Literature Review

See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/5129702

Choices for part-time jobs and the impacts on the wage differentials. A

comparative study for Great Britain and the Netherlands

Article · April 2003

Source: RePEc

CITATIONS

15 READS

52

2 authors, including:

Some of the authors of this publication are also working on these related projects:

WISCO database of occupations View project

Zeggenschap over arbeidstijden - Control over working hours View project

Kea Tijdens

University of Amsterdam

392 PUBLICATIONS   1,068 CITATIONS   

SEE PROFILE

All content following this page was uploaded by Kea Tijdens on 30 May 2014.

The user has requested enhancement of the downloaded file.

IRISS at CEPS/INSTEAD

An Integrated Research Infrastructure in the Socio-Economic

Sciences

CHOICES FOR PART-TIME JOBS AND THE IMPACTS ON THE WAGE DIFFERENTIALS

A COMPARATIVE STUDY FOR GREAT BRITAIN AND THE NETHERLANDS

by

Yongjian Hu & Kea Tijdens

IRISS WORKING PAPER SERIES

No. 2003-05

Please pay a visit to http://www.ceps.lu/iriss

RESEARCH GRANTS for individual or collaborative research projects

(grants awarded for periods of 2-12 weeks)

What is IRISS-C/I?

IRISS-C/I is a project funded by the European Commission in its ‘Access to Major Research Infrastructures’ programme. IRISS-C/I funds short visits at CEPS/Instead for researchers willing to undertake collaborative and/or internationally comparative research in economics and other

social sciences.

Who may apply?

We encourage applications from all interested individuals (doing non-proprietary research in a European institution) who want to carry out their research in the fields of expertise of

CEPS/Instead.

What is offered by IRISS-C/I?

Free access to the IRISS-C/I research infrastructure (office, computer, library…); access to the CEPS/Instead archive of micro-data (including e.g. the ECHP); technical and scientific assistance;

free accommodation and a contribution towards travel and subsistence costs.

Research areas

Survey and panel data methodology; income and poverty dynamics; gender, ethnic and social inequality; unemployment; segmentation of labour markets; education and training; social

protection and redistributive policies; impact of ageing populations; intergenerational relations; regional development and structural change.

Additional information and application form

IRISS-C/I B.P. 48 L-4501 Differdange (Luxembourg)

Email: [email protected] Homepage: http://www.ceps.lu/iriss

IRISS-C/I An Integrated Research Infrastructure in the

Socio-Economic Sciences at CEPS/Instead, Luxembourg

Project supported by the European Commission and the Ministry of Culture, Higher Education and Research

(Luxembourg)

1

Choices for part-time jobs and the impacts on the wage differentials

A comparative study for Great Britain and the Netherlands

Yongjian Hu

the department of general economics

University of Amsterdam

the Netherlands

[email protected] Kea Tijdens

the department of general economics

University of Amsterdam

the Netherlands

[email protected]

March 2003

Acknowledgement: This research was (co-) funded by a grant of the European Commission under the

Transnational Access to major Research Infrastructures contract HPRI-CT-2001-00128

hosted by IRISS-C/I at CEPS/INSTEAD Differdange (Luxembourg).

2

Abstract This paper uses the European Household Panel (the ECHP) to analyze individuals’

choices on part-time jobs and their impacts on the wage differentials. Our study is a

comparative study between Great Britain and the Netherlands. In contrast to most of the

previous researches on part-time employment, we make a distinction between short part-

time and long part-time jobs. The results show that overall women were more likely to

take part-time jobs in both countries, but the effect was much stronger in the Netherlands

than it was in Great Britain. We find that there was no substantial wage gap between long

part-time and full-time jobs in the Netherlands, working long part-time were more likely

to be treated as full-time jobs, which may suggest the presence of “ retention part-time

jobs “ described by Tilly (1996). On the other hand, the results show that part-time

workers in Great Britain suffered relatively larger wage penalties, yet, working short part-

time was not significantly different from working long part-time because they both

received lower wage rates compared to full-time jobs.

Keywords: short part-time job, long part-time job and wage differential(gap) JEL Classification: J22,J31

3

Choices for part-time jobs and the impacts on the wage differentials

A comparative study for Great Britain and the Netherlands

______________________________________________________

1. Introduction During the past two decades the increase in part-time employment represented one of the

most striking structural changes experienced by many EU member countries. Particularly,

there has been a rapid expansion of part-time employment in the Dutch labour market,

where part-time job as a percentage of the total employment remained the level of 28 per

cent in the first half of the 1990s and reached 30 per cent by the late 1990s. The stronger

growth in part-time employment also characterized the British labour market, the share of

working part-time relative to the total jobs varied from 20 to 25 per cent in the

1990s(OECD, Labor Force Statistics, 2000). For both countries, part-time working’s

contribution to job creation has been become important as a whole. Between 1987 and

1997, in the Dutch labour market, part-time jobs accounted for 42 per cent of the total

growth of employment, while in the British labour market, 53 per cent of the job creation

was attributed to the growth of part-time working (OECD Economic Outlook, 1999).

In an optimistic view, part-time employment functions as an effective means to meet the

needs of both employees and employers. It not only gives a chance to have a continuity

of labor market attachment for those who want to combine their home responsibilities

and earn additional income, but also provides firms with a certain flexibility of adjusting

their workforce to copy with the fluctuation of market for their products.

As part-time jobs rise rapidly, there has been a growing concern about the quality of part-

time jobs. The common criticism is that part-time workers earned lower wage rate than

those of full-time workers (Dekker et al., 2000; Blank, 1990; Ermisch et al., 1992,1991),

moreover, part-time workers were less well protected, in particular, they received low

fringe benefits (Montgomery et al, 1993) and were expected to have a low incidence of

firm-specific training compared to full-time workers. Tilly (1996) argued that majority of

par-time workers had no little chance for promotion and career perspective.

4

Most of the previous empirical studies on part-time employment treated part-time job as

an undifferentiated mass. Their major assumptions were: (1) overall, full-time workers

have higher levels of human capital than do part-time workers, as a result, part-time

workers earn less than full-time workers (2) female full-time and part-time workers may

differ on a host of unobservable traits, women with part-time jobs are less productive

than those whose orientation are toward full-time employment and toward labour market

attachment more generally (3) part-time jobs are “bad jobs”, which have been identified

possessing all characteristics in secondary labour market.

However, in a study conducted by Tilly (1996), part-time workers are seen as a

heterogeneous group instead of a homogenous one. According to Tilly, there are two

types of part-time jobs: secondary part-time job and retention part-time job. Secondary

part-time jobs are similar with jobs in secondary labour market, in which workers

normally have low skills and compensation compared to their full-time counterparts.

Those who do secondary part-time jobs have little prospect of advancement, hence,

secondary part-time jobs are really “ bad jobs”. On the other hand, retention part-time

jobs are found in primary labor markets and tend to be created for those who have skills,

they are paid at the level comparable to or above those of full-timers.

Therefore, part-time job is not as simple as we thought it be, we need go further and

explore the differences within the group of part-time workers. The aims of our study are:

first, we make a comparative study between the Netherlands and Great Britain. We intend

to identify what the major factors could affect an individual’s decision on taking part-

time jobs and what differences are likely to exist between two countries. Second, we

break part-time jobs into two categories: short part-time job and long part-time job. By

distinguishing between short part-time job and long part-time job, we shed light on the

nature of voluntarily part-time jobs and their impacts on the wage gap.

Some of the previous studies have taken into account the differences between short part-

time and long part-time jobs (see, e.g. Dekker, et al., 2000; Tam, 1997; Tijdens, 1997). In

their studies, those who worked less than 12 hours per week were also classified as part-

time jobs. In contrast to their classifications, we truncates those who worked less than 12

hours per week since these kinds of jobs are usually regarded as involuntarily part-time

jobs or marginal jobs. With respect to the econometric specification, we allow for the

5

sample selection and estimate the wage equations by two-step method. While two-regime

endogenous switching regressions are widely used, for our study, since three categories

are defined, i.e. short part-time job, long part-time job and full-time job, we estimate the

ordered probit model instead of probit model, all standard errors for wage equations are

corrected for accordingly.

The remainder of this paper is organized as follows: Part two provides an overview of the

existing literature on explaining why there could be wage differential between full-time

and part-time jobs. The recent empirical researches are reviewed in Part three. We

introduce the data used for this study in Part four and outline the econometric

specification in Part five. The estimation results are discussed in Part six. We summarize

our findings and make conclusions in the last part.

For ease of reference when comparing two countries in the following parts we frequently

use terms “the Dutch model (or the NL model) “ and “the British model (or the UK

model)”. The former refers to the models and the estimations based on the data for the

Netherlands, while the latter refers to the models and the results drawn from the data for

Great Britain.

2. Theoretical background Although the wage gap between full-time and part-time jobs has been the subject of

enormous empirical investigations since the 1980s, a systematic theoretical framework is

hardly found in any standard labour economics. Most of explanations heavily rely on the

extensions of standard theories in labour economics, where the theory of endogenous

wage setting, the compensation wage and the dual labour market theory are widely cited.

We take up the theories listed above as follows.

2.1 The endogenity of wage setting The neoclassical analysis of labor supply describes individual allocation of time between

the market and non-market as the outcome of maximizing utility subject to a budget

constraint. The relevant price for each individual is an hourly wage rate that is

exogenously determined and does not vary with the number of hours supplied (see, e.g.

Killingsworth 1983). Each employee chooses to work the number of hours for which the

marginal rate of substation of leisure for consumption is equal to the wage rate, while

each employer is a price-taker and pays employee according to the market wage rate and

6

the productivity of the employee. Within this theoretical framework, full-time and part-

time work are not distinguished from each other as distinct forms of wage work, instead,

they are regarded as merely numbers of working hours, which can be incrementally

increased or decreased by unit (Tam, 1997). The real wage rate shall be the same

regardless he (she) is working part-time or full-time.

However, the exogeneity of wage setting (constant wage) is often criticized and

questionable. For example, in a study on the determination of daily hours and wages,

Barzel (1973) argued that a worker’s productivity usually starts slowly at the beginning

of a working day and gradually rises. Even at the last hour of a normal work day, a

worker’s productivity still exceeds the average daily productivity, as a result, the length

of working hours could effectively affect a worker’s marginal product (MP), the shorter

hours individuals choose to work, the lower average hourly rate. The implication of the

linkage between working hours and productivity is that part-time workers should be paid

a lower rate rather than full-time workers’ rate.

In other studies quasi-fixed labor cost is found to play an important role in explaining

interdependence between wage and working hours. Quasi-fixed labor cost refers to those

costs with the employment of labor, which do not vary with the amount of output (for

example, hiring and training cost, and some employee benefits). The existence of quasi-

fixed labor cost makes a firm’s decision about its optimal combination for employment

and hours no different from the one about the usage of any two factors of production

(Ehrenberg and Smith, 2000). The firm must adjust both its employment level and

working hours so that the costs of producing an added unit of output equal for each. If the

hourly labor costs or the quasi-fixed costs of part-time workers fall relative to those full-

time workers, part-time employment should expand relative to those of full-time workers.

Once we admit to the existence of quasi-fixed labor cost, the assumption of independence

between the wage and hours leads to absurd results, i.e. in order to minimize the cost per

unit of labor input, the employer must set the average number of hours per worker equal

to the total time in the period. A more reasonable assumption is that there should be a

whole locus of possible wage-hour combinations, the wage-hours locus must slope

upward in order for the cost-minimizing selection of a wage-hours combination (Rosen,

1976).

7

Overall, the theory of endogenity of wage setting emphasizes the effects of working

hours on productivity and the roles of quasi-fixed labor cost on the mixing each type of

labor. As Leeds (1990) points out, since the endogenity hypothesis holds for any job or

skill level, there is no reason to expect part-time workers to differ significantly with

regard to training or education variables. This implies that part-time job is paid less

simply because they are less productive than full-time workers are. The lower

productivity is not caused by less education level or on the job training, but by shorter

working time horizons than full-time workers are.

2.2 Compensation wage and the theory of efficiency wage Compensation theory predicts that job characteristics that workers consider undesirable

will raise the pay to compensate for the unpleasant conditions, those that are desired

should be purchased by employers in lower wages. Higher wage for full-time job is seen

as the existence of a wage premium. Such wage premium lies in that a large proportion of

women regard part-time as a “desired form “of mixing their home responsibilities and

earning additional income. Employers who can offer these convenient working hours are

in a good position to bargain and reduce the wage rate of part-time workers.

Spatial constraints for part-time workers could also cause compensation wage.

Considering small number of working hours, women find it much convenient to combine

their part-time jobs with their housework if the firm is easily accessible from their home.

Therefore, their labor supplies are likely to be less elastic than the supply of full-time

workers (Ermisch and Wright, 1991). If firms make use of their monopsony power in the

local market, profit maximization entails paying lower wages to part-time workers.

Apart from the idea of compensation wage, the theory of efficiency wage is also utilized

to explain the differences in wage between part-time and full-time jobs. The fundamental

idea underlying the theory of efficiency wages is that firms may gain some benefit from

paying their workers more than the marginal product they produced. High wage could be

more recouped from the additional effort than it generates since it can prevent workers

from shirking and turnover.

The implication of efficiency wage is that paying high wage for full-time employees can

effectively render them more productive, this is because full-time employees take risk

that they can not count on getting a new full-time job at the above market wages if they

8

are fired. On the other hand, part-time workers, who face no threat of lower wages after

being fired will be less productive. Hence, part-time workers are not inherently less

productive than full-timers, just because of different set incentives.

2.3 The dual labor market theory The theory of dual labour market claims that the overall labor market is divided into two

segments: primary market and secondary market. Primary labour market offers jobs with

high wages, good working conditions, employment stability and chances for

advancement. On the contrary, secondary market has jobs, which, relative those in the

primary sector, are decidedly less attractive. They tend to involve low wages, poor

working conditions, and considerable variability. Hence, primary market is characterized

by so-called “good job” while the secondary market is full of “bad job”.

There are institutional barriers between the segments, mobility between two sectors is

limited. The secondary market tends to be filled by groups whose attachment to paid

employment is weaker, such as females, youths and part-timers; most of them constantly

switch between low-paid jobs. The jobs in primary market tend to be the filled by prime-

age cored workers and skilled contingent workers, their mobility are more likely to be

intra-segment rather than inter-segment.

Tilly (1996) applied the theory of dual labor market to the study on part-time job market.

In contrast to the most previous researches that classified all part-time jobs into “bad

jobs”, Tilly reexamined the nature of part-time job and made a distinction between “

good part-time job and “bad part-time job “.

Tilly’s dual conception on part-time job rests on the assumption that part-time work

comprises of three broad categories: short part-time job, secondary part-time job and

retention part-time job. He argued that short part-time job occurs when, instead of laying

workers off during a business downturn, an employer temporarily reduces workers’ hours

until sales revive, this is what we call “ involuntarily part-time jobs.

Compared to short part-time job, Tilly viewed secondary part-time employment and

retention part-time job as the most important part-time jobs. Secondary part-time jobs are

“bad part-time jobs” characterized by low skill requirements, low pay and fringe benefits

and high turnover. Secondary part-time employment thus represents one form of

“secondary labour market” in the dual labour market. On the other hand, retention part-

9

time jobs are “good part-time jobs” created to retain valued employees whose life

circumstances prevent them from working full-time. Retention part-time arrangements

tend to be offered only to workers in relatively skilled jobs, most of workers in this group

are so called “ core workers “. According to Tilly, unlike secondary part-time

employment, retention part-time workers normally possess firm-specific training and

remain internalized; firms can not easily replace them with several secondary part-time

jobs.

3. Review of the recent empirical studies Dekker et al. (2000) used SEP (the Dutch Social-Economic Panel) to analyze part-time

work in the Netherlands. When estimating the multinomial model, they classified part-

time work into short part-time and long part-time jobs. Those who worked less than or

equal to 12 hours per week were regarded as short part-time jobs, while those who

worked greater than 12 hours but less than 33 hours were defined as long part-time jobs.

The threshold between part-time and full-time jobs is 33 working hours per week. The

results show that in the Netherlands married men were more likely to work and especially

to work full-time, women with children were significantly less likely to work long part-

time or full-time. The findings also suggested that short part-time jobs were quite distinct

in terms of the attached wages, but to what extent such distinction existed, it was unclear

from their paper.

Tijdens (1997) classified working arrangement into four categories: full-time job is

defined as working over 35 hours per week, long part-time jobs are those jobs working

between 24-35 hours, medium-sized part-time jobs are 16-23 hours and short part-time

jobs are those working less than 16 hours. The main conclusion drawn from the study is

that no substantial wage gaps were found among the different groups in the Netherlands.

Furthermore, Tijdens argued that the Dutch female part-time workforce consists of two

groups. The first group is created as an employer strategy, a disproportionate amount of

temporary work with lower wage characterizes this group. Another group working in

part-time job is created due to family responsibilities; high skilled women are likely to be

over represented in this group. While the former group has been decreasing over time, the

latter group has been increasing.

10

By using the British data, Ermisch et al. (1992) constructed an order probit model and

estimated wage equations for part-time and full-time jobs. The order probit model was

estimated for the three-way decision of whether to work full-time, part-time and not at

all. They found that the lower rate of return to human capital in part-time employment

made a relatively small contribution to women’s lower pay relative to men’s in the

British labor market.

Tam (1997) made an extensive study on part-time jobs in the UK. His classification

between short-hour part-time and long-hour part-time was motivated by his argument that

part-time workers who work fewer than 17 hours per week is the group who is minimally

protected under the existing legal framework in the UK. Unfortunately, his analysis for

the wage was based on combining all part-time workers instead of separating short hour

part-time from long hour part-time; hence, the conclusion was very general.

In a comparative study between Great Britain and other OECD countries, Bardasi et al.

(2000) constructed a multinomial logit model and estimated the wage equations based on

the data of Luxembourg Income Studies. Three categories, i.e. part-time jobs, full-time

jobs and non-employment are defined in the multinomial logit model. Their results

revealed that that in the UK, the hourly wage gap between part-time and full-time jobs

was about 15 per cent, the differences in observable characteristics between part-time and

full-time workers explained almost the entire unadjusted wage gap.

4.The data 4.1 The ECHP data The data used in this study come from the ECHP-UDB released in December 2001 by

Eurostat. The ECHP-UDB is short for “the users’ database of European Community

Household Panel”(hereinafter called “the ECHP”). As a standardized yearly based-

survey, it aims at forming a coordinated system of household surveys across EU member

states, providing comparable information on diverse economic and social indicators

concerning living conditions of private households and persons. The first full-scale

survey begun in 1994, the latest wave was conducted in 2001, there are currently 14

member countries involved in the survey.

While each member country carries its own labor force survey, some countries usually

combine their data collection for the ECHP with the existing national panels. Most

11

surveys for the ECHP are based on two-stage random sampling: sample areas are selected

in the first stage, followed by the selection of a small number of addresses or households

within each selected area. Note, however, in part of the Netherlands and Great Britain,

direct sampling method was used, for example, in city like Amsterdam. The sample size

for each country is determined on the basis of various theoretical and practical

considerations and the available budget.

In the ECHP-UDB released in December 2001, five waves for the years 1994-98 are

available. The data sets used for our study are drawn from the panels of Great Britain and

the Netherlands for the year 1998. As both countries have been on scale in part-time job

market, the numbers who reported that they were working part-time in the surveys are

relatively larger than the ones obtained from other countries. Moreover, each category

(i.e. short part-time, long part-time and full-time job, see the discussion in the following

sub-section) considerably varies within the categories of every chosen discrete

independent variable, so that the relative models can be estimated without predicting

perfectly.

The data contains 4962 households and 8826 persons in the year 1998 for the

Netherlands, while there are 4996 households and 8868 persons in the year 1998 for

Great Britain. Since we excluded those respondents aged over 65 or below 15 and

categorized part-time jobs into short part-time and long part-time jobs, the number of

observations is decreased when estimating models, especially estimating wage equations.

Moreover, the treatment for missing values by econometric software (STATA) we used

could also lead to a substantial decrease in the number of observations. Note that the

number of observations used in estimation is always listed along with the report for the

results.

4.2 Classification of part-time working Given the emphasis on examining differences among short part-time and long part-time

jobs, it is essential that the classification of part-time jobs under our study combine this

requirement. Two steps are needed to achieve this goal. At first, based on weekly

working hours reported by respondents, we define full-time jobs as those working 30

hours or more, those less than 30-hours are seen as part-time jobs. Second, we divide

part-time jobs into two categories: short part-time job and long part-time job. Short part-

12

time job is defined as those working at least 12 but not more than 21 hours per week, a

job are a long part-time job if the weekly working hour is between 22 and 29.

As mentioned in Part one, Dekker (2000), Tam (1997) and Tijden (1997) also divided

part-time jobs into short part-time and long part-time jobs. Table 4.1 compares our

classification with theirs.

Table4.1 Comparisons with other authors’ classifications

Short part-time Long part-time Full-time

In our study 12-21 hours 22-29 hours >=30 hours

In Dekker ’s paper (1999) <= 12 hours 13-32 hours >=33 hours

In Tam ’s paper (1997)

In Tijdens ’s paper (1997)*

<=16 hours

<=15 hours

17-29 hours

24-35 hours

>=30 hours

>= 36 hours

Note: according to Tijdens’ classification, persons who work between 16 and 23 hours are classified

as medium-sized part-time jobs; see Tijdens (1997), p178.

The classification under our study is different from others in two major aspects: first,

adopting 30-hours to differentiate between part-time and full-time job; second, truncating

those who reported that they worked less than 12 hours per week. The treatment for those

who reported their working less than 12 hours per week merits attention. Working less

than 12 hours are usually called “involuntarily part-time job”, the central bureau for

statistics in the Netherlands explicitly defines these people as unemployed persons, which

are not considered in our study. Further, considering the possible inaccurate reporting of

working hours in the survey and the fact that marginal jobs are relatively fewer in the

data, dropping them from the group of part-timers in the data does not cause any

meaningful selection bias.

5. Econometric specification 5.1 The ordered probit model and the ordered inverse Mills’ ratios Assume that there are three observed choices: the respondent works either short part-

time, or long part-time, or full-time. Accordingly, let y takes on the value, 1,2,3 to denote

these three outcomes. The choices observed are assumed to be ordinal and mutually

exclusive.

An ordered probit model can be derived in the form of the propensity index function

model, which is analytically convenient than the methods motivated by the utility

13

function. Conventionally, the latent variable can be denoted by y *, the asterisk implies

its nature of latent variable.

jjj uZy += '* γ

The propensity index function shows that individuals have their own preference,

depending on certain measurable factors Z, and the unobservable factors ju . The error

term in the function ju is assumed to have standard normal distribution with unity

variance. The setting of unity variance is for identification due to the unobservable nature

of the lateen variable y *.

Although y *i is a latent variable and unobservable, we do observe:

y =1 if y *< µ the individual works short-part-time

y =2 if µ ≤ y *< 1µ the individual works long-part-time

y =3 if 1µ ≤ y * the individual works full time

µ and 1µ are two thresholds that the individuals have to cross over to make y

observable, therefore, there are two truncated points reflecting three intervals of each

individual’s propensity: ),( µ−∞ , ),[ 1µµ and ),[ 1 +∞µ . Long (1999) explained that

while the assumption of unity variance can identify variance in the index function, the

mean of the latent variable is still unidentified, therefore, parameterization of model is

necessary. Following Green (2000)’s specification, we set the first cut-point be zero, as a

result, we leave only one threshold to estimate and remain the intercept in the model. If a

person’s propensity index is less than zero, he (she) is observed to take short part-time,

otherwise, he (she) will be observed to work long part-time job or full-time job.

Estimation of ordered probit model functions twofold: first, examine the major factors

that could affect individuals’ choices of part-time jobs; second, use the predicted

probabilities of observing different choices to compute the inverse Mills’ ratios, which is

denoted by )(iΛ .

Allowing three choices in an ordered probit model implies that the distribution is doubly

truncated and not one point (from above or below), as is necessary to apply the standard

Heckman selection model, the ordered inverse Mills’ ratios can be computed as follows:

14

the inverse Mills ratio for short part-time job:

)]'(1/[)'()1( γγφλ ZZ Φ−−=

the inverse Mills ratio for long part-time job:

)]'()'(/[)]'()'([)2( 11 γγµγµφγφλ ZZZZ −Φ−−Φ−−−=

the inverse Mills ratio for full-time job:

)]'(1/[)'()3( 11 γµγµφλ ZZ −Φ−−=

The computed inverse Mills’ ratios are plugged into the wage equations to correct for

selectivity-bias. Since the coefficient matrix γ is used to compute the inverse Mills’ ratio

for every observation, its impact will be accounted for when computing covariance

matrix for the wage equations, as explained in the following section.

5.2 Wage equations and correction for selectivity-bias Following Main et al. (1993)’s notation, we write the original wage equations as follows:

3 ' 333

2 ' 222

1 ' 111

εβ

εβ

εβ

+=

+=

+=

XW

XW

XW

the subscripts i=1,2,3 designate short part-time job, long part-time job and full-time job

respectively. For simplicity, the subscript for individual j is suppressed. iX is a matrix

consisting of an intercept and explanatory variables; iβ is a coefficient matrix including

a constant , iε is an error term.

Substituting the relative inverse Mills’ ratios into the original wage equations, we have:

111 ' 111 )1( νλσρβ ++= XW

222 ' 222 )2( νλσρβ ++= XW

333 ' 333 )3( νλσρβ ++= XW

the error term iν is assumed to be homogenous; iρ is the coefficient for the correlation

between the error term in the ordered probit model and the error term in the wage

equations; iσ is the standard error for the ith wage equation.

15

Least squared regression of iW on exogenous variables iX and )(iλ would produce

consistent estimates. The consistent estimate of the error variance for the ith wage

equation is calculated as follows:

22 ˆ' 1ˆ iiiii been

δσ +=

where:

∑ =

= 3

1

ˆ1 i

iji n δδ

)ˆ')(ˆ()(ˆˆ γλλδ ijjjij Zii +=

The squared correlation coefficients between the error term in the ordered probit model

and the ith wage equation is given by:

2

2 2

i

i i

b σ

ρ =

The corrected-standard errors for the wage regressions are calculated according to: 1

* ' **

2' *

1 *

' *

2 ]][)1([)(),( −− +−= XXQXXXXbVar iδρσβ

Note that *X is a matrix including λ , ))(ˆ()( * ''

* 2 XZCovVArZXQ γρ −= . The term Q is

designed to account for the fact that the same estimate of γ is used to compute )(ˆ iλ . For

the detailed description, see Green (2000) and Main et al. (1993).

5.3 The decomposition of wage differentials The Decomposition of wage differentials is based on the method developed by Oaxaca

(1973), the extension here is that the self-selection is included in the augments.

)()()(loglog SSSFFFSFSSFFSF XXXww λσρλσρβββ −+−+−=−

)()()(loglog LLLFFFLFLLFFLF XXXww λσρλσρβββ −+−+−=−

)()()(loglog SSSLLLSLSSLLSL XXXww λσρλσρβββ −+−+−=−

The subscripts SLF ,, denote full-time job, long part-time job and short part-time job

receptively, β is the coefficients for explanatory variables in the wage equations and

X is the average values for each explanatory variable. The left hand side for each

iiib ρσ=

16

decomposing equation shows the differentials of the average log hourly wage. The right

hand side consists of three parts that contribute to the wage gap: the first term is the

contribution due to the differences in the average characteristic between part-time and

full-time employees; the second term represents the contribution arising from the

different wage offers for part-time and full-time jobs. The final part is taken as a

selection-bias contribution, in which λ is the average inverse Mills’ ratio, the product of

ρ and σ is the estimated coefficient for the inverse Mills’ ratios in wage equation.

6. The results Two major components build up this part: we first introduce the variables included in the

ordered probit model and the wage equations, a more detailed explanation for these

variables are given in the Appendix A. We then present the empirical results and make a

comparative analysis between the Netherlands and Britain.

6.1 Variables 6.1.1 The key variables in the ordered probit model

The dependent variable in the ordered probit model is a categorical variable representing

three outcomes, taking on the value 1 if one works short part-time; 2 if one works long

part-time; and 3 if one works full-time. The numbers are assumed to be ordinal in both

working time and utility, short-part-time job is less than long part-time job, long part-

time is less than full-time job.

We expect a greater influence of social and demographic characteristics on individual

choices of working part-time. Though, part-time jobs are strongly associated with

women, the share of male part-timers has risen recently. Yet, excluding male workers

from the sample will lead to a decrease in the number of observations when estimating

the models. Hence, a dummy variable for gender is defined in the model to test what

differences are likely to exist between men and women.

Decision of individual labor supply is enhanced by jointly considering household and the

number of dependent children in the family. The dummies for martial status and children

in the model allow us to further understand how individual’s decision is constrained

within the family framework. In addition, the interaction items among the variables for

gender, martial status and the number of children are constructed in the model.

17

As a proxy for the income effect imposed by family other members, we define a

categorical variable describing the number of actively working persons in a family. By

dosing so, we expect that larger number of active working persons, higher income a

family possesses, and more likely one takes part-time job.

The segregated nature of the labour market by occupation and industry adds a further

impact on the patterns of working time. Many clerical jobs and the jobs in service sectors

are more likely to be taken by women and part-timers, whereas other occupations are

more likely to work full-time. We make use of the information provided by the

respondents in the ECHP and define a categorical variable to capture the effect of

occupation. In addition, the dummies reporting one’s job level are also concluded in the

model.

The ordered probit model under our study can be viewed as a ‘choice model” in the sense

that all these factors are set for the employees (supply side). In order to analyze the roles

of the firms (demand side), we need more firm-oriented variables from the ECHP.

Unfortunately, the ECHP is a household survey instead of employee-employer match

data; we are not able to consider the effects of firm demand on part-time jobs in the

choice model.

6.1.2 The key variables in the wage equations

The dependent variable is the natural logarithm of the hourly wage expressed in its own

national currency (Dutch guilder and British pound), it is defined by the reported current

monthly earnings divided by the number of actual hours worked in that respective period.

The variable for age is used as a proxy for one’s potential working experience; the

dummy for tenure is a measure of firm-specific experiences. To control for the possible

wage gaps among different industries and the effects due to the size of firm, we include

dummies for industry and the size of firms. In addition, as explained in Part 5, we use the

computed inverse Mills’ ratios to correct for selectivity-bias in the wage equations.

6.2 The results of estimated ordered probit model 6.2.1 The choices for part-time jobs by gender and age

Table 6.1 summarizes the results of the estimated ordered probit model for the

Netherlands and Great Britain, the signs of the coefficients reveal the direction of effects

associated with the average individual demographic characteristics and the job related

18

Table 6.1 The estimates of ordered probit model

NL Model UK Model Constant 1.289** 1.395** (0.032) (0.027) Gender (female=1) -0.571** -0.290** (0.105) (0.109) Age 0.073** 0.054** (0.023) (0.018) Age squared (/10) -0.013** -0.009** (0.003) (0.002) Education (high level as ref. Group) Middle level -0.368 -0.003 (0.320) (0.103) Basic level -0.206 -0.105 (0.220) (0.073) Marriage (married=1) 0.552** 0.360** (0.134) (0.146) Gender x married -0.969** -0.795** (0.147) (0.155) Children (=1 if at least one child) -0.197 0.135 (0.133) (0.173) Gender x children -0.837** -1.053** (0.149) (0.183) Health problem (health=1) -0.087 0.704** (0.242) (0.353) Income effect (>3 as a ref. Group) Active working family members: 0-1 0.442** -0.004 (0.165) (0.125) Active working family members: 2-3 0.200 0.086 (0.155) (0.113) Unemployment (=1 if previously unemployed) -0.199** -0.086 (0.071) (0.140) Occupation (operatives as a ref. group) Professional workers -0.010 -0.649** (0.160) (0.203) Technician -0.167 -0.656** (0.154) (0.198) Clerical workers -0.154 -0.894** (0.159) (0.186) Service workers -0.510** -1.356** (0.162) (0.186) Craftsmen 0.086 -0.066 (0.189) (0.267) Elementary occupation -0.449** -1.292** (0.180) (0.197) Job level (basic level as a ref. group) Supervisory 0.554** 0.640** (0.187) (0.134) Intermediate 0.280** 0.562** (0.098) (0.090) No. of observations: 2638 3346 Log likelihood -1436.38 -1416.55 Pseudo R2 0.274 0.243

Standard errors in parenthesis, ** significant at 5 %.

19

attributes (occupation and job level). The tests for joint significances are listed in table

B2. We present the effects of a change in an explanatory variable on the probability

(marginal effect) in table B3.

It can be seen from table 6.1 that the coefficient for gender was significant in both

models, indicating that the women were more likely to take part-time jobs compared to

men. At the mean level, being a woman, the probability of working full-time decreased

while the probability for working short part-time and long part-time increased. By

comparing the probabilities for the variable of gender in table B3, we find that the Dutch

women had higher propensity to work part-time than the British women, the increase in

probability of a woman working part-time in the Netherlands was 0.151 (0.073+0.078),

two times higher than a woman’s in Britain.

As demonstrated in table 6.1, a married person was more likely to work part-time in both

models, which is consistent with the findings for the effect of martial status by Dekker

(2000) and Bardasi et al. (2000). In order to test for the difference between the married

women and men in taking part-time jobs, we interact the variables of gender and the

martial status. The coefficients for such interaction are found to be significant, implying

that compared to the married men, the married women had higher likelihood of working

part-time. Again, it is important to note that the increase in probabilities of working part-

time for the married women in the NL model was greater than their counterpart’s in the

UK model, as illustrated in table B3.

We also find a significant interaction for the variables of gender and children in table 6.1,

suggesting that the effect of children on women was stronger than men’s. Therefore, the

presence of children put more constrains on women’s labor supplies than men’s, the

adjustment for working time was mainly undertaken by women’s choice in part-time job.

Compared to the UK model, the NL model demonstrates that the effects of children for

the Dutch women were larger in both working short and long part-time (see table B3).

Based on the estimated ordered probit model, a simulation is made to examine the

combined impacts of gender, martial status and age on part-time choices. Suppose that

there are hypothetical persons aged 35 and 45 years respectively, they are married and

have at least one child under the age of 12. In addition, they have middle level education

and enjoy good health, other factors are set to be on the average.

20

Table6.2 Simulation: the probabilities of choosing part-time and full-time jobs

Short PT Long PT Full-time job Men (age=35) 0.0751 0.1294 0.7955 NL model Men (age=45) 0.0150 0.0446 0.9404 Women (age=35) 0.1928 0.2067 0.6004 Women (age=45) 0.0548 0.1070 0.8382 Men (age=35) 0.0145 0.0255 0.9600 UK model Men (age=45) 0.0033 0.0079 0.9889 Women (age=35) 0.0360 0.0501 0.9138 Women (age=45) 0.0075 0.0154 0.9770

Under these conditions, we compute the predicted probabilities and present them in table

6.2. As anticipated, in every age group women’s probabilities of working part-time jobs

are larger than men’s. For the NL model, among the women aged 35, the likelihood of

their taking short part-time job is 0.1928, one and a half time larger than men’s, yet, such

differential in the prime-age group is enlarged more than two times. On the other hand,

the ratios in probabilities of working long part-time between women and men for two age

groups are 1.59 and 2.40 respectively. Therefore, the simulation for the NL model

demonstrates that the gap in probabilities between women and men are increased from

the age group of 35 to the group of 45. However, the same pattern can not be found in the

UK model, where the ratios do not substantially vary with the age.

This result may be explained by the assumption that in the Netherlands the composition

of male part-timers is not fixed, most of them may use working part-time as a “bridge”

and eventually work full-time. On the other hand, in Britain the population of working

part-time is inelastic due to the limitation of social and demographic characteristics, the

age effect seems relatively smaller compared to the one for the Netherlands.

Table 6.3 Comparison between the U.S. and some European countries

Full-time job Part-time job Canada 0.596 0.142

The United States 0.585 0.171

Germany 0.515 0.233

Italy 0.502 0.129

The Netherlands* 0.600 0.230

Great Britain* 0.914 0.086

Note: for the countries without asterisks, the figures are from Bardasi et al (2000).

21

Table 6.3 makes a comparison in the predicted probabilities between the United States

and some major European countries including the Netherlands and Great Britain. For the

countries without asterisks, the figures are copies from Bardasi et al. (2000)’s paper; the

last two rows come from our own simulation in table 6.2. Note that the figures from

Bardasi et al.’s prediction refer to those married women aged 35,with the medium

education level and one child of ages 12-17.

The higher proportion of part-time in Germany seems to be associated with higher

unemployment in its own labor market; working par-time may not be individual

voluntary choice. In contrast, the recent studies suggest that the high incidence of part-

time jobs and flexible work in the Netherlands correspond to personal preferences (see,

e.g. Pot et al., 2001). Britain has a relatively higher ratio of working part-time in its

overall labor market, the lowest part-time rate for the British women in table 6.3 may

indicate that a large proportion of part-timers may be concentrated in other age group.

6.2.2 The choices for part-time jobs and education

The signs of education shown in table 6.1 were negative, indicating that relative to those

with high-level education, people with medium and basic levels of education were more

likely to work part-time jobs. However, the results of single and joint tests are found to be

insignificant in the model. Since the job level and the occupation were positively

associated with one’s education level, entering them in the model might cause

multicollinearity and make the variable for education insignificant, we reestimate the

model without the variables for job level and occupation. The re-estimation shows that in

the NL model, the education (medium and basic levels) was significant only at 10% level,

whereas in the UK model, the basic education level was significantly associated with

working part-time jobs at 5% level, but the coefficient for the middle education level was

insignificant.

Once we accept the assumption that the education level has no impact on taking part-time

jobs, the choice of part-time undertaken by women are quite likely motivated by combing

their participation in labor market and traditional roles in family production.

6.2.3 The choices for part-time jobs and the income effect

The signs of income effect shown in table 6.1 are positive in the NL model, implying that

people with less family income were more likely to work longer hours. However, the

22

significant effect is only found among those having at most one family member actively

involved in working. With 2-3 family members actively working, the effect becomes

insignificant. In the case of the UK model, all coefficients for the effect of income and

the joint test are shown insignificant.

6.2.4 The choice for part-time jobs and the characteristics related to work

A careful examination on table 6.1 and table B3 shows that job level had a significant

effect on individual choice in working part-time. The higher job level, the less likely one

worked part-time. For example, in the Dutch model, being in a supervisory position

decreased the probability of working short part-time and long part-time by 0.07 and 0.08

respectively, being in an intermediate position would decrease the probability by 0.03 and

0.04 respectively. Compared the last two rows in table B3, we see that there was a

significant variation in the changes of probability for a person from the supervisory

position to the intermediate position in the NL model, while for the UK model, the

marginal effects remain almost the same.

By and large, the signs of coefficients for the occupation variable are in line with our

expectation. The joint test for occupation show a significant result for both models. Note

that in the models, our reference group of comparison is operatives who were supposed to

be male-dominated occupation and usually work in the organized consecutive production,

hence, we assume that they are less likely to work part-time. In the UK model, compared

to the operatives, all other occupations were more likely to work part-time, the

coefficients were significant with the exception of craftsmen’s. Among the different

occupation, the NL model shows that, service and elementary occupations, which are

female dominated, were significantly associated with working part-time. The sign of

craftsman was positive, implying that they were more likely to work full-time.

Similar pattern can be found in both models that service workers and elementary

occupations had higher probabilities of working short part-time jobs. Note, however, in

the NL model, the likelihood remained almost the same when comparing the probabilities

between short part-time and long part-time jobs undertaken by the service and elementary

occupations. By contrast, the probabilities of working long part-time jobs by service

workers and elementary occupations in the UK model were decreased compared to the

23

ones of working short part-time jobs, suggesting that those occupations were more likely

to take short part-time jobs rather than long part-time jobs.

6.3 The results of estimating wage equations and decomposing wage

differentials 6.3.1 The analysis of wage equations

The regression results are reported in tables B4 and B5 in the appendix, all standard

errors have been adjusted according to the formula shown in the econometric

specification. Coefficients related to selection bias λ , ρ , and the standard errors of the

wage equations are reported in the lower panel in the table.

As the proxy variables, the age and the quadratic in age are used to capture the effect of

one’s potential working experience. Our results are in line with the basic assumption of

human capital theory and the wage function is concave in the effect of age. It can be seen

from the tables that the effect of age for full-time job was stronger in both models. For

example, in the UK model, holding other variable constant, the change in age by one year

would increase the wage by 8 percent for full-time jobs, but the wage rose only by 5.4 per

cent and 3.7 per cent for long part-time and full-time jobs.

Table B4 displays that the wage level for part-timers in the NL model was significantly

associated with their educational attainments, workers with high education received

higher wage. Furthermore, the effect of education on wage for part-timers was much

stronger than that for full-time jobs, although the coefficients for full-time jobs were

shown insignificant.

A significant effect of education on full-time job is found in the UK model, workers with

basic education level earned 24 per cent less than those with higher education in the

group of full-time jobs. Given each level of education table B5 shows that there was a

large difference in wage between short part-time and long part-time jobs. For example,

suppose a person with basic education works long part-time, his (her) wage will be 21 per

cent lower than that received by high educated part-timers, however, if this person works

short part-time job, his (her) wage would be substantially decreased and be paid 32.41 per

cent less than the level for higher educated part-timers in the same group.

The variables of industry and the sizes of firms are designed to capture the possible

compensation wage across industries and firms. Most coefficients were insignificant. As

24

to the effect of firm size, a similar pattern can be found in both models, i.e. working in

large- sized firms were paid higher than working in small and medium-sized firms. Note

that there was evidence shown in the tables that the effect of firm size was stronger in the

group of part-time jobs. For example, in the UK model, working short and long time in

the small-sized firms was paid 24.72 and 30.76 per cent less than the level of wage

working in large-sized firms. By contrast, the wage for those working full-time in small-

sized firm was only 19.77 per cent lower than the wage in large-sized firms.

Based on tables B4 and B5, we see that the hypothesis of no selection bias is rejected in

the Dutch model. For the UK model, we are able to accept the hypothesis of no selection

bias only for full-time job, but the hypothesis is rejected for part-time jobs. Our findings

for the UK model are in agreement line with Ermisch et al. (1992)’s analysis on the

British part-time workers.

The implication of rejecting hypothesis of no selection bias is that the standard errors for

the estimates of the parameters in the wage equation are biased upwards, the adjustment

of standard errors has to be corrected for. The lower panels in both tables also present the

coefficient ρ , which is a measure of the correlation between the error term in the choice

model and the one in the wage equation. The negative ρ suggests that, for given

measured characteristics, part-time or full-time employees who were observed in the

wage equations received lower wage offers than those excluded. On the other hand, a

positive ρ implies that, among those with similar observed characteristics, those (part-

timer or full-timer) who were excluded from the wage equations had lower wage offers

than those included.

6.3.2 The decomposition of wage differentials

The unadjusted wage differentials listed in table 6.4 are computed by taking anti-log of

the difference between the average wages rate for short (long) part-time job and full-time

job and subtracting 1 from it.

The figures in the second row of table 6.4 indicate that the wage gap between short part-

time and full-time jobs was about 11 per cent in the Dutch model, working short part-

time received 7.25 per cent lower wage than working long part-time job. Two things

deserve our attention when examining the long part-time job in the Dutch model. First,

the wage gap between full-time and long part-time jobs was only about 3 per cent;

25

second, the differential in wage between long part-time job and full-time jobs was

significantly smaller than the one between full-time and short part-time jobs.

Table 6.4 The unadjusted wage differentials (%)

NL model UK model

Full-time vs. short part-time jobs 10.52 31.52

Long-time vs. short part-time jobs 7.25 1.61

Full-time vs. long part-time jobs 3.05 29.43

The most interesting finding is the contrast between two models. A direct comparison

explicitly suggests that in general there was quite a large wage differential between full-

time and part-time jobs in the UK model. For example, short part-time workers and long

part-time workers earned 31 per cent and 29 per cent less than those working full-time

job respectively, much larger gaps than their counterparts in the NL model. The

magnitude of the effects of part-time jobs on the wage was quite striking, yet, it is

important to note that the wage gap between short part-time and long part-time jobs in the

UK model existed but quite small compared to their gaps with full-time jobs.

The results of decomposing the wage gaps are reported in table 6.5 The wage

differentials are decomposed into three components: the first part is due to the differences

in the average productivity-enhancing characteristics (endowments); the second part is

attributable to the differences in the average sample selection; the third part is what we

call “ return differences “ revealed by differences in estimated coefficients. Three pairs

are made when decomposing the wage differentials: full-time and short part-time jobs,

long part-time and short part-time jobs, and full-time and long part-time jobs.

To the extent the characteristics affected the wage gap, it can be seen from table 6.5 that

the differences helped reduced the wage gap between short (long) part-time and the full-

time jobs in the NL model. One possible explanation for this is that the older people were

over represent in the sample of part-timers for the NL model (recall that we use age as a

proxy for ones’ potential working experience).

Moreover, as we argued before, in the Netherlands, those working part-time jobs might

not be necessarily lower educated, people with high education were quite actively

involved in the part-time job market, especially work long part-time jobs. Thus, the

differences in education level for those contained in our sample reduced the wage

26

differentials. Similar explanation can be applied to the effect of narrowing wage gap

between long part-time and full-time jobs in the UK model, although such effect was

minor compare to the one in the NL model.

Table 6.5 The decomposition of wage gap

NL Model UK Model Full time

vs. Short PT

Long PT vs.

Short PT

Full time vs.

Long PT

Full time vs.

Short PT

Long time vs.

Short PT

Full time vs.

Long PT Wage gap

0.100 0.067 0.032 0.274 0.016 0.258

Due to: Characteristics -0.042 0.050 -0.088 0.039 0.036 -0.001

Selection -0.119 -0.195 0.008 -0.253 -0.199 -0.063

Return 0.328 0.215 0.113 0.488 0.169 0.322

Both models in table 6.5 show the positive effect of “ return”, indicating part-time jobs

are discriminated against full-time jobs. But a comparison between two models displays

the following patterns: (1) for the pair of full-time and short part-time jobs, the effect of

return in the UK model was stronger than the one in the NL model, the differences

accounts for 0.488 in the wage gap for the UK model as compared to 0.328 for the NL

model. (2) the differences in return contributed to the wage gap between full-time and

long part-time jobs in the NL model was much less than it is in the UK model.(3) the

differences in return in the NL model contributed more to the wage gap between long

part-time and short part-time jobs than it did in the counterpart for the UK model.

Such observation indicates that in Britain for those working short and long part-time, they

both suffered larger wage penalties, long part-timers were more likely to be treated as

short part-timers in terms of their wage rate. In contrast, those working long part-time in

the Netherlands were more likely treated as those working full-time. On the other hand,

in both countries, working short part-time jobs were paid lower, but the effect was

stronger in Great Britain, as shown in table 6.5.

7. Summary In this paper, we have made a comparative study on part-time jobs for the Netherlands

and Great Britain. The data sets from the European Household Panel (ECHP) are used.

27

We classify part-time job into short part-time and long part-time jobs and tend to explore

their impacts on the wage differentials.

In most comparisons between two countries, we find that employees in the Netherlands

had higher likelihood of taking part-time jobs than their counterparts in Great Britain. As

expected, marriage increased the probabilities of working part-time, but the effect

imposed on women was stronger than it did on men. The same conclusion can be applied

to the effect caused by the presence of children. Our simulation shows that given a certain

of constrains, the proportion of working part-time by the Dutch women is relatively

higher than the level in the comparable group in Great Britain.

The relationship between the choices of part-time jobs and the job related characteristics

are captured by the distribution of occupations and job levels. The result shows that

working in service sector increased the likelihood of taking part-time jobs, being

supervisory position or intermediate job level, their likelihood of working full-time

significantly rose. These conclusions hold for the Dutch model and the British model.

Surprisingly, although a large proportion of women were working part-time in the

Netherlands, the wage differentials between part-timers and full-time employees are

relatively smaller than the ones in Britain. More importantly, we find that the pattern of

the wage gap in the Netherlands was quite distinct from the one in Great Britain. The fact

that there was only 3 per cent of the unadjusted wage gap between long part-time and

full-time jobs in the Dutch model might single the presence of the ‘retention part-time job

“ characterized by Tilly (1996). Yet, in the Dutch model, the average characteristics for

those taking long part-time job is higher than those of full-time employees, as a result, the

difference in the average characteristics made a contribution of reducing the wage gap.

Compared to the Netherlands, working part-time in Britain suffered much wage penalty.

Our finding is in line with Tam (1997)’s arguments about the British part-time workers.

The analysis for the UK model suggests that in Great Britain there was no obvious

distinction between long part-time and short part-time jobs in terms of their wage. The

wage penalty for long part-time job was almost same as the one received by the short

part-time job.

Our study is based on the selected wave from ECHP; we only observe individual choice

on part-time job at one point in time. In order to capture the dynamic change of

28

individual choice and its effect on the wage, we need to make a panel study. Moreover,

the quality of part-time jobs actually covers many aspects such as the benefit and training

opportunities received by part-timers and their career prospects, etc., which really call for

a rich data set containing more information on part-timers. All these will be left for our

further study in the future.

29

Appendix A The definition of variables in the ordered probit and the wage equations

(1) The variable for part-time jobs: a dependent variable with three categories in the

ordered probit model. The variable is coded 1 if weekly working time is between 12 and

21 hours (short part-time job), 2 if weekly working time is between 22 and 29 hours (long

part-time job), 3 if weekly working time is greater than 29 hours (Full-time job).

(2) Hourly wage: a dependent variable in the wage equations. Hourly wage is calculated

according to the reported current gross wage divided by the number of actual working

hours in that respective period. We follow the convention and take the logarithm of

hourly wage.

(3) Gender: gender=1 for male, zero otherwise.

(4) Age: a continuous variable in years.

(5) Age squared: squared age divided by 10.

(6) Education: a categorical variable describing individuals’ education level. Education

=1 if it is a high education level, education =2 if it is a middle education level, education

=3 if it is a basic education level. Accordingly, three dummy variables are created, the

dummy for the basic education level is chosen as a reference group.

(7) Marriage: marriage=1 for married person, zero otherwise.

(8) Children: a dummy variable indicating the number of children under the age of 12 in a

household .It is coded 1 if there is at least one child, 0 if no child.

(9) Health: a dummy variable reporting the respondent’s healthy condition. Heatlth=1 if

one is healthy, zero otherwise.

(10) Income effect: The income effect is proxyed by the variable indicating the active-

working members in a household. It is coded 1 if there is at most one person actively

working, 2 if there are two or three actively working persons, 3 if there are four or five

persons actively working. Three dummies are created, the third group is chosen as a

reference group.

(11) Unemployment: a dummy variable, 1 if one was previously unemployed, zero

otherwise.

(12) Tenure: a categorical variable used for the estimation of wage equation. Tenure=1if

it is less than five years; tenure=2 if it is between 5 and 10 years, tenure=3 if it is greater

30

than 10 years. Three dummies are created, the third group is chosen for a comparison

group.

(13) Occupation:

Occupation=1 if one is a professional worker

Occupation=2 if one is a technician

Occupation=3 if one is a clerical worker

Occupation=4 if one is a service worker

Occupation=5 if one is a craftsmen

Occupation=6 if one is an operative

Occupation=7 if one’s job is elementary

Seven dummies are created, the group of operatives is chosen as a reference category.

(14) Job level:

Job level=1: supervisory

Job level =2: intermediate

Job level=3: basic

Three dummies are constructed, the third group is chosen as a reference group.

(15) Firm size:

Firm size=1 if there are less than 11 employees, small-sized firm

Firm size=2 if the number of employees is between 11 and 100,medium-szie firm

Firm size=3 if the number of employees is over 100, large-sized firm

Three dummies are created, large sized-firms are defined as a reference group.

(16) Industry:

Industry=1: agriculture

Industry=2: manufacturing

Industry=3: service

Three dummies are created, the dummy for agriculture is chosen as a reference group.

(17) Private: a dummy variable, indicating that the firm where the respondent worked is a

private or public sector. Public sector is chosen as a reference group.

31

Appendix B Table B1 The means and standard deviations of the variables in the ordered probit model NL UK Variable Mean Std. Dev. Mean Std. Dev. Gender 0.473 0.499 0.521 0.500 Age 36.340 9.299 36.147 11.535 Age squared (/10) 140.701 70.957 143.961 88.713 Middle education level 0.013 0.115 0.088 0.283 Primary education level 0.959 0.198 0.252 0.434 Marriage (married=1) 0.596 0.491 0.530 0.499 Gender X marriage 0.263 0.440 0.277 0.447 Children 0.367 0.482 0.292 0.454 Children x gender 0.146 0.354 0.144 0.352 Health 0.986 0.118 0.995 0.071 Active members:0-1 0.292 0.455 0.253 0.435 Active members: 2-3 0.675 0.468 0.673 0.469 Unemployment 0.221 0.415 0.056 0.230 Professional workers 0.205 0.404 0.162 0.369 Technician 0.271 0.444 0.159 0.366 Clerical workers 0.167 0.373 0.227 0.419 Service workers 0.120 0.325 0.172 0.378 Craftsmen 0.109 0.311 0.113 0.317 Elementary occupations 0.056 0.231 0.078 0.268 Job level: supervisory 0.060 0.238 0.107 0.309 Job level: intermediate 0.131 0.338 0.188 0.390 Table B2 The LR test for joint significances in the ordered probit model NL Model UK Model LR test for joint significance of education 1.40 2.26 LR test for joint significance of income effect 12.91** 1.78 LR test for joint significance of occupation 34.36** 151.32** LR test for joint significance of job level 16.85** 58.64** ** Significant at 5 % level. Note: the income effect refers to the number of actively working members in a family.

32

Table B3 The marginal effects based on the estimated ordered probit model NL model UK model Short PT Long PT Full-time Short PT Long PT Full-time Gender 0.073 0.078 -0.151 0.022 0.022 -0.040 Age -0.009 -0.010 0.019 -0.004 -0.004 0.008 Age squared (/10) 0.002 0.002 -0.004 0.001 0.001 -0.002 Middle education level 0.045 0.051 -0.096 0.0002 0.0002 -0.0004 Primary education level 0.025 0.028 -0.054 0.008 0.008 -0.016 Marriage (married=1) -0.075 -0.076 0.151 -0.027 -0.027 0.054 Gender x married 0.170 0.129 -0.299 0.059 0.060 -0.120 Children 0.025 0.028 -0.053 -0.010 -0.01 0.020 Gender x children 0.158 0.112 -0.271 0.153 0.097 -0.249 Health problem 0.010 0.012 -0.022 -0.053 -0.053 0.106 Active member: at most one -0.048 -0.058 0.106 0.0003 0.0003 -0.006 Active member: 2-3 -0.026 -0.028 0.054 -0.06 -0.007 0.013 Unemployment 0.025 0.028 -0.052 0.006 0.007 -0.013 Professional workers 0.001 0.001 -0.003 0.048 0.049 -0.097 Technician 0.021 0.023 -0.044 0.049 0.050 -0.099 Clerical workers 0.019 0.021 -0.040 0.067 0.068 -0.134 Service workers 0.063 0.070 -0.133 0.101 0.103 -0.204 Craftsmen -0.011 -0.012 0.022 0.005 0.005 -0.010 Elementary occupation 0.055 0.062 -0.118 0.096 0.098 -0.194 Job level: supervisory -0.068 -0.077 0.145 -0.048 -0.048 0.096 Job level: intermediate -0.035 -0.039 0.073 -0.042 -0.043 0.085 Note: due to rounding, the sum of marginal effect for some variables may not be equal to zero.

33

Table B4 The estimation of wage equation for the NL model

Explanatory variable Short PT coefficient Long PT

Coefficient Full-time

Coefficient Constant 1.494 2.037 1.097 (0.450) (0.406) (0.153) Age 0.091** 0.094** 0.109** (0.015) (0.014) (0.006) Age squared (/10) -0.010** -0.011** -0.012** (0.002) (0.002) (0.001) Education (ref.group: high) Middle level -0.560** -0.844** -0.012 (0.261) (0.248) (0.082) Basic level -0.430** -0.564** -0.007 (0.206) (0.174) (0.044) Tenure (ref.group: <=5 years) 6-10 years 0.238** 0.143** 0.038** (0.050) (0.048) (0.019) More than 10 years 0.196** 0.135** 0.043 (0.058) (0.055) (0.022) Unemployment (unemployed=1) -0.070* -0.059 -0.092** (0.047) (0.048) (0.019) Firm size (ref.group: large) Small -0.110** -0.204** -0.113** (0.050) (0.050) (0.022) Medium -0.131 -0.078 -0.077** (0.071) (0.066) (0.023) Industry (ref.group: agriculture) Manufacturing 0.062 0.072 0.122 (0.240) (0.261) (0.096) Service 0.061 0.093 0.183** (0.233) (0.249) (0.050) Private (ref.group: public) -0.147** -0.111** -0.093** (0.047) (0.044) (0.020) Lambuda -0.146** -0.131** -0.143** (0.052) (0.029) (0.017)

σ 0.394 0.339 0.370 ρ 0.370 0.386 0.386

No. of observations 371 293 1914 Adjusted R-squared 0.268 0.335 0.376

The corrected standard errors in parenthesis, ** significant at 5%, * significant at 10 %. Note: short PT: short part-time job; long PT: long part-time job.

34

Table B5 The estimation of wage equation for the UK model

Explanatory variable Short PT

Coefficient Long PT

Coefficient Full -time

Coefficient Constant 0.411 0.799 0.620 (0.471) (0.357) (0.134) Age 0.037** 0.050** 0.080** (0.011) (0.013) (0.004) Age squared (/10) -0.004** -0.005** -0.009** (0.001) (0.002) (0.001) Education (ref.group: high) Middle level -0.090 -0.124* -0.086** (0.074) (0.077) (0.027) Basic level -0.281** -0.197** -0.212** (0.051) (0.056) (0.018) Tenure (ref.group: <=5 years) 6-10 years 0.112* 0.122* 0.045** (0.068) (0.071) (0.022) More than 10 years 0.138* 0.046 0.029 (0.088) (0.104) (0.032) Unemployment (unemployed=1) 0.056 -0.149 -0.030 (0.108) (0.114) (0.031) Firm size (ref.group: large) Small -0.221** -0.268** -0.180 (0.058) (0.059) (0.017) Medium -0.118* -0.066 -0.105** (0.063) (0.068) (0.018) Industry (ref.group: agriculture) Manufacturing 0.233 0.054 0.056 (0.404) (0.225) (0.105) Service 0.263 0.139 0.058 (0.395) (0.212) (0.104) Private (ref.group: public) -0.269** -0.114** -0.109* (0.052) (0.055) (0.018) Lambuda -0.131* -0.158** -0.253 (0.075) (0.040) (0.015)

σ 0.392 0.348 0.441 ρ 0.333 0.453 0.574

No. of observations 324 219 2738 Adjusted R-squared 0.284 0.296 0.355

The corrected standard errors in parenthesis, ** significant at 5 %, * significant at 10 %. Note: short PT: short part-time job; long PT: long part-time job.

35

References Barzel, Y. (1973). “The determination of daily hours and wages.” Quarterly Journal of

Economics, May, 220-38.

Bardasi, Elena and Janet C.Gornick (2000). “Women and part-time employment:

workers’ ‘choices’ and wage penalties in five industrialized countries.” discussion paper

2000-11, Institute for Social and Economic Research (ISER), University of Essex.

Blank, Rebecca (1990). “ Are part-time jobs bad jobs?” In: Gary Burtless (ed.), A future

of louse jobs? Brookings institution, Washington, D.C.

Dekker, Ronald, Muffels, Ruud and Elena Stancanelli (2000). “A longitudinal analysis

part-time work by women and men in the Netherlands.” In: Gustafsson, S. et al. (ed.),

Gender and the labour market, Econometric evidence on the obstacles in achieving

gender equality. MacMillan.

Ehrenberg, R.G. and Robert S.Smith (2000). Modern labour economics: theory and

public policy. Addison-Wesley.

Ermisch, F. John and Robert E.Wright (1991). ” Women offers and full-time and part-

time employment by British women.” The journal of human resources XXVIII, 112-

133.

Ermisch, F.John and Robert E.Wright (1992). “Differential returns to human capital in

Full-time and part-time employment.” In: Folbre, Nancy. et al. (ed.) Women’s work in

the world economy. Macmillan.

Eurostat (2001). ECHP UDB manual: waves 1 to 5,survey years 1994 to 1998.

Greene, H.W. 2000. Econometric analysis. Prentice Hall International, Inc.

Killingsworth, Mark R. (1983). Labour supply. Cambridge: Cambridge University Press.

Leeds, Michael A. (1990). “Part-time status and hourly earnings of black and white

men.” Economic inquiry. Vol.XXVIII, 543-54.

Lettau, Michael K. (1997). ”Compensation in part-time jobs verse full-time jobs what

if the job is the same?” Economic Letters 56, 101-106.

Llloyd, B.C. and Beth N.T. (1979). The economics of sex differentials. Columbia

University Press.

Long, J.S. (1999). Regression models for categorical and limited dependent variables.

Sage Publications.

36

Main, Brain G.M. and Barry Reilly (1993) “ The Employed size-wage gap: evidence for

Britain.” Economica 60,125-142.

Maddala, G.S.(1983). Limited dependent and qualitative variable in econometrics.

Cambridge: Cambridge University Press.

Moffitt, R. (1984). “The estimation of a joint wage-hours labor supply model “.

Journal of Labor Economics, October 550-66.

Montgomery and James Cosgrove (1993). “Fringe benefits and the demand for part-

timers.” Industrial and labour relations review, October 472-84.

Montgomery and James Cosgrove (1995). “Are part-time women paid less? A model

with firm- specific effects.” Economic Inquiry, 33(1), 119-133.

OECD (2000). Labour Force Statistics 1979-1999.

OECD (1999). Economic Outlook No.65, Paris.

Oaxaca, Ronald (1973). “Male-female wage differentials in urban labour market.”

International Economic Review, Vol.14 No.3, October.

Pot, F., Koene, B.and Jaap Paauwe(2001).“Contingent employment in the Netherlands.”

ERIM report series Research Management, University of Erasmus.

Rosen,H.S.(1976).” Taxes in a labour supply model with joint wage-hours

determination.” Econometrica 44(3), 485-507.

Tam, May (1997). Part-time employment: a bridge or a trap? Ashgate Publishing Ltd.

Tilly, Chris (1996). Half a job: Bad and good part-time jobs in a changing labour

market. Temple University Press.

Tijdens, Kea (1997). “Are part-time and full-time jobs really so different? Explaining

women’s working time from family, housework, individual and work characteristics.”

In: Tijdens et al.(ed). Time allocation and gender: the relationship between paid labour

and household work. Tilburg University Press.

Tijdens, Kea. (2002). “Gender roles and labour use strategy: women’s part-time work in

the European Union.” Feminist Economics, 8(1), 71-99.

IRISS-C/I is currently supported by the European Community under the

Transnational Access to Major Research Infrastructures action of the Improving the Human Research Potential and the Socio-

Economic Knowledge Base programme (5th framework programme)

[contract HPRI-CT-2001-00128]

Please refer to this document as

IRISS Working Paper 2003-05, CEPS/INSTEAD, Differdange, G.-D. Luxembourg

(CEPS/INSTEAD internal doc. #07-03-0020-E)

View publication statsView publication stats