Male-female wage gap and firm effect
The case of young Italian workers
(Running title: Male-female wage gap and firm effect)
Capellari Saveria, Chies Laura, Zaccarin Susanna
Dipartimento di Scienze Economiche e Statistiche
Università degli Studi di Trieste
Piazzale Europa, 1
34127 Trieste (Italy)
Preliminary draft
Please do not cite without permission
Email: saveriac@econ.univ.trieste.it,
laurac@econ.univ.trieste.it
susannaz@econ.univ.trieste.it
Corresponding author: Saveria Capellari
1
Abstract. The idea that wages are determined by firm and individual characteristics leads to
gather that there is a firms’ effect that influences wage differentials. This paper presents the
results of an empirical analysis of gender wage differentials - based on INPS data firms for
people between the ages of 20 and 25 employed in the private sector in 1996 - which takes
into account the characteristics of workers and firms using a two level random effect model.
Firm variables proved to be significant, and firm female proportion showed a negative effect
on wage of both man and woman.
1. Introduction1
In recent years extensive use has been made of employer-employee datasets for the
development of an international line of research on male-female segregation and
discrimination at the workplace.
Linking company and employee characteristics, these datasets are particularly suitable for an
analysis of discrimination and segmentation, making it potentially possible to attribute wage
differentials unfavourable to women - observed in all OECD countries - to characteristics of
the employee and the workplace. It is a matter of understanding whether women’s
disadvantages depend, to use Groshen’s terms, on “who you are, what you do or where you
work” (Groshen, 1991). There is a broad consensus to the fact that in Italy, as in many other
OECD countries, gender-based wage differentials are substantial1. These differentials could be
explained in terms of differences within or between groups of men and women.
Although Italian literature on wage differentials is rich in empirical studies (Iter, 2001;
Favaro, Magrini, 2002; Borgarello, Devicienti, 2002), the field of the interaction between
1
We are grateful for the use of the INPS Panel to Laboratorio Revelli of Moncalieri (Turin, Italy) and for
research assistance from Vania Colladel.
2
gender discrimination and the presence of segregation or selection at a company or local
labour market level has yet to be fully explored.
This study presents the results of an empirical analysis of gender wage differentials based on
the INPS Panel covering the period 1986-1996 for people between the ages of 20 and 25
employed in the private sector in 1996. The aim is to analyse the factors determining malefemale wage differentials, taking account of the characteristics of workers and firms.
The wage equation was estimated separately for males and females, using a two-level random
effects model to allow for firm heterogeneity (Bryk, Raudenbush, 1992; Goldstein, 1995;
Snijders, Bosker, 1999). This specification is particularly suitable for the analysis of
hierarchical data such as those in the employer-employee dataset, which contains information
on people employed in the same firm. A hierarchy consists of units grouped at different
levels, thus employees are the level 1 units clustered within firms employers that are the level
2 units. It is not unreasonable to assume that a firm will set an individual’s wage relative to
the other workers in the firm, so the standard assumption of independent observations in
regression analysis is violated and models that account properly for the grouping effect have
to be specified.
The results we obtained are directly comparable with those presented by Reilly and Wirjanto
(1999), who use a similar approach to emphasise the importance of company segregation, not
only individual characteristics, as an explanation of wage differentials.
Wage differential analysis is traditionally performed by breaking down the gap into two
components (Oaxaca, 1973). One is associated with human capital, described by the observed
characteristics of the individuals, and the other is ascribable to the rewards given to these
characteristics, and may thus be attributed to discrimination in a broad sense. In our study the
female wage differential was calculated taking separate account of individual and firm
characteristics.
3
The paper is organized as follows. The first section presents a discussion of the link between
wage differentials and discrimination and sets out the main results achieved in this field in the
international literature. Section 2 provides a brief illustration of the features of the data-set
used in the current analysis. Section 3 presents the results of the estimates obtained with the
model. Section 4 puts forward a decomposition of wage differentials which specifies the
effects of individual and firm characteristics. Some concluding remarks are made in Section 5.
2.
Brief survey of recent literature
The idea that wages are determined more by the firm than by individual characteristics leads
in turn to the idea that there is a firms’ effect that influences wage differentials in general.
There may be sectors or size of firm where men (or women) employees are numerically
prevalent and within the sector or company no gender discrimination is observed, but it
becomes evident when female concentrations are more marked.
Theoretical explanations of a possible effect of this kind have been advanced by literature on
the segmentation of the labour market and by discrimination theories (Becker, 1971; Arrow,
1985). In the former case it is postulated that entry into the job market takes place through job
competition (which means that outsiders do not constitute a random group of entrants, and
employers’ choices are made on the basis of lower training costs, which are determined by the
characteristics of the worker). In the latter, discriminatory preferences are held to make it
cheaper for a company to employ males. Both types of model imply that at an aggregate level
women are subject to segregation in particular firms and occupations.
Recent literature contains an increasing body of evidence that segregation is the biggest factor
in creating problems of equity in general. This conviction has given rise to the adoption of
comparable worth policies in many western countries. The growth of female occupation in the
4
1980s and 90s was not accompanied by a proportional development of desegregation (Bettio,
Villa, 1996). Confirmation that the movement towards desegregation (where it actually exists)
is slow has been provided by comparisons between young and adult employees (Favaro,
2003). Using the European Consumer Household Panel (ECHP) and the European
Standardised Expenditure Survey (ESES), Bettio (2002) has recently pointed out that
segregation in the European Union is vertical, that is to say on the basis of occupational rank,
but that it leads to such a level of concentration between sectors as to entail horizontal
segregation too.
The effect of segregation on wages in Italy has been empirically estimated by Reilly and
Lucifora (1990) on the basis of ENI-IRI figures. Their results show that gender differentials
decrease as the proportion of women in a company increases, and that the presence of female
segregation entails lower wages for both sexes.
More recent studies (Gubta, Rothstein, 1999; Reilly, Wirjanto, 1999) confirm the hypothesis
that in addition to an effect of segregation by sector and occupation there is a specific effect
caused by the number of women in a company. Reilly and Wirjanto (1999) test the predictions
of the Arrow model of heterogeneous employer discrimination. They use the 1979 crosssection of the General Segmentation Survey (GSS) from the Maritime Provinces of Canada.
They document the heterogeneity at the establishment level of the gender composition. The
establishment female proportion was shown to be negatively correlated with the average
establishment wage and positively correlated with establishment size2. Their results showed,
however, that the proportion of women in an establishment had a negative effect on men’s and
women’s wages alike, a result which is consistent with other previous results and a hypothesis
of crowding.
The empirical analysis in our study takes separate account of the effects of firms’ sector and
size and of the proportion of female employees in the firm. It thus enables us to assess
whether and to what extent differences between companies count in the determination of
5
wages and what effect a high proportion of female employees in a firm has on men’s and
women’s wages. As a consequence of concentrating the analysis on a group of young
employees the wage differentials will be smaller than those found in other studies (Addis,
Waldmann, 1996; Iter, 2001; Blau, Khan, 1997), since women’s wage gaps seem to increase
with age and working experience.
3.
The INPS Panel
The employer-employee data used in the study were drawn from the INPS Panel constructed
by LaborRevelli of Turin3. Our sample covers all workers born between 1971 and 1976 who
began working between 1986 and 1991 and were employed in the private sector on December
1st 1996.
Information on individuals collected in the Panel comprise gender, age, occupation4, type of
contract (national, local, firm specific, individual), hierarchical level, duration of employment,
province of employment, earnings and some aspects of the working histories (changes in
occupation, employer and sector; dates of entry; non-working periods). From this information
working experience was defined as the total number of paid months and propensity to job
mobility as the number of times a worker had changed employer5.
The dataset does not provide information on education level and family background
(especially number and age of children). As regards education an inference may be made as to
the possession of a high-school diploma by reference to first-job entry after the age of 18.
Information on companies is confined to the average size of the firm, sector of activity
according to the three-digit ATECO81 code, the firm’s location at the province level and the
age of the firm6.
ATECO sectors were grouped into eight categories, set out in Appendix A (Table A.1).
6
Agriculture and public administration were excluded on the grounds of scarcity of data. It was
also decided to group firm employee numbers into three size classes (1-9, 10-99, and more
than 100) and four macro-areas (north-east, north-west, centre and south). Since the Panel
data used do not provide a gender breakdown of employees by firm, for an analysis of the
effect of proportions of female employees recourse was made to the figures provided by 1997
Intermediate Industry Census. Firms in the Census were classified by very detailed
information on size ( …….chiedere Vania), sector (three-digit ATECO81 code) and province
of location. The mean female proportion obtained within each cell were then imputed to the
firms in the INPS sample.
We also used the proportion of females per company unit provided by the 1991 Industry
Census. The preliminary analysis produced consistently better results with the 1997 Census,
so this is the one featured in the estimates presented from now on.
3.1
Employee and firm characteristics
The sample of workers born from 1971 to 1976 in the 1996 dataset comprised 14,5967 people,
6,164 women and 8,432 men, working in 13,136 companies8 (see Appendix A, Table A.1,
A.3). Males accounted for about 58% of the total. One of the most important differences
between males and females was education: 78% of females had a high-school diploma,
compared with 66% of males. This difference was reflected in occupational rank, with a
higher concentration of women in white-collar professions and a greater proportion of men in
blue-collar occupations. The part-time percentage was low, but was made up for the most part
of women, as was to be expected. Overall, 14% of workers were employed on trainee
contracts; in this regard the proportion of men was slightly higher.
7
The other working history elements in the sample showed basic similarities between the
genders: specifically, the values for experience and mobility were very close.
Firm sizes were highly concentrated in the smallest category (less than 10 employees), despite
the fact that this size was under-represented in this sample as compared with the census
figures.
Distribution by sector showed a concentration of women in traditional manufacturing (food,
textiles, leather, wood and furniture), commerce and personal services, while men were
predominant in heavy and light engineering (metal products, electrical and electronic
equipment, industrial machinery).
Taking account of the results of the initial analyses and observations made recently on INPS
Panel data (Ginzburg et al., 1998; Gavosto, Rossi, 1999; Contini, 2002), gender gap was
calculated on annual earnings (Section B of the INPS form) which are proportional to the
number of days worked9.
Gender earnings ratio is about 90%. Though rather low, this differential is not in contradiction
with other estimates made for Italy, since it is known that differentials increase with age as a
result of, for example, women’s discontinuous presence in the labour market and possible
vertical segregation.
Significant differences emerged in terms of the distribution of males and females classified on
the proportion of females employed in the firm10 (see Appendix A, Tables A.1 and A.2). In the
classes with the lowest and the highest proportion of female employees the mean percentage
was 19% and 81% respectively, which confirmed the presence of segregation in Italian
companies (Table A2). The employees of companies with a high proportion of women had
greater experience (an average of about twelve months more), but despite this had wage levels
16% lower than employees in male-dominated firms (Table A2). Furthermore, in firms with
the highest proportion of females, training and occupational ranks were lower (Table A.2) and
the firms themselves tended to be of small size and over 70% of them operated in traditional
8
manufacturing and about 18% in the personal services (Table A.3). As far as firm
characteristics are concerned over 65% of the sample was made up of firms with a low
proportion of female employees. Firms with higher proportions of females tended to be
located in north-east (Table A.3).
4.
Estimation results
For an assessment of the effects of individual and company characteristics on individuals’
earnings, the wage equation was set out (separately for males and females) as follows:
Wij = Xij+ Zj + uj + eij
i = 1,…, nj , j = 1,…, J;
[1]
eij = N (0, 2e)
uj = N (0, 2u)
where Wij is the logarithm of the annual earnings of worker i employed in company j, Xi and Z
j
are the vectors of individual and firm characteristics respectively and and are the vectors
of the associated coefficients. In equation [1] the effect of individual and company
characteristics are assumed to be constant for all workers, whereas variability between firms is
given by the random term uj. The coefficient estimates were obtained using the residual
maximum likelihood (REML) method (Snijders, Bosker, 1999)11; the results are reported in
Table A.412.
The proportion of variance of the annual paid wage attributable to the firm (measured by the
intra-class correlation coefficient and evaluated starting from the initial model without any
explanatory variables, “null” model) is equal to 46.5% for males and 42.2% for females
9
respectively. After controlling for the workers’ characteristics, the firm’s variables explain
60% of firm’s variance for males and 59% for females.
The effect of individual variables was as expected: wages increased with the possession of a
high-school diploma (+5% for women and +13% for men), a white-collar position (+26.6%
for women and +21.3% for men) and experience (+20% for women and +22% for men),
which is consistent with the results generally presented in the literature. Mobility between
firms proved to be significant, penalising employees of both genders (-5.6% for women and –
3.7% for men). Though this result has appeared in other studies it was not a foregone
conclusion - since the sample deals with young people it might be expected that greater
mobility would be motivated by a search for a better job. As emerged in other analyses of
INPS data, part-time employment proved to penalise both genders, especially in blue-collar
occupations, though the effect in our estimates was more marked. For blue-collar workers,
male and female alike, the wage was halved, while for white-collar employees it went down
by about 30% compared to full-time employees. Part-time work, however did not appear to
penalise women (-48.5%) more than men, as also reported by Favaro and Magrini (2002). The
effect of being employed on a trainee contract was positive with the exception of male bluecollar part-time employees (of whom there were very few). This is explained by the fact that
the sample includes workers with discontinuous work experience. Compared to the latter
category, young workers on trainee contracts stay longer in their jobs and are better paid, as
found by ISFOL (1998)13.
With a few exceptions, company variables proved to be significant (Table A.4). Wages were
lower for both genders not working in heavy and light engineering, with the exception of the
energy and chemical industries. Working in services, especially transport and communications
and personal services, proved to penalise women. Both for males and females, the location of
a firm in southern Italy as opposed to north-west proved to have a negative effect on wages.
Significant gender differences did emerge in some respects: company location in north-eastern
10
Italy proved to be important for women (+5%) and in central Italy negatively for men (-5.8%).
Firm size had a positive, and slightly more marked, effect for men (respectively +9,4% and
+21%, as against 7,6% and 18.5% for women).
The hypothesis that the proportion of women employees in a firm has an effect on wages was
confirmed by the data on men (-11.9%) and women (-9.3%) alike, as in Reilly and Wirjanto
(1999). The proportion of women employees did not appear to interact significantly with the
company size variable and with the economic sectors.
5.
The decomposition of male-female differential
The analysis of gender-based wage differentials is usually carried out following an Oaxaca
decomposition (1973), which imputes the difference in the average wage between women and
men to two elements: differences between the human capital characteristics possessed by the
two groups and differences between the returns of those characteristics, which may be
interpreted as a measure of discrimination. The decomposition is based on wage equations for
males and females, while discrimination is addressed on the difference between the
coefficients estimated for males and females. The underlying assumption is that the
coefficients calculated for males would be prevalent in the absence of discrimination.
Although the distinction between the human capital component and discrimination may not be
considered a very clear one14, with some caution the decomposition seems to produce relevant
results.
Over the years the Oaxaca method has been extensively used with a series of variants that
have attempted to take account not only of the limitations that the traditional technique
appears to have, but also of models better suited to the question in hand15.
11
One of the most critical problems to be highlighted is the possible selection bias determined
by the fact that wages are observed exclusively for employed people. The selection bias is
usually assessed by the use of a Heckman procedure16. This problem, certainly relevant to our
study, could not be solved because the data available did not contain the further information
required to estimate participation.
A recent estimate of the various components determining the wage gap (Flabbi, 1997), based
on Bank of Italy data, shows that wage differentials have been diminishing over time and that
this has been accompanied by a loss of explanatory power on the part of individual
characteristics caused by their growing homogeneity between the genders. The differential is
consequently explained by differences in returns, especially on education, for women. In line
with other studies on Italy, Flabbi’s paper does not take specific account of firm’s effects.
In what follows traditional decomposition has been used to highlight the effects of individual
and company characteristics on wage differentials. It is formulated as follows:
ln Wm  ln Wf  X f (ˆ m  ˆ f )  ( X m  X f )ˆ m  ˆ f ( Zm  Zf )  Zm (ˆ m  ˆ f )
[2]
The first two terms of the right-hand component shows the effects on differentials of the
human capital variables ( X m  X f ) ̂ and the associated different returns for men and women
X f ( ˆm  ˆ f ) . The second two terms sums up the effects of the various firm characteristics,
highlighting the differences in the distribution of workers among firms ˆ f ( Z m  Z f ) and the
different returns given to males and females within a single firm type Z m (ˆ m  ˆ f ) .
The results of the decomposition are presented in the Appendix A (Table A.5).
12
Concentrating on human capital possessed by the two groups and on its returns, it emerges
that women in the sample appear to be in a favourable position in terms of composition by
occupation groups, education and experience (an unusual result, probably connected with their
young age), while returns have the following diversified effects: white-collar occupation
favours women more than men, while the possession of a high-school diploma and experience
gives more advantages to men. Mobility among firms, which has been observed to bring down
wages, penalises women more, as in Favaro and Magrini (2002).
Considering individual characteristics, the conclusion is that the women’s group on average
possesses an endowment of human capital that would justify higher wages than those earned
by men, and that the effect of discrimination on the returns given to education and experience
has a decisive weight. The contribution of the discriminatory component in explaining the
wage differential is 65%.
Firm’s characteristics, expressed by the variables sector, size and proportion of female
employees, play a significant - though less important - role, accounting for about 11% of the
overall differential. A more detailed examination produces a series of interesting points. The
proportion of women employees in a firm has a significant negative impact on men’s and
women’s wages. Women are penalised as a result of their greater concentration in companies
with high proportions of women employees, but also because they receive a lower wage in
companies with a medium proportion of women. In firms with a high proportion of women,
men prove to be more heavily penalised.
As pointed out above, working in sectors other than energy and chemicals has a negative
impact on the wage received by all the workers in our sample, and this impact is in all cases
greater for women. Also in terms of company size, which has a marked positive impact on the
wage of all individuals, the impact is less favourable for women.
To sum up, the employer factors which go furthest to explain wage differentials to the
detriment of women seem to stem from their greater concentration in firms employing a large
13
proportion of women, especially in the personal services sector, and from lower wages for
women than for men in firms with a medium proportion of women and medium-to-large sized
firms. These results point to the need for further exploration. While a role is evidently played
by segregation in determining the gender gap, it is also clear that there are wage disparities
within single firm types whose causes have yet to be deeply investigated.
6.
Conclusions
The initial results of the quantitative analysis configure an important role for the gender
composition of a firm in determining the male-female wage differential, which is consistent
with the results of research carried out in other countries.
The application of the random effect model has provided convincing evidence in this regard,
which points the way to further exploration of the impact of the proportion of women in firms
on pay. The role of individual characteristics still seems to be of considerable significance,
however, especially their different returns, which may be an index of discriminatory
behaviour.
In this regard it seems that experimentation is required with models that can take account of
the fact that identical individual characteristics may have different effects in the various firm
contexts. To this end it is essential to have precise information on proportions of women
employees which is not imputed, as in this application, but drawn directly from the INPS data
set. The fact that firm characteristics have an effect on wage differentials also seems to require
an enrichment of data with more qualitative information on firms’ employment strategies.
The simple fact of a negative impact on women’s wage determined by firms clearly does not
necessarily imply the presence of discriminatory behaviour - it may also be a result of
individual preferences and technology. In the absence of a structural model able to provide a
14
simultaneous explanation of decisions on recruitment policy and choices regarding the gender
composition of firms, the effects observed cannot be attributed only to one of the hypotheses
mentioned.
Although the results obtained here have given a first answer to the initial question regarding
“who you are, what you do or where you work” (Groshen, 1991), further research in this
direction would benefit greatly from an extension of the dataset with more information on
firms (part of which is already being gathered) and individual characteristics.
15
APPENDIX A
Table A.1: Descriptive statistics
Individual characteristics
Complete sample
No
Number of cases
%
Males
No
Females
%
No
%
14596
100
8432
57.77
6164
42.23
4225
28.95
2849
33.79
1376
22.32
10371
71.05
5583
66.21
4788
77.68
2520
17.27
851
10.09
1669
27.08
White collar full time trainee contract
565
3.87
235
2.79
330
5.35
White collar part time
482
3.3
80
0.95
402
6.52
95
0.65
20
0.24
75
1.22
Blue collar full time
7506
51.43
5187
61.5
2319
37.62
Blue collar full time trainee contract
1309
8.97
1017
12.06
292
4.74
697
4.78
245
2.91
452
7.33
64
0.44
29
0.34
35
0.57
1358
9.3
768
9.11
590
9.57
52
0.36
38
0.45
14
0.23
661
4.53
483
5.73
178
2.89
3208
21.98
2386
28.3
822
13.34
3594
24.62
1685
19.98
1909
30.97
Construction
1309
8.97
1218
14.44
91
1.48
Wholesale and retail trade, Hotels
3510
24.05
1762
20.9
1748
28.36
Transportation and communication
461
3.16
330
3.91
131
2.13
1156
7.92
380
4.51
776
12.59
645
4.42
150
1.78
495
8.03
North West
5330
36.52
3008
35.67
2322
37.67
North East
4551
31.18
2559
30.35
1992
32.32
Centre
2451
16.79
1444
17.13
1007
16.34
South
2264
15.51
1421
16.85
843
13.68
<10
5602
38.38
3280
38.9
2322
37.67
10-99
5935
40.66
3331
39.5
2604
42.25
> 99
3059
20.96
1821
21.6
1238
20.08
Education
Comprehensive
Upper secondary
Occupation
White collar full time
White collar part time trainee contract
Blue collar part time
Blue collar part time trainee contract
Apprentice
Industry (ATECO81)
Electric and gas
Mining and chemistry
Metal products, electrical and electronic equipment,
industrial machinery
Food, textile, leather, wood and furniture
Finance, banking, business services
Personal services
Area
Firm size
16
Firm female proportion
Low (up to 40%)
9695
66.42
6754
80.1
2941
47.71
Medium (between 40 and 70%)
3785
25.93
1492
17.69
2293
37.2
High (higher then 70%)
1116
7.65
186
2.21
930
15.09
Mean
St.dev.
Mean
St.dev.
Mean
St.dev.
45.04
29.43
44.79
29
45.38
30
2.25
1.53
2.38
1.63
2.08
1.35
Annual wage (It. Lire 1996 in thousands)
21543
8995
22546
9054
20172
8731
Firm female proportion
32.27
22.48
24.6
18.33
42.76
23.38
Experience (No of paid months)
Mobility (No of changes of employer)
17
Tab.A.2: Individual characteristics by firm female proportion
Firm female proportion
low
n
medium
%
n
high
%
n
%
Gender
Males
6.754
69.66
1.492
39.42
186
16.67
Females
2.941
30.34
2.293
60.58
930
83.33
Comprehensive
2.789
28.77
1.008
26.63
428
38.35
Upper secondary
6.906
71.23
2.777
73.37
688
61.65
1.654
17.06
741
19.58
125
11.2
White collar full time trainee contract
392
4.04
154
4.07
19
1.7
White collar part time
318
3.28
147
3.88
17
1.52
74
0.76
19
0.5
2
0.18
5.115
52.76
1.742
46.03
649
58.16
Blue collar full time trainee contract
978
10.09
242
6.39
89
7.97
Blue collar part time
290
2.99
350
9.25
57
5.11
32
0.33
28
0.74
4
0.36
842
8.68
362
9.56
154
13.8
52
0.54
-
-
-
-
574
5.92
81
2.14
6
0.54
2.877
29.68
310
8.19
21
1.88
1.688
17.41
1.119
29.56
787
70.52
Construction
1.309
13.5
-
-
-
-
Wholesale and retail trade, Hotels
1.964
20.26
1.491
39.39
55
4.93
Transportation and communication
416
4.29
41
1.08
4
0.36
Finance, banking, business services
634
6.54
470
12.42
52
4.66
Personal services
181
1.87
273
7.21
191
17.11
North west
3.525
36.36
1.504
39.74
301
26.97
North east
2.834
29.23
1.267
33.47
450
40.32
Centre
1.612
16.63
677
17.89
162
14.52
South
1.724
17.78
337
8.9
203
18.19
<10
3.617
37.31
1.353
35.75
632
56.63
10-99
3.895
40.18
1.688
44.6
352
31.54
> 99
2.183
22.52
744
19.66
132
11.83
Education
Occupation
White collar full time
White collar part time trainee contract
Blue collar full time
Blue collar part time trainee contract
Apprentice
Industry (ATECO81)
Electric and gas
Mining and chemistry
Metal products, electrical and electronic
equipment, industrial machinery
Food, textile, leather, wood and furniture
Area
Firm size
18
Experience (No of paid months)
Mobility (No of changes of employer)
Annual wage (It. Lire 1996 in thousands)
Firm female proportion
Mean
St.dev.
Mean
St.dev.
Mean
St.dev.
44.04
28.75
43.98
29.27
57.27
32.83
2.21
1.49
2.35
1.64
2.23
1.42
22360.12
9161.29
20288.22
8819.34
18702.48
6803.88
18.95
11.17
51.97
8.2
81.13
7.36
19
Table A.3: Firm Characteristics
Firm female proportion
Complete sample
N
Number of cases
%
13136
Low
N
100
Medium
%
N
High
%
N
%
8616
100
3460
100
1060
100
9 695
66.4
3 785
25.9
1 116
7.6
Gender composition of employees
Total
14 596
Males
8 432
57.8
6 754
80.1
1 492
17.7
186
2.2
Females
6 164
42.2
2 941
47.7
2 293
37.2
930
15.1
29
0.22
29
0.34
Mining and chemical industry
158
1.2
76
2.2
76
2.2
6
0.57
Metal products, electrical and electronic equipment, industrial machinery
591
4.5
286
8.27
286
8.27
19
1.79
Food, textile, leather, wood and furniture
2814
21.42
1036
29.94
1036
29.94
742
70
Construction
1286
9.79
1286
14.93
Wholesale and retail trade, Hotels,..
3144
23.93
1744
20.24
1350
39.02
50
4.72
Transportation and communication
348
2.65
315
3.66
30
0.87
3
0.28
Finance, banking, business services
973
7.41
501
5.81
421
12.17
51
4.81
Personal services
602
4.58
152
1.76
261
7.54
189
17.83
North West
4687
35.68
3060
35.52
1346
38.9
281
26.51
North East
4144
31.55
2551
29.61
1168
33.76
425
40.09
Centre
2224
16.93
1433
16.63
635
18.35
156
14.72
South
2081
15.84
1572
18.25
311
8.99
198
18.68
<10
5391
41.04
3483
40.42
1300
37.57
608
57.36
10-99
5899
44.91
3870
44.92
1679
48.53
350
33.02
Industry (ATECO 81)
Electric and gas
Area
Firm size
20
> 99
1846
14.05
1263
14.66
481
13.9
102
9.62
21
Table A.4: Log wage regression: final model, two-level REML estimates
Females
Males
Individual Variables
Coefficients
Constant
P-value
Coefficients
P-value
9.3055
<.0001
9.2716
<.0001
0.05268
0.0003
0.131
<.0001
White collar full time
0.2662
<.0001
0.2134
<.0001
White collar full time, trainee contract
0.2857
<.0001
0.2312
<.0001
-0.3005
<.0001
-0.323
<.0001
White collar part time, trainee contract
-0.234
<.0001
-0.1905
0.0166
Blue collar trainee contract
0.0705
0.0031
0.0379
0.0034
-0.4852
<.0001
-0.5102
<.0001
-0.199
0.0019
-0.3799
<.0001
-0.05086
0.0063
-0.2103
<.0001
0.2031
<.0001
0.2225
<.0001
-0.1244
<.0001
-0.1282
<.0001
-0.05575
<.0001
-0.03747
<.0001
0.04241
0.1665
0.07901
<.0001
Food, textile, leather, wood and furniture
-0.04974
0.0048
-0.03555
0.0038
Construction
-0.04347
0.3053
-0.0418
0.0019
Wholesale and retail trade, Hotels
0.008854
0.6163
-0.02443
0.0529
-0.1671
<.0001
-0.1375
<.0001
-0.08169
<.0001
-0.0133
0.5516
-0.157
<.0001
-0.1022
0.0014
0.05004
<.0001
0.01578
0.1138
Centre
-0.02154
0.1355
-0.05858
<.0001
South
-0.1528
<.0001
-0.1397
<.0001
0.07684
<.0001
0.09382
<.0001
0.185
<.0001
0.2094
<.0001
Medium
-0.04964
<.0001
-0.02198
0.0592
High
-0.09295
<.0001
-0.1191
<.0001
Education
Upper secondary
Occupation
White collar part time
Blue collar part time
Blue collar part time trainee contract
Apprentice
Experience/10
Experience /100
2
Mobility
Firm variables
Industry
Electric, gas, mining and chemical industry
Transportation and communication
Finance, banking, business services
Personal services
Area
Northeast
Firm size
10-99
> 99
Firm female proportion
22
Fit test
Female
Male
Final
2e
2u
-2logL
Null Model P-value
0.1499
<0.0001
0.1095
<0.0001
Final Model P-value
0.09549
<0.0001
0.04522
<0.0001
Null Model P-value
9049.1
5475.8
11520.8
0.1258
0.1093
<0.0001
<0.0001
Model
0.08724
0.04389
P-value
<0.0001
<0.0001
6843.6
23
Tav. A.5: Log wage decomposition
Xm-Xf
m-f
(Xm-Xf)*m
(m-f)*Xf
-0.1147
0.07832
-0.0150257
0.06083898
-0.1699
-0.0528
-0.03625666
-0.01429824
-0.0256
-0.0545
-0.00591872
-0.00291575
-0.0557
-0.0225
0.0179911
-0.001467
-0.0098
0.0435
0.0018669
0.0005307
0.0732
-0.0326
0.00277428
-0.00154524
-0.0442
-0.025
0.02255084
-0.0018325
-0.0023
-0.1809
0.00087377
-0.00103113
-0.0046
-0.15944
0.00096738
-0.01525841
-0.059
0.0194
-0.0131275
0.0880372
-0.0059
-0.0038
0.00075638
-0.00172444
0.3
0.01828
-0.011241
0.0380224
Zm-Zf
m-f
(Zm-Zf)*m
(m-f)*Zf
0.0306
0.0366 0.00241771
0.00114192
-0.1099
0.01419 0.00390695
0.00439464
0.1296
0.00167 -0.00541728
2.4716E-05
Wholesale and retail trade, Hotels
-0.0746
-0.033284 0.00182248
-0.00943934
Transportation and communication
0.0178
0.0296 -0.0024475
0.00063048
Finance, banking, business services
-0.0808
0.06839 0.00107464
0.0086103
Personal services
-0.0625
0.0548 0.0063875
0.00440044
-0.0197
-0.03426 -0.00031087
-0.01107283
Centre
0.0079
-0.03704 -0.00046278
-0.00605234
South
0.0317
0.0131 -0.00442849
0.00179208
-0.0275
0.01698 -0.00258005
0.00717405
0.0152
0.0244 0.00318288
0.00489952
Medium
0.3239
0.02766 0.00711932
0.01319659
High
-0.1288
-0.02615 0.01534008
-0.00394604
Individual Variables
Upper secondary
White collar full time
White collar full time, trainee contract
White collar part time
White collar part time, trainee contract
Blue collar trainee contract
Blue collar part time
Blue collar part time trainee contract
Apprentice
Experience/10
Esxperience2/100
Mobility
Firm Variables
Industry
Electric, gas, mining and chemical industry
Food, textile, leather, wood and furniture
Construction
Area
Northeast
Firm size
10-99
> 99
Firm female proportion
Notes
24
1
Estimates put them at between 13% and 20% from the mid-1980s to the mid-1990s.
If the hypothesis of this model were confirmed, it would support the contention that occupational
segregation may not necessarily be negative for women. For a discussion of this point see Bettio
(2001).
2
3
The INPS Panel is a Longitudinal Sample of Workers and Employers based on the data of the
National Social Security Institute (INPS) collected for institutional purposes on employers and
employees. The Panel is based on a sample of employees born on the 10th of March, June,
September and December of each year from 1985 to 1996. The company’s longitudinal records are
associated to each worker in the sample. Each yearly sample includes approximately 100,000
employees and their firms.
4
The INPS codes for occupation classification are extremely broad: white-collar or blue-collar,
apprentice, part-time or full-time and fixed-term training contract (Contratti di formazione e lavoro).
5
An attempt was made to make use of other information contained in the Panel, specifically
concerning periods of temporary leaves during employment and types of contract. Temporary leaves
are generally due to temporary unemployment benefit (CIG) and maternity benefit. Such periods
mark a discontinuation of work that may be a source of wage differences. In the preliminary analysis
the variable did not prove to be significant (as was to be expected, given the age range) and was not
considered anymore. Types of contract are national/company versus individual/local. 90% of
contracts were national, however individual contracts were found to have a negative, though not
greatly significant effect for both genders, especially for women, so the variable was not included in
the final model.
6
An assessment was also made of the possible effects of other information about companies, such as
how long companies were active in the period 1986-96 and employment growth trends, but none of
this information proved significant.
7
The workers in the dataset used for this study do not include those employed in agriculture and
public administration, seasonal workers (here defined as having a work experience of three month
or less) and workers to whose firms it was not possible to attribute a proportion of females.
8
It should be pointed out that because of the particular construction of the INPS Panel (Contini,
2002), derived essentially from a population of workers, the dataset contains a very high proportion
of firms represented by a single worker. Empirical analysis conducted on the sole basis of firms
with at least two 20-25 year-old workers in the dataset and estimates run on the entire dataset
breaking down companies by size confirm the company effect results obtained from the totality of
20-25 year-old workers in the 1996 archive.
9
Daily wages, obtained by dividing total paid wages by the number of declared working days, may
be subject to the objections raised by Ginzburg et al. (1999). See Capellari and Chies (2003).
10
The classes are: low proportion of females (up to 40%), medium proportion (40-70%), and high
proportion (more than 70%).
11
As implemented in the MIXED procedure of SAS.
12
Table A.4 shows the final version obtained after the various tests carried out during the research
work which took account of interactions between occupational rank and certain individual variables
(education, experience) which proved to be non-significant.
13
The positive impact on pay found here does not appear to contradict results obtained in other
studies on groups of workers with more continuous working experience. Indeed, if we concentrate
exclusively on the stayers in our sample, the impact is negative.
14
This is because the difference in individual characteristics is determined by choices that may be
the result of pre-market discrimination, and the discriminatory component also includes the effects
of possible omitted variables of preferences and productivity which may be correlated to variables
included in the analysis.
15
Pooled wage regression coefficients (instead of the male coefficient alone) have often been used
as benchmarks for the evaluation of discrimination: in this case males may be overpaid and females
underpaid. Attention has also been devoted to the identification problem that arises when
categorical variables are used (Oaxaca and Ramson, 1999) and the effect of wage structure changes,
especially in the studies on the evolution over time of the wage gap in the wake of the paper
published by Juhn, Murphy and Pierce (1991).
16
Neuman and Oaxaca (1998) show that decomposition is sensitive to how the selection bias term is
interpreted. The problem is that it should be further decomposed into the two components of human
capital and discrimination.
References
Addis E., Waldmann R. (1996), “La struttura salariale e I differenziali per sesso in Italia”,
Economia & Lavoro, Vol. XXX, n. 1-2, 87-103.
Arrow, K. J. (1985), “The Theory of Discrimination”, Collected Papers of Kenneth J. Arrow,
Applied Economics, 6, 143-64.
Becker G. S. (1971), The Economics of Discrimination, Chicago University Press, Chicago.
Bettio F., Villa P. (1996), Trends and Prospects for Women’s Employment in the 90s: Italy, EC
Network on the Situation of Women in the Labour Market, UMIST, University of
Manchester, Manchester.
Bettio F. (2002), “The Pros and Cons of Occupational Gender Segregation in Europe”, Canadian
Public Policy, Supplement n. 1, s65-s84.
Blau, F. D., Kahn, L. M. (1997), “Swimming Upstream: Trends in the Gender Wage Differential in
1980s”, Journal of Labor Economics, Part 1,15, 1, 1-42.
Borgarello A., Devicienti F. (2002), “Tendenze nella distribuzione dei salari”, in Contini (a cura di),
Osservatorio sulla mobilità del lavoro in Italia, Il Mulino, Bologna.
Bryk A.S., Raudenbush S.W. (1992), Hierarchical Linear Models, Applications and Data Analysis
Methods, Sage Publications, Newbury Park, CA.
Capellari S., Chies L. (2003), “Il Panel INPS per l’analisi dei differenziali salariali di genere:
problemi e potenzialità”, DISES WP n. 100, Trieste.
Contini et al. (2002), “Safari nella giungla dei salari”, Laboratorio Revelli, WPS n. 3.
Contini B. (2002), Osservatorio sulla mobilità del lavoro in Italia, Il Mulino, Bologna.
Favaro D., Magrini S. (2002), Caratteristiche della discriminazione di genere, mimeo.
Favaro D. (2003), “Discriminazione di genere e segregazione occupazionale”, in Lucifora C. (a cura
di) Mercato occupazione e salari: la ricerca sul lavoro in Italia, Ed. Mondadori
Università, Perugia.
Flabbi L. (1997), “Discriminazione di Genere e Rendimenti dell'Istruzione: un'analisi su microdati
individuali”, Rivista di Politica Economica, 12, 187-234.
Flabbi L. (2001), “La Discriminazione: evidenza empirica e teoria economica”, in: Brucchi Luchino
(ed.) Manuale di Economia del Lavoro, Bologna: Il Mulino.
Flabbi L. (2003), “Gender Discrimination Estimation in a Search Model with Matching and
Bargaining”, Job Market Paper.
Gavosto A., Rossi F. (1999), “Giornate retribuite e differenziali salariali nei dati INPS.” Replica
all’articolo di A. Ginzburg et al.”Un nuovo autunno caldo nel Mezzogiorno? Note in
Margine al dibattito sui differenziali salariali territoriali”, Politica economica, XV, n 2.
Ginzburg A., Scaltriti M., Solinas, Zoboli R. (1998), “Un nuovo autunno caldo nel Mezzogiorno?
Note in margine al dibattito sui differenziali salariali territoriali”, Politica Economica,
XIV, n. 3, 377-410.
Goldstein H. (1995), Multilevel Statistical Models, 2nd Ed, Edward Arnold, London.
Groshen, E. L (1991), “The Structure of the Female/Male Wage Differential: Is It Who you are,
What You do, or Where You Work?”, Journal of Human Resources, 26, 3, 457-472.
Gubta, N.D, Rothstein, D. S., 2001, The Impact of Worker and Establishment-level Characteristics
on Male-Female Wage Differentials: Evidence from Danish Matched Employee-Employer
Data, CLS Working Papers Aarhus School of Business, Centre for Labour Market and
Social Research.
ISFOL (1998), Il lavoro in Italia: profili, percorsi politiche, Franco Angeli, Milano
ITER (2001), I differenziali salariali di genere, Ministero del Lavoro.
Juhn, C., Murphy K. M., Pierce, B. A. (1993), “Wage Inequality and the Rise in Returns to Skill”,
Journal of Political Economy, 101, 410-442.
Lucifora C., Reilly B. (1990), “Wage Discrimination and Female Occupational Intensity”, Labour,
Vol. 4 (2), 147-168.
Oaxaca, Ronald L. (1973), “Male-Female Wage Differentials in Urban Labor Markets”,
International Economic Review, 9, 693-709.
Oaxaca R., Neuman S. (1998), “Estimating Labor Market Discrimination with Selectivity Corrected
Wage Equations: Methodological Considerations and An Illustration From Israel”,
Discussion Paper No. 1915, Centre for Economic Policy Research.
OECD (2002), Employment Outlook, Paris.
Reilly K. T., Wirjanto T. S. (1999), “Does More Mean Less? The male/female wage gap and the
proportion of females at the establishment level”, The Canadian Journal of Economics,
vol.32, 4, 906-929.
Snijders T. A. B., Bosker R. J. (1999), Introduction to Basic and Advanced Multilevel Modelling,
Sage, London.