Measuring the effect of gender segregation on the gender gap in time-related underemployment

Juan Acosta-Ballesteros; María del Pilar Osorno-del Rosal; Olga María Rodríguez-Rodríguez

doi:10.1186/s12651-021-00305-0

Measuring the effect of gender segregation on the gender gap in time-related underemployment

Acosta-Ballesteros, Juan; Osorno-del Rosal, María del Pilar; Rodríguez-Rodríguez, Olga María 2021-12-01 00:00:00 This paper focuses on the impact that gender segregation in the labour market exerts on the underemployment gender gap for young adult workers in Spain. In order to analyse the relative importance of segregation in this gap, we develop a methodology based on two counterfactual simulations that provides a detailed decomposition of the gap into endowments and coefficients effects as well as the interaction of these effects. To the best of our knowl‑ edge, we are the first to perform a decomposition using bivariate probit models with sample selection. Using annual samples of the Spanish Labour Force Survey 2006–2016, the results show that working in female‑ dominated occupa‑ tions or industries hinders working as many hours as desired, especially for women. Furthermore, we conclude that the gender gap in underemployment is mainly due to the different distribution of male and female workers across occupations and industries. Additionally, the different impact by gender that working in the same gender ‑typing jobs exerts on the risk of underemployment contributes to widening the gap. Keywords: Gender gap, Occupational and Industry segregation, Time‑related underemployment, Counterfactual simulations, Decomposition analysis JEL Classification: J16, J22 1 Introduction industry segregation may be an important factor, as sug- Time-related underemployment, which refers to those gested by Barret and Doiron (2001), since a higher under- workers who would like to work more hours than avail- employment rate has been linked to female-dominated able, is a persistent problem in labour markets and the occupations and industries (Kjeldstad and Nymoen Spanish one is no exception. Moreover, this problem 2012a, Kjeldstad and Nymoen, 2012b, and Kamerāde and increased during the Great Recession in many countries Richardson, 2018). Moreover, as Spain has experienced (Bell and Blanchflower 2013; Acosta-Ballesteros et al. higher levels of segregation than other European coun- 2018). tries in a persistent way (Iglesias-Fernández et al. 2012), Furthermore, women experience this situation more analysing the effect of segregation on the underemploy - often than men do (Weststar 2011; Kjeldstad and ment gender gap in the Spanish labour market emerges Nymoen 2012a, b; Vuluku et al. 2013; Acosta-Ballesteros as an interesting research issue. et al. 2018), therefore, we must pay attention to the rea- To the best of our knowledge, only Vuluku et al. (2013) sons for this gender gap. Particularly, occupational and have tried to explain the underemployment gender gap, but they did not include any occupational and industry segregation indicators in their study. Thus, to overcome *Correspondence: orguez@ull.es this shortcoming in the literature, the main objective Departamento de Economía, Contabilidad y Finanzas, Universidad de La of this article is to carry out an in-depth analysis of the Laguna (ULL), La Laguna ( Tenerife), Spain underemployment gender gap. Specifically, we intend Full list of author information is available at the end of the article © The Author(s) 2021. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/. 22 Page 2 of 16 J. Acosta‑Ballesteros et al. to test whether segregation plays an important role in some employed people would like to work more hours explaining it, as well as to quantify how much is due to than available. men and women working in different industries and The demographic and job factors that determine time- occupations, and how much is due to men and women related underemployment have been previously analysed facing different underemployment risks when they work in the literature (Hakim 1997; Weststar 2011; Prause and in the same gender-typing jobs. Dooley 2011; McKee-Ryan and Harvey 2011; Kjeldstad To do this, first, we quantify the effect of occupational and Nymoen 2012a, 2012b; Wilkins 2006; Acosta-Ball- and industry segregation on workers’ underemployment esteros et al. 2018). Particularly, significant differences in risk using a detailed measure of gender segregation. Esti- underemployment have been found across occupations mating this impact through bivariate models with selec- and industries (Kjeldstad and Nymoen 2012a, 2012b; Val- tion enables us to handle the potential sample selection letta et al. 2016). However, very few studies have linked bias due to estimating the probability of underemploy- these differences to occupational and industry gender ment just for employed people. Second, as we are not segregation in the labour market. u Th s, Kjeldstad and aware of a methodology that allows decomposing a gap Nymoen (2012a) and Kjeldstad and Nymoen (2012b) using this kind of model, we develop one that is inspired find a higher underemployment risk in those occupations by the Fairlie technique (1999, 2005, and 2017). It is based and sectors that are traditionally female-dominated, with on two counterfactual simulations that provide a detailed a stronger effect for men, although they do not include decomposition of the gap into effects due to workers hav - specific variables for segregation in their econometric ing different characteristics, effects due to these charac - model. Kamerāde and Richardson (2018) consider seg- teristics having different returns, and the interaction of regation measures in their analysis, and they also find a both these effects. higher likelihood of underemployment in female-dom- We focus on young workers because this collective is inated occupations; however, this effect is not so clear especially affected by underemployment. Data from the across industries. Additionally, Dueñas-Fernández et al. Spanish Labour Force Survey (LFS) indicate that in 2017 (2016), who do not focus specifically on this issue, ana - the underemployment rate for workers under 35 was lyse involuntary part-time work in Spain and find that 14.7%, while the figure was 8.5% for workers older than segregation, especially occupational, is strongly related to 34 years old. Additionally, by looking at people right at part-time work (particularly for women). the beginning of their careers we can avoid many of the There are several reasons that explain higher under - cumulative advantages/disadvantages that people may employment risks in female-dominated occupations and have experienced throughout their careers. Thus, focus - industries. In this sense, as Kamerāde and Richardson ing on young workers allows us to have a current view of (2018) point out, women are mainly employed in labour underemployment patterns and of the gender gap in it, intensive jobs where employers can change the number avoiding possible gender differences from the past. of hours their employees work to adapt to fluctuations This paper is organised as follows. It begins with the in demand. Therefore, part-time or short-schedule jobs conceptual framework on time-related underemploy- are more likely to be found in female-dominated occupa- ment and puts forward our working hypotheses. The next tions, which may lead to underemployment. Moreover, section describes the methodological approach used. female-dominated occupations usually require low quali- Then, the data and variables used in the econometric fications. In addition, female workers tend to cluster in model are presented. This is followed by the results, while industries that offer comparatively low payment for the a discussion of these findings is provided at the end. same level of qualification, such as in education, health and social work activities (Boll et al. 2016). This pattern 2 Conceptual framework and hypotheses also translates into a frequent desire to work more hours. According to neoclassical theory, individuals can choose Conversely, male-dominated occupations are typically their working hours freely from a continuous time dis- characterised by better-paid jobs and are usually related tribution; these hours are chosen by maximising a util- to more stable, full-time contracts (Hegewisch et al. ity function subject to a particular budget constraint. 2010). Furthermore, male workers are overrepresented Nevertheless, employers and trade unions’ decisions, in industries that offer high rewards for the same level the degree of labour mobility and economic conditions of qualification (particularly manufacturing). Therefore, determine the actual hours offered to employees (Simic male-dominated occupations and industries often lead to 2002). Therefore, workers’ preferred and actual hours low underemployment rates. may not coincide, so some individuals will work either Despite these arguments suggesting the important role more (overemployed) or less (underemployed) than they that occupational and industry segregation plays in the want. Thus, time-related underemployment means that likelihood of underemployment, accurate estimates of Measuring the effect of gender segregation on the gender gap in time‑related underemployment Page 3 of 16 22 its impact have not been achieved in the aforementioned is mainly due to the different distribution of male research. In this article, we overcome this shortcoming and female workers across occupations and indus- using a more suitable estimation strategy. Specifically, we tries. propose and test the following hypothesis: Hypothesis 2b: The different impact that working in certain occupations and industries exerts on the risk Hypothesis 1: working in female-dominated occu- of male and female underemployment contributes to pations and industries implies a higher probabil- widening the gender gap. ity of time-related underemployment than being employed in male-dominated ones, both for men and women.3 Methodology As a first step, to analyse the effect of occupational and As women face a higher risk of underemployment industry segregation on time-related underemploy- than men do, there is a gender gap regarding this handi- ment, we estimate two bivariate probit selection mod- cap and occupational and industry segregation may els (Greene 2012), one for men and another for women. have an important impact on it. Furthermore, this effect These models enable us to handle the potential sample may be partially due to the uneven distribution of men selection bias due to estimating the probability of under- and women across different jobs, as suggested by Bar - employment just for employed people, as Acosta-Balles- rett and Doiron (2001). Additionally, differences in the teros et al. (2018) have already shown. returns that working in female or male-dominated jobs ∗ ∗ Let us define y and y as the latent variables reflecting 1 2 imply should also be considered. These authors, as a sim - the likelihood of being underemployed and employed, ple exercise, give women the average male distribution respectively. Thus, the model can be specified as follows: across occupations and industries and conclude that the ∗ ∗ main reason that explains women being involuntary part- y = x γ + ε , y = 1 if y > 0, 0 otherwise i1 1 i1 i1 i1 i1 timers more often than men is simply being employed in (1) different industries and occupations. ∗ ∗ y = x γ + ε , y = 1 if y > 0, 0 otherwise i2 2 i2 i2 i2 i2 Interestingly, previous research highlights the fact that (2) men may benefit from their minority status in female- with (y , x ) observed only when y = 1. i2 i1 i1 dominated jobs in several ways (Simpson 2004). In this In these equations, y indicates if worker i is underem- i1 sense, as reviewed in Lupton (2006), men progress more ployed and y if the individual is employed; row vector i2 quickly than women do to senior positions avoiding the x contains the variables explaining underemployment; i1 problem of the “glass ceiling” inherent in vertical segre- x reflects the variables determining employment. As i2 gation. Additionally, men may be channelled into cer- usual, the independent variables that have a qualitative tain specialties in occupations that are regarded as more nature are included in the model as dummy variables or appropriate to their gender. As a third advantage, men as groups of them. Finally, ε and ε are the error terms, i1 i2 are paid more than women are in female-dominated which follow a bivariate normal distribution with mean occupations (Torre 2018). By contrast, women may face zero, variance equal to 1 and covariance ρ negative outcomes in male-dominated jobs (Simpson To test if working in female-dominated occupations 1997, 2000). Thus, for example, as Martin and Barnard and industries implies a higher probability of under- (2013) find, formal and covert organisational practices, employment than working in male-dominated ones which maintain gender discrimination and bias, are the (Hypothesis 1), we analyse the estimated marginal effects main challenges that women face. These arguments may of occupational and industry segregation on the prob- also apply regarding underemployment, so female work- ability of underemployment. Since the model is bivariate ers may face a higher risk of underemployment than men with selection, these partial effects (like those regarding both in female and male-dominated jobs. the rest of variables) are obtained using the conditional Nevertheless, to the best of our knowledge, only probability of underemployment given employment. Vuluku et al. (2013) have tried to identify the reasons In addition, the marginal effects on the probability of behind the underemployment gender gap, though they employment are computed using the selection equation. do not include any measure of gender segregation in their To simplify notation, we redefine the variables and analysis and use univariate models, which can lead to coefficients in Eqs. (1) and (2) as follows. The variables biased estimations. We fill this gap in the literature using a new methodology that allows us to propose and test the following hypotheses: Hypothesis 2a: The gender gap in underemployment The interdependency caused by groups of dummies has been taken into account in order to calculate the marginal effects. 22 Page 4 of 16 J. Acosta‑Ballesteros et al. considered in both equations for individual i are gath- non-linear models, the independent contribution of one ered in x , which is a row vector including vectors x and variable to the gap depends on the value of the other i i1 x . Additionally, vector β contains the estimated values variables, which may imply a path dependence problem. i2 1 for γ (γ ) and takes value zero for those variables in x Moreover, Fairlie methodology does not identify the 1 1 i2 which are not included in x . In a similar way, β includes coefficients effect corresponding to a specific variable, i1 2 components equal to zero for those variables consid- which is required to test Hypothesis 2b. ered in Eq. (1) but not in Eq. (2). Thus, x β ≡ x γ and The aggregate decomposition in our methodol - i 1 i1 1 x β ≡ x γ . ogy, which is a direct extension of Fairlie’s, is defined i 2 i2 2 According to this notation, the estimated probability of by Eqs. (4) to (6), where E reflects the endowments being underemployed conditioned to being employed for effect using as weights women’s coefficients, and C individual i is: quantifies the coefficients effect using as weights men’s characteristics: j j BVN(x β , x β , ρ ) j j i i j 1 2 F x β , x β , ρ = i i i (3) Gap = C + E 1 2 M W (4) �(x β ) M M M M M where Φ is the cumulative standard normal distribution F x β , x β , ρ ∀Men i 1 i 2 C = function and BVN is the joint cumulative distribution of (5) M W M W W the bivariate normal. Superscript j refers to men (M) or F x β , x β , ρ ∀Men i 1 i 2 women (W). As stated above, our main objective is to identify the most relevant factors explaining the gender gap in M W M W F x β , x β , ρ ∀Men i 1 i 2 underemployment and, more specifically, to test if gen - E = der segregation accounts for an important portion of (6) W W W W F x β , x β , ρ it. To achieve this goal and test Hypotheses 2a and 2b, a ∀Women i 1 i 2 detailed decomposition of the gap is required. The traditional Oaxaca-Blinder two-fold decomposi - Summations in (5) and (6) are across the subsample of tion (Blinder, 1973 and Oaxaca, 1973) of the gap into the employed, as we decompose differences in the average endowments (portion of the gap due to group differences predicted probabilities of being underemployed condi- in observable characteristics) and coefficients effects W M tioned to being employed. Thus, N and N indicate the (the “unexplained” portion of the gap) cannot be applied sample size for employed women and men, respectively. because our model is not linear. Previous research (Even An alternative decomposition (Eq. 7) with each com- and Macpherson, 1990; Doiron and Riddell, 1994; Fairlie ponent evaluated using as weights the other gender coef- 1999, 2005, 2017; Yun, 2004, 2008; Powers et al., 2011; ficients or endowments is also possible. However, we do and Bazen et al., 2017) has decomposed the gap in pro- not define and explain it here because it is symmetric to bit and logit models, with the Fairlie and Yun techniques this one. being the two most widely applied. However, as we esti- mate a nonlinear model with two equations, we develop a Gap = E + C M W (7) new procedure to decompose the gap, which extends the To obtain a detailed decomposition, we develop a Fairlie technique to this kind of model. We have chosen methodology based on two counterfactual simulations the Fairlie approach as our starting point because it uses that identify the contribution of each variable to both a non-linear function to obtain the gap decomposition, E and C . These simulations can be used together to while in the Yun procedure, the curvature of the corre- W M approximate the total impact of a specific variable on the sponding function is not considered. underemployment gender gap. According to the Fairlie technique, the contribution of The first one provides a detailed decomposition of the each observable variable to the explained portion of the endowments effect and has been designed for discrete gap is equal to the change in the average predicted prob- variables (as most of the variables in the labour market ability from replacing (for instance) the female distribu- tion with the male distribution of that variable (keeping constant the rest). The procedure he proposed is match - Fairlie does not focus on the "unexplained" portion of the gap because of the ing one-to-one individuals in the female and male sub- difficulty in interpreting results. As we explain below, this shortcoming can be samples and switching the distributions of variables overcome using the Kim (2013) methodology. sequentially from a woman to a man. Nevertheless, the For a continuous variable, we propose equalling the distribution of fre- quencies by gender as described, but considering the variable is discrete order of switching is potentially important because in with a very large number of categories. Measuring the effect of gender segregation on the gender gap in time‑related underemployment Page 5 of 16 22 literature). It is inspired by Fairlie, who pointed out that differences in Taylor expansion remainders (see Appen - a potential solution to the path dependence problem “is dix 1 in Additional file 1). to estimate each contribution by switching the variable To calculate the detailed decomposition of the coeffi - of interest first” (Fairlie 2005, page 313), as our method cients effect, we propose a second counterfactual simu - does. Specifically, we calculate the contribution of a sin - lation following a similar procedure to that used in the gle variable k as the change in women’s average condi- first one. Oaxaca and Ransom (1999) show that this tional probability of underemployment resulting from decomposition is destined to suffer from an identifica - switching women from the categories where they are tion problem, since the detailed coefficients effect attrib - over-represented to those where they are under-repre- uted to dummy or categorical variables is not invariant sented. This procedure is carried out until women’s rel - to the choice of reference groups. Gardeazábal and Ugi- ative frequencies across the categories of k are equal to dos (2004) and Yun (2005) propose methods to solve this men’s ones. The selection of women who are switched is problem. Despite being widely used, these approaches random, so the procedure is repeated 50 times to ensure show some limitations. u Th s, we use the grand-mean consistency, and then the results are averaged. As the method that Kim (2013) proposes. This method appears changes described affect 10% of the observations or less to be a good option for analyses regarding labour market for most variables, the change in the probability of under- outcomes because it accurately estimates the extent to employment is due to a relatively small change in the which each variable contributes to the group differences. data. Additionally, it gives a meaning to the intercept term and Specifically, the contribution of a single variable k, to the coefficient component of each dummy variable. denoted as θ (k) , can be computed as described in Specifically, we calculate the coefficient effect related Eq. (8): to a specific variable k,θ (k), k = 1, . . . n , as the change W→M W W→M W W W W W W W F x (k)β , x (k)β , ρ F x β , x β , ρ i i ∀Women ∀Women W i 1 i 2 i 1 i 2 (8) θ (k) = − W W N N W→M W where x (k) contains the same information as x in men’s average conditional probability of underemploy- i i but variable k has been modified as described and ment if the parameter of a specific characteristic were k = 1,2…n, where n refers to the number of categorical that of women. Additionally, it is necessary to include the variables included in x . change corresponding to parameter ρ,θ . These effects M M ∼ ∼ The sum of the individual contributions of all the varia - are described in Eqs. (10) and (11), where β and β 1 2 bles does not exactly equal the endowments effect. Thus, are the estimated transformed coefficients of the model the summing up property, which the method proposed M→W according to Kim (2013) method. In β (k) and by Fairlie has, does not satisfy. So we can write: 1 M→W β (k) , men’s coefficients corresponding to the vari - W W able k have been replaced by women’s ones. E = θ (k) + D E E (9) k=1 M→W M→W ∼ ∼ M M F x β (k), x β (k), ρ W 1 2 ∀Men i i An approximation error ( D ) emerges because the θ (k) = endowments effect ( E ) in the aggregate decomposition W N M M ∼ ∼ is computed by switching all the variables simultaneously. M M F x β , x β , ρ 1 2 ∀Men i i Conversely, in our simulation, we switch only one vari- able at a time. As the conditional probability of underem- ployment is not linear, both results are slightly different. (10) Although when both expressions, E and θ (k) , k=1 E are linearised they coincide, a disparity emerges from the According to Fortin et al. (2011) and Kim (2013), these normalizations have several limitations: they may leave the estimation and decomposition with- out a simple meaningful interpretation; they will likely be sample specific and make comparisons across studies impossible; and they are sensitive to the number of categories and to the grouping method. The estimated coefficients are transformed by subtracting from each of them the grand-mean weighted sum of the coefficients of each vari- This number of repetitions was selected after an analysis of sensitivity. We M M − − decided to choose 50 because the average difference in the results found with able β (k) or β (k), k = 1, . . . , n . It is also necessary to sum −5 1 2 respect to using 200 was around 10 and the standard errors could be com- M M − − n n puted in a reasonable time. β (k) or β (k) to the intercepts in order to transform k=1 1 k=1 2 Note that each discrete variable k is included as a set of dummies in x . them. i 22 Page 6 of 16 J. Acosta‑Ballesteros et al. M M ∼ ∼ point, so they are more easily interpreted. This three-fold M M W β β 9 F x , x , ρ ∀Men 1 2 i i decomposition when the starting point is women can be θ = ρ easily obtained from Eqs. (4) or (7) and can be expressed (11) as: M M ∼ ∼ M M M β β F x , x , ρ 1 2 ∀Men i i Gap = E + C +(C − C ) W W M W (13) Gap = E + C +(E − E ) W W M W (14) Again, summing the individual contributions of the variables does not exactly equal the coefficients effect, The new term, C − C = E − E , can be inter- M W M W C . An approximation error ( D ) emerges for the same C preted as an interaction component that indicates the reasons already explained (see Appendix 2 in Additional portion of the gap that occurs when both endowments file 1). Thus, we can write : and coefficients change simultaneously. Alternatively, it is the portion of the gap that remains after controlling for M M M the endowments and coefficients effects. This interaction C =− θ (k) + θ + D (12) β ρ C component is more difficult to interpret than the first two k=1 and is often disregarded. However, we believe, as Etezady Since our detailed decomposition of the gap is the et al. (2021), that neglecting it provides a substantially sum of both expressions (E and C ), the approxima- W M incomplete picture of the total influences of endowments tion errors imply that the sum of individual contributions and coefficients to the gap. u Th s, our analysis is based on of all the variables does not equal the gap. To assess the Eq. (14). magnitude of this disparity, in the Results section, we dis- Some final comments regarding our methodology play the approximation errors of our decomposition. need to be pointed out. First, to obtain the standard Despite our decomposition of the gap not being exact, errors for the results of both counterfactual simulations, it provides technical advantages compared to Fairlie which are necessary to test if the corresponding changes decomposition procedure, as well as being applicable to in the probability of underemployment are statistically a bivariate probit model. Thus, its economic interpreta - significant, Krinsky and Robb’s (1986) method has been tion is straightforward, and it avoids the path dependence applied, as Dowd et al. (2014) explain. problem, since it always uses the same starting point, real Second, the survey structure of our data has been taken women (or men) in the sample, and only one characteris- into account in the methodology. Thus, the bivariate pro - tic is modified. Moreover, our approach does not require bit selection models have been estimated considering a one-to-one matching of individuals, since we replicate sample weights and cluster-robust standard errors. Addi- the distribution of each specific variable and the num - tionally, the sample weights have been considered in both ber of women who have a specific characteristic changed counterfactual simulations by replicating each observa- is just those strictly necessary, so we keep almost real tion according to its weight. individuals. Conversely, in Fairlie decomposition tech- Third, our methodology is displayed for bivariate pro - nique, each woman is randomly matched with a man in bit models with sample selection, but it can also be easily the sample, and she takes his characteristics sequentially applied to single equation models like the probit or logit until she becomes that man. As the sequential change ones. This fact allows us to carry out some robustness of characteristics is made, it is likely that the remaining analyses. Thus, we specify univariate probit models to combination of characteristics will be unreal. In addition, explain underemployment and we obtain the three-fold our methodology offers a simulation that allows us to decomposition of the gap. These results are compared to approximate a detailed decomposition of the coefficients effect. Even though we could test our hypotheses using a two- Equations 13 and 14 can also be proposed for men. fold decomposition, it is increasingly common in the lit- In studies based on survey data not only the outcome variable but also erature to use a three-fold one (Daymont and Andrisani, the predictors are subject to sampling variation (Jann 2008). It implies that 1984), which has the advantage that endowments and the standard errors may be underestimated, especially those regarding the endowment component. However, the results of the models shown in Addi- coefficients effects are computed from the same starting tional file 2 seem to indicate that this is not the case, since the standard errors estimated using our methodology are very similar to those obtained using the Oaxaca-Blinder or Yun methodologies. This process is required in the first simulation in order to switch the Note that the minus sign is required in Eq. (12) because in the simulation value of a specific variable from those categories where women (men) are the average man is the starting point, while when computing C , the average over-represented. In the second simulation, weighting each observation man is the final point. according to its raising factor is enough. Measuring the effect of gender segregation on the gender gap in time‑related underemployment Page 7 of 16 22 those obtained using Fairlie and Yun methodologies for a criterion, we obtain a band similar to the one in Hakim probit model and to those achieved using Oaxaca-Blinder (1998), where gender-integrated occupations are charac- technique for a linear model. Table S1 in Additional file 2 terised by a proportion of women ten percentage points shows these results, which are similar to those obtained around the percentage of women in total employment. with our methodology. Our gender segregation measures have been com- puted using the three-digit codes from both occupations 4 Data and variables (according to the National Classification of Occupations, In this article, we use the definition of time-related 1994 and 2011) and industries (National Classification underemployment directly provided by the Spanish Sta- of Economic Activities, 1993 and 2009). However, when tistical Office. Specifically, the criteria applied in the the number of people working in a certain occupation Spanish LFS to classify workers as underemployed (in or industry is less than 50, segregation has been defined line with the International Labor Organization Bureau according to two-digit or one-digit codes. Additionally, of Statistics recommendations) are: they would like to given the methodological change in both classifications, work more hours, they are available to do so, and they it has been necessary to calculate the value of each seg- work less than the usual weekly hours of full-timers in regation variable for two different sub-periods. As both their industry. u Th s, underemployment is a more accu - gender segregation measures are correlated, we have rate indicator of labour underutilization than involuntary solved the collinearity problem by defining an interaction part-time employment. It reflects non-desired workdays variable with nine categories that integrates both occupa- for all types of workers, capturing the preference of both tional and industry segregation. part-timers and full-timers to have longer workdays. As education plays an important role in the risk of The data used come from the 2006–2016 annual sam - underemployment (Acosta-Ballesteros et al. 2018), we ples of the Spanish LFS. Therefore, our database is a define 43 educational categories using the information pool of cross-sectional annual observations, since each provided by the LFS on education level and field of study, individual is included only once in the annual sam- and according to the National Classification of Education ple. Our sample contains young people aged 16 to 34 (2000 and 2014). Ten specializations for vocational train- who were active. The few individuals with inconsistent ing and university degrees are distinguished. Moreover, answers or who do not provide the necessary informa- whether workers took longer than usual in completing tion for the analysis have been removed. The final sample their studies is also considered. includes 70,445 women: 73.6% are employed, and among The remaining explanatory variables include nation - them, 16.5% are underemployed. The corresponding fig - ality, having children under 16, and some additional ures for the male subsample are 80,962, 75.1% and 11.8%, regressors reflecting household composition; whether respectively. the individual is enrolled in formal studies is also taken The independent variables included in the economet - into account. We also consider professional status (self- ric analysis (displayed in Table 1) reflect the main fac - employed or employed in the public or private sector tors previously found to determine underemployment. with a fixed-term or permanent contract), the size of In order to classify occupations and industries as gender- the firm, having a recent job (tenure up to 12 months dominated or integrated, we follow the relative concept and depending on the worker’s age), the unemployment of Anker (1998). Thus, the dividing line between gender- rate by gender in the Autonomous Regions, as well as dominated and integrated occupations (or industries) a dummy variable that takes value one if the observa- is established in relation to the average percentage of tion corresponds to the period after the labour reform of female workers in the labour force as a whole (44% over 2012. the period analysed). Specifically, we consider female- dominated occupations or industries are those hav- This fact only occurs in a few occupations (industries) that account for 0.15% (0.89%) of workers in our sample. ing more than 1.25 times the mean percentage female, Before the Bologna Process, the Spanish education system distinguished while male-dominated ones are those having less than short-cycle (three years) and long-cycle (more than three years, usually 0.75 times the mean percentage female. If the percent- five) university degrees. The new degrees under the European Higher Edu- age of women is between both limits, the occupation or cation Area are included as short-cycle programmes. This is the only continuous variable in the model. As it is already defined industry is labelled as gender-integrated. Applying this by gender, it is not necessary to develop the procedure explained above. us Th , in the first simulation, the change in this characteristic has been car - ried out by simply attributing each woman the unemployment rate she would face if she were a man (and vice-versa). The accuracy of the results in this paper using these data is our sole respon - sibility. This reform, among other measures, allows firms to reduce the working The frequencies of the independent variables are provided in Table S2 in hours of their employees more easily than before, and may partially explain Additional file 2. the relatively high underemployment rate observed since 2012 in Spain. 22 Page 8 of 16 J. Acosta‑Ballesteros et al. Table 1 Average marginal effects: underemployment and employment by gender Variable Women Men Under Employ Under Employ Educational attainment (omitted: primary education or less) Compulsory secondary − 0.006 0.077*** 0.003 0.082*** Non‑ compulsory secondary − 0.024*** 0.141*** − 0.013 0.142*** Lower vocational training Education − 0.026 0.116*** 0.049 0.181*** Arts and humanities 0.012 0.170*** 0.066 0.116*** Social sciences − 0.092** 0.342*** − 0.109*** 0.149 Business, administration, law − 0.014 0.134*** − 0.009 0.133*** Sciences − 0.012 0.194*** 0.036 0.029 ICT − 0.052 0.124** 0.060* 0.135*** Technology − 0.026 0.096*** 0.002 0.138*** Agriculture 0.007 0.121** − 0.021 0.163*** Health − 0.036*** 0.183*** 0.005 0.127*** Social services − 0.031*** 0.145*** − 0.006 0.122*** Higher vocational training Education − 0.019 0.171*** 0.073* 0.190*** Arts and humanities − 0.016 0.136*** − 0.011 0.138*** Social sciences 0.025 0.153*** 0.020 0.179*** Business, administration, law − 0.034*** 0.171*** − 0.024* 0.169*** Sciences − 0.054** 0.174*** − 0.045 0.160*** ICT − 0.028 0.187*** − 0.042*** 0.206*** Technology − 0.005 0.174*** − 0.016* 0.179*** Agriculture − 0.005 0.030 0.020 0.156*** Health − 0.038*** 0.181*** − 0.003 0.164*** Social services − 0.036** 0.153*** − 0.013 0.155*** Short‑ cycle university Education − 0.028** 0.224*** 0.021 0.185*** Arts and humanities 0.062* 0.125*** − 0.070*** 0.117*** Social sciences 0.012 0.132*** 0.055 0.147*** Business, administration, law − 0.059*** 0.169*** − 0.042*** 0.189*** Sciences − 0.046 0.146*** − 0.022 0.170*** ICT − 0.060** 0.161*** − 0.060*** 0.214*** Technology 0.007 0.152*** − 0.054*** 0.174*** Agriculture − 0.020 0.160*** − 0.058** 0.114** Health − 0.040*** 0.230*** − 0.004 0.200*** Social services − 0.067*** 0.182*** − 0.033 0.204*** Long‑ cycle university Education 0.000 0.203*** 0.008 0.167*** Arts and humanities 0.015 0.127*** − 0.001 0.112*** Social sciences − 0.017 0.130*** − 0.008 0.125*** Business, administration, law − 0.074*** 0.192*** − 0.069*** 0.183*** Sciences − 0.030* 0.180*** − 0.070*** 0.184*** ICT − 0.099*** 0.270*** − 0.098*** 0.246*** Technology − 0.012 0.149*** − 0.057*** 0.201*** Agriculture − 0.048* 0.169*** − 0.093*** 0.147*** Health − 0.075*** 0.272*** − 0.078*** 0.267*** Social services − 0.027 0.210*** − 0.051*** 0.161*** Student − 0.011* − 0.058*** − 0.001 − 0.072*** Measuring the effect of gender segregation on the gender gap in time‑related underemployment Page 9 of 16 22 Table 1 (continued) Variable Women Men Under Employ Under Employ Time to complete studies (omitted: within appropriate time) Less than three years late 0.013*** − 0.044*** 0.009 − 0.028*** Three or more years late 0.047*** − 0.074*** 0.010 − 0.052*** Age (omitted: 16–19) 20–24 0.019 0.158*** 0.010 0.084*** 25–29 0.008 0.232*** 0.009 0.135*** 30–34 0.008 0.269*** 0.003 0.153*** Live with parents 0.018*** − 0.099*** − 0.004 − 0.092*** Live in couple − 0.006 − 0.010 − 0.028*** 0.048*** Children under 16 0.003 − 0.086*** 0.017** − 0.029*** Non‑Spanish 0.050*** − 0.057*** 0.071*** − 0.076*** Occupational segregation (omitted: male‑ dominated) Integrated 0.007 − 0.001 Female‑ dominated 0.046*** 0.028*** Industry segregation (omitted: male‑ dominated) Integrated 0.023*** 0.026*** Female‑ dominated 0.045*** 0.016** Tenure 12 months or less 0.039*** 0.041*** Professional status (omitted: self‑ employed) Public sector employee with permanent contract − 0.000 0.001 Public sector employee with fixed‑term contract 0.072*** 0.067*** Private sector employee with permanent contract 0.040*** 0.029*** Private sector employee with fixed‑term contract 0.124*** 0.092*** Firm size (omitted: up to 10 workers) More than 10 workers − 0.040*** − 0.021*** Unknown − 0.042*** − 0.041*** Unemployment rate 0.002*** 0.003*** NUTS Region (omitted: Northwest) Northeast − 0.003** 0.043*** − 0.002*** 0.019*** Madrid − 0.002** 0.032*** − 0.001 0.010 Centre 0.002** − 0.028*** − 0.000 0.002 East − 0.002** 0.026*** 0.000 − 0.001 South 0.003** − 0.056*** 0.006*** − 0.059*** Canary Islands 0.003** − 0.045*** 0.007*** − 0.064*** After 2012 labour reform 0.041*** 0.026*** The year when the survey was carried out is controlled for, but not reported Under: underemployed. Employ: employed Unemployment rate by gender and year in the Autonomous Region of residence p < .10. **p < .05. ***p < .01 Finally, the employment equation includes variables as 5 Results educational attainment, taking longer than usual to grad- 5.1 Marginal effects uate, age, household composition, enrolled in school, the In this section, we present the results of the bivariate area of residence (which reflects the general conditions probit selection models for both subsamples, men and of local demand for work) and the year when the worker women. The estimations indicate that the rho coefficients was interviewed. are 0.207 (p-value = 0.01964) and 0.361 (p-value = 0.0000), 22 Page 10 of 16 J. Acosta‑Ballesteros et al. respectively, justifying the use of these models to handle male peers. Thus, the underemployment gender gap the sample selection bias due to estimating the underem- (−0.039) is not negligible. ployment probability just for employed people. As stated above, the main objective of this article is to The marginal effects of the explanatory variables have carry out an in-depth analysis of this gap and to deter- been computed from the estimated coefficients and are mine the impact of gender segregation on it. To do this, displayed in Table 1. The ones corresponding to the we have carried out a detailed three-fold decomposi- employment equation (Columns 2 and 4) follow the same tion of the gap obtained from Eq. (14) and the results are direction as in previous studies. shown in Table 2. The results obtained for underemployment are shown According to the last row in Table 2, the overall endow- in Columns 1 and 3. Focusing on the role of gender seg- ments effect, E , is very important and seems to explain regation, we observe that workers in female-dominated the underemployment gender gap (130.8%), while the occupations are those with the highest probability of coefficient effects, C , is not significant. Moreover, the underemployment, with a larger impact for women. Spe- interaction term, which is positive (0.0126), reduces the cifically, we find that young women working in female- gap (− 32.7%). dominated occupations are 4.6 percentage points more Regarding the detailed decomposition of the gap, the likely to be underemployed than those in male-dominated sum of the individual contributions of all the variables occupations; the corresponding increase in the male sub- to each component of the gap shows that the errors sample is 2.8 points. When segregation is defined in rela - terms from our counterfactual simulations are small and tion to activity sectors, our results indicate that women not statistically significant. These results validate our in female-dominated industries also show the highest approach. likelihood of underemployment (an increase of 4.5 per- The results in columns 1–3 allow us to conclude that centage points). Men in female-dominated activities are the different distribution of men and women across jobs also more often underemployed than in male-dominated is crucial to explain the gender gap, as stated in Hypoth- ones. However, men in gender-balanced industries are esis 2a. Thus, if women were distributed across occu - the most prone to suffer this handicap. Altogether, these pations and industries as men are (maintaining female results support Hypothesis 1, confirming a higher under - coefficients), the gap would be reduced to the greatest employment risk associated with working in female- extent (− 0.0453). This result suggests that gender segre - dominated jobs than in male-dominated ones. gation leads to a kind of discrimination against women. The marginal effects of the remaining explanatory vari - Indeed, it cannot be argued that women work more fre- ables are not discussed due to space constraints. We only quently in certain (female-dominated) jobs because they want to briefly point out some of the results regarding prefer shorter work-schedules. Conversely, their larger educational attainment, due to the relevance of this vari- risk of underemployment, which means they would like able in almost any labour market outcome. The figures in to work more hours than available more often than their Table 1 suggest that having a long-cycle university degree male peers, is mainly due to the kind of jobs they work in. in almost any field reduces underemployment. Overall, Focusing on the coefficients effect (Columns 4–6 in business, administration and law, and information and Table 2), the contribution of the intercept term of the communication technologies (ICTs) seem to be the best underemployment equation (Intercept 1) quantifies the specializations at most education levels. Additionally, it extent to which women are, on average, treated differ - is worth noting that education means greater differences ently to men. This contribution can be interpreted as the in the risk of being underemployed for women than for average extent of discrimination (Kim 2013). Our results men. Some additional comments are included in the sen- indicate that the contribution of Intercept 1 to the gap is sitivity section. small and not significant, so women and men would face the same average risk of underemployment. However, 5.2 A nalysis of the underemployment gender gap the coefficient effect of gender segregation indicates an According to the results of the bivariate probit models, important deviation from the mean discriminatory level. women’s estimated conditional probability of underem- Specifically, the gender gap in underemployment would ployment is 0.163; the corresponding figure for men is reduce (− 0.0146) if women had men’s returns in the 0.124. Therefore, young female workers in Spain are 1.31 same gender-typing jobs (maintaining female character- times more likely of being underemployed than their istics). Therefore, the different impact on underemploy - ment that working in certain occupations and industries exerts on the risk of experiencing it contributes to widen- The marginal effects corresponding to each gender segregation measure ing the gender gap, supporting our Hypothesis 2b. In fact, have been obtained from the estimated parameters of the interaction variable this is the most important contribution to the coefficient capturing both occupational and industry segregation. Measuring the effect of gender segregation on the gender gap in time‑related underemployment Page 11 of 16 22 Table 2 Three‑fold decomposition of underemployment gender gap Variable Endowments Coefficients Interaction Total Contribution SE %Gap Contribution SE %Gap Contribution SE %Gap %Gap Educational attainment 0.0073** 0.00314 − 18.9 0.0053** 0.00218 − 13.8 − 0.0061*** 0.00364 15.7 − 17.0 Student 0.0003* 0.00018 − 0.9 0.0001 0.00007 − 0.2 − 0.0003 0.00024 0.9 − 0.2 Time to complete studies − 0.0001*** 0.00004 0.3 − 0.0003** 0.00012 0.7 0.0001 0.00006 − 0.2 0.8 Age − 0.0021*** 0.00004 5.4 − 0.0001* 0.00005 0.2 − 0.0052* 0.00029 13.4 19.0 Live with parents 0.0016*** 0.00054 − 4.2 0.0019** 0.00078 − 4.8 − 0.0018* 0.00072 4.8 − 4.2 Live in couple 0.0006 0.00063 − 1.7 − 0.0006* 0.00033 1.5 0.0019* 0.00087 − 4.8 − 5.0 Children under 16 − 0.0003 0.00047 0.7 0.0006 0.00049 − 1.6 − 0.0011 0.00072 2.8 1.9 Non− Spanish − 0.0009*** 0.00012 2.4 0.0009 0.00057 − 2.3 − 0.0001 0.00015 0.2 0.3 Gender segregation − 0.0453*** 0.00355 117.0 − 0.0146*** 0.0032 37.8 0.0217*** 0.00473 − 56.2 98.6 Professional status − 0.0028*** 0.00038 7.3 − 0.0041*** 0.00091 10.7 0.0005** 0.00058 − 1.2 16.8 Recent job − 0.0008*** 0.00022 2.1 − 0.0003 0.0008 0.7 0.0000 0.00026 0.1 2.9 Firm size − 0.0030*** 0.00032 7.7 − 0.0016*** 0.00056 4.2 0.0011* 0.00042 − 2.8 9.1 Unemployment rate − 0.0059*** 0.00077 15.1 0.0057 0.01078 − 14.7 − 0.0002** 0.0011 0.6 1.0 NUTS Region 0.0000 0.00001 0.0 0.0000 0.00003 0.1 0.0000 0.00002 − 0.1 0.0 After 2012 reform − 0.0009*** 0.0001 2.2 − 0.0013*** 0.00047 3.2 0.0005*** 0.00012 − 1.2 4.2 Year − 0.0003** 0.00012 0.8 0.0005** 0.00022 − 1.2 − 0.0002 0.00016 0.5 0.1 Intercept 1 0.0016 0.00894 − 4.1 − 4.1 Intercept 2 − 0.0025** 0.00102 6.4 6.4 0.0114*** 0.00936 − 29.3 − 29.3 Approximation error 0.0017 0.00441 − 4.4 − 0.0033 0.0046 8.5 0.00199 0.0054 − 5.1 − 1.0 Aggregate − 0.0506*** 0.0037 130.8 − 0.0007 0.00454 1.9 0.0126* 0.0051 − 32.7 100.0 Standard errors (SE) have been computed using Krinsky and Robb’s method (1,000 draws). %Gap: Change as a percentage of the total underemployment gender gap effect affecting the underemployment gender gap, rein -5.3 Sensitivity analysis forcing our previous finding regarding the discriminating In this subsection, we carry out different sensitivity anal - effect of segregation against women. yses to check the robustness of our results. Table 3 shows The interaction term captures the effect of changing the estimated gender gap in underemployment and the endowments and coefficients simultaneously. The posi - specific contribution of occupational and industry segre - tive sign of this portion suggests that the risk of under- gation to it, according to Eq. (14). employment in female-dominated jobs, where women In Model I, we re-estimate our baseline model using an are indeed highly represented, is significantly larger for alternative definition of gender segregation. Specifically, women than for men. Therefore, this interaction term we have re-defined integrated occupations and industries reflects that, once the returns have been changed from as those where the female percentage is between 0.5 and women to men, the additional contribution of distribut- 1.5 times the average female share of employment in the ing women across jobs as men is only − 0.0236 (which is labour force. The following two models consider gender 0.0217 smaller than the contribution of segregation to the segregation only in occupations (Model II) or indus- endowments effects). tries (Model III) to analyse their separate effects. u Th s, Regarding the total impact of each individual variable we can check if our results change when no threshold is to the underemployment gender gap, figures in the last established to classify occupations and industries as gen- column in Table 2 show gender segregation is the most der-dominated or integrated. Additionally, the baseline important one explaining it (98.6%). Moreover, gender model is re-estimated splitting the original sample into differences in age also widen the gap, explaining 19% of two subsamples: workers with tertiary education (Model it. A similar conclusion is obtained for professional status IV) and those without (Model V). Therefore, taking into (16.8%). Conversely, educational attainment reduces the account the marginal effects already analysed, we can test underemployment gap (− 17%). our hypotheses for more homogeneous groups of educa- tion. Model VI is the same as the benchmark model but includes inactive people together with unemployed ones 22 Page 12 of 16 J. Acosta‑Ballesteros et al. Table 3 Sensitivity analysis: Contribution of occupational and industry segregation to the underemployment gender gap Gap Endowments Coefficients Interaction Total Contribution SE %Gap Contribution SE %Gap Contribution SE %Gap %Gap Baseline model − 0.0387 − 0.0453 0.00355 117.0 − 0.0146 0.0032 37.8 0.0217 0.00473 − 56.2 98.6 Model I − 0.0387 − 0.0439 0.00422 113.4 − 0.0147 0.00368 38.1 0.0215 0.00539 − 55.5 96.0 Model II − 0.0387 − 0.0386 0.00313 99.9 − 0.0118 0.0029 30.5 0.0184 0.00426 − 47.5 82.9 Model III − 0.0387 − 0.0295 0.00239 76.1 − 0.0102 0.00222 26.4 0.0166 0.00313 − 43.0 59.5 Model IV − 0.0369 − 0.0361 0.00322 98.1 − 0.0007 0.00368 1.9 0.0083 0.00530 − 22.5 77.5 Model V − 0.0536 − 0.0524 0.00608 97.9 − 0.0254 0.01318 47.4 0.0351 0.00750 − 65.5 79.8 Model VI − 0.0387 − 0.0451 0.00355 116.6 − 0.0143 0.00419 37.0 0.0215 0.00473 − 55.5 98.1 Model VII − 0.0742 − 0.0615 0.00237 82.8 − 0.0124 0.00359 16.7 0.0301 0.00332 − 40.5 59.0 Model VIII − 0.0387 − 0.0429 0.00371 111.0 − 0.0163 0.00368 42.1 0.0170 0.05017 − 44.0 109.1 Model IX − 0.0384 − 0.0466 0.00364 121.2 − 0.0154 0.00329 40.0 0.0232 0.00482 − 60.4 100.8 Model X − 0.0387 − 0.0453 0.00355 117.0 − 0.0179 0.00417 46.2 0.0217 0.00473 − 56.2 107.0 Standard errors (SE) have been computed using Krinsky and Robb’s method (1,000 draws). %Gap: Change as a percentage of the total underemployment gender gap. All the values for the contribution of segregation are significant at 99% except the coefficient and interaction components in Model IV, which are not significant when estimating the employment equation. Model VII As expected, in Models II and III the three effects are includes the same independent variables as the baseline smaller than those obtained in the baseline model. It is model but only involuntary part-time workers are consid- noteworthy that the portion of the gap explained by seg- ered underemployed, as is often the case in the literature. regation is larger when we consider only occupational The last three models include some methodological segregation. Some differences are also found if only invol - changes. In Model VIII, we adjust the variable gender seg- untary part-time workers are considered underemployed regation last instead of first to check if the order of switch - (Model VII). This decision implies estimating our models ing affects the results. As Heckman-type selection models using an alternative endogenous variable, so the results are mostly identified by assumptions about error distribu - obtained are likely to change significantly. However, the tions, in Model IX we use Inverse Probability Weighting different distribution of men and women across occupa - (IPW) based on a probit model as an alternative technique tions and industries is what still drives the gap. to address sample selection problems (see Seaman and The most interesting results are those obtained for White 2013 for a review). Finally, Model X is estimated workers with different levels of education (Models IV and using the methodology proposed in Gardeazábal and Ugi- V). It is worth noting that the estimated underemploy- dos (2004) to address the identification problem that arises ment gap is larger for workers without tertiary education, for categorical variables (instead of Kim’s approach). as might be expected. Gender segregation, however, is the In general terms, the results in the first column of factor that mainly explains the gap regardless of workers’ Table 3 are quite similar to those from the main analysis educational level, explaining almost the same percentage and few differences can be found. In particular, the larg - of it in both subsamples. Moreover, the different distri - est estimated underemployment gender gap is found for bution of men and women across jobs widens the gap to workers without tertiary education (Model V) and when a larger extent for less educated workers than for more only involuntary part-timers are classified as underem - educated ones. Additionally, only in the sample of work- ployed (Model VII). ers without tertiary education, the different returns asso - When we carry out the detailed three-fold decom- ciated with women and men widen the gap. Thus, we can position, our main conclusions remain unchanged. conclude that tertiary education reduces the gap not only because women and men are more similarly distributed across jobs (in comparison with less educated workers), but also because working in the same gender-typing jobs To facilitate comparisons with the baseline model, the reweighting leads to a similar risk of underemployment for male and approach has been applied only when estimating the coefficients of the probit that explains underemployment, but not when the decomposition of the gap female. has been computed. Therefore, the results obtained with IPW are conditioned Regarding the last three models, there are some to being employed. small differences with the baseline in the three compo - This result is not surprising, since involuntary part-time work in Spain nents, especially in the coefficients effect (that seems to is mainly a female problem. In fact, the average percentage of involuntary part-timers among male workers in the last 15 years is just 3.5%, while this increase). These differences translate into a larger pro - figure is 11.6% for women. portion of the gap explained by segregation in the three Measuring the effect of gender segregation on the gender gap in time‑related underemployment Page 13 of 16 22 models. However, our main conclusions still hold, so our Krinsky-Robb approach could lead to underestimating procedure is robust to the methodological changes they the standard errors. Despite our results indicating this is include. not the case, nonparametric bootstrap could be consid- Overall, the estimated underemployment gender gap ered as an alternative. is mainly explained by the different distribution of male The procedure has been tested in several ways and the and female workers across occupations and industries in main conclusions still hold. Firstly, the results obtained every specification of the model, which clearly supports applying our procedure to a univariate probit model are Hypothesis 2a. Moreover, Hypothesis 2b is also confirmed very similar to those obtained through other decompo- because, in most cases, the different impact that working sition techniques like Fairlie, Yun (2004) and Oaxaca- in certain gender-typing jobs exerts on the risk of male Blinder (Oaxaca 1973 and Blinder 1973). Secondly, the and female underemployment contributes to widening main results of our baseline model are robust to meth- the gender gap. Hence, the results in this subsection give odology changes, definition of some variables and to the robustness to our findings. sample used. Our results demonstrate that working in female- 6 Conclusions and discussion dominated occupations and industries implies a higher This article provides evidence of the crucial impact of probability of time-related underemployment than in occupational and industry segregation on the time- male-dominated ones, confirming our first hypothesis. related underemployment gender gap for people aged 16 Moreover, we find that the disadvantage in terms of to 34 using the annual samples of the Spanish LFS from underemployment that implies working in a female-dom- 2006–2016. It is worth noting that despite the Spanish inated occupation or industry is greater for women than labour market being deeply gender segregated, the effect for men. of this feature on the gap has not been addressed before Furthermore, according to our results, the estimated in the literature. Furthermore, to the best of our knowl- underemployment gender gap for young workers in edge, we are the first to perform a decomposition using Spain is 3.9 percentage points. As underemployment has bivariate probit models with sample selection. To do this, a negative impact on income, welfare dependency and we have developed a methodology based on two coun- life satisfaction (Wilkins 2007), the higher underemploy- terfactual simulations that provides a three-fold detailed ment risk faced by women implies negative consequences decomposition of the underemployment gender gap into related to experience, earnings, and possibly promotions endowments and coefficients effects as well as the inter - (Weststar, 2011). Therefore, designing effective policies action of these effects. leading to more gender equality will only be possible if Our methodology, inspired by the Fairlie technique the factors behind the gender gap in underemployment (1999, 2005, and 2017), has several advantages. Thus, it are clearly identified. To the best of our knowledge, the allows identifying the coefficients effect correspond - reasons for this difference have been little investigated. ing to a specific variable (while Fairlie technique does As an exception, Barrett and Doiron (2001) affirm that not). In fact, this effect can be easily interpreted using the main reason that explains women being involuntary Kim’s (2013) method to solve the identification problem. part-timers more often than men is the fact they are Moreover, we keep almost real individuals for two rea- employed in different industries and occupations. Never - sons. First, our approach does not require a one-to-one theless, Vuluku et al. (2013), who are the only ones who matching of individuals, since we replicate the distribu- have decomposed the underemployment gender gap, did tion of each specific variable, and the number of women not include occupational or industry segregation as an who have a specific characteristic adjusted is just those explanatory factor, while we do. This fact could be one strictly necessary. Second, the proposed procedure of the reasons why they find that only 5.4% of the gap always uses the same starting point, and we estimate each is explained by female-male differences in characteris - contribution by switching the variable of interest first (as tics, while 94.6% is unexplained, while our results are the suggested by Fairlie 2005). Thus, only one characteris - opposite. tic is modified at a time. Although we are aware that the The results obtained from our simulations lead us order of switching the variables is potentially important, to conclude that the fact that men and women work in this decision provides a potential solution to the path different industries and occupations is what widens dependence problem already pointed out by Fairlie. the underemployment gap to the greatest extent. This The methodology proposed has some limitations. effect is even increased due to women facing a different It does not show the summing up property of the Fair- risk of underemployment than men when they work in lie method, however, the estimated approximation the same gender-typing jobs. So, Hypotheses 2a and 2b errors are small and not significant. Moreover, using the are supported and we can state that the gender gap in 22 Page 14 of 16 J. Acosta‑Ballesteros et al. underemployment would be largely reduced if men and linked to a lower risk of experiencing underemployment. women were more evenly distributed across occupations This result is in line with the literature attributing ICTs and industries. Thus, segregation (especially occupa - the potential capacity to reduce gender inequalities, since tional) is not only a source of gender differences in terms they improve the occupational and professional posi- of wages and job quality (Stier and Yaish 2014), but also a tion of women (Castaño et al. 1999), particularly in the key factor explaining the underemployment gender gap. Spanish labour market (Iglesias-Fernández et al. 2010a, Moreover, as the World Economic Forum (2017) states, 2010b). Specifically, ITCs reduce both the need for man - a crucial factor for further progress in reducing the over- ual labour and physical effort in favour of knowledge, all global gender gap is the closing of occupational gender teamwork and communication skills (WWW-ITC 2004). gaps. Therefore, policy measures should be designed and This fact, in turn, promotes changes in the sectoral dis - implemented to fight against segregation in the labour tribution and educational requirement of jobs and the market in order to achieve gender equality in Spain. demand for occupations (Iglesias-Fernández et al. 2012), In this respect, education is an important factor to be leading to more opportunities for women. However, the highlighted. Thus, we can consider two different ways in ratio of female/male graduates in ICTs is only 0.14 in which education can influence underemployment. Firstly, Spain (World Economic Forum 2017). education has a direct impact on underemployment since Although educational segregation by gender plays a the marginal effects show that some educational attain - significant role in shaping gender segregation within the ments contribute to reduce the risk of underemploy- labour market, as Smyth and Steinmetz (2008) point out, ment. Moreover, the results from our simulations using women and men who choose similar fields do not have the baseline model, as well as those obtained after split- exactly the same occupational outcomes. Thus, educational ting the sample into workers with tertiary education and policies should be complemented with other instruments. without it, allow us to affirm that education reduces the For instance, reinforcing policies that encourage employ- underemployment gender gap. Specifically, our results ers to hire female workers in male intensive occupations suggest a higher educational attainment helps women to and industries could also help reduce both segregation escape from this disadvantage in the labour market. and the gender gap in underemployment. Additionally, Second, education may affect segregation in the mar - the economic conditions of female occupations should ketplace due to the link between educational presorting be improved to raise both men’s and women’s interest in and occupational and industry segregation (Borghans and female-dominated occupations. In order to achieve an Groot 1999; Shauman 2006; Smyth and Steinmetz 2008). egalitarian distribution of men and women across jobs, The statistical evidence on the strength of the link between eradicating the disincentives to work in female-dominated segregation in education and in employment is mixed occupations is necessary (Torre 2018). Finally, any other (Bettio et al. 2009). Nevertheless, appropriate remedies that measures devoted to fighting against gender stereotypes address the barriers women experience to enter male domi- and discrimination would be welcome to reduce inequality nated jobs should include changes in the education and in the Spanish labour market. training of women and girls, such as introducing gender aware career counselling/guidance (National Foundation Abbreviation for Australian Women 2017). Particularly, young women LFS: Labour Force Survey. should be encouraged to enrol in previously male-domi- nated education programmes in order to gain access to a Supplementary Information wider range of jobs. Additionally, encouraging young men The online version contains supplementary material available at https:// doi. to join female-dominated specialties could translate into org/ 10. 1186/ s12651‑ 021‑ 00305‑0. a more mixed gender educational profiles. In fact, as Bet - tio et al. (2009) point out, since women are outperforming Additional file 1: Appendix 1. Approximation error in the endowments effect. Appendix 2. Approximation error in the coefficients effect. men in levels of education attained—up to the first stage of Additional file 2: Table S1. Comparison of methods for the three ‑fold tertiary education—choice of field is the primary channel decomposition of the underemployment gender gap. Table S2. Frequen‑ through which education can influence de-segregation in cies (%) of the independent variables by gender: not underemployed the labour market in the future. (NU), underemployed (U), or unemployed (Unem) Especially desirable would be more women specialis- ing in ICTs since our results show this field seems to be Acknowledgements We thank three anonymous referees for their insightful comments and sug‑ gestions, which have helped to improve this article considerably. Authors’ contributions Analysing the causality relation between educational and occupational seg- All authors wrote, read and approved the final manuscript. regation is beyond the scope of this study. Measuring the effect of gender segregation on the gender gap in time‑related underemployment Page 15 of 16 22 Funding Daymont, T.N., Andrisani, P.J.: Job preferences, college major, and the gender Authors declare that they have received no specific funding to conduct this gap in earnings. J. Hum. Resour. 19, 408–428 (1984) research. Doiron, D.J., Riddell, W.C.: The impact of unionization on male‑female earnings differences in Canada. J. Hum. Resour. 29(2), 504–534 (1994) Availability of data and materials Dowd, B.E., Greene, W.H., Norton, E.C.: Computation of standard errors. The data that support the findings of this study are available from Instituto Health Serv. Res. 49(2), 731–750 (2014) Nacional de Estadística (INE). Purchase terms do not allow authors sharing Dueñas‑Fernández, D., Iglesias‑Fernández, C., Llorente ‑Heras, R.: La involun‑ these data. tariedad en el empleo a tiempo parcial y la Gran Recesión: un análisis de género en España. EKONOMIAZ. Revista Vasca De Economía 90(02), Code availability 284–317 (2016) The results have been obtained using STATA. Etezady, A., Shaw, F.A., Mokhtarian, P.L., Circella, G.: What drives the gap? Applying the Blinder‑ Oaxaca decomposition method to examine gen‑ erational differences in transportation‑related attitudes. Transportation Declarations 48, 857–883 (2021) Even, W.E., Macpherson, D.A.: Plant size and the decline of unionism. Econ. Ethics approval and consent to participate Lett. 32(4), 393–398 (1990) The authors of this study declare this manuscript does not infringe the copy‑ Fairlie, R.W.: The absence of the African‑American owned business: an right, moral rights, or other intellectual property rights of any other person. analysis of the dynamics of self‑ employment. J. Labor Econ. 17(1), The authors of this study testify that this paper is original and has not incurred 80–108 (1999) in any type of plagiarism. The authors of this study declare that the article Fairlie, R.W.: An extension of the Blinder‑ Oaxaca decomposition technique is original, has not already been published in a journal, and is not currently to logit and probit models. J Econ. Soc. Meas. 30(4), 305–316 (2005) under consideration by another journal. The authors of this study agree to the Fairlie, R.W.: Addressing path dependence and incorporating sample terms of the SpringerOpen Copyright and License Agreement. weights in the nonlinear Blinder‑ Oaxaca decomposition technique for logit, probit and other nonlinear models. Stanford Institute for Consent for publication Economic Policy Research, Working Paper No. 17–013, University of Not applicable. California (2017) Fortin, N., Lemieux, T., Firpo, S.: Decomposition methods in economics. In: Competing interests Ashenfelter, O., Card, D. (eds.) Handbook of labor economics, Vol 4a, Authors declare that no competing interests exist. Handbooks in Economics, pp. 1–102. North Holland, Great Britain (2011) National Foundation for Australian Women Gender segregation in the work‑ Author details place and its impact on women’s equality. Report submitted to Senate Departamento de Economía, Contabilidad y Finanzas, Universidad de La Finance and Public Administration References Committee. https:// www. Laguna (ULL), La Laguna ( Tenerife), Spain. Facultad de Economía, Empresa y aph. gov. au/ Parli ament ary_ Busin ess/ Commi ttees/ Senate/ Finan ce_ and_ Turismo, Universidad de La Laguna (ULL), Campus de Guajara, Apartado 456, Public_ Admin istra tion/ Gende rsegr egati on/ Report (2017) Postal Code 38200, La Laguna ( Tenerife), Spain. Gardeazábal, J., Ugidos, A.: More on identification in detailed wage decompo ‑ sitions. Rev. Econ. Stat. 86(4), 1034–1036 (2004) Received: 27 July 2020 Accepted: 11 November 2021 Greene, W.H.: Econometric analysis, 7th edn. Pearson, Upper Saddle River (2012) Hakim, C.: A sociological perspective on part‑time work. In: Blossfeld, H.P., Hakim, C. (eds.) Between equalization and marginalization: women working part‑time in Europe and the United States of America, pp. 22–70. References Oxford University Press, Oxford (1997) Acosta‑Ballesteros, J., Osorno ‑ del Rosal, M.P., Rodríguez‑Rodríguez, O.M.: Hakim, C.: Social Change and Innovation in the Labour Market: Evidence from Underemployment and employment among young workers and the the Census SARs on Occupational Segregation and Labour Mobility, Part‑ business cycle in Spain: the importance of education level and specialisa‑ time Work and Student Jobs. Homework and Self‑ employment. Oxford tion. J. Educ. Work 31(1), 28–46 (2018) University Press, Oxford (1998) Anker, R.: Gender and jobs: sex segregation of occupations in the world. Inter‑ Hegewisch, A., Liepmann, H., Hayes, J., Hartmann, H.: Separate and not equal? national Labour Office, Geneva (1998) Gender segregation in the labor market and the gender wage gap. Barrett, G.F., Doiron, D.J.: Working part time: by choice or by constraint. Can. J. Institute for Women’s Policy Research, Briefing Paper No. 377, Washington Econ. 34(4), 1042–1065 (2001) (2010) Bazen, S., Joutard, X., Magdalou, B.: An Oaxaca decomposition for nonlinear Iglesias‑Fernández, C., Llorente ‑Heras, R., Dueñas‑Fernández, D.: Diferencias de models. J. Econ. Soc. Meas. 42(2), 101–121 (2017) Género en el Empleo TIC. Cuad. Econ. 33(92), 105–138 (2010b) Bell, D., Blanchflower, D.: Underemployment in the UK Revisited. Natl. Inst. Iglesias‑Fernández, C., Llorente ‑Heras, R. Dueñas‑Fernández, D.: Profundizando Econ. Rev. 224(1), F8–F22 (2013) en la segregación laboral. Sectores, ocupación y TIC en España. Instituto Bettio, F., Verashchagina, A., Mairhuber, I., Kanjuo‑Mrčela, A.: Gender segrega‑ Universitario de Análisis Económico y Social. Serie de Documentos de tion in the labour market: Root causes, implications and policy responses Trabajo 07/2012, Universidad de Alcalá (2012) in the EU. Publications Office of the European Union, Luxembourg (2009) Iglesias‑Fernández, C., Llorente ‑Heras, R., Dueñas‑Fernández, D.: Mujer y Blinder, A.S.: Wage discrimination: reduced form and structural estimates. J. Empleo TIC en España. In C. Castaño (Dir.), Género y TIC. Presencia, Hum. Resour. 8(4), 436–455 (1973) Posición y Políticas, pp. 83–106. UOC Ediciones, Barcelona (2010a) Boll, C., Leppin, J., Rossen, A., Wolf, A.: Magnitude and impact factors of the Jann, B.: The Blinder‑ Oaxaca decomposition for linear regression models. Stata gender pay gap in EU countries. Report prepared for and financed by the J. 8(4), 453–479 (2008) European Commission–Directorate‑ General for Justice, European Union Kamerāde, D., Richardson, H.: Gender segregation, underemployment and (Hrsg.), Hamburg (2016) https:// op. europa. eu/ en/ publi cation‑ detai l/‑/ subjective well‑being in the UK labour market. Hum. Relat. 71(2), publi cation/ aa27a c5b‑ fc83‑ 11e5‑ b713‑ 01aa7 5ed71 a1/ langu age‑ en 285–309 (2018) Borghans, L., Groot, L.: Educational presorting and occupational segregation. Kim, C.: Detailed wage decompositions: revisiting the identification problem. Labour Econ. 6, 375–395 (1999) Sociol. Methodol. 43, 346–363 (2013) Castaño, C., Iglesias, C., Mañas, E., Sánchez‑Herrero, M.: Diferencia o discrimi‑ Kjeldstad, R., Nymoen, E.H.: Part‑time work and gender: worker versus job nación: La situación de las mujeres en el mercado de trabajo y el impacto explanations. Int. Labour Rev. 151(1–2), 85–107 (2012a) de las tecnologías de la información. Consejo Económico y Social, Madrid Kjeldstad, R., Nymoen, E.H.: Underemployment in a gender‑segregated labour (1999) market. Econ. Ind. Democracy 33(2), 207–224 (2012b) 22 Page 16 of 16 J. Acosta‑Ballesteros et al. Krinsky, I., Robb, A.L.: On approximating the statistical properties of elasticities. Stier, H., Yaish, M.: Occupational segregation and gender inequality in job qual‑ Rev. Econ. Stat. 68(4), 715–719 (1986) ity: a multi‑level approach. Work Employ. Soc. 28(2), 225–246 (2014) Lupton, B.: Explaining men’s entry into female‑ concentrated occupations: Torre, M.: Stopgappers? The occupational trajectories of men in female‑ domi‑ issues of masculinity and social class. Gender Work Organ. 13(2), 103–128 nated occupations. Work Occup. 45(3), 283–312 (2018) (2006) Valletta, R.G., Bengali, L., Van der List, C.: Cyclical and Market Determinants of Martin, P., Barnard, A.: The experience of women in male‑ dominated occupa‑ Involuntary Part‑time Employment. IZA Discussion Paper No.9738, Bonn tions: A constructivist grounded theory inquiry. Journal of Industrial (2016) Psychology 39(2), 1–12 (2013) Vuluku, G., Wambugu, A., Moyi, E.: Unemployment and underemployment in McKee‑Ryan, F.M., Harvey, J.: I have a job, but…: a review of underemploy‑ Kenya: a gender gap analysis. Economics 2(2), 7–16 (2013) ment. J. Manage. 37(4), 962–996 (2011) Weststar, J.: A review of women’s experiences of three dimensions of under‑ Oaxaca, R.: Male‑female wage differentials in urban labor markets. Int. Econ. employment. In: Feldman, D., Maynard, D. (eds.) Underemployment: Rev. 14(3), 693–709 (1973) psychological, economic, and social challenge, pp. 105–125. Springer, Oaxaca, R., Ransom, M.: Identification in detailed wage decompositions. Rev New York (2011) Econ Stat 81(1), 154–157 (1999) Wilkins, R.: Personal and job characteristics associated with underemployment. Powers, D.A., Yoshioka, H., Yun, M.S.: mvdcmp: multivariate decomposition for Aust. J. Labour Econ. 9(4), 371–393 (2006) nonlinear response models. Stata J. 11(4), 556–576 (2011) Wilkins, R.: The consequences of underemployment for the underemployed. J. Prause, J., Dooley, D.: Youth Underemployment. In: Feldman, D., Maynard, D. Ind. Relat. 49(2), 247–275 (2007) (eds.) Underemployment: psychological, economic, and social challenge, World Economic Forum: The Global Gender Gap Report. http:// www3. wefor pp. 59–80. Springer, New York (2011)um. org/ docs/ WEF_ GGGR_ 2017. pdf (2017) Seaman, S., White, I.R.: Review of inverse probability weighting for dealing with WWW‑ITC: Conceptual framework and state of the art. Widening women’s missing data. Stat. Methods Med. Res. 22(3), 278–295 (2013) work in information and communication technology. http:// www. ftu‑ Shauman, K.: Occupational sex segregation and earning of occupations: what namur. org/ fichi ers/ D12‑ print. pdf (2004) causes the link among college‑ educated workers? Soc. Sci. Res. 35, Yun, M.S.: Decomposing differences in the first moment. Econ. Lett. 82(2), 577–619 (2006) 275–280 (2004) Simic, M.: Underemployment and overemployment in the UK. Labour Market Yun, M.S.: A simple solution to the identification problem in detailed wage Trends 110(8), 399–414 (2002) decompositions. Econ. Inq. 43, 766–772 (2005) Simpson, R.: Have times changed? Career barriers and the token woman Yun, M.S.: Identification problem and detailed Oaxaca decomposition: a gen‑ manager. Brit. J. Manage. 8(s1), 121–130 (1997) eral solution and inference. J. Econ. Soc. Meas. 33(1), 27–38 (2008) Simpson, R.: Gender mix and organizational fit: how gender imbalance at dif‑ ferent levels of the organization impacts on women managers. Women Publisher’s Note Manag. Rev. 15(1), 5–20 (2000) Springer Nature remains neutral with regard to jurisdictional claims in pub‑ Simpson, R.: Masculinity at work: the experiences of men in female dominated lished maps and institutional affiliations. occupations. Work Employ. Soc. 18(2), 349–368 (2004) Smyth, E., Steinmetz, S.: Field of study and gender segregation in European labour markets. Int. J. Comp. Sociol. 49(4–5), 257–281 (2008) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Journal for Labour Market Research Springer Journals http://www.deepdyve.com/lp/springer-journals/measuring-the-effect-of-gender-segregation-on-the-gender-gap-in-time-Uxhwg20wKk

Loading next page...

Publisher: Springer Journals
Copyright: Copyright © The Author(s) 2021
ISSN: 1614-3485
eISSN: 2510-5027
DOI: 10.1186/s12651-021-00305-0
Publisher site: See Article on Publisher Site

Abstract

This paper focuses on the impact that gender segregation in the labour market exerts on the underemployment gender gap for young adult workers in Spain. In order to analyse the relative importance of segregation in this gap, we develop a methodology based on two counterfactual simulations that provides a detailed decomposition of the gap into endowments and coefficients effects as well as the interaction of these effects. To the best of our knowl‑ edge, we are the first to perform a decomposition using bivariate probit models with sample selection. Using annual samples of the Spanish Labour Force Survey 2006–2016, the results show that working in female‑ dominated occupa‑ tions or industries hinders working as many hours as desired, especially for women. Furthermore, we conclude that the gender gap in underemployment is mainly due to the different distribution of male and female workers across occupations and industries. Additionally, the different impact by gender that working in the same gender ‑typing jobs exerts on the risk of underemployment contributes to widening the gap. Keywords: Gender gap, Occupational and Industry segregation, Time‑related underemployment, Counterfactual simulations, Decomposition analysis JEL Classification: J16, J22 1 Introduction industry segregation may be an important factor, as sug- Time-related underemployment, which refers to those gested by Barret and Doiron (2001), since a higher under- workers who would like to work more hours than avail- employment rate has been linked to female-dominated able, is a persistent problem in labour markets and the occupations and industries (Kjeldstad and Nymoen Spanish one is no exception. Moreover, this problem 2012a, Kjeldstad and Nymoen, 2012b, and Kamerāde and increased during the Great Recession in many countries Richardson, 2018). Moreover, as Spain has experienced (Bell and Blanchflower 2013; Acosta-Ballesteros et al. higher levels of segregation than other European coun- 2018). tries in a persistent way (Iglesias-Fernández et al. 2012), Furthermore, women experience this situation more analysing the effect of segregation on the underemploy - often than men do (Weststar 2011; Kjeldstad and ment gender gap in the Spanish labour market emerges Nymoen 2012a, b; Vuluku et al. 2013; Acosta-Ballesteros as an interesting research issue. et al. 2018), therefore, we must pay attention to the rea- To the best of our knowledge, only Vuluku et al. (2013) sons for this gender gap. Particularly, occupational and have tried to explain the underemployment gender gap, but they did not include any occupational and industry segregation indicators in their study. Thus, to overcome *Correspondence: orguez@ull.es this shortcoming in the literature, the main objective Departamento de Economía, Contabilidad y Finanzas, Universidad de La of this article is to carry out an in-depth analysis of the Laguna (ULL), La Laguna ( Tenerife), Spain underemployment gender gap. Specifically, we intend Full list of author information is available at the end of the article © The Author(s) 2021. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/. 22 Page 2 of 16 J. Acosta‑Ballesteros et al. to test whether segregation plays an important role in some employed people would like to work more hours explaining it, as well as to quantify how much is due to than available. men and women working in different industries and The demographic and job factors that determine time- occupations, and how much is due to men and women related underemployment have been previously analysed facing different underemployment risks when they work in the literature (Hakim 1997; Weststar 2011; Prause and in the same gender-typing jobs. Dooley 2011; McKee-Ryan and Harvey 2011; Kjeldstad To do this, first, we quantify the effect of occupational and Nymoen 2012a, 2012b; Wilkins 2006; Acosta-Ball- and industry segregation on workers’ underemployment esteros et al. 2018). Particularly, significant differences in risk using a detailed measure of gender segregation. Esti- underemployment have been found across occupations mating this impact through bivariate models with selec- and industries (Kjeldstad and Nymoen 2012a, 2012b; Val- tion enables us to handle the potential sample selection letta et al. 2016). However, very few studies have linked bias due to estimating the probability of underemploy- these differences to occupational and industry gender ment just for employed people. Second, as we are not segregation in the labour market. u Th s, Kjeldstad and aware of a methodology that allows decomposing a gap Nymoen (2012a) and Kjeldstad and Nymoen (2012b) using this kind of model, we develop one that is inspired find a higher underemployment risk in those occupations by the Fairlie technique (1999, 2005, and 2017). It is based and sectors that are traditionally female-dominated, with on two counterfactual simulations that provide a detailed a stronger effect for men, although they do not include decomposition of the gap into effects due to workers hav - specific variables for segregation in their econometric ing different characteristics, effects due to these charac - model. Kamerāde and Richardson (2018) consider seg- teristics having different returns, and the interaction of regation measures in their analysis, and they also find a both these effects. higher likelihood of underemployment in female-dom- We focus on young workers because this collective is inated occupations; however, this effect is not so clear especially affected by underemployment. Data from the across industries. Additionally, Dueñas-Fernández et al. Spanish Labour Force Survey (LFS) indicate that in 2017 (2016), who do not focus specifically on this issue, ana - the underemployment rate for workers under 35 was lyse involuntary part-time work in Spain and find that 14.7%, while the figure was 8.5% for workers older than segregation, especially occupational, is strongly related to 34 years old. Additionally, by looking at people right at part-time work (particularly for women). the beginning of their careers we can avoid many of the There are several reasons that explain higher under - cumulative advantages/disadvantages that people may employment risks in female-dominated occupations and have experienced throughout their careers. Thus, focus - industries. In this sense, as Kamerāde and Richardson ing on young workers allows us to have a current view of (2018) point out, women are mainly employed in labour underemployment patterns and of the gender gap in it, intensive jobs where employers can change the number avoiding possible gender differences from the past. of hours their employees work to adapt to fluctuations This paper is organised as follows. It begins with the in demand. Therefore, part-time or short-schedule jobs conceptual framework on time-related underemploy- are more likely to be found in female-dominated occupa- ment and puts forward our working hypotheses. The next tions, which may lead to underemployment. Moreover, section describes the methodological approach used. female-dominated occupations usually require low quali- Then, the data and variables used in the econometric fications. In addition, female workers tend to cluster in model are presented. This is followed by the results, while industries that offer comparatively low payment for the a discussion of these findings is provided at the end. same level of qualification, such as in education, health and social work activities (Boll et al. 2016). This pattern 2 Conceptual framework and hypotheses also translates into a frequent desire to work more hours. According to neoclassical theory, individuals can choose Conversely, male-dominated occupations are typically their working hours freely from a continuous time dis- characterised by better-paid jobs and are usually related tribution; these hours are chosen by maximising a util- to more stable, full-time contracts (Hegewisch et al. ity function subject to a particular budget constraint. 2010). Furthermore, male workers are overrepresented Nevertheless, employers and trade unions’ decisions, in industries that offer high rewards for the same level the degree of labour mobility and economic conditions of qualification (particularly manufacturing). Therefore, determine the actual hours offered to employees (Simic male-dominated occupations and industries often lead to 2002). Therefore, workers’ preferred and actual hours low underemployment rates. may not coincide, so some individuals will work either Despite these arguments suggesting the important role more (overemployed) or less (underemployed) than they that occupational and industry segregation plays in the want. Thus, time-related underemployment means that likelihood of underemployment, accurate estimates of Measuring the effect of gender segregation on the gender gap in time‑related underemployment Page 3 of 16 22 its impact have not been achieved in the aforementioned is mainly due to the different distribution of male research. In this article, we overcome this shortcoming and female workers across occupations and indus- using a more suitable estimation strategy. Specifically, we tries. propose and test the following hypothesis: Hypothesis 2b: The different impact that working in certain occupations and industries exerts on the risk Hypothesis 1: working in female-dominated occu- of male and female underemployment contributes to pations and industries implies a higher probabil- widening the gender gap. ity of time-related underemployment than being employed in male-dominated ones, both for men and women.3 Methodology As a first step, to analyse the effect of occupational and As women face a higher risk of underemployment industry segregation on time-related underemploy- than men do, there is a gender gap regarding this handi- ment, we estimate two bivariate probit selection mod- cap and occupational and industry segregation may els (Greene 2012), one for men and another for women. have an important impact on it. Furthermore, this effect These models enable us to handle the potential sample may be partially due to the uneven distribution of men selection bias due to estimating the probability of under- and women across different jobs, as suggested by Bar - employment just for employed people, as Acosta-Balles- rett and Doiron (2001). Additionally, differences in the teros et al. (2018) have already shown. returns that working in female or male-dominated jobs ∗ ∗ Let us define y and y as the latent variables reflecting 1 2 imply should also be considered. These authors, as a sim - the likelihood of being underemployed and employed, ple exercise, give women the average male distribution respectively. Thus, the model can be specified as follows: across occupations and industries and conclude that the ∗ ∗ main reason that explains women being involuntary part- y = x γ + ε , y = 1 if y > 0, 0 otherwise i1 1 i1 i1 i1 i1 timers more often than men is simply being employed in (1) different industries and occupations. ∗ ∗ y = x γ + ε , y = 1 if y > 0, 0 otherwise i2 2 i2 i2 i2 i2 Interestingly, previous research highlights the fact that (2) men may benefit from their minority status in female- with (y , x ) observed only when y = 1. i2 i1 i1 dominated jobs in several ways (Simpson 2004). In this In these equations, y indicates if worker i is underem- i1 sense, as reviewed in Lupton (2006), men progress more ployed and y if the individual is employed; row vector i2 quickly than women do to senior positions avoiding the x contains the variables explaining underemployment; i1 problem of the “glass ceiling” inherent in vertical segre- x reflects the variables determining employment. As i2 gation. Additionally, men may be channelled into cer- usual, the independent variables that have a qualitative tain specialties in occupations that are regarded as more nature are included in the model as dummy variables or appropriate to their gender. As a third advantage, men as groups of them. Finally, ε and ε are the error terms, i1 i2 are paid more than women are in female-dominated which follow a bivariate normal distribution with mean occupations (Torre 2018). By contrast, women may face zero, variance equal to 1 and covariance ρ negative outcomes in male-dominated jobs (Simpson To test if working in female-dominated occupations 1997, 2000). Thus, for example, as Martin and Barnard and industries implies a higher probability of under- (2013) find, formal and covert organisational practices, employment than working in male-dominated ones which maintain gender discrimination and bias, are the (Hypothesis 1), we analyse the estimated marginal effects main challenges that women face. These arguments may of occupational and industry segregation on the prob- also apply regarding underemployment, so female work- ability of underemployment. Since the model is bivariate ers may face a higher risk of underemployment than men with selection, these partial effects (like those regarding both in female and male-dominated jobs. the rest of variables) are obtained using the conditional Nevertheless, to the best of our knowledge, only probability of underemployment given employment. Vuluku et al. (2013) have tried to identify the reasons In addition, the marginal effects on the probability of behind the underemployment gender gap, though they employment are computed using the selection equation. do not include any measure of gender segregation in their To simplify notation, we redefine the variables and analysis and use univariate models, which can lead to coefficients in Eqs. (1) and (2) as follows. The variables biased estimations. We fill this gap in the literature using a new methodology that allows us to propose and test the following hypotheses: Hypothesis 2a: The gender gap in underemployment The interdependency caused by groups of dummies has been taken into account in order to calculate the marginal effects. 22 Page 4 of 16 J. Acosta‑Ballesteros et al. considered in both equations for individual i are gath- non-linear models, the independent contribution of one ered in x , which is a row vector including vectors x and variable to the gap depends on the value of the other i i1 x . Additionally, vector β contains the estimated values variables, which may imply a path dependence problem. i2 1 for γ (γ ) and takes value zero for those variables in x Moreover, Fairlie methodology does not identify the 1 1 i2 which are not included in x . In a similar way, β includes coefficients effect corresponding to a specific variable, i1 2 components equal to zero for those variables consid- which is required to test Hypothesis 2b. ered in Eq. (1) but not in Eq. (2). Thus, x β ≡ x γ and The aggregate decomposition in our methodol - i 1 i1 1 x β ≡ x γ . ogy, which is a direct extension of Fairlie’s, is defined i 2 i2 2 According to this notation, the estimated probability of by Eqs. (4) to (6), where E reflects the endowments being underemployed conditioned to being employed for effect using as weights women’s coefficients, and C individual i is: quantifies the coefficients effect using as weights men’s characteristics: j j BVN(x β , x β , ρ ) j j i i j 1 2 F x β , x β , ρ = i i i (3) Gap = C + E 1 2 M W (4) �(x β ) M M M M M where Φ is the cumulative standard normal distribution F x β , x β , ρ ∀Men i 1 i 2 C = function and BVN is the joint cumulative distribution of (5) M W M W W the bivariate normal. Superscript j refers to men (M) or F x β , x β , ρ ∀Men i 1 i 2 women (W). As stated above, our main objective is to identify the most relevant factors explaining the gender gap in M W M W F x β , x β , ρ ∀Men i 1 i 2 underemployment and, more specifically, to test if gen - E = der segregation accounts for an important portion of (6) W W W W F x β , x β , ρ it. To achieve this goal and test Hypotheses 2a and 2b, a ∀Women i 1 i 2 detailed decomposition of the gap is required. The traditional Oaxaca-Blinder two-fold decomposi - Summations in (5) and (6) are across the subsample of tion (Blinder, 1973 and Oaxaca, 1973) of the gap into the employed, as we decompose differences in the average endowments (portion of the gap due to group differences predicted probabilities of being underemployed condi- in observable characteristics) and coefficients effects W M tioned to being employed. Thus, N and N indicate the (the “unexplained” portion of the gap) cannot be applied sample size for employed women and men, respectively. because our model is not linear. Previous research (Even An alternative decomposition (Eq. 7) with each com- and Macpherson, 1990; Doiron and Riddell, 1994; Fairlie ponent evaluated using as weights the other gender coef- 1999, 2005, 2017; Yun, 2004, 2008; Powers et al., 2011; ficients or endowments is also possible. However, we do and Bazen et al., 2017) has decomposed the gap in pro- not define and explain it here because it is symmetric to bit and logit models, with the Fairlie and Yun techniques this one. being the two most widely applied. However, as we esti- mate a nonlinear model with two equations, we develop a Gap = E + C M W (7) new procedure to decompose the gap, which extends the To obtain a detailed decomposition, we develop a Fairlie technique to this kind of model. We have chosen methodology based on two counterfactual simulations the Fairlie approach as our starting point because it uses that identify the contribution of each variable to both a non-linear function to obtain the gap decomposition, E and C . These simulations can be used together to while in the Yun procedure, the curvature of the corre- W M approximate the total impact of a specific variable on the sponding function is not considered. underemployment gender gap. According to the Fairlie technique, the contribution of The first one provides a detailed decomposition of the each observable variable to the explained portion of the endowments effect and has been designed for discrete gap is equal to the change in the average predicted prob- variables (as most of the variables in the labour market ability from replacing (for instance) the female distribu- tion with the male distribution of that variable (keeping constant the rest). The procedure he proposed is match - Fairlie does not focus on the "unexplained" portion of the gap because of the ing one-to-one individuals in the female and male sub- difficulty in interpreting results. As we explain below, this shortcoming can be samples and switching the distributions of variables overcome using the Kim (2013) methodology. sequentially from a woman to a man. Nevertheless, the For a continuous variable, we propose equalling the distribution of fre- quencies by gender as described, but considering the variable is discrete order of switching is potentially important because in with a very large number of categories. Measuring the effect of gender segregation on the gender gap in time‑related underemployment Page 5 of 16 22 literature). It is inspired by Fairlie, who pointed out that differences in Taylor expansion remainders (see Appen - a potential solution to the path dependence problem “is dix 1 in Additional file 1). to estimate each contribution by switching the variable To calculate the detailed decomposition of the coeffi - of interest first” (Fairlie 2005, page 313), as our method cients effect, we propose a second counterfactual simu - does. Specifically, we calculate the contribution of a sin - lation following a similar procedure to that used in the gle variable k as the change in women’s average condi- first one. Oaxaca and Ransom (1999) show that this tional probability of underemployment resulting from decomposition is destined to suffer from an identifica - switching women from the categories where they are tion problem, since the detailed coefficients effect attrib - over-represented to those where they are under-repre- uted to dummy or categorical variables is not invariant sented. This procedure is carried out until women’s rel - to the choice of reference groups. Gardeazábal and Ugi- ative frequencies across the categories of k are equal to dos (2004) and Yun (2005) propose methods to solve this men’s ones. The selection of women who are switched is problem. Despite being widely used, these approaches random, so the procedure is repeated 50 times to ensure show some limitations. u Th s, we use the grand-mean consistency, and then the results are averaged. As the method that Kim (2013) proposes. This method appears changes described affect 10% of the observations or less to be a good option for analyses regarding labour market for most variables, the change in the probability of under- outcomes because it accurately estimates the extent to employment is due to a relatively small change in the which each variable contributes to the group differences. data. Additionally, it gives a meaning to the intercept term and Specifically, the contribution of a single variable k, to the coefficient component of each dummy variable. denoted as θ (k) , can be computed as described in Specifically, we calculate the coefficient effect related Eq. (8): to a specific variable k,θ (k), k = 1, . . . n , as the change W→M W W→M W W W W W W W F x (k)β , x (k)β , ρ F x β , x β , ρ i i ∀Women ∀Women W i 1 i 2 i 1 i 2 (8) θ (k) = − W W N N W→M W where x (k) contains the same information as x in men’s average conditional probability of underemploy- i i but variable k has been modified as described and ment if the parameter of a specific characteristic were k = 1,2…n, where n refers to the number of categorical that of women. Additionally, it is necessary to include the variables included in x . change corresponding to parameter ρ,θ . These effects M M ∼ ∼ The sum of the individual contributions of all the varia - are described in Eqs. (10) and (11), where β and β 1 2 bles does not exactly equal the endowments effect. Thus, are the estimated transformed coefficients of the model the summing up property, which the method proposed M→W according to Kim (2013) method. In β (k) and by Fairlie has, does not satisfy. So we can write: 1 M→W β (k) , men’s coefficients corresponding to the vari - W W able k have been replaced by women’s ones. E = θ (k) + D E E (9) k=1 M→W M→W ∼ ∼ M M F x β (k), x β (k), ρ W 1 2 ∀Men i i An approximation error ( D ) emerges because the θ (k) = endowments effect ( E ) in the aggregate decomposition W N M M ∼ ∼ is computed by switching all the variables simultaneously. M M F x β , x β , ρ 1 2 ∀Men i i Conversely, in our simulation, we switch only one vari- able at a time. As the conditional probability of underem- ployment is not linear, both results are slightly different. (10) Although when both expressions, E and θ (k) , k=1 E are linearised they coincide, a disparity emerges from the According to Fortin et al. (2011) and Kim (2013), these normalizations have several limitations: they may leave the estimation and decomposition with- out a simple meaningful interpretation; they will likely be sample specific and make comparisons across studies impossible; and they are sensitive to the number of categories and to the grouping method. The estimated coefficients are transformed by subtracting from each of them the grand-mean weighted sum of the coefficients of each vari- This number of repetitions was selected after an analysis of sensitivity. We M M − − decided to choose 50 because the average difference in the results found with able β (k) or β (k), k = 1, . . . , n . It is also necessary to sum −5 1 2 respect to using 200 was around 10 and the standard errors could be com- M M − − n n puted in a reasonable time. β (k) or β (k) to the intercepts in order to transform k=1 1 k=1 2 Note that each discrete variable k is included as a set of dummies in x . them. i 22 Page 6 of 16 J. Acosta‑Ballesteros et al. M M ∼ ∼ point, so they are more easily interpreted. This three-fold M M W β β 9 F x , x , ρ ∀Men 1 2 i i decomposition when the starting point is women can be θ = ρ easily obtained from Eqs. (4) or (7) and can be expressed (11) as: M M ∼ ∼ M M M β β F x , x , ρ 1 2 ∀Men i i Gap = E + C +(C − C ) W W M W (13) Gap = E + C +(E − E ) W W M W (14) Again, summing the individual contributions of the variables does not exactly equal the coefficients effect, The new term, C − C = E − E , can be inter- M W M W C . An approximation error ( D ) emerges for the same C preted as an interaction component that indicates the reasons already explained (see Appendix 2 in Additional portion of the gap that occurs when both endowments file 1). Thus, we can write : and coefficients change simultaneously. Alternatively, it is the portion of the gap that remains after controlling for M M M the endowments and coefficients effects. This interaction C =− θ (k) + θ + D (12) β ρ C component is more difficult to interpret than the first two k=1 and is often disregarded. However, we believe, as Etezady Since our detailed decomposition of the gap is the et al. (2021), that neglecting it provides a substantially sum of both expressions (E and C ), the approxima- W M incomplete picture of the total influences of endowments tion errors imply that the sum of individual contributions and coefficients to the gap. u Th s, our analysis is based on of all the variables does not equal the gap. To assess the Eq. (14). magnitude of this disparity, in the Results section, we dis- Some final comments regarding our methodology play the approximation errors of our decomposition. need to be pointed out. First, to obtain the standard Despite our decomposition of the gap not being exact, errors for the results of both counterfactual simulations, it provides technical advantages compared to Fairlie which are necessary to test if the corresponding changes decomposition procedure, as well as being applicable to in the probability of underemployment are statistically a bivariate probit model. Thus, its economic interpreta - significant, Krinsky and Robb’s (1986) method has been tion is straightforward, and it avoids the path dependence applied, as Dowd et al. (2014) explain. problem, since it always uses the same starting point, real Second, the survey structure of our data has been taken women (or men) in the sample, and only one characteris- into account in the methodology. Thus, the bivariate pro - tic is modified. Moreover, our approach does not require bit selection models have been estimated considering a one-to-one matching of individuals, since we replicate sample weights and cluster-robust standard errors. Addi- the distribution of each specific variable and the num - tionally, the sample weights have been considered in both ber of women who have a specific characteristic changed counterfactual simulations by replicating each observa- is just those strictly necessary, so we keep almost real tion according to its weight. individuals. Conversely, in Fairlie decomposition tech- Third, our methodology is displayed for bivariate pro - nique, each woman is randomly matched with a man in bit models with sample selection, but it can also be easily the sample, and she takes his characteristics sequentially applied to single equation models like the probit or logit until she becomes that man. As the sequential change ones. This fact allows us to carry out some robustness of characteristics is made, it is likely that the remaining analyses. Thus, we specify univariate probit models to combination of characteristics will be unreal. In addition, explain underemployment and we obtain the three-fold our methodology offers a simulation that allows us to decomposition of the gap. These results are compared to approximate a detailed decomposition of the coefficients effect. Even though we could test our hypotheses using a two- Equations 13 and 14 can also be proposed for men. fold decomposition, it is increasingly common in the lit- In studies based on survey data not only the outcome variable but also erature to use a three-fold one (Daymont and Andrisani, the predictors are subject to sampling variation (Jann 2008). It implies that 1984), which has the advantage that endowments and the standard errors may be underestimated, especially those regarding the endowment component. However, the results of the models shown in Addi- coefficients effects are computed from the same starting tional file 2 seem to indicate that this is not the case, since the standard errors estimated using our methodology are very similar to those obtained using the Oaxaca-Blinder or Yun methodologies. This process is required in the first simulation in order to switch the Note that the minus sign is required in Eq. (12) because in the simulation value of a specific variable from those categories where women (men) are the average man is the starting point, while when computing C , the average over-represented. In the second simulation, weighting each observation man is the final point. according to its raising factor is enough. Measuring the effect of gender segregation on the gender gap in time‑related underemployment Page 7 of 16 22 those obtained using Fairlie and Yun methodologies for a criterion, we obtain a band similar to the one in Hakim probit model and to those achieved using Oaxaca-Blinder (1998), where gender-integrated occupations are charac- technique for a linear model. Table S1 in Additional file 2 terised by a proportion of women ten percentage points shows these results, which are similar to those obtained around the percentage of women in total employment. with our methodology. Our gender segregation measures have been com- puted using the three-digit codes from both occupations 4 Data and variables (according to the National Classification of Occupations, In this article, we use the definition of time-related 1994 and 2011) and industries (National Classification underemployment directly provided by the Spanish Sta- of Economic Activities, 1993 and 2009). However, when tistical Office. Specifically, the criteria applied in the the number of people working in a certain occupation Spanish LFS to classify workers as underemployed (in or industry is less than 50, segregation has been defined line with the International Labor Organization Bureau according to two-digit or one-digit codes. Additionally, of Statistics recommendations) are: they would like to given the methodological change in both classifications, work more hours, they are available to do so, and they it has been necessary to calculate the value of each seg- work less than the usual weekly hours of full-timers in regation variable for two different sub-periods. As both their industry. u Th s, underemployment is a more accu - gender segregation measures are correlated, we have rate indicator of labour underutilization than involuntary solved the collinearity problem by defining an interaction part-time employment. It reflects non-desired workdays variable with nine categories that integrates both occupa- for all types of workers, capturing the preference of both tional and industry segregation. part-timers and full-timers to have longer workdays. As education plays an important role in the risk of The data used come from the 2006–2016 annual sam - underemployment (Acosta-Ballesteros et al. 2018), we ples of the Spanish LFS. Therefore, our database is a define 43 educational categories using the information pool of cross-sectional annual observations, since each provided by the LFS on education level and field of study, individual is included only once in the annual sam- and according to the National Classification of Education ple. Our sample contains young people aged 16 to 34 (2000 and 2014). Ten specializations for vocational train- who were active. The few individuals with inconsistent ing and university degrees are distinguished. Moreover, answers or who do not provide the necessary informa- whether workers took longer than usual in completing tion for the analysis have been removed. The final sample their studies is also considered. includes 70,445 women: 73.6% are employed, and among The remaining explanatory variables include nation - them, 16.5% are underemployed. The corresponding fig - ality, having children under 16, and some additional ures for the male subsample are 80,962, 75.1% and 11.8%, regressors reflecting household composition; whether respectively. the individual is enrolled in formal studies is also taken The independent variables included in the economet - into account. We also consider professional status (self- ric analysis (displayed in Table 1) reflect the main fac - employed or employed in the public or private sector tors previously found to determine underemployment. with a fixed-term or permanent contract), the size of In order to classify occupations and industries as gender- the firm, having a recent job (tenure up to 12 months dominated or integrated, we follow the relative concept and depending on the worker’s age), the unemployment of Anker (1998). Thus, the dividing line between gender- rate by gender in the Autonomous Regions, as well as dominated and integrated occupations (or industries) a dummy variable that takes value one if the observa- is established in relation to the average percentage of tion corresponds to the period after the labour reform of female workers in the labour force as a whole (44% over 2012. the period analysed). Specifically, we consider female- dominated occupations or industries are those hav- This fact only occurs in a few occupations (industries) that account for 0.15% (0.89%) of workers in our sample. ing more than 1.25 times the mean percentage female, Before the Bologna Process, the Spanish education system distinguished while male-dominated ones are those having less than short-cycle (three years) and long-cycle (more than three years, usually 0.75 times the mean percentage female. If the percent- five) university degrees. The new degrees under the European Higher Edu- age of women is between both limits, the occupation or cation Area are included as short-cycle programmes. This is the only continuous variable in the model. As it is already defined industry is labelled as gender-integrated. Applying this by gender, it is not necessary to develop the procedure explained above. us Th , in the first simulation, the change in this characteristic has been car - ried out by simply attributing each woman the unemployment rate she would face if she were a man (and vice-versa). The accuracy of the results in this paper using these data is our sole respon - sibility. This reform, among other measures, allows firms to reduce the working The frequencies of the independent variables are provided in Table S2 in hours of their employees more easily than before, and may partially explain Additional file 2. the relatively high underemployment rate observed since 2012 in Spain. 22 Page 8 of 16 J. Acosta‑Ballesteros et al. Table 1 Average marginal effects: underemployment and employment by gender Variable Women Men Under Employ Under Employ Educational attainment (omitted: primary education or less) Compulsory secondary − 0.006 0.077*** 0.003 0.082*** Non‑ compulsory secondary − 0.024*** 0.141*** − 0.013 0.142*** Lower vocational training Education − 0.026 0.116*** 0.049 0.181*** Arts and humanities 0.012 0.170*** 0.066 0.116*** Social sciences − 0.092** 0.342*** − 0.109*** 0.149 Business, administration, law − 0.014 0.134*** − 0.009 0.133*** Sciences − 0.012 0.194*** 0.036 0.029 ICT − 0.052 0.124** 0.060* 0.135*** Technology − 0.026 0.096*** 0.002 0.138*** Agriculture 0.007 0.121** − 0.021 0.163*** Health − 0.036*** 0.183*** 0.005 0.127*** Social services − 0.031*** 0.145*** − 0.006 0.122*** Higher vocational training Education − 0.019 0.171*** 0.073* 0.190*** Arts and humanities − 0.016 0.136*** − 0.011 0.138*** Social sciences 0.025 0.153*** 0.020 0.179*** Business, administration, law − 0.034*** 0.171*** − 0.024* 0.169*** Sciences − 0.054** 0.174*** − 0.045 0.160*** ICT − 0.028 0.187*** − 0.042*** 0.206*** Technology − 0.005 0.174*** − 0.016* 0.179*** Agriculture − 0.005 0.030 0.020 0.156*** Health − 0.038*** 0.181*** − 0.003 0.164*** Social services − 0.036** 0.153*** − 0.013 0.155*** Short‑ cycle university Education − 0.028** 0.224*** 0.021 0.185*** Arts and humanities 0.062* 0.125*** − 0.070*** 0.117*** Social sciences 0.012 0.132*** 0.055 0.147*** Business, administration, law − 0.059*** 0.169*** − 0.042*** 0.189*** Sciences − 0.046 0.146*** − 0.022 0.170*** ICT − 0.060** 0.161*** − 0.060*** 0.214*** Technology 0.007 0.152*** − 0.054*** 0.174*** Agriculture − 0.020 0.160*** − 0.058** 0.114** Health − 0.040*** 0.230*** − 0.004 0.200*** Social services − 0.067*** 0.182*** − 0.033 0.204*** Long‑ cycle university Education 0.000 0.203*** 0.008 0.167*** Arts and humanities 0.015 0.127*** − 0.001 0.112*** Social sciences − 0.017 0.130*** − 0.008 0.125*** Business, administration, law − 0.074*** 0.192*** − 0.069*** 0.183*** Sciences − 0.030* 0.180*** − 0.070*** 0.184*** ICT − 0.099*** 0.270*** − 0.098*** 0.246*** Technology − 0.012 0.149*** − 0.057*** 0.201*** Agriculture − 0.048* 0.169*** − 0.093*** 0.147*** Health − 0.075*** 0.272*** − 0.078*** 0.267*** Social services − 0.027 0.210*** − 0.051*** 0.161*** Student − 0.011* − 0.058*** − 0.001 − 0.072*** Measuring the effect of gender segregation on the gender gap in time‑related underemployment Page 9 of 16 22 Table 1 (continued) Variable Women Men Under Employ Under Employ Time to complete studies (omitted: within appropriate time) Less than three years late 0.013*** − 0.044*** 0.009 − 0.028*** Three or more years late 0.047*** − 0.074*** 0.010 − 0.052*** Age (omitted: 16–19) 20–24 0.019 0.158*** 0.010 0.084*** 25–29 0.008 0.232*** 0.009 0.135*** 30–34 0.008 0.269*** 0.003 0.153*** Live with parents 0.018*** − 0.099*** − 0.004 − 0.092*** Live in couple − 0.006 − 0.010 − 0.028*** 0.048*** Children under 16 0.003 − 0.086*** 0.017** − 0.029*** Non‑Spanish 0.050*** − 0.057*** 0.071*** − 0.076*** Occupational segregation (omitted: male‑ dominated) Integrated 0.007 − 0.001 Female‑ dominated 0.046*** 0.028*** Industry segregation (omitted: male‑ dominated) Integrated 0.023*** 0.026*** Female‑ dominated 0.045*** 0.016** Tenure 12 months or less 0.039*** 0.041*** Professional status (omitted: self‑ employed) Public sector employee with permanent contract − 0.000 0.001 Public sector employee with fixed‑term contract 0.072*** 0.067*** Private sector employee with permanent contract 0.040*** 0.029*** Private sector employee with fixed‑term contract 0.124*** 0.092*** Firm size (omitted: up to 10 workers) More than 10 workers − 0.040*** − 0.021*** Unknown − 0.042*** − 0.041*** Unemployment rate 0.002*** 0.003*** NUTS Region (omitted: Northwest) Northeast − 0.003** 0.043*** − 0.002*** 0.019*** Madrid − 0.002** 0.032*** − 0.001 0.010 Centre 0.002** − 0.028*** − 0.000 0.002 East − 0.002** 0.026*** 0.000 − 0.001 South 0.003** − 0.056*** 0.006*** − 0.059*** Canary Islands 0.003** − 0.045*** 0.007*** − 0.064*** After 2012 labour reform 0.041*** 0.026*** The year when the survey was carried out is controlled for, but not reported Under: underemployed. Employ: employed Unemployment rate by gender and year in the Autonomous Region of residence p < .10. **p < .05. ***p < .01 Finally, the employment equation includes variables as 5 Results educational attainment, taking longer than usual to grad- 5.1 Marginal effects uate, age, household composition, enrolled in school, the In this section, we present the results of the bivariate area of residence (which reflects the general conditions probit selection models for both subsamples, men and of local demand for work) and the year when the worker women. The estimations indicate that the rho coefficients was interviewed. are 0.207 (p-value = 0.01964) and 0.361 (p-value = 0.0000), 22 Page 10 of 16 J. Acosta‑Ballesteros et al. respectively, justifying the use of these models to handle male peers. Thus, the underemployment gender gap the sample selection bias due to estimating the underem- (−0.039) is not negligible. ployment probability just for employed people. As stated above, the main objective of this article is to The marginal effects of the explanatory variables have carry out an in-depth analysis of this gap and to deter- been computed from the estimated coefficients and are mine the impact of gender segregation on it. To do this, displayed in Table 1. The ones corresponding to the we have carried out a detailed three-fold decomposi- employment equation (Columns 2 and 4) follow the same tion of the gap obtained from Eq. (14) and the results are direction as in previous studies. shown in Table 2. The results obtained for underemployment are shown According to the last row in Table 2, the overall endow- in Columns 1 and 3. Focusing on the role of gender seg- ments effect, E , is very important and seems to explain regation, we observe that workers in female-dominated the underemployment gender gap (130.8%), while the occupations are those with the highest probability of coefficient effects, C , is not significant. Moreover, the underemployment, with a larger impact for women. Spe- interaction term, which is positive (0.0126), reduces the cifically, we find that young women working in female- gap (− 32.7%). dominated occupations are 4.6 percentage points more Regarding the detailed decomposition of the gap, the likely to be underemployed than those in male-dominated sum of the individual contributions of all the variables occupations; the corresponding increase in the male sub- to each component of the gap shows that the errors sample is 2.8 points. When segregation is defined in rela - terms from our counterfactual simulations are small and tion to activity sectors, our results indicate that women not statistically significant. These results validate our in female-dominated industries also show the highest approach. likelihood of underemployment (an increase of 4.5 per- The results in columns 1–3 allow us to conclude that centage points). Men in female-dominated activities are the different distribution of men and women across jobs also more often underemployed than in male-dominated is crucial to explain the gender gap, as stated in Hypoth- ones. However, men in gender-balanced industries are esis 2a. Thus, if women were distributed across occu - the most prone to suffer this handicap. Altogether, these pations and industries as men are (maintaining female results support Hypothesis 1, confirming a higher under - coefficients), the gap would be reduced to the greatest employment risk associated with working in female- extent (− 0.0453). This result suggests that gender segre - dominated jobs than in male-dominated ones. gation leads to a kind of discrimination against women. The marginal effects of the remaining explanatory vari - Indeed, it cannot be argued that women work more fre- ables are not discussed due to space constraints. We only quently in certain (female-dominated) jobs because they want to briefly point out some of the results regarding prefer shorter work-schedules. Conversely, their larger educational attainment, due to the relevance of this vari- risk of underemployment, which means they would like able in almost any labour market outcome. The figures in to work more hours than available more often than their Table 1 suggest that having a long-cycle university degree male peers, is mainly due to the kind of jobs they work in. in almost any field reduces underemployment. Overall, Focusing on the coefficients effect (Columns 4–6 in business, administration and law, and information and Table 2), the contribution of the intercept term of the communication technologies (ICTs) seem to be the best underemployment equation (Intercept 1) quantifies the specializations at most education levels. Additionally, it extent to which women are, on average, treated differ - is worth noting that education means greater differences ently to men. This contribution can be interpreted as the in the risk of being underemployed for women than for average extent of discrimination (Kim 2013). Our results men. Some additional comments are included in the sen- indicate that the contribution of Intercept 1 to the gap is sitivity section. small and not significant, so women and men would face the same average risk of underemployment. However, 5.2 A nalysis of the underemployment gender gap the coefficient effect of gender segregation indicates an According to the results of the bivariate probit models, important deviation from the mean discriminatory level. women’s estimated conditional probability of underem- Specifically, the gender gap in underemployment would ployment is 0.163; the corresponding figure for men is reduce (− 0.0146) if women had men’s returns in the 0.124. Therefore, young female workers in Spain are 1.31 same gender-typing jobs (maintaining female character- times more likely of being underemployed than their istics). Therefore, the different impact on underemploy - ment that working in certain occupations and industries exerts on the risk of experiencing it contributes to widen- The marginal effects corresponding to each gender segregation measure ing the gender gap, supporting our Hypothesis 2b. In fact, have been obtained from the estimated parameters of the interaction variable this is the most important contribution to the coefficient capturing both occupational and industry segregation. Measuring the effect of gender segregation on the gender gap in time‑related underemployment Page 11 of 16 22 Table 2 Three‑fold decomposition of underemployment gender gap Variable Endowments Coefficients Interaction Total Contribution SE %Gap Contribution SE %Gap Contribution SE %Gap %Gap Educational attainment 0.0073** 0.00314 − 18.9 0.0053** 0.00218 − 13.8 − 0.0061*** 0.00364 15.7 − 17.0 Student 0.0003* 0.00018 − 0.9 0.0001 0.00007 − 0.2 − 0.0003 0.00024 0.9 − 0.2 Time to complete studies − 0.0001*** 0.00004 0.3 − 0.0003** 0.00012 0.7 0.0001 0.00006 − 0.2 0.8 Age − 0.0021*** 0.00004 5.4 − 0.0001* 0.00005 0.2 − 0.0052* 0.00029 13.4 19.0 Live with parents 0.0016*** 0.00054 − 4.2 0.0019** 0.00078 − 4.8 − 0.0018* 0.00072 4.8 − 4.2 Live in couple 0.0006 0.00063 − 1.7 − 0.0006* 0.00033 1.5 0.0019* 0.00087 − 4.8 − 5.0 Children under 16 − 0.0003 0.00047 0.7 0.0006 0.00049 − 1.6 − 0.0011 0.00072 2.8 1.9 Non− Spanish − 0.0009*** 0.00012 2.4 0.0009 0.00057 − 2.3 − 0.0001 0.00015 0.2 0.3 Gender segregation − 0.0453*** 0.00355 117.0 − 0.0146*** 0.0032 37.8 0.0217*** 0.00473 − 56.2 98.6 Professional status − 0.0028*** 0.00038 7.3 − 0.0041*** 0.00091 10.7 0.0005** 0.00058 − 1.2 16.8 Recent job − 0.0008*** 0.00022 2.1 − 0.0003 0.0008 0.7 0.0000 0.00026 0.1 2.9 Firm size − 0.0030*** 0.00032 7.7 − 0.0016*** 0.00056 4.2 0.0011* 0.00042 − 2.8 9.1 Unemployment rate − 0.0059*** 0.00077 15.1 0.0057 0.01078 − 14.7 − 0.0002** 0.0011 0.6 1.0 NUTS Region 0.0000 0.00001 0.0 0.0000 0.00003 0.1 0.0000 0.00002 − 0.1 0.0 After 2012 reform − 0.0009*** 0.0001 2.2 − 0.0013*** 0.00047 3.2 0.0005*** 0.00012 − 1.2 4.2 Year − 0.0003** 0.00012 0.8 0.0005** 0.00022 − 1.2 − 0.0002 0.00016 0.5 0.1 Intercept 1 0.0016 0.00894 − 4.1 − 4.1 Intercept 2 − 0.0025** 0.00102 6.4 6.4 0.0114*** 0.00936 − 29.3 − 29.3 Approximation error 0.0017 0.00441 − 4.4 − 0.0033 0.0046 8.5 0.00199 0.0054 − 5.1 − 1.0 Aggregate − 0.0506*** 0.0037 130.8 − 0.0007 0.00454 1.9 0.0126* 0.0051 − 32.7 100.0 Standard errors (SE) have been computed using Krinsky and Robb’s method (1,000 draws). %Gap: Change as a percentage of the total underemployment gender gap effect affecting the underemployment gender gap, rein -5.3 Sensitivity analysis forcing our previous finding regarding the discriminating In this subsection, we carry out different sensitivity anal - effect of segregation against women. yses to check the robustness of our results. Table 3 shows The interaction term captures the effect of changing the estimated gender gap in underemployment and the endowments and coefficients simultaneously. The posi - specific contribution of occupational and industry segre - tive sign of this portion suggests that the risk of under- gation to it, according to Eq. (14). employment in female-dominated jobs, where women In Model I, we re-estimate our baseline model using an are indeed highly represented, is significantly larger for alternative definition of gender segregation. Specifically, women than for men. Therefore, this interaction term we have re-defined integrated occupations and industries reflects that, once the returns have been changed from as those where the female percentage is between 0.5 and women to men, the additional contribution of distribut- 1.5 times the average female share of employment in the ing women across jobs as men is only − 0.0236 (which is labour force. The following two models consider gender 0.0217 smaller than the contribution of segregation to the segregation only in occupations (Model II) or indus- endowments effects). tries (Model III) to analyse their separate effects. u Th s, Regarding the total impact of each individual variable we can check if our results change when no threshold is to the underemployment gender gap, figures in the last established to classify occupations and industries as gen- column in Table 2 show gender segregation is the most der-dominated or integrated. Additionally, the baseline important one explaining it (98.6%). Moreover, gender model is re-estimated splitting the original sample into differences in age also widen the gap, explaining 19% of two subsamples: workers with tertiary education (Model it. A similar conclusion is obtained for professional status IV) and those without (Model V). Therefore, taking into (16.8%). Conversely, educational attainment reduces the account the marginal effects already analysed, we can test underemployment gap (− 17%). our hypotheses for more homogeneous groups of educa- tion. Model VI is the same as the benchmark model but includes inactive people together with unemployed ones 22 Page 12 of 16 J. Acosta‑Ballesteros et al. Table 3 Sensitivity analysis: Contribution of occupational and industry segregation to the underemployment gender gap Gap Endowments Coefficients Interaction Total Contribution SE %Gap Contribution SE %Gap Contribution SE %Gap %Gap Baseline model − 0.0387 − 0.0453 0.00355 117.0 − 0.0146 0.0032 37.8 0.0217 0.00473 − 56.2 98.6 Model I − 0.0387 − 0.0439 0.00422 113.4 − 0.0147 0.00368 38.1 0.0215 0.00539 − 55.5 96.0 Model II − 0.0387 − 0.0386 0.00313 99.9 − 0.0118 0.0029 30.5 0.0184 0.00426 − 47.5 82.9 Model III − 0.0387 − 0.0295 0.00239 76.1 − 0.0102 0.00222 26.4 0.0166 0.00313 − 43.0 59.5 Model IV − 0.0369 − 0.0361 0.00322 98.1 − 0.0007 0.00368 1.9 0.0083 0.00530 − 22.5 77.5 Model V − 0.0536 − 0.0524 0.00608 97.9 − 0.0254 0.01318 47.4 0.0351 0.00750 − 65.5 79.8 Model VI − 0.0387 − 0.0451 0.00355 116.6 − 0.0143 0.00419 37.0 0.0215 0.00473 − 55.5 98.1 Model VII − 0.0742 − 0.0615 0.00237 82.8 − 0.0124 0.00359 16.7 0.0301 0.00332 − 40.5 59.0 Model VIII − 0.0387 − 0.0429 0.00371 111.0 − 0.0163 0.00368 42.1 0.0170 0.05017 − 44.0 109.1 Model IX − 0.0384 − 0.0466 0.00364 121.2 − 0.0154 0.00329 40.0 0.0232 0.00482 − 60.4 100.8 Model X − 0.0387 − 0.0453 0.00355 117.0 − 0.0179 0.00417 46.2 0.0217 0.00473 − 56.2 107.0 Standard errors (SE) have been computed using Krinsky and Robb’s method (1,000 draws). %Gap: Change as a percentage of the total underemployment gender gap. All the values for the contribution of segregation are significant at 99% except the coefficient and interaction components in Model IV, which are not significant when estimating the employment equation. Model VII As expected, in Models II and III the three effects are includes the same independent variables as the baseline smaller than those obtained in the baseline model. It is model but only involuntary part-time workers are consid- noteworthy that the portion of the gap explained by seg- ered underemployed, as is often the case in the literature. regation is larger when we consider only occupational The last three models include some methodological segregation. Some differences are also found if only invol - changes. In Model VIII, we adjust the variable gender seg- untary part-time workers are considered underemployed regation last instead of first to check if the order of switch - (Model VII). This decision implies estimating our models ing affects the results. As Heckman-type selection models using an alternative endogenous variable, so the results are mostly identified by assumptions about error distribu - obtained are likely to change significantly. However, the tions, in Model IX we use Inverse Probability Weighting different distribution of men and women across occupa - (IPW) based on a probit model as an alternative technique tions and industries is what still drives the gap. to address sample selection problems (see Seaman and The most interesting results are those obtained for White 2013 for a review). Finally, Model X is estimated workers with different levels of education (Models IV and using the methodology proposed in Gardeazábal and Ugi- V). It is worth noting that the estimated underemploy- dos (2004) to address the identification problem that arises ment gap is larger for workers without tertiary education, for categorical variables (instead of Kim’s approach). as might be expected. Gender segregation, however, is the In general terms, the results in the first column of factor that mainly explains the gap regardless of workers’ Table 3 are quite similar to those from the main analysis educational level, explaining almost the same percentage and few differences can be found. In particular, the larg - of it in both subsamples. Moreover, the different distri - est estimated underemployment gender gap is found for bution of men and women across jobs widens the gap to workers without tertiary education (Model V) and when a larger extent for less educated workers than for more only involuntary part-timers are classified as underem - educated ones. Additionally, only in the sample of work- ployed (Model VII). ers without tertiary education, the different returns asso - When we carry out the detailed three-fold decom- ciated with women and men widen the gap. Thus, we can position, our main conclusions remain unchanged. conclude that tertiary education reduces the gap not only because women and men are more similarly distributed across jobs (in comparison with less educated workers), but also because working in the same gender-typing jobs To facilitate comparisons with the baseline model, the reweighting leads to a similar risk of underemployment for male and approach has been applied only when estimating the coefficients of the probit that explains underemployment, but not when the decomposition of the gap female. has been computed. Therefore, the results obtained with IPW are conditioned Regarding the last three models, there are some to being employed. small differences with the baseline in the three compo - This result is not surprising, since involuntary part-time work in Spain nents, especially in the coefficients effect (that seems to is mainly a female problem. In fact, the average percentage of involuntary part-timers among male workers in the last 15 years is just 3.5%, while this increase). These differences translate into a larger pro - figure is 11.6% for women. portion of the gap explained by segregation in the three Measuring the effect of gender segregation on the gender gap in time‑related underemployment Page 13 of 16 22 models. However, our main conclusions still hold, so our Krinsky-Robb approach could lead to underestimating procedure is robust to the methodological changes they the standard errors. Despite our results indicating this is include. not the case, nonparametric bootstrap could be consid- Overall, the estimated underemployment gender gap ered as an alternative. is mainly explained by the different distribution of male The procedure has been tested in several ways and the and female workers across occupations and industries in main conclusions still hold. Firstly, the results obtained every specification of the model, which clearly supports applying our procedure to a univariate probit model are Hypothesis 2a. Moreover, Hypothesis 2b is also confirmed very similar to those obtained through other decompo- because, in most cases, the different impact that working sition techniques like Fairlie, Yun (2004) and Oaxaca- in certain gender-typing jobs exerts on the risk of male Blinder (Oaxaca 1973 and Blinder 1973). Secondly, the and female underemployment contributes to widening main results of our baseline model are robust to meth- the gender gap. Hence, the results in this subsection give odology changes, definition of some variables and to the robustness to our findings. sample used. Our results demonstrate that working in female- 6 Conclusions and discussion dominated occupations and industries implies a higher This article provides evidence of the crucial impact of probability of time-related underemployment than in occupational and industry segregation on the time- male-dominated ones, confirming our first hypothesis. related underemployment gender gap for people aged 16 Moreover, we find that the disadvantage in terms of to 34 using the annual samples of the Spanish LFS from underemployment that implies working in a female-dom- 2006–2016. It is worth noting that despite the Spanish inated occupation or industry is greater for women than labour market being deeply gender segregated, the effect for men. of this feature on the gap has not been addressed before Furthermore, according to our results, the estimated in the literature. Furthermore, to the best of our knowl- underemployment gender gap for young workers in edge, we are the first to perform a decomposition using Spain is 3.9 percentage points. As underemployment has bivariate probit models with sample selection. To do this, a negative impact on income, welfare dependency and we have developed a methodology based on two coun- life satisfaction (Wilkins 2007), the higher underemploy- terfactual simulations that provides a three-fold detailed ment risk faced by women implies negative consequences decomposition of the underemployment gender gap into related to experience, earnings, and possibly promotions endowments and coefficients effects as well as the inter - (Weststar, 2011). Therefore, designing effective policies action of these effects. leading to more gender equality will only be possible if Our methodology, inspired by the Fairlie technique the factors behind the gender gap in underemployment (1999, 2005, and 2017), has several advantages. Thus, it are clearly identified. To the best of our knowledge, the allows identifying the coefficients effect correspond - reasons for this difference have been little investigated. ing to a specific variable (while Fairlie technique does As an exception, Barrett and Doiron (2001) affirm that not). In fact, this effect can be easily interpreted using the main reason that explains women being involuntary Kim’s (2013) method to solve the identification problem. part-timers more often than men is the fact they are Moreover, we keep almost real individuals for two rea- employed in different industries and occupations. Never - sons. First, our approach does not require a one-to-one theless, Vuluku et al. (2013), who are the only ones who matching of individuals, since we replicate the distribu- have decomposed the underemployment gender gap, did tion of each specific variable, and the number of women not include occupational or industry segregation as an who have a specific characteristic adjusted is just those explanatory factor, while we do. This fact could be one strictly necessary. Second, the proposed procedure of the reasons why they find that only 5.4% of the gap always uses the same starting point, and we estimate each is explained by female-male differences in characteris - contribution by switching the variable of interest first (as tics, while 94.6% is unexplained, while our results are the suggested by Fairlie 2005). Thus, only one characteris - opposite. tic is modified at a time. Although we are aware that the The results obtained from our simulations lead us order of switching the variables is potentially important, to conclude that the fact that men and women work in this decision provides a potential solution to the path different industries and occupations is what widens dependence problem already pointed out by Fairlie. the underemployment gap to the greatest extent. This The methodology proposed has some limitations. effect is even increased due to women facing a different It does not show the summing up property of the Fair- risk of underemployment than men when they work in lie method, however, the estimated approximation the same gender-typing jobs. So, Hypotheses 2a and 2b errors are small and not significant. Moreover, using the are supported and we can state that the gender gap in 22 Page 14 of 16 J. Acosta‑Ballesteros et al. underemployment would be largely reduced if men and linked to a lower risk of experiencing underemployment. women were more evenly distributed across occupations This result is in line with the literature attributing ICTs and industries. Thus, segregation (especially occupa - the potential capacity to reduce gender inequalities, since tional) is not only a source of gender differences in terms they improve the occupational and professional posi- of wages and job quality (Stier and Yaish 2014), but also a tion of women (Castaño et al. 1999), particularly in the key factor explaining the underemployment gender gap. Spanish labour market (Iglesias-Fernández et al. 2010a, Moreover, as the World Economic Forum (2017) states, 2010b). Specifically, ITCs reduce both the need for man - a crucial factor for further progress in reducing the over- ual labour and physical effort in favour of knowledge, all global gender gap is the closing of occupational gender teamwork and communication skills (WWW-ITC 2004). gaps. Therefore, policy measures should be designed and This fact, in turn, promotes changes in the sectoral dis - implemented to fight against segregation in the labour tribution and educational requirement of jobs and the market in order to achieve gender equality in Spain. demand for occupations (Iglesias-Fernández et al. 2012), In this respect, education is an important factor to be leading to more opportunities for women. However, the highlighted. Thus, we can consider two different ways in ratio of female/male graduates in ICTs is only 0.14 in which education can influence underemployment. Firstly, Spain (World Economic Forum 2017). education has a direct impact on underemployment since Although educational segregation by gender plays a the marginal effects show that some educational attain - significant role in shaping gender segregation within the ments contribute to reduce the risk of underemploy- labour market, as Smyth and Steinmetz (2008) point out, ment. Moreover, the results from our simulations using women and men who choose similar fields do not have the baseline model, as well as those obtained after split- exactly the same occupational outcomes. Thus, educational ting the sample into workers with tertiary education and policies should be complemented with other instruments. without it, allow us to affirm that education reduces the For instance, reinforcing policies that encourage employ- underemployment gender gap. Specifically, our results ers to hire female workers in male intensive occupations suggest a higher educational attainment helps women to and industries could also help reduce both segregation escape from this disadvantage in the labour market. and the gender gap in underemployment. Additionally, Second, education may affect segregation in the mar - the economic conditions of female occupations should ketplace due to the link between educational presorting be improved to raise both men’s and women’s interest in and occupational and industry segregation (Borghans and female-dominated occupations. In order to achieve an Groot 1999; Shauman 2006; Smyth and Steinmetz 2008). egalitarian distribution of men and women across jobs, The statistical evidence on the strength of the link between eradicating the disincentives to work in female-dominated segregation in education and in employment is mixed occupations is necessary (Torre 2018). Finally, any other (Bettio et al. 2009). Nevertheless, appropriate remedies that measures devoted to fighting against gender stereotypes address the barriers women experience to enter male domi- and discrimination would be welcome to reduce inequality nated jobs should include changes in the education and in the Spanish labour market. training of women and girls, such as introducing gender aware career counselling/guidance (National Foundation Abbreviation for Australian Women 2017). Particularly, young women LFS: Labour Force Survey. should be encouraged to enrol in previously male-domi- nated education programmes in order to gain access to a Supplementary Information wider range of jobs. Additionally, encouraging young men The online version contains supplementary material available at https:// doi. to join female-dominated specialties could translate into org/ 10. 1186/ s12651‑ 021‑ 00305‑0. a more mixed gender educational profiles. In fact, as Bet - tio et al. (2009) point out, since women are outperforming Additional file 1: Appendix 1. Approximation error in the endowments effect. Appendix 2. Approximation error in the coefficients effect. men in levels of education attained—up to the first stage of Additional file 2: Table S1. Comparison of methods for the three ‑fold tertiary education—choice of field is the primary channel decomposition of the underemployment gender gap. Table S2. Frequen‑ through which education can influence de-segregation in cies (%) of the independent variables by gender: not underemployed the labour market in the future. (NU), underemployed (U), or unemployed (Unem) Especially desirable would be more women specialis- ing in ICTs since our results show this field seems to be Acknowledgements We thank three anonymous referees for their insightful comments and sug‑ gestions, which have helped to improve this article considerably. Authors’ contributions Analysing the causality relation between educational and occupational seg- All authors wrote, read and approved the final manuscript. regation is beyond the scope of this study. Measuring the effect of gender segregation on the gender gap in time‑related underemployment Page 15 of 16 22 Funding Daymont, T.N., Andrisani, P.J.: Job preferences, college major, and the gender Authors declare that they have received no specific funding to conduct this gap in earnings. J. Hum. Resour. 19, 408–428 (1984) research. Doiron, D.J., Riddell, W.C.: The impact of unionization on male‑female earnings differences in Canada. J. Hum. Resour. 29(2), 504–534 (1994) Availability of data and materials Dowd, B.E., Greene, W.H., Norton, E.C.: Computation of standard errors. The data that support the findings of this study are available from Instituto Health Serv. Res. 49(2), 731–750 (2014) Nacional de Estadística (INE). Purchase terms do not allow authors sharing Dueñas‑Fernández, D., Iglesias‑Fernández, C., Llorente ‑Heras, R.: La involun‑ these data. tariedad en el empleo a tiempo parcial y la Gran Recesión: un análisis de género en España. EKONOMIAZ. Revista Vasca De Economía 90(02), Code availability 284–317 (2016) The results have been obtained using STATA. Etezady, A., Shaw, F.A., Mokhtarian, P.L., Circella, G.: What drives the gap? Applying the Blinder‑ Oaxaca decomposition method to examine gen‑ erational differences in transportation‑related attitudes. Transportation Declarations 48, 857–883 (2021) Even, W.E., Macpherson, D.A.: Plant size and the decline of unionism. Econ. Ethics approval and consent to participate Lett. 32(4), 393–398 (1990) The authors of this study declare this manuscript does not infringe the copy‑ Fairlie, R.W.: The absence of the African‑American owned business: an right, moral rights, or other intellectual property rights of any other person. analysis of the dynamics of self‑ employment. J. Labor Econ. 17(1), The authors of this study testify that this paper is original and has not incurred 80–108 (1999) in any type of plagiarism. The authors of this study declare that the article Fairlie, R.W.: An extension of the Blinder‑ Oaxaca decomposition technique is original, has not already been published in a journal, and is not currently to logit and probit models. J Econ. Soc. Meas. 30(4), 305–316 (2005) under consideration by another journal. The authors of this study agree to the Fairlie, R.W.: Addressing path dependence and incorporating sample terms of the SpringerOpen Copyright and License Agreement. weights in the nonlinear Blinder‑ Oaxaca decomposition technique for logit, probit and other nonlinear models. Stanford Institute for Consent for publication Economic Policy Research, Working Paper No. 17–013, University of Not applicable. California (2017) Fortin, N., Lemieux, T., Firpo, S.: Decomposition methods in economics. In: Competing interests Ashenfelter, O., Card, D. (eds.) Handbook of labor economics, Vol 4a, Authors declare that no competing interests exist. Handbooks in Economics, pp. 1–102. North Holland, Great Britain (2011) National Foundation for Australian Women Gender segregation in the work‑ Author details place and its impact on women’s equality. Report submitted to Senate Departamento de Economía, Contabilidad y Finanzas, Universidad de La Finance and Public Administration References Committee. https:// www. Laguna (ULL), La Laguna ( Tenerife), Spain. Facultad de Economía, Empresa y aph. gov. au/ Parli ament ary_ Busin ess/ Commi ttees/ Senate/ Finan ce_ and_ Turismo, Universidad de La Laguna (ULL), Campus de Guajara, Apartado 456, Public_ Admin istra tion/ Gende rsegr egati on/ Report (2017) Postal Code 38200, La Laguna ( Tenerife), Spain. Gardeazábal, J., Ugidos, A.: More on identification in detailed wage decompo ‑ sitions. Rev. Econ. Stat. 86(4), 1034–1036 (2004) Received: 27 July 2020 Accepted: 11 November 2021 Greene, W.H.: Econometric analysis, 7th edn. Pearson, Upper Saddle River (2012) Hakim, C.: A sociological perspective on part‑time work. In: Blossfeld, H.P., Hakim, C. (eds.) Between equalization and marginalization: women working part‑time in Europe and the United States of America, pp. 22–70. References Oxford University Press, Oxford (1997) Acosta‑Ballesteros, J., Osorno ‑ del Rosal, M.P., Rodríguez‑Rodríguez, O.M.: Hakim, C.: Social Change and Innovation in the Labour Market: Evidence from Underemployment and employment among young workers and the the Census SARs on Occupational Segregation and Labour Mobility, Part‑ business cycle in Spain: the importance of education level and specialisa‑ time Work and Student Jobs. Homework and Self‑ employment. Oxford tion. J. Educ. Work 31(1), 28–46 (2018) University Press, Oxford (1998) Anker, R.: Gender and jobs: sex segregation of occupations in the world. Inter‑ Hegewisch, A., Liepmann, H., Hayes, J., Hartmann, H.: Separate and not equal? national Labour Office, Geneva (1998) Gender segregation in the labor market and the gender wage gap. Barrett, G.F., Doiron, D.J.: Working part time: by choice or by constraint. Can. J. Institute for Women’s Policy Research, Briefing Paper No. 377, Washington Econ. 34(4), 1042–1065 (2001) (2010) Bazen, S., Joutard, X., Magdalou, B.: An Oaxaca decomposition for nonlinear Iglesias‑Fernández, C., Llorente ‑Heras, R., Dueñas‑Fernández, D.: Diferencias de models. J. Econ. Soc. Meas. 42(2), 101–121 (2017) Género en el Empleo TIC. Cuad. Econ. 33(92), 105–138 (2010b) Bell, D., Blanchflower, D.: Underemployment in the UK Revisited. Natl. Inst. Iglesias‑Fernández, C., Llorente ‑Heras, R. Dueñas‑Fernández, D.: Profundizando Econ. Rev. 224(1), F8–F22 (2013) en la segregación laboral. Sectores, ocupación y TIC en España. Instituto Bettio, F., Verashchagina, A., Mairhuber, I., Kanjuo‑Mrčela, A.: Gender segrega‑ Universitario de Análisis Económico y Social. Serie de Documentos de tion in the labour market: Root causes, implications and policy responses Trabajo 07/2012, Universidad de Alcalá (2012) in the EU. Publications Office of the European Union, Luxembourg (2009) Iglesias‑Fernández, C., Llorente ‑Heras, R., Dueñas‑Fernández, D.: Mujer y Blinder, A.S.: Wage discrimination: reduced form and structural estimates. J. Empleo TIC en España. In C. Castaño (Dir.), Género y TIC. Presencia, Hum. Resour. 8(4), 436–455 (1973) Posición y Políticas, pp. 83–106. UOC Ediciones, Barcelona (2010a) Boll, C., Leppin, J., Rossen, A., Wolf, A.: Magnitude and impact factors of the Jann, B.: The Blinder‑ Oaxaca decomposition for linear regression models. Stata gender pay gap in EU countries. Report prepared for and financed by the J. 8(4), 453–479 (2008) European Commission–Directorate‑ General for Justice, European Union Kamerāde, D., Richardson, H.: Gender segregation, underemployment and (Hrsg.), Hamburg (2016) https:// op. europa. eu/ en/ publi cation‑ detai l/‑/ subjective well‑being in the UK labour market. Hum. Relat. 71(2), publi cation/ aa27a c5b‑ fc83‑ 11e5‑ b713‑ 01aa7 5ed71 a1/ langu age‑ en 285–309 (2018) Borghans, L., Groot, L.: Educational presorting and occupational segregation. Kim, C.: Detailed wage decompositions: revisiting the identification problem. Labour Econ. 6, 375–395 (1999) Sociol. Methodol. 43, 346–363 (2013) Castaño, C., Iglesias, C., Mañas, E., Sánchez‑Herrero, M.: Diferencia o discrimi‑ Kjeldstad, R., Nymoen, E.H.: Part‑time work and gender: worker versus job nación: La situación de las mujeres en el mercado de trabajo y el impacto explanations. Int. Labour Rev. 151(1–2), 85–107 (2012a) de las tecnologías de la información. Consejo Económico y Social, Madrid Kjeldstad, R., Nymoen, E.H.: Underemployment in a gender‑segregated labour (1999) market. Econ. Ind. Democracy 33(2), 207–224 (2012b) 22 Page 16 of 16 J. Acosta‑Ballesteros et al. Krinsky, I., Robb, A.L.: On approximating the statistical properties of elasticities. Stier, H., Yaish, M.: Occupational segregation and gender inequality in job qual‑ Rev. Econ. Stat. 68(4), 715–719 (1986) ity: a multi‑level approach. Work Employ. Soc. 28(2), 225–246 (2014) Lupton, B.: Explaining men’s entry into female‑ concentrated occupations: Torre, M.: Stopgappers? The occupational trajectories of men in female‑ domi‑ issues of masculinity and social class. Gender Work Organ. 13(2), 103–128 nated occupations. Work Occup. 45(3), 283–312 (2018) (2006) Valletta, R.G., Bengali, L., Van der List, C.: Cyclical and Market Determinants of Martin, P., Barnard, A.: The experience of women in male‑ dominated occupa‑ Involuntary Part‑time Employment. IZA Discussion Paper No.9738, Bonn tions: A constructivist grounded theory inquiry. Journal of Industrial (2016) Psychology 39(2), 1–12 (2013) Vuluku, G., Wambugu, A., Moyi, E.: Unemployment and underemployment in McKee‑Ryan, F.M., Harvey, J.: I have a job, but…: a review of underemploy‑ Kenya: a gender gap analysis. Economics 2(2), 7–16 (2013) ment. J. Manage. 37(4), 962–996 (2011) Weststar, J.: A review of women’s experiences of three dimensions of under‑ Oaxaca, R.: Male‑female wage differentials in urban labor markets. Int. Econ. employment. In: Feldman, D., Maynard, D. (eds.) Underemployment: Rev. 14(3), 693–709 (1973) psychological, economic, and social challenge, pp. 105–125. Springer, Oaxaca, R., Ransom, M.: Identification in detailed wage decompositions. Rev New York (2011) Econ Stat 81(1), 154–157 (1999) Wilkins, R.: Personal and job characteristics associated with underemployment. Powers, D.A., Yoshioka, H., Yun, M.S.: mvdcmp: multivariate decomposition for Aust. J. Labour Econ. 9(4), 371–393 (2006) nonlinear response models. Stata J. 11(4), 556–576 (2011) Wilkins, R.: The consequences of underemployment for the underemployed. J. Prause, J., Dooley, D.: Youth Underemployment. In: Feldman, D., Maynard, D. Ind. Relat. 49(2), 247–275 (2007) (eds.) Underemployment: psychological, economic, and social challenge, World Economic Forum: The Global Gender Gap Report. http:// www3. wefor pp. 59–80. Springer, New York (2011)um. org/ docs/ WEF_ GGGR_ 2017. pdf (2017) Seaman, S., White, I.R.: Review of inverse probability weighting for dealing with WWW‑ITC: Conceptual framework and state of the art. Widening women’s missing data. Stat. Methods Med. Res. 22(3), 278–295 (2013) work in information and communication technology. http:// www. ftu‑ Shauman, K.: Occupational sex segregation and earning of occupations: what namur. org/ fichi ers/ D12‑ print. pdf (2004) causes the link among college‑ educated workers? Soc. Sci. Res. 35, Yun, M.S.: Decomposing differences in the first moment. Econ. Lett. 82(2), 577–619 (2006) 275–280 (2004) Simic, M.: Underemployment and overemployment in the UK. Labour Market Yun, M.S.: A simple solution to the identification problem in detailed wage Trends 110(8), 399–414 (2002) decompositions. Econ. Inq. 43, 766–772 (2005) Simpson, R.: Have times changed? Career barriers and the token woman Yun, M.S.: Identification problem and detailed Oaxaca decomposition: a gen‑ manager. Brit. J. Manage. 8(s1), 121–130 (1997) eral solution and inference. J. Econ. Soc. Meas. 33(1), 27–38 (2008) Simpson, R.: Gender mix and organizational fit: how gender imbalance at dif‑ ferent levels of the organization impacts on women managers. Women Publisher’s Note Manag. Rev. 15(1), 5–20 (2000) Springer Nature remains neutral with regard to jurisdictional claims in pub‑ Simpson, R.: Masculinity at work: the experiences of men in female dominated lished maps and institutional affiliations. occupations. Work Employ. Soc. 18(2), 349–368 (2004) Smyth, E., Steinmetz, S.: Field of study and gender segregation in European labour markets. Int. J. Comp. Sociol. 49(4–5), 257–281 (2008)

Journal

Journal for Labour Market Research – Springer Journals

Published: Dec 1, 2021

Keywords: Gender gap; Occupational and Industry segregation; Time-related underemployment; Counterfactual simulations; Decomposition analysis; J16; J22

References

Underemployment and employment among young workers and the business cycle in Spain: the importance of education level and specialisation

Acosta-Ballesteros, J; Osorno-del Rosal, MP; Rodríguez-Rodríguez, OM
Working part time: by choice or by constraint

Barrett, GF; Doiron, DJ
An Oaxaca decomposition for nonlinear models

Bazen, S; Joutard, X; Magdalou, B
Underemployment in the UK Revisited

Bell, D; Blanchflower, D
Wage discrimination: reduced form and structural estimates

Blinder, AS
Educational presorting and occupational segregation

Borghans, L; Groot, L
Job preferences, college major, and the gender gap in earnings

Daymont, TN; Andrisani, PJ
The impact of unionization on male-female earnings differences in Canada

Doiron, DJ; Riddell, WC
Computation of standard errors

Dowd, BE; Greene, WH; Norton, EC
La involuntariedad en el empleo a tiempo parcial y la Gran Recesión: un análisis de género en España

Dueñas-Fernández, D; Iglesias-Fernández, C; Llorente-Heras, R
What drives the gap? Applying the Blinder-Oaxaca decomposition method to examine generational differences in transportation-related attitudes

Etezady, A; Shaw, FA; Mokhtarian, PL; Circella, G
Plant size and the decline of unionism

Even, WE; Macpherson, DA
The absence of the African-American owned business: an analysis of the dynamics of self-employment

Fairlie, RW
An extension of the Blinder-Oaxaca decomposition technique to logit and probit models

Fairlie, RW
Decomposition methods in economics

Fortin, N; Lemieux, T; Firpo, S; Ashenfelter, O; Card, D
More on identification in detailed wage decompositions

Gardeazábal, J; Ugidos, A
A sociological perspective on part-time work

Hakim, C; Blossfeld, HP; Hakim, C
Diferencias de Género en el Empleo TIC

Iglesias-Fernández, C; Llorente-Heras, R; Dueñas-Fernández, D
The Blinder-Oaxaca decomposition for linear regression models

Jann, B
Gender segregation, underemployment and subjective well-being in the UK labour market

Kamerāde, D; Richardson, H
Detailed wage decompositions: revisiting the identification problem

Kim, C
Part-time work and gender: worker versus job explanations

Kjeldstad, R; Nymoen, EH
Underemployment in a gender-segregated labour market

Kjeldstad, R; Nymoen, EH
On approximating the statistical properties of elasticities

Krinsky, I; Robb, AL
Explaining men’s entry into female-concentrated occupations: issues of masculinity and social class

Lupton, B
The experience of women in male-dominated occupations: A constructivist grounded theory inquiry

Martin, P; Barnard, A
I have a job, but…: a review of underemployment

McKee-Ryan, FM; Harvey, J
Male-female wage differentials in urban labor markets

Oaxaca, R
Identification in detailed wage decompositions

Oaxaca, R; Ransom, M
mvdcmp: multivariate decomposition for nonlinear response models

Powers, DA; Yoshioka, H; Yun, MS
Youth Underemployment

Prause, J; Dooley, D; Feldman, D; Maynard, D
Review of inverse probability weighting for dealing with missing data

Seaman, S; White, IR
Occupational sex segregation and earning of occupations: what causes the link among college-educated workers?

Shauman, K
Underemployment and overemployment in the UK

Simic, M
Have times changed? Career barriers and the token woman manager

Simpson, R
Gender mix and organizational fit: how gender imbalance at different levels of the organization impacts on women managers

Simpson, R
Masculinity at work: the experiences of men in female dominated occupations

Simpson, R
Field of study and gender segregation in European labour markets

Smyth, E; Steinmetz, S
Occupational segregation and gender inequality in job quality: a multi-level approach

Stier, H; Yaish, M
Stopgappers? The occupational trajectories of men in female-dominated occupations

Torre, M
Unemployment and underemployment in Kenya: a gender gap analysis

Vuluku, G; Wambugu, A; Moyi, E
A review of women’s experiences of three dimensions of underemployment

Weststar, J; Feldman, D; Maynard, D
Personal and job characteristics associated with underemployment

Wilkins, R
The consequences of underemployment for the underemployed

Wilkins, R
Decomposing differences in the first moment

Yun, MS
A simple solution to the identification problem in detailed wage decompositions

Yun, MS
Identification problem and detailed Oaxaca decomposition: a general solution and inference

Yun, MS

Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Measuring the effect of gender segregation on the gender gap in time-related underemployment

Measuring the effect of gender segregation on the gender gap in time-related underemployment

Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Measuring the effect of gender segregation on the gender gap in time-related underemployment

Measuring the effect of gender segregation on the gender gap in time-related underemployment

Abstract

Journal

Recommended Articles

References

Our policy towards the use of cookies