 2 (2004) Te simultaneous caracter (participation and te extent of labor supply) of te labor supply decisions as been addressed in a large number of studies and te same olds for te econometric issues associated wit sample selection, i.e., te fact tat wages offered are not observed for non participants. Te early work of Heckman on te development of sample selection bias-corrected tecniques played an important role in more recent labor supply studies (see Heckman, 1974, 1979; Killingswort, 1983; and Killingswort and Heckman, 1986, for an extensive and detailed review of te issues involved). Assuming tat tere are no fixed costs of entering and leaving te labor market and tat te ours of work vary freely, te married woman s labor supply beaviour is saped by two relations. Te first is associated wit te market wage rate, w im, and te second wit te sadow value of time in non market activities, w ir. Following Franz (1985) we assume tat te market wage w im is determined by a set of independent variables X wi, w im = X wi â w + å wi (1) were, å wi is a normally distributed random error. Te sadow wage equation (at zero ours of work), w ir, is also a linear function of X ri : w ir = X ri â r + å ri (2) were, å ri represents a random normally distributed error. Taking into consideration te decision rule for participation in te labor market, i.e., w im > w ir, te reduced form equation of te labor supply function is as follows: w im = i = { X wi â w + å wi 0 if w ri $ w im X â + å if w ri < w im (3) were, i represents working ours for te sub-sample of working women, X is te set of te explanatory variables associated wit i, â and å 3 are te corresponding parameters and errors, respectively. Given te simultaneous caracter of bot, te 3. Note tat å = å wi - å ri. Since å wi and å ri are jointly normally distributed, ten å and å wi are also jointly normally distributed.

5 J. DAOULI, et al., Sout-Eastern Europe Journal of Economics 2 (2004) participation decision and te extent of labor supply, if te observed variables in X wi include one variable wc does not appear in X ri, ten te system of equations (1) and (2) is identified (Wooldridge, 2002). Te beavioral model of te woman s labor supply scedule represented in (3) cannot be estimated wit simple regression tecniques since market wage rates for non-working women are not observed. Terefore, if te estimation of te wage rate equation in (3) is restricted to working women only, ten te conditional expected value of is given by: E ( wim > 0) = E( wim ε > X â ) = X â + E( å å > X â ) wi ρ w = X wi âw + M i σ φ( z were, i ) ( Χβ ) M i is te Inverse Mill s ratio and Ζi. Φ( Ζi ) σ w wi (4) * i Tus, te market wage function to be estimated is: = ρ + w w im X wi âw M i + σ å mi (5) Since te value of M i is not known, equation (5) cannot be estimated via conventional metods. A consistent estimate of M i can be obtained by maximizing te probability tat a married woman is working, conditional on te variables included in Z i. Te estimated coefficient Mˆ for eac woman in te entire sample is ten included as an i additional explanatory variable in te market wage rate equation (Wooldridge, 2002). Given te estimation results of te wage equation (5), we can proceed wit te estimation of te structural labor supply function for te entire population of married women, (i.e., working women and non-working ones). Since te distribution of ours is truncated at zero ours, we use te Tobit maximum likeliood tecnique to estimate te labor supply function (Franz, 1985; Averrett and Hotckiss, 1997). Ts procedure requires te use of a latent variable, wc is determined as follows: * i = X = â + å 0 * i if if * > 0 0 * i (6)

6 24 J. DAOULI, et al., Sout-Eastern Europe Journal of Economics 2 (2004) Maximizing te underlying likeliood function of expression (6) we are able to obtain estimates of and, as well as te probability of participation: L 2 * ( β, σ ) = P( = 0) P( = ) = 0 0 > 0 X Φ β σ > 0 X β φ σ (7) Finally, in order to obtain an estimate of te implied labor supply elasticity wit respect to te market wage we first need to distinguis te determinants of te labor supply decision from te determinants of te market wage rate. Ts is necessary since we lack reliable data on te determinants of market wages, i.e., experience, vocational training, etc. Tus, te parameter estimate of te predicted market wage in te Tobit equation is expected to be biased. To avoid ts problem and following Heckman (1980) and Franz (1985) we need to determine one variable tat clearly affects market wages and ten use it as a proxy for wages in te labor supply equation. 4 In ts case, te market wage equation and te labor supply equation can be estimated independently. 3. Data and variable definitions Te data utilized in te empirical analysis are extracted from te NHBS, wc is carried out by te National Statistical Service of Greece. Tey refer to women between te ages of 15 and 64, married, 5 wit spouse present and reporting (te spouse) positive income from work in te previous year (1998). We ave dropped from te sample a) tose women wo were employed in temporary/part-time jobs or in farming in 1998, since we do not know te extent of ts employment, i.e., ours of work on a yearly basis, b) self-employed women, and c) students. Te total number of women dropped from te sample, for te reasons above, is Tus, a 4. Heckman, (1980) and Franz, (1985) use experience and vocational training, respectively. In te present application we use education, wc is readily available. 5. Tere is no doubt tat te decisions regarding marriage and labor supply are interrelated. However, te simultaneous investigation of tese decisions requires te use of panel data wc are not, at present, at our disposal. 6. Ts figure also includes tose married women wo worked in 1998 but didn t provide information on eiter teir annual earnings from labor or teir working ours.

7 J. DAOULI, et al., Sout-Eastern Europe Journal of Economics 2 (2004) woman is classified as working if se as positive income from labor in 1998 and satisfies te aforementioned sample selection criteria. After tese adjustments are made, te data set consists of 606 married women wo worked in full time jobs in 1998 and of 854 non-working ones. 7 Te comparison between working and nonworking women reveals tat working women are younger, more educated, ave more cldren below te age of 13, ave less property income and more capital income tan non-working ones and most of tem live in te greater Atens conurbation. Te participation decision is represented by a dummy variable (PART) wc takes te value of 1 if te married woman reported positive income from labor in 1998 and zero oterwise, (Killingswort, 1983 pp. 236 and 239). In ts application te important ours of work variable (HOURS) is measured by te variable usual ours of work on a weekly basis reported in te survey. Te equally important market ourly wage rate (WAGE) for working women was obtained by first transforming annual earnings from labor to weekly earnings from labor and ten by dividing troug by HOURS. 8 Altoug ts indirect metod of obtaining wage rates reduces te efficiency of te estimates of labor supply parameters, it does not introduce bias or inconsistency (Killingswort, 1983). In te Mincerian tradition, te (logaritm of) market wage rate is assumed to depend on scooling and labor market experience. In our case, scooling is measured by te education variable (EDUC), wc is a continuous variable of te completed years of scooling. Furtermore, empirical evidence suggests tat a married woman wit ger education is more likely to be in te labor force. Since information on labor market experience, as suc, is not available we will attempt to capture te effect of labor market experience on wages by employing te age (AGE) variable, wc is readily available. A large number of ouseolds in our sample, more tan 50 per cent, report income from sources oter tan work. Ts non-work income is composed, mainly, of property income (PROPERTY) and capital income, (CAPITAL). Property income consists exclusively of income from rents wle capital includes income from interest payments and dividends. We examine separately te two sources of non-work income because, in te static framework adopted, property income is a more permanent source of income tan capital income and, tus, only te former is expected to ave a significant effect on labor supply decisions. In general, and if we assume tat 7. A non-working married woman is one wo does not report income from labor in 1998 but satisfies all te remaining selection criteria, i.e., age between 15-64, spouse present and reporting positive income from work. 8. Overtime pay is included in annual earnings from labor but ours of overtime are not available.

8 26 J. DAOULI, et al., Sout-Eastern Europe Journal of Economics 2 (2004) leisure is a normal good, an increase in non-work income is expected to reduce bot te likeliood of participation in te labor market and te extent of labor supply. In te context of a ouseold/family labor supply model, te income earned by te usband plays an important role in determining te labor supply coices of s wife. To capture ts interrelationsp between te labor supply decisions of te family members we ave employed te variable wc represents te usband s annual earnings from labor, (HUSLAB). It is reasonable to expect a negative relationsp between HUSLAB and te likeliood of te wife being in te labor force, ceteris paribus. Te composition of te ouseold is also expected to be an important factor in te analysis of te wife s time allocation decisions. We ave constructed te variable (NKIDL13) to represent te number of cldren of less tan 13 years old in te ouseold. Ts variable is expected to exert a negative impact on te likeliood of labor force participation of married women and especially of young married women, ceteris paribus. Finally, te region of residence is also expected to affect preferences for work. Publised data for Greece sow tat te LFP for women exbits a drastic increase in major conurbations (Kanellopoulos and Mavromaras, 2002). In te present application, te dummy variable (ATTIKI) is constructed to reflect regional effects. It takes te value of 1 if te ouseold is situated in te prefecture of Attiki and zero oterwise. Table 2 presents te sample means of all te exogenous and endogenous variables wc are used for te estimation of te participation equation, te market wage function and te ours of work equation.

9 J. DAOULI, et al., Sout-Eastern Europe Journal of Economics 2 (2004) Table 2. Descriptive Statistics Non-Workers Workers Total Sample Number of observations PART (0.49) HOURS (9.35) WAGE ( ) NKIDL (0.91) 0.82 (0.86) 0.76 (0.89) PROPERTY 2a (545953) (413614) (495590) CAPITAL 2a (722213) ( ) ( ) EDUC (3.73) (4.18) (4.33) AGE (10.52) (7.69) (9.49) HUSLAB 2a ( ) ( ) ( ) ATTIKI (0.50) 0.63 (0.48) 0.56 (0.56) Source: National Houseold Budget Survey-Greece, In Years, 2 In dracmas, 3 Percentage * Standard deviations in parenteses.

10 28 J. DAOULI, et al., Sout-Eastern Europe Journal of Economics 2 (2004) Empirical results Te statistical package LIMDEP 7.0 was used for te estimation of te model. Table 3 presents te results of te participation equation and Table 4 te estimated wage and labor supply functions for married women in Greece. Te participation model predicts correctly te status of 71.5% of te sample, i.e., of bot workers and nonworkers. For te sub-sample of workers te correct predictions were relatively lower (58.1%) tan tat of non workers (81.1%). For te participation equation a goodness of fit measure based on te likeliood ratio test statistic (or McFadden s R 2 ) is computed to be 0.181, wc represents a satisfactory fit of our model. For te wage selection regression te conventional R 2 measure is indicating a rater poor fit. It is owever, muc better tan tose reported in te international literature were, R 2 values for wage-generating functions seldom exceed te 10% mark. A mis-specification Cragg test as also been conducted and it did not reveal any significant specification problems of our labor supply-tobit model. Several factors appear to influence te married woman s labor force participation decisions, sown in Table 3. Te presence of cldren of less tan 13 years old tends, as expected, to increase te ousewife s value of time in te ouseold, tus affecting negatively te probability of participation in te labor market in periods in wc te need for cldcare or ousework is g. 9 Te relevant parameter estimate as te expected sign and it is statistically significant. Te estimated coefficients on education and age are also statistically significant wit te expected sign. Ts is in agreement wit te finding reported by Megr et al., (1989) and supports te ypotesis tat existing uman capital endowments and fertility issues affect te probability of participation in Greece. 10 One can also observe tat usband s earnings ave a negative effect on te wife s probability of participation in full time employment, indicating tat intra-family time allocation decisions are interdependent. As expected, property income as a strong negative effect on participation wle te effect of capital income is positive and ardly significant. 11 Tree factors exbit a strong positive effect on participation. First, more educated married women are more likely to be employed in a full time job tan less educated ones. Second, a ger probability is also observed for women tat live in te greater Atens conurbation and, trd, te age-participation profile does exist and it is non-linear. 9. More age groups for cldren and more regions were originally incorporated. However, a joint statistical test revealed tat a finer age and region classification could not be supported by te data. 10. In Megr et al., (1989), p. 400, Table1, te estimated coefficients for te effect of education and cldren of less tan six years on participation are and , respectively. 11. Neverteless, te positive sign of te CAPITAL variable is interesting and it certainly requires furter investigation.

11 J. DAOULI, et al., Sout-Eastern Europe Journal of Economics 2 (2004) Table 3. Estimated Results of te Participation Equation Variables Participation Equation Marginal Effects Estimated Coefficients t-values Estimated Coefficients Intercept *** *** NKIDL * -.033* PROPERTY *** -.102*** CAPITAL *.018* EDUC ***.053*** AGE ***.090*** AGE squared *** -.001*** HUSLAB *** -.016*** ATTIKI **.057** Observations Log-Likeliood Mc Fadden C squared (8d.f.) Source: National Houseold Budget Survey-Greece, Asterisks indicate statistical significance, *** at 1%, ** at 5%, * at 10%.

12 30 J. DAOULI, et al., Sout-Eastern Europe Journal of Economics 2 (2004) Turning now to te estimated results of te wage selection equation, Table 4, we observe tat bot education and age play a statistically important role in wage formation. Te estimated coefficients ave te expected signs and indicate tat years of education and age exert a strong positive effect on wage rates. Lastly, it sould be noted tat our test for te presence of selectivity bias could not be rejected by te data indicating tat suc a bias does really exist. Te g and positive value of te selection term indicates tat te unobserved variables tat raise te probability of participation also increase te market wage (Heckman, 1980, p.230). Table 4. Estimated Results of te Wage amd Labor Supply Functions Variables Hourly Wage Hours Worked Marginal Effects Estimated t-values Estimated t-values Estimated Coefficients Coefficients Coefficients Intercept *** *** *** NKIDL ** ** PROPERTY *** *** CAPITAL **.2585** EDUC *** ** 1.635*** * AGE *** *** 3.232*** AGE squared *** *** -.041*** HUSLAB *** -.493*** ATTIKI *** 2.773*** Lamda *** R-squared.268 Log likeliood Source: National Houseold Budget Survey-Greece, Asterisks indicate statistical significance, *** at 1%, ** at 5%, * at 10%.

13 J. DAOULI, et al., Sout-Eastern Europe Journal of Economics 2 (2004) Te Tobit estimates of te labor supply equation are also reported in Table 4. Te estimated coefficients ave te expected signs and tey are statistically significant. Tey sow tat te extent of labor supply is positively affected by age, education and region of residence and negatively affected by non-work income, usband s earnings and te number of cldren of less tan 13 years old. Te two age variables ave te expected signs indicating tat te lifetime labor supply profile is non-linear. Furtermore, usband s earnings exert a negative impact on te wife s labor supply scedule, indicating tat significant substitutabilities do exist witn te Greek ouseolds. From te obtained results we were able to calculate a measure of te implied labor supply elasticity wit respect to ourly wage rates. Te elasticity is evaluated at sample means and is found to be Ts value does not differ considerably from tose elasticities reported in te review article by Killingswort and Heckman (1986) and it provides evidence for te rater inelastic caracter of te labor supply function of married women in te Greek labor market. Finally, column 3 of Table 4 presents te marginal effects of te labor supply function at sample means. Te marginal effects are obtained by multiplying te labour supply estimates of Table 4 (column 2) by an adjustment factor. Ts adjustment factor is estimated at te mean values of all te explanatory variables and sows te estimated probability of observing a positive response, (>0). In our case, te adjustment factor is equal to 0.44, wc means tat te estimated probability of a woman being in te labor force is about 0.44, (Wooldridge, 2002). 13 Tus, te effect of one more year of scooling on ours supplied is about 1.6 additional ours per week. Similarly, a 10 per cent increase in property income reduces te wife s expected supplied ours by 3.3 ours per week. 5. Conclusions Te main purpose of ts paper was to apply te well establised teoretical and empirical metodology of labor supply analysis, developed over te last 30 years, in order to estimate te labor supply scedule of married women in Greece. Te major 12. Te elasticity estimate is calculated according to te instructions of J. Heckman (1980, p. 228, Table 5.3), i.e., by dividing te EDUC coefficient in te wage function into te EDUC coefficient in te labor supply function, and dividing by average weekly ours supplied by working women (36.61 ours). See also Killingswort and Heckman, 1986; Franz, From equation 6, te partial effects at sample means are given by: E( ) ˆ β ^ i Xi X φ = ˆ β = ( > 0 ) ˆ β ˆ σ Prob X Xi

14 32 J. DAOULI, et al., Sout-Eastern Europe Journal of Economics 2 (2004) motive for ts project, apart from te extremely limited amount of related researc in Greece, is tat te required micro data ave only recently started to be collected in ouseold budget surveys, wc are carried out in a systematic way by te National Statistical Service of Greece. Using data from te 1998/99 NHBS and modern econometric tools we were able to identify te major factors tat influence female labor market participation, te offered market wages as well as te sensitivity of te labor supply to market wage rate canges, in Greece. Our findings are teoretically plausible and, in general, consistent wit earlier labor supply studies for married women worldwide. Our results indicate tat te labor force participation decision of Greek married women is saped by uman capital variables, i.e., education and age, by te composition of te ouseold and, in particular, te presence of cldren of primary scool age, te existence of property income, te size of usband s labor income and by geograpcal factors. Wit respect to te market wage, our findings indicate tat tey are influenced by te same uman capital variables and by unobserved factors, i.e., te presence of selectivity could not be rejected by te data. Finally, te estimation of a standard labor supply equation revealed tat ours of work in full time paid employment by married women are determined by te same factors tat sape te participation decision. To te extent tat an increase in female labor force participation is a target of employment policy, our results warrant furter investigation of measures tat reduce te time requirements of cld care in te ouseold. Public and private daycare centers and te all day primary scool, recently introduced by te Ministry of Education in te Greek public scool system are cases tat deserve furter investigation. Te same olds for te extent of labor supply in full time paid employment. Finally, te estimated labor supply elasticity indicates tat an increase in te market wage will cause an increase in ours supplied, albeit less tan analogous, i.e., labor supply is wage inelastic. Undoubtedly, more complete and informative labor and employment data sets are needed in te case of Greece. In addition, since te decisions concerning various labor supply aspects are not of a static caracter, empasis sould be placed on te more dynamic panel data bases, currently created in Greece and te EU. Te use of panel data is expected to sed more ligt on te interdependencies of labor supply and time allocation, wit issues concerning marriage, fertility, education, retirement, etc.

15 J. DAOULI, et al., Sout-Eastern Europe Journal of Economics 2 (2004) References Averett, S., and J. L., Hotckiss, 1997, Female Labor Supply wit a Discontinuous, Nonconvex Budget Constraint: Incorporation of a Part-Time/Full-Time Wage Differential, Te Review of Economics and Statistics, 79, Becker, G. S., 1965, A Teory of te Allocation of Time, Economic Journal, 75, Franz, W.,1985, An Economic Analysis of Female Work Participation, Education, and Fertility: Teory and Empirical Evidence for te Federal Republic of Germany, Journal of Labor Economics, 3, Greene, W., 1995, LIMDEP version 7.0, Bellport, NY: Econometric Software. Heckman, J. J., 1974, Sadow Prices, Market Wages and Labour Supply, Econometrica, 2, Heckman, J.J., 1979, Sample Selection Bias as a Specification Error, Econometrica, 47, Heckman, J.J., 1980, Sample Selection Bias as a Specification Error wit an Application to te Estomation of Labor Supply Functions, in Smit, J., (ed.), Female Labor Supply, Teory and Estimation, Te Rand Corporation. Heckman, J. J., 1993, Wat as been learned about labor supply in te past twenty years?, AEA Papers and Proceedings, 83, Kanellopoulos, C., and Mavromaras, K., 2002, Male- Female Labour Market Participation and Wage Differentials in Greece, Labour, 16, Kanellopoulou, K., 1994, Human Resources: Population, Labour Market, Education, KEPE, Atens, 10, (In Greek). Killingswort, R. M., 1983, Labor Supply, Cambridge University Press. Killingswort, R. M., and J. J. Heckman, 1986, Female Labor Supply, in Ansenfelter, C. O., and Layard, R., (eds.), Handbook of Labor Economics, 1, Nort-Holland. Megr, C., Y., Ioannides, C., Pissarides, 1989, Female Participation and Male Unemployment Duration in Greece, Evidence from te Labour Force Survey, European Economic Review, 33, Mroz, A. T., 1987, Te Sensitivity of an Empirical Model of Married Women s Hours of Work to Economic and Statistical Assumptions, Econometrica, 55, Pietro-Rondriquez J., and Rodriquez-Gutierrez C., 2003, Participation of Married Women in te European Labour Market and te Added Worker Effect, Journal of Socio-Economics, 32, Pollak A. R., 2003, Gary Becker s Contributions to Family and Houseold Economics, Review of Economics of te Houseold, 1, Wooldridge M. J., 2002, Econometric Analysis of Cross Section and Panel Data, London: Te MIT Press, England.