LABOUR MARKET STRUCTURE AND DETERMINANTS OF EARNINGS...
Transcript of LABOUR MARKET STRUCTURE AND DETERMINANTS OF EARNINGS...
Chapter 6
LABOUR MARKET STRUCTURE AND DETERMINANTS OF EARNINGS
6.1 The Proposition
The discussions in Chapter 2 (Sections 2.1 and 2.2) reveal that the segmented labour
market theory focuses mainly on the socio-institutional factors influencing job access
patterns and wages. According to this theory, given a structured labour market, the
mechanisms which determine wages are quite different in primary and secondary
segments. In the primary segment, wages are determined within the structured internal
labour markets by rules, customs and procedures are unresponsive to economic factors, and
wages are attached to jobs rather than personal productivity or ability of the individual
worker. In the secondary segment, on the other hand, wages are detennined by aggregate
market supply of and demand for a particular category of workers. In this segment,
variations in earnings are not caused due to variations in productivity-related
characteristics across workers, because employers at the time of recruitment, do not give
weightage to the productivity-related characteristics such as level of education of the
secondary segment worker.
So far as the role of education is concerned, the segmentation theorists argue that
education is important at the 'port of entry' into the primary segment of the labour market
(Reich, et.al., 1973; Carnoy and Rumberger, 1976). In fact, employers use education as a
'screening device' while making recruitments to primary segment jobs. Since wages are
attached to jobs rather than to individuals, job entry becomes crucial in determining one's
wages in the primary segment of the labour market. However, the educational level of an
individual is not important in gaining access to secondary segment jobs. Apart from this,
the theory argues that, given a segmented labour market, the socio-economic background
of the worker, to a large extent, influences his/her level of earnings. Since there is a
correlation between socio-economic backgrounds and educational levels of employees
(Bowles, 1972), there can be a correlation between education and earnings, particularly in
the primary segment of the labour market. To be specific, the segmentation theorists claim
that the returns to human capital such as education, experience and training in the primary
and secondary segments are different, and in the secondary segment, there is little or no
returns to schooling (Dickens and Lang, 1985 ). It has also been found in many empirical
studies that, if the labour market is segmented, the determinants of earnings and the size of
estimated coefficients (of common explanatory variables) in the secondary segment
earnings equation are significantly different from those in the primary segment earnings
equation (see Section 2.3, Chapter 2).
Given the theoretical propositions of the segmented labour market theory, most of the
empirical studies attempting to test for existence of segmented labour markets use
multivariate analysis to examine the determinants of earnings in both primary and
secondary labour market segments (see Section 2.3, Chapter 2). Specifically, these studies
use Ordinary Least Squares method of estimating separate earnings functions for primary
and secondary segments. Moreover, in most of such studies modified Mincerian type
human capital earnings functions are estimated separately for each segment of the labour
market. In the present study, we have used the above mentioned methodology for
estimating earnings functions separately for each labour market segment. Specifically, in
this chapter the attempt is to : (i) identify the determinants of earnings in the whole sample
and in each labour market segment; and (ii) examine the significance of education as an
explanatory variable in the earnings equations for both primary and secondary segments.
6.2 The Model
It has already been mentioned in Chapter 2 (Section 2.3) that almost all empiricists use
the 'human capital model' to estimate earnings functions in different labour market
segments. Their specification of earnings function is based on the Mincerian earnings
model (1962), though their actual earnings functions are the modified version of the
Mincerian type. This modified version which is often termed as the exploratory approach.
besides considering basic human capital variables such as years of schooling and years of
196
I I
labour market experience, includes any such variable that may be expected to influence
earnings (Andrisani, 1973; Blaug, 1974; Osterman, 1975; Rosenberg, 1975; Camoy and
Rumberger, 1976; Camoy, Girling and Rumberger 1976; Psacharapoulos, 1977; Mac
Nabb, 1987; Liu, 1975; Velloso, 1975; Toledo, 1979; Lobo, 1977; Uthoff, 1986; Singh,
1984; Baneljee and Knight, 1985; Snooks, 1983; Gasper, 1995; Ammugan and
Nagarajan, 1994). According to the exploratory approach, variables relating to worker's
personal characteristics, socio-economic background, occupation and labour market
conditions, and geographical conditions are considered in the earnings functions.
Moreover, almost all the empirical studies use the ordinary least squares (OLS) method of
estimating earnings functions for different labour market segments.
The general model of earnings function used in this study is as follo\vs :
Yi = / (HP, QS, FB, SB, OL, GL), where,
Yi refers to the monthly total earnings of a worker from the cunent job held;
HP refers to the vector of human capital variables other than QS and other personal characteristics of the individual worker;
QS refers to the vector of variables relating to quality of schooling of the individual worker;
FB refers to the vector of variables related to the family background of the individual worker;
SB refers to vector of variables representing the social background of the individual worker;
OL refers to the vector of variables pertaining to occupation and the labour market; and
GL refers to variables explaining the geographical condition of residence of the worker.
197
Given the general earnings function, our specific model in this study is the ordinary
least squares with continuous regressors. The specific model is :
, Yj = b0 + ~1bj Xjj + Uj,
where, j stands for an individual, i for a variable, Yi is the continuous dependent variable,
Xj (i=I...n) are n independent variables, continuous or dummy, b0 and bj are
parameters to be estimated, and Uj are random, unobserved disturbances with zero mean
and constant- i.e. unknown variance.
In this study, the total sample size is 413, and we have divided these san1ple workers
into primary independent, primary subordinate and secondary segments on the basis of
'years of schooling' of the sample workers and 'protection and autonomy' of jobs. The
number of the primary segment workers, which is the sum of the workers in the primary
independent and primary subordinate segments, is 116, and the secondary segment has 297
workers.
Given the sample size and the size of individual labour market segments, we have tried
to estimate separate earnings functions for each labour market segment and for the entire
sample. Accordingly, we have estimated earnings functions like the one mentioned above
for each group of workers. Therefore, the earnings function which has been estimated for
the entire sample workers is :
H n In Y· = b + L bW X· + u·
I 0 j =l' IJ J w
where, In Yi is the log of total monthly earnings before tax on the current job of the
sample worker.
The earnings function which has been estimated for the secondary segment workers is :
s I;
ln Yi = b0 + t: M XiJ. + UJ·· t: 1 I
s where, In Yi is the log of total monthly earnings before tax on the current job of the
secondary segment worker.
198
Similarly, the earnings function which has been estimated for the primary segment
workers is:
P n In Y· = b + '\' bfx·· + U· 1 0 f;'l I lJ J'
p where, In Yi is the log of total monthly earnings before tax on the current job of the
primary segment worker.
In our mode!, we have specified the earnings functions in the semi-logarithmic fom1,
that is, the logarithm of earnings is regressed on the potential explanatory variables.
Because, this has some definite advantages. Often, income distributions are found to be
approximately log-normally distributed. Therefore, the semi-logarithmic form of earnings
function provides a better fit than other functional forms involving the same explanatory
variables. Moreover, the human capital theoretic reasoning itself argues for semi
logarithmic form since the investment cost of schooling and post-schooling are treated in
time- equivalent values. Moreover, use of semi-logarithmic form enables in interpreting
the regression coefficients as the percentage effect of a unit change in the explanatory
variables on earnings (Mincer, 1974; Becker and Chiswick, 1966; Snooks, 1983; Fields,
1980; Blaug, 1987).
6.2.1. Definition of Variables
We have used several relevant variables in our model, and these variables are measured
in various ways. The point to be noted here is that the number and the type of variables
used in different regression equations are different. However, all or most of the basic
independent variables given in Table 6.1 have been chosen to estimate alternative earnings
equations separately for individual labour market segments, and for the entire sample
population of workers.
199
Table 6.1. : Definitions of Variables
Sl. No. Variable Name
1. Age (E2)
2. sex (E3 l
3. Caste (ES)*
4. Religion (E6)*
5. Marital status (EB)
6. Family size (E10)
7. Place of origin (E15)
8. Parents' education (PAEDU)
9. Father's occupation (FAOCC)*
200
Value Labels/Description
measured in years
0 male 1 female
1
2
scheduled caste backward caste
3 higher caste (Dummy variable::; C1 ~ 1 if
backward caste, 0 if otherwise; C2 = 1 if higher caste, 0 if otherwise; excluded category is Scheduled Caste)
1 Hindu 2 Muslim 3 Christic;.n
(Dummy variables : R1 = 1 if Hindu, 0 if otherwise; R2 = 1 if Muslim, 0 if otherwise;
0 1
excluded Christian)
unmarried married
category
measured in total numbe-c of members in the family, including the respondent
0 rural 1 urban
measured in years of schooling of both father and mother of worker.
measured in seven categories:' 0 unclassified workers 1 unskilled manual worker 2 cultivator 3 petty shopkeeper 4 skilled worker 5 clerical & related worker
is
6 administrative & professional (Dummy variables F1 = 1 if
cultivator, 0 if otherwise; F2 1 if manual worker, 0
10 Average annual family income (E21)
11. Place of schooling (E22)
12. Educational achievement (E25A) of the worker
13. Education squared
14. Vocational educational (E26) achievement
15. Vocational education squared (VEDNQ)
16. Mode of entry to Ist job (FJMODE)
17. First job of the worker (E29)*
201
if otherwise; F3 1 if petty shopkeeper, 0 if otherwise; F4 = 1 if skilled worker, 0 if otherwise; FS = 1 if clerical and related worker, 0 if otherwise; F6 = 1 if administrative or professional worker, 0 if otherwise; excluded category is unclassified workers)
measured in thousands of rupees and includes income of all the earning members in the family, except that of the individual worker
0 no schooling 1 rural 2 urban
measured in years ·of schooling
square of E25A
measured in years of vocational schooling
square of E26
0
1
informal channel of entry or selection formal channel of entry or selection
measured in seven categories & ranked on the basis of mean earnings of the last month of the 1st job
1 unskilled manual 2 skilled 3 clerical & related 4 sales 5 supervisors 6 technical & Professional 7 administrative &
managerial Jl = 1 (Dummy variables
unskilled manual if otherwise; J2
worker, 1
skilled worker, 0 otherwise; J3 1
if 0
if i.f if
18. Average number of hours worked daily (E54)
19. On-the-job training facility on the current job (E57)
20. Job changes or relative stability of employment (V4)
21. Labour market structure (V6)
22. Labour market experience (LMEX)
23. Labour market experience squared (LMEXQ}
24. Farm size (Pll}
25. Education-experience interaction (EDEXP}
clerical and related worker, 0 if otherwise; J4 1 if supervisor, 0 if otherwise; JS 1 if technical or professional worker, 0 if otherwise; J6 1 if sales worker, 0 if otherwise; excluded category is proprietors)
measures in number of hours worked daily
0 no 1 yes
1 first job is current job 2 changed job twice 3 changed job thrice 4 changed job 4 times 5 changed job 5 times 6 changed job 6 times
0 secondary segment 1 primary segment (i.e. primary independent segment plus primary subordinate segment)
measured in actual years of labour market experience in all the jobs held by the individuals worker
Square of LMEX
measured in terms of total annual turnover of the firm in rupees (in lakhs)
years of education multiplied by years of labour market experience
* Later in the multiple regression analysis we have used dummy varia.bles (i.e. derived regressors} instead of these basic variables.
The dependent variable 'In Y' is the log of total monthly earnings before ta'\ on the
current job held by the worker. It also includes other types of payments such as overtime
202
wages, incentive bonus etc. The variables considered in different regressions can broadly
be grouped into the following categories :
Human Capital Variables : This category of variables includes investment in the form of schoo_ling and post- schooling investment in the form of labour market experience. It also includes investment in on-the-job training of the worker, either by the worker himself or by the firm. In order to account for the parabolic effect of schooling on the earnings profiles of the worker over the years, we have considered 'education squared' as an independent variable in the earnings equations. Moreover, the post-schooling investment component ofhuman capital is measured in tenns of years of experience in the labour market In our sample, in many cases, employment is not continuous and there are periods of job search and unemployment. Therefore, instead of following the Mincerian way of calculation of experience in the labour market, we have considered in the model the actual years of experience in the labour market as indicated by the respondents. Besides, as age is proxy for labour market experience of the worker, we have preferred actual years of labour market experience as an independent variable in the model. In fact, it is better to discard age of the worker in favour of ·actual years of expcrien::e in the labour market', as age itself is negatively correlated with length of schooling (Mincer, 1972; Blaug, 1987). Besides, in our model, the parabolic effect of declining 'experience-earnings' after a given years of labour market experience is observed by treating experience in quadratic form (i.e. experience plus experience squared) (Mincer, 1974).
Other Personal Characteristics and Social Background Variables : This category of independent variables includes sex, caste, religion and marital status of the worker. These, variables do influence the earnings structure of workers in the labour market. Therefore we have considered these variables in our model. Besides, 'marital status' is a good indicator of the stable behaviour of the \Vorker, because married worked are considered to be responsible and stable workers, at least by the employers, and therefore marital status exerts considerable effect on their earnings.
Family Backgrotmd Variables : It has been observed in empirical literature that family background greatly influences one's educational attainment and occupational attainment, and ultimately one's earnings level (Behrman, et.al, 1980; Leibowits, 1974; Perl, 1973; Fishlow, 1972; Mincer & Polacheck, 1974). This category of independent variables includes family size, parents education measured in terms of years of schooling of both father and mother (i.e. mother's schooling plus father's schooling), fathers occupation and average annual family income of the worker, excluding the income of the worker.
Occupation and Labour Market Related Variables : It is a \Vel! known fact that the mode of entry to the first job and the first job itself are important detem1inants
203
of the current job of the worker, and ultimately his/her earnings level. Therefore, we have considered 'mode of entry to the first job' and 'the first job' of the respondent as independent variables in the model. Besides, to capture the effects of structural factors in the labour market on one's earnings level we have included -labour market structure (i.e. segment of employment), which captures the effects of a whole lot ofjob characteristics. So far as the firm level variables are concerned, we have included size of the firm measured in average annual turnover of the fitm (in lakhs of rupees) as an mdependent variable in various regression ec;uations.
Quality of Schooling Variables : In our sample, all the educated workers have their schooling from government schools. However, we have used ·location of schools' (rural or urban schools) as an independent variable in the analysis.
Geograp!Iical Location Related Variables : The place of one's origin - i.e. rural or urban - affects one's accessibility to an environment of schooling and job market. Such an accessibility determines one's capacity to face the challenges in education and employment, which ultimately determines one's earnings potentiai (Birdsall & Behrman, 1984). Therefore, to capture the dfect of one's place of origin on one's earnings, place of origin of the worker is included as an indepcnuent variable in the analysis.
6.3 Empirical Analysis
The present study is based on a total sample of 413 workers of 11 manufacturing fim1s
in Delhi, and after grouping these workers, on the ba<>is of 'years of schooling' of workers
and 'protection and autonomy of employment' we have derived three distinct labour
market segments - i.e. primary independent, primary subordinate and secondary. After
dividing the sample workers, it was found that only 16 workers fail into the primary
independent segment, 100 workers into the primary subordinate segment, and 297 workers
into the secondary segment of the labour market. Given the small size of the primary
independent segment, we have clubbed this segment with the primary subordimte sep'1ent
for the purpose of empirical analysis, which raises the number of workers falling into the
primary segment of the labour market to 116. Then we estimated separate ean1ings
functions for each labour market segment and for the entire sample population of workers
in an attempt to find out the best statistical explanation of variations in earnings between
individuals.
204
In each case, we have estimated the equation by a stepwise regression procedure.
Perhaps, it is pertinent here to mention that to estimate the regression equations we have
used the Statistical Package for Social Sciences (SPSS) computer programme. While
estimating the earniqgs equation by a stepwise regression procedure, the computer
programme enters an independent variable in each step on the basis of highest partial
correlation coefficient, and it accepts or rejects an independent variable on the basis of a
partial F-test with 95 percent confidence limit. However, before reaching at the final
regression equations for each labour market segment and for the sample population of
workers, we have checked for high correlation among independent variables and selected
the explanatory variables of each equation carefully. Before running the regressions, we
have calculated the arithmetic mean and standard deviation of all independent variables
and the simple Pearson's correlation coefficients between all pairs of variables, including
the dependent variable. However, some of the variables used in our correlation analysis
are categorical. In this case, one can also calculate the correlation matrix using polychoric
or polyserial correlation method. Table 6.1 shows the list of independent variables which
we used in different alternative regressions. However, some regressions do not include all
the independent variables listed in Table 6.1. Depending on the nature of the individual
labour market segment, we have dropped a few independent variables in various specific
regressions. For example, in the equation for the secondary segment we have not included
variables such as 'on-the-job training' 'years of vocational schooling' as none of the
workers in this segment has on-the-job training or vocational schooling. For each
individual labour market segment and for the entire sample population of workers we have
estimated several alternative regressions, and finally reached at 3 regre~sion equations -
one each for the primary labour market segment, secondary labour market segment, and the
entire sample population - which provide best statistical explanations of the variations in
earnings. We have estimated several alternative regressions before finally selecting the
three equations because of the fact that many of the independent variables considered in
different regression equations are highly correlated with each other. Running of alternative
regressions helped to differentiate and establish the actual contribution of the independent
variables to earnings variations among individuals. All the regression models attempted
205
here are highly significant, and confirm to the partial F-test with 95 percent confidence
limit. In the following sections, the attempt is to examine the determinants of earnings in
different labour market segments and in the entire sample population of workers in the
manufacturing sector i~ Delhi.
6.3.1 Determinants of Earnings in the Secondary Segment
Initially, in an attempt to explain total earnings of a worker in the secondary segment,
we ran several alternative regressions which included 16 original variables and 18 dummy
variables. As has already been mentioned in Table 6.1 we have used dummy variables for
'caste', 'religion' 'father's occupation' and 'first job' ofthe worker. This is to be noted here
that while using dummy variables in multiple regression we always get rid of at least one
subcategory of the original variables, and the effect of this subcategory i:o, then c~ptured in
the constant term of the regression equation. In other words, while using dummy variables,
we must drop one sub-group to avoid 'singularity' of the moment-matrix of independent
variables (Blaug, 1987). It is to mention here that we have chosen the method of not
constraining the constant term of the regression equation to zero. Since we have not
constrained the constant term to equal zero, the regression coefficients associated with
dummy variables must be interpreted as giving the percentage difference in earnings of an
individual belonging to a particular category rather thar1 to the excluded category, after
holding all other variables constant. However, our final regression equation for the
secondary segment included 7 independent variables : (i) labour market experience; (ii)
labour market experience squared; (iii) religion; (iv) marital status; (v) parents' education;
(vi) average annual family income; and (vii) first job ofthe worker.
The correlation analysis for the secondary segment reveals that 'age of the \vo:~.;:er' is
highly correlated with log earnings (see Table 6.2). Age is also highly and significantly
correlated with total labour market experience of the worker, which in tum is highly
correlated with log earnings. Given the high and significant correlation coefficient
between 'age' and 'labour market experience', i.e .. 5647 we have dropped age in favour of
206
Table 6.2 : Correlation matrix for secondary segment
LOG EAR E2 AGEQ E3 E5 E6 E8 EJO El5 PAEDU FAOCC E21
LOG EAR 1.0000
E2 AGEQ .3329 .. .3172**
1.0000 .9883 .. 1.0000
EJ -.0203 -.0103 -.0164 1.0000
E5 .0291 .0578 .0511 .1164
1.0000
E6 E8 .3955 .. .3070•.
-.0652 .5404** -.0624 .4925•• -.0688 .1233 -.0923 .0371 1.0000 -.0071
1.0000
207
E10 E15 PAEDU FAOCC E2l .1812* .0520 .2377•• .2749 .. .23JO• • .1265 -.0439 -1046 -.0110 -.0173 .0946 -.04R6 -.1119 -.0168 -.0280
-.0972 .6293*' 3029•• .305J.·· .3004'" -.0869 .1299 .1278 .1676• .04-0 .1084 -.0174 .1343 .1851• .0033
-.2163•• .1494• -.1442 .0465 -.0247 1.0000 -.1369 .0370 .0436 .231J9H
1.0000 .3033•• .. 3543** .21 os· • 1.0000 ,6031 .. .4588·.
1.0000 .46J2H 1.0000
Table 6.2 (Contd.)
E22 E25A EDUQ FJMODE E29 E54 LMEX LMEXQ PI\ --·----------LOG EAR -.0130 -.0155 .0110 .0573 .4580** -.1838* .479JH .4049.0 -.0689 E2 -.0936 -.0862 -.0684 -.0041 -.0606 .0838 .5647*• .5205** .0982 AGEQ -.1084 -1138 -.0949 -.0119 -.0504 .0634 .5713*" .55J()U .0851 E3 .4912** .1419* .1491 -.0169 -1124 -.2242'* -.0544 ·.05.3,1 -.0890 E5 .1379 .2019** .2\\6H .0538 ·.0910 . 0262 -.onn -.0776 -.1016 E6 -.0274 -.1075 -.11 I 3 .1090 .3929 .. -.1283 ()()(,'.) 0083 -.1480 E8 .1659• .0944 .0666 .0495 -.0782 -.0361 ·.4122*. -.2832 .. .0617
E10 -.0980 .0550 . 0671 -.0410 .0199 .0420 .1255 .0624 -.0481 E15 . 7431** .0941 .0743 .1266 -0063 -.1867* -.010:i .0140 -.0182 PAEDU .3040** .3294** .3360*. .2367+* .I.'S 16* -.2161** - 05) 1 -.0535 -1722•
FAOCC . 3584 .. . ?857 .. .2fi21U .0675 . 1940* • -.2332 .. .0194 .0161 -.1544*
E21 .1734* .1837~ .1905"' .0206 .1187 -.1534• .0680 .0·106 .0492
E22 1.0000 .4533 .. .2.9J3H. .1044 .f)174 .1357 -.0934 -.0959 -0205
E25A 1.0000 .9542* 1 .0753 -.0004 -.1477 -.li\3 ·. i 193 -.0842
EDUQ 1.0000 .0775 .0014 -.1582* -.0865 -.0854 -0857
FJMODE 1.0000 .I \JR .0244 -0239 -.0292 -.0237
E29 !.0000 -.3634•. .1 :1.11• 1463 -.216)H
E54 1.0000 -1)3!• - 1563• .4140**
LMEX .1.0000 .9410** -.0449
LMEXQ 1.0000 ·.0284
P11 1.0000 ----·-·---------··----------------------·-----------~---·--·-·---------·----·---·-···-------·------------------·--
Number d C<l:><:s : 297 2-tailcd Signif: • -- .01 ** -- .00 I
::'08
labour market experience in the regression equation for the secondary segment. Moreover,
age is a proxy fOi work experience, and if data on actual ye.ars of work experience are
available, it is better to use work experience than age of the worker in the earnings function
(Mincer, 1972; Blaug, i 987). Similarly, we have dropped ·age squared' in favour of
'labour market experience squared' in the equation. 'Marital status' of the worker is
positively and significantly correlated with log earnings, and it means that if the worker is
married, his/her earnings is higher than the worker who is unmarried.
It can be seen in Table 6.2 that the correlation coefficients between ·parent::' <:ducation'
and fathers' occupation', parents' education and 'average annual family income', and
'fathers' occupation' and 'average annuai family income' are high and significar1t. This
implies that if the worker has less educated parents, his/ber father is into low paid
occupation, and in tum his/her average family income is also less. Similarly, a worker
whose average family income is high has better educated parents and his/her father is also
into a relatively better paid occupation. The degree of correlation between 'j;,thers'
occupation' and 'parents' education' is .6031, which is also significant. Therefore, in our
final equation for the secondary segment 'parents' education' captures the effects of
'fathers' occupation' on earnings, and therefore 'fathers' occupation' does not figure in the
final equation. Moreover, given the positive and significantly high degree of conelation
between 'place of origin' of the worker and his/ber 'average a~mual family income', the
latter variable captures the effect of the 'place of origin' on earnings. It also implies that
workers having rural origin belong to relatively poor families in terms of average <L"L'l.ual
family income.
Table 6.3: Mean & standard deviation of earnings schooling and experience for the secondary segment
Statistics
Mean Standard deviation No. of cases
Log earnings
7.20 .41
297
Years of schooling
209
7.38 2.86
297
Years of labour market experience
4.77 3.38
297
In the secondary segment, the 'place of origin' of the worker and his/her 'place of
schooling' is highly and significantly correlated with each other. This simply means that
workers having rural origin have their education from rural government schools, and those
having urban origin ~ave their education from urban government schools. Besides, the
correlation coefficient between the 'place of schooling' and 'years of schooling' of the
worker is .4533 which is also significant thereby implying that workers in the secondary
segment who have their education from rural schools have less years of schooling than that
of workers schooled in urban government schools. Therefore, when years of schooling of
the worker is included as an independent variable in the equation, 'place of schooling' is
dropped from the equation on the basis of partial F-test. However, we will find latter that
the 'years of schooling' of the worker is also not included in the fmal equation for the
secondary segment, as it is not significantly correlated with log earnings. Same is ilie case
with ·education squared' as an independent variable in the equation.
In our regression equation for the secondary segment, we have included the ·first job'
of the worker as a regressor, because it is positively and significantly correlated with log
earnings and the explanatory power of this variable is also significant. It can be seen in
Table 6.2 that the 'first job' of the worker is negatively and significantly correlated ·with
'average number of hours worked daily' by the worker and the 'firm size'. This means that
workers in the low paid secondary segment jobs have to work for longer hours daily, and
they are mostly concentrated in relatively small firms, measured in terms of average annual
turnover of the firm. So the 'first job' of the worker as an independent variable in the
regression equation for the secondary segment captures the effects of the other two
correlated variables on earnings.
210
Table 6.4: Correlation matrix for final equation for secondary segment
LOG EAR LMEX LMEXO PAEDU E8 E21 Rl Ji
LOG EAR 1.000 .4793** .4049** .2377** .3070** .2330** -.3955"* -.4580 .. LMEX 1.0000 .9410 .. -.0531 .4122** .0680 -.0069 -.1531* LMEXQ 1.0000 -.0535 .2832""* .0406 -.0083 -.1463 PAEDU 1.0000 -.1442 .4588"'* -.1343 -.1516• E8 1.0000 -.0247 -.0071 -.0782 E21 1.0000 -.0033 -.1187
Rl 1.0000 .3929*•
J 1 1.0000 _____ , _____
Number of cases : 297 2 - tailed significance : * - .01 * * - .001
211
Table 6.5 Regression results for the secondary segment (Dependent variable : log earnings)
Vairable 8-Co-efficient Standard Error p - coefficient T- Value Significance of of 8-coefficient T-Value
Labour market .78168 .016305 .651242 4.794 .0000 experience
Labour market -.001947 .907824E-04 -.276137 -2.145 .0328 experience squared
Religion Hindu -.369436 .600168 -.274846 -6.140 .0000 Marital status .113702 .045546 .120298 2.496 .0131 Parents' education .014717 .004516 .153719 3.259 .0013 Average annual 2.637586E-06 1.19916E-06 .102341 2.200 .0286
family income First job
Unskilled manual -.209691 .038767 -.245828 -5.409 .0000
Constant 7.210028 Multiple R .72038 R-Square .51895 Adjusted R-square .50730 Standard Error .28480
F "' 44.53858 Regression 7 Residual 2.89 No. of cases 297
2l2
J
The final earnings equation for the secondary segment of the labour market includes
five original variables and two derived regressors or dummy variables. Thus, the final
equation for the secondary labour market segment which best fits the data is as follows (see
Table 6.5):
5 In Yi = 7.210028 + .078168 (labour market experience) - .369436 (religion
Hindu) - .209691 (first job : unskilled manual worker) + .014717 (parents' education) + .113702 (marital status) + 2.637586E-06 (average annual family income) - .001947 (labour market experience squared)
The above earnings equation explains nearly 52 (i.e. R2 = .51895) percent of earnings
variance in the secondary segment of the labour market, and all the B-coefficients are
significant at 5 percent level, and even some coefficients are significant at one percent
level. Among all independent variables in the regression equation for the secondary
segment, the explanatory power of the 'labour market experience' is the highest - i.e. it
explains nearly 23 per cent of earnings variance in the secondary segment (R2 =.22976).
·Labour market experience' of the worker also enters at a very early stage in a stepwise
regression procedure. 'Labour market experience' of the worker and 'religion' of the
worker, which has been divided into two dummy variables, together explain nearly 38
percent of earnings differentials in the secondary segment. In this case, R2 is .38355, and
R2 change is .15379, which implies that religion of the worker alone explains nearly 15
percent variance in earnings. When ·first job' of the worker, which is again divided into 6
dummy variables is brought in the equation, it along with earlier two variables explain
about 45 percent of variance in earnings (R2 = .44834). Here, R2 change is .6479. Thus,
the explanatory power of' first job' of the worker is relatively low in the equation. Then, in
the next step, inclusion of 'parents' education' as an explanatory variable in the equation
raises the value ofR2 marginally from .44834 to .48204. This variable together with earlier
three variables explain about 48 percent of earnings variations in the secondary segment.
The contribution of 'parents' education' alone in explaining the earnings variance in the
secondary segment is about 3.4 percent. 'Marital status' ofthe worker along \vith the earlier
four explanatory variables explain nearly 50 percent of earnings differentials in the
secondary segment (R2 = .50195). The individual contribution of 'marital status' as an
213
independent variable in explaining total earnings in this segment is only about 2 percent.
When the 'average annual family income' of the worker enters the equation, nearly 51
percent of earnings variance is explained (R2=.51 129). Finally, 'labour market experience
squared' when brough~ in the equation raises the value of R2 to .51895, and the earnings
equation for the secondary segment explains nearly 52 percent of earnings differentials.
However, it can be seen in Table 6.5 that the B-coefficient of 'labour market experienced
squared' is negative which captures the declining experience-earnings profiles after a
certain period of work experience of the secondary segment worker. Thus, we find that
total earnings in the secondary segment can be explained in terms of : (I) labour market
experience; (2) religion; (3) first job; ( 4) parents education; (5) average annual family
i~come; (6) marital status; and (7) labour market experience squared. It should be noted
here that in the secondary segment 'years of schooling' of the worker has no market
premium, though work experience has.
Before we move on to discuss our earnings equation for the primary segment of the
labour market, it is perhaps important here to mention that multicollinearity among the
independent variables is a problem in any regression equation. Since it is a property of the
sample data and not of the population, one cannot strictly speaking, test for its existence
(Blaug, 1987). But the correlation matrix of our best regression discussed above does not
show that multicollinearity affects our results (see Table 6.4). Multicollirearity manifests
in high R-square coupled with low t-values of almost all the coefficients. That is not the
case here.
6.3.2 Determinants of Earnings in the Primary Segment
In the primary segment of the labour market, we have explained total earnings in terms
of: (1) father's occupation; (2) labour market experience; (3) labour market experience
squared; (4) on-the-job training; (5) sex; (6) years of schooling squared; and (7) years of
vocational schooling squared. Initially, we regressed log earnings on 20 original and 16
dummy variables. We ran several alternative regressions to decide about the best
explanatory variables in the earnings equation for the primary segment. The final
214
l
regression equation which best fits the primary segment data includes six original variables
and one dummy variable - i.e. 'administrative or professional occupation' of the father of
the worker. Our final regression for the primary segment is significant, and the B
coefficient of all the explanatory variables in the final regression are also highly
significant, with 95 percent confidence limit. We have also checked for the problem of
multicollinearity between independent variables, and our final correlation Table 6.8 does
not suggest existence of multicollinearity among explanatory variables. The final earnings
equation for the primary segment is as follows (see Table 6.9): p
In Yi = 6.852817 + .700172 (father's occupation : administrative/professional) + .092363 (labour market experience) - .002426 (labour market experience squared) + .288124 (on-the-job training) - .232952 (sex) + .002679 (years of schooling squared) + .022613 (years of vocational schooling squared).
We reached at the above final earnings equation for the primary segment after carefully
examining all the independent variables initially considered in the equation. Like the
secondary segment equation, we preferred to drop 'age' and 'age squared' in favour of
'labour market experience' and 'labour market experience squared'. If we examine the
correlation matrix for the primary segment of the labour market, we will find that ·sex of
the worker' is negatively and significantly correlated with log earnings (see Table 6.6).
This means that if the worker is a female, she earns less than her male counterpart in the
primary segment of the labour market. However, 'sex of the worker' is also negatively and
significantly correlated with 'marital status of worker'. The correlation coefficient between
these two variables is -.4882, which implies that most of the female workers in the
primary segment are unmarried. Besides, 'marital status' of the worker is significantly
correlated with many other independent variables (see Table 6.6), and the correlation
coefficient between 'labour market experience' and 'marital status' of the worker is .3821,
which is significant too. This simply means that, in the primary segment of the labour
market, most of the female workers are unmarried, and the married workers have relatively
longer labour market experience. Therefore, when 'labour market experience' is included
in the equation, the 'marital status' does not enter the equation as an explanatory variable
on the basis of partial F -test.
215
l
Table 6.6 : Correlation matrix for primary segment
LOG EAR E2 AGEQ E3 E5 E6 E8 EIO E15 FAOCC E21
LOG EAR 1.0000 .6449° 0 .6259° 0 -.3136•. .1194 -.0487 .3172•• .2240 .2309 .5337•• .5192°* E2 1.0000 .9912° 0 ·.2976° .0235 -.0238 .4849°. .0484 .0067 .2632" .3575° 0
AGEQ 1.0000 ·.2630° .0191 -.0261 .4418 .. .0365 .0240 .2649° .353o••
E3 1.0000 .1229 .1695 -.4882°. -.1440 .2229 -.0763 .0189
E5 1.0000 .0327 -.2060 -.0659 .2675• .0991 .1610
E6 1.0000 -.0149 1088 -1540 -.0371 .0450
E8 1.0000 .1656 -.3646•. .1002 .0943
EIO I 0000 -.1272 .1462 .2590°
E15 1.0000 .1314 .383&··
FAOCC 1.0000 .456o••
E21 1.0000
216
Table 6.6 (Contd.)
E22 E25A EDUQ E26 VEDNQ FJMODE E29 E54 E57 V4 PAEDU LMEX LMEXQ PI I LOOEAR .2264 .2339 .2426° .3321° 0 .4397° 0 -.0470 .4009° 0 -.0634 .5773° 0 .2746° .3168** .4932** .3679** .0193 E2 .0017 .0275 .0307 .2080 .3005* -.!351 .0691 .0947 .3964** .6577** -.0!31 .8200** .7139** -.0298 AGEQ .0204 .0323 .0352 .1993 .2917* -.1233 .0572 .0836 .3924* 0 .6475** -.0137 .8363** .7662** -.0562 E3 .2229 .1620 .1470 -.2270 -.3056** .3184** -.3115* 0 -.1892 -.1542 -.2114 .1538 -.1795 -.1354 .0960 E5 .2675* .1678 .2040 -.0321 -.0306 .1337 .1034 -.1651 .1184 -.0627 .3639** -.0590 -.0884 -.1469 E6 -.1540 .0529 .0566 -.0298 -.0209 .2041 .0190 -.0055 -.1153 .0134 1073 -0364 -.0403 .1139 E8 .3646** .3397** -.3323** -.3078°* .3590** -.2664* -.0515 .2993* -.2030 -.3713* 0 .2715* .382.1** .2572* .1057 EIO -.1272 -.1351 -.1338 .1204 .1470 -.1835 .1732 .0982 .0277 -.0625 -.0114 -.1220 -.IS68 -.0336 El5 .9654** .4813 .. .4771 .. -.1657 -.1301 .4274* 0 .1809 -.3487** .1021 -.0624 .5345** .0413 .0442 -.0966 FAOCC .1711 .0886 .0919 .3032** .3431** -.0069 .2240 .1868 .3528*. -.0428 .3041** .1250 .1149 .1610 E21 .3786** .2691* .2681* .2871* .3693** .2427* .2240 -.1653 .3193** .0770 .4668*• .1956 .1342 .2623* E22 1.0000 .4419 .. .4379** -.1774 -.1333 .3920** .2350 -.3154* 0 .1021 -.0417 .4963*• .0328 .0415 -.0759 E2SA 1.0000 .9987** -.4506** -.3748** .3473** .1700 -.4849** .0623 -.1984 .4673** -.0231 -.0144 -.1587 EDUQ 1.0000 -.4497** -.3739** .3471° 0 .0546 -.4834° 0 .0582 -.1979 .4731** -.0229 -.0152 -.1543
E26 1.0000 .9660** -.0831 .0715 .2422* .3549** .0969 .0515 .1826 .1697 .3001.
VEDNQ 1.0000 -.0927 .3096** .1665 .4253** .1329 .0931 .2514* .2300 .2722•
FJMODE 1.0000 .3511 •• -.2748* -.0095 -.1202 .4933 .. -.1086 -.0637 .1472
E29 1.0000 -.0184 .2841° -.2214 .3330*• -.1247 -.0914 .0906
E54 1.0000 -.0153 .1328 -.4151** .0382 .0175 .2246
E57 1.0000 .2016 .1861 .3969** .3463** .0764
V4 1.0000 -.2Wl .7485** .6396** .0202
PAEDU 1.0000 -.1013 -.0711 .0734
LMEX .1.0000 .9248** .0439
LMEXQ 1.0000 -.0019
PII 1.0000
Number of cases: 116 2-tailed Signif: • -- .01 •• -- .001
217
Table 6.7: Mean & standard deviation of earnings, schooling and experience for the primary segment
Statistics
Mean Standard deviation No. of cases
Log earnings
7.95 .59
116
Years of schooling
13. I I 2.20
I 16
Years of labour market expenence
6.3I 4.79
I I 6
It can be seen in Table 6.6 that 'parents' education', 'father's occupation' and 'average
annual family income' of the worker are highly positively and significantly correlated with
each other. Each of these variables is also positively and significantly correlated with log
earnings. This implies that as the parents' educational status increases, the occupational
status of the father of the primary segment worker increases, and so also his/her average
annual family income. It also means that workers having higher socio-economic
background earn more than the workers having lower socio-economic background in the
primary segment of the labour market. Among these three variables, 'fathers' occupation'
is highly positively correlated with log earnings, and this variable captures the effects of
other two variables in explaining the earnings variance in the primary segment. Therefore,
when 'father's occupation' as an explanatory variable is included in the regression equation
for the primary segment, the explanatory power of the model improves over the
explanatory power of other models which includes either of the other two highly correlated
variables, i.e. 'parents' education' and 'average annual family income'. Though 'years of
schooling' of the worker is not significantly correlated with log earnings, it is positively
and significantly correlated with 'father's occupation' implying thereby that as the father's
occupational status increases, the educational status of the worker also increases.
However, 'years of vocational schooling' is positively and significantly correlated with log
earnings. This means that every additional years of vocational schooling of the worker
increases earnings in the primary segment of the labour market.
218
Table 6.8 Correlation matrix for final equation for the primary segment
LOG EAR E3 E57 EDUQ LMEX LMEXQ VEDNQ F6 LOG EAR 1.0000 -.3 136 .. .5773** .2426* .4932 .. .3679•• .4397•• .7046•• E3 1.0000 -.1542 .1470 -.1795 -.1354 -.3056•• -.1600 E57 1.0000 .0582 .3969*• .3463•. .4253•• .4419•• EDUQ 1.0000 -.0229 -.0152 -.3739•. .2150 LMEX 1.0000 .9248•• .2514• .2610• LMEXQ 1.0000 .2300 .2309 VEDNQ 1.0000 .3768• F6 1.0000
Number of cases : 116 2 -tailed significance : • - .0 I • • - .00 I
219
Table 6.9: Regression results for the primary segment (Dependent variable: log earnings)
Vairable B-coefficient Standard Error of B-coefficient
Father's occupation Administrative or Professional .700172 .102529
Labour market experience .092363 .016286 Labour market experience squared -.002426 5.81015E-04 On-the-job training .288124 .105840 Sex -.232952 .088201 Years of schooling squared .002679 6.22593E-04 Years of vocational shoo ling squared .022613 .007282 Constant 6.852817 Multiple R .86407 R- square .74661 Adjusted R- square .73019 Standard Error .30812 F= 45.45971 Regression 7 Residual 108 No. of cases 116
220
~-coefficient T-Value Significance ofT
.408791 6.829 .0000
. 746214 5.671 .0000 -.534730 -4.175 .0001 .163690 2.722 .0076
-.136008 -2.641 .0095 .250173 4.303 .0000 .203380 3.106 .0024
We have dropped 'first job' of the worker as an explanatory variable from the regression
equation for the primary segment, though this variable is highly positively correlated to log
earnings. This we have done because of the fact that examination of sample data on first
and current jobs of workers in the primary segment of the labour market reveals that the
·first job' of most of the workers is also their current job. Therefore, to neutralize the effect
of current job on the earnings level of workers on the current job, we preferred to drop
'first job' as an explanatory variable from the equation. However, the effect of the first job
is captured by the 'years of vocational schooling' variable (see Table 6.6).
Thus, when log earnings is regressed on the final 6 original variables and one dummy
variable included in the equation for primary segment, the model explains nearly 75
percent of earnings variance (see Table 6.9). The B-coefficients of all the explanatory
variables in the equation are also highly significant at 5 percent level. Among all the
explanatory variables in the equation, the explanatory power of 'father's occupation' is the
highest having R-square value of .49647. This variable alone explains about 50 percent
earnings differentials in the primary segment. 'Father's occupation' along with 'labour
market experience' of the worker explains nearly 60 percent of variation in earnings in the
primary segment. In this case, R-square change is .1 0263, thereby implying that inclusion
of' labour market experience' in the equation increases the explanatory power of the model
by about 10 percent. When 'labour market experience squared' is brought in the equation,
the explanatory power of the model increases by around 5 percent (R2 change = .04624),
and the model explains nearly 65 percent of earnings differentials in the primary segment.
However, the negative B- coefficient of 'labour market experience squared' variable
implies that, after a certain years of labour market experience, every additional year of
labour market experience yields a negative return in the primary segment of the labour
market.
In the next step, with inclusion of ·on-the-job training' facility on the current job as an
explanatory variable in the regression raises the R2-value of the model to .68235, which
221
means that the contribution of this variable alone in explaining variations in earnings in the
primary segment is about 4 percent (i.e. R2 -change = .03701 ). 'Sex' of the worker along
with earlier variables explains 70 percent of variations in earnings. The individual
contribution of 'sex' as an explanatory variable in raising the explanatory power of the
model is marginal, i.e. only around 2 percent.
When 'years of schooling squared' is brought in the regression equation for the primary
segment, the explanatory power of the model increases by nearly 2 percent (i.e. R-square
change is .02382), and the model explains about 72 percent of earnings variance is the
primary segment. Finally, 'years of vocational schooling squared' along with the other six
variables increases the explanatory power of the model to about 75 percent, i.e. R2 =
.74661. Thus, our final regression model suggests that the major determinants of earnings
in the primary segment are 'father's occupation' ·labour market experience' ·sex' ·on-the
job training' and 'general and vocational education' of the worker. It should be noted here
that education of the worker has a positive market premium in the primary segment of the
labour market, and not in the secondary segment. Moreover, 'on-the-job training' has also
a positive markeepremium in the primary segment of the labour market.
6.3.3 Determinants of Earnings in the whole Sample
We have also estimated several alternative regressions to identify the maJor
determinants of earnings in the entire sample population of workers. The final regression
model for the whole sample is highly significant, and the B-coefficients of the explanatory
variables in the fmal regression equation are also significant at 5 percent level. To begin
with, we regressed log earnings on 22 original variables and 16 dummy variables. But, our
final aggregate earnings equation includes 6 original variables and 3 dummy variables. The
aggregate earnings equation which best fits the sample data is as follows (see Table 6.13) :
w In Yi = 6.815261 + .277726 (labour market segment of employment) + .084892
(labour market experience) + .743984 (fathers' occupation administrative or professional) + .399225 (religion : Muslim) + .016425 (parents' education) - .002071 (labour market experience squared) -.173946 (first job : unskilled manual worker) + .374189 (on-the-job training)+ .112043 (marital status).
222
Before going to discuss about the final earnings equation for the entire population, it is
perhaps necessary here to mention the process of selection of the variables. Like the earlier
two earnings functions, we have also preferred here to drop 'age' and 'age squared' in
favour of'labour mark~t experience' and 'labour market experience squared'. It can be seen
in Table 6.10 that 'caste of the worker' is positively and significantly correlated with 'years
of schooling', thereby meaning that workers belonging to lower caste categories arc
relatively less educated than the higher caste workers. The, ·religion of the worker' is
positively and significantly correlated with log earnings, and as we can see this variable
remains as an important explanatory variable in our final earnings function. 'Marital status'
of the worker is positively and significantly correlated with log earnings and 'labour
market experience'. This implies that unmarried workers in the sample have less years of
labour market experience, and they earn less than the married workers in the small scale
manufacturing sector in Delhi. Similarly, the degree of correlation between 'place of
origin' of the worker and log earnings is also positive and significant, which means that
workers having urban origin earn more than the workers having rural origin, in the
manufacturing sector. However, the 'place of origin' of the worker variable is significantly
correlated with many other independent variables, particularly with 'parents' education.
'Parents' education' captures the effect of the 'place of origin' on the earnings of the
individual worker. (see Table 6.1 0). Table 6.10 shows that in our sample, workers having
rural background are a disadvantaged group in the small scale manufacturing sector in
Delhi. Their parents are less educated and the occupational status of their fathers is also
low. Moreover, these workers belong to poor households, having considerably low
average annual family income. So far as their labour market experience is concerned these
workers had access to their low paid first job through informal channels of recruitment.
And, they are mostly found working in the lower segment of the manufacturing labour
market of the small scale manufacturing sector. Thus, the 'labour market structure'
variable or the labour market segment in which the individual worker is employed captures
the influence of a whole lot of socio-economic variables 'on the earnings of the worker.
223
Table 6.10 : Correlation matrix for the whole sample
LOG EAR E2 AGEQ E3 E5 E6 E8 EIO El5 PAEDU FAOCC E21 LOG EAR 1.0000 .5285** .5258** -.0294 .1938** .1782** -.2070** .0962 .3953** .5780** .5425** .5723** E2 1.0000 .9849** -.0814 .1260 -.0247 -.4759** .0660 .1305* .1684** .2286** .3083** AGEQ 1.0000 -.0856 .1157 -.0235 .4247** .. 0389 .1346* .1633** .2404** .3224** E3 1.0000 .1374* .0703 -.2779** -.1273* .431 o•• .2776** .1395* .1730** E5 1.0000 -.0369 -.0682 -.1036 .2413** .2915** .2128** .1812*. E6 1.0000 -.0066 .0976 -.0475 .1297* .0743 .0595 E8 1.0000 -.2066*. -.2376*. -.1836** -.0079 -.0022 EIO 1.0000 -.1602* -.0607 .0357 .1259
El5 1.0000 .6196** .3764** .4952*•
PAEDU 1.0000 .5610** .632o••
FAOCC 1.0000 .5607••
E21 1.0000
224
Table 6.10 (Contd.)
E22 E25A F.OUQ F.26 VEDNQ EDEXP f-JMODE E29 E54 E57 V4 V6 LMEX LMEXQ PI\ LOG EAR .3490 .. .44!2•• .5048** . 5305** .5434 .. .5960• .4091** .6624** -.4007*. .4817** .2306** .5892** .4910 .. .3920** .1169 E2 .0946 .1648** .2028 .. .2757•• .3073 .. .6396• .1541• .2474*"' -.0804 .3123*. .3358** .2923** .6802** .6082*• .I 173 AGEQ .0956 .1529• .1929*• .2779•• .3171•• .6740* .1459* .2430** -.0909 .3380*. .3439** .2850** .7045** .6748 .. .0917 E3 .3941•• .2230** .2266** .0009 -.0701 .0101 .2790** .0040 -.2600** -.0524 -.1376* .1744** -.0806 -.0720 .0378 E5 .2493** .3299** .3269•• .1996•• .1747•• .1319* .2260** .2170** -.1766** .1069 -.1578* .2822** .0000 -.0095 -.0205 E6 -.0446 .0093 .0292 .0187 .0216 .0049 .1606* .1272* -.0992 -.0600 .1235 .0647 -.0024 -.0104 -.0127 E8 -.2401*• -.1581• -.1595• .0839 -.1225 .2319* -.1576* -.0251 .1315* .-.0861 .2930** -.0695 -.3779*• .2383 .. -.0213 EIO -.1421• -.0619 -.0787 -.0106 .0067 -.0606 -.1439* -.0298 .1011 -.0175 .0867 -.1010 .0161 -.0656 -.0661 E15 .8803•• .48oo•• .5336•. .2587** .2326** .3029* .5596*. .4691*. -.4379** .2192** -.1030 .5021** .1026 .1068 .0763 PAEDU .5860** .6725•• .7243*"' .5194** .4808** .3632* .7023** .6905** -.5399*. .3130*"' -.1265 .7176** .0757 .0693 .1325* FAOCC .3974•• .4025** .4065** .4442** .4458** .3211* .2936** .4442** -.2424** .3736** -.0642 .4072** .1403* .1373• .1043
E21 .4559•• .4635** .5107** .5261** .5459** .4067* .4943** .523 I** -.3629*. .3950*. -.0255 .5122** .2086** .1732** .2580~ ~
E22 1.0000 .6097** .5683** .2545** .2309** .3417* .5070** .4461** -.3952** .2049 .. -.1231 .4833 •• .0520 .0584 .OS 10
E25A 1.0000 .9638** .3891** .3452** . 5092* .5755"* .5710** -.4953 •• .2333** -.1733** .6916** .0651 .0610 .0957
EDUQ 1.0000 .3944** .3504** .5100* .6464** .629R** -.5398"* . 2583 .. -1702*• .7547 .. .0<>65 .0895 .1105
E26 1.0000 .9731 .. .3627* .4351** . 6563** -.2712** .4457** -.0627 .6591 •• . 1971** .1990** .2754" •
VEDNQ 1.0000 . 3874* .3 775 •• .63 13 •• -.2517*. .5011** -.0407 .5960** .2296** .2365** .2628**
EDEXP 1.0000 .3124** . 3534** -.3253 •• .4163 •• .3494** .4623** .8174** .7993** .0882
FJMODE 1.0000 .6504** -.4479** .2154** -.1266 .7132** .0760 .0745 .2279**
E29 1.0000 -.4922** .3943** -.0568 .8128** .1306* .1103 .1825*"
E54 1.0000 -.1672** -.1278* -.5176*. -.1640** -.1405* .167!**
E57 1.0000 .0381 .31 06** .3012** .3145** .1\80
V4 1.0000 -.1247 .5912** .4348 .. -.1155
V6 1.0000 .1782** .1575* .2417$$
LMEX 1.0000 .9094** .0345
LMEXQ 1.0000 .0247
PII 1.0000
Number of cases: 413 2-tailed Sign if: • -- .0 I ... -- .001
225
Table 6.10 also shows that the degree of correlation between 'place of schooling' of the
worker and log earnings is positive and high, meaning thereby that workers schooled is
rural government schools earn less than the workers schooled in urban government schools
in the manufacturing labour market. These workers are also relatively less educated.
Besides, 'years of schooling' of the worker is positively and significantly correlated with
log earnings. This means that, in our sample, workers having more years of schooling
earn more than the workers having less years of schooling. However, we will find that in
our final equation even ·year of schooling' is not considered, rather ·on-the-job training'
proves to be one of the explanatory variables of earnings variations in the whole sample.
Similarly, 'years of vocational schooling' and log earnings are also highly and significantly
correlated with each other, which means that workers having more years of vocational
schooling are paid more than the workers having less years of vocational schooling. The
positive and high correlation between 'labour market structure' and 'years of vocational
schooling' suggests that workers in the upper segment of the labour market have high
vocational educational status than the workers in the lower segments of the labour market.
In fact, in our sample, in the secondary segment of the labour market, workers have no
vocational schooling at all. Thus, 'labour market structure' also captures the influence of
vocational education of the worker on his/her earnings.
In the manufacturing labour market, 'average number of hours worked daily' by the
worker is negatively correlated with log earnings, which implies that workers who work for
more number of hours a day are paid less than the workers who work for less number of
hours a day (see Table 6.1 0). Similarly, workers who work for a longer duration in a day
are found in the lower segments of the labour market. Here too 'labour market structure'
captures the influence of· daily work duration' of the worker on his/her earnings. ·On-the
job training' is paid in the manufacturing labour market and those workers working in the
upper segment of the labour market enjoy the on-the-job training facility. Thus, 'labour
market structure' as an explanatory variable in the earnings equation for the whole sample
226
capture the effects of many socio-economic, educational and labour market related
variables on the earnings of the worker.
Table 6.11 : ¥ean & standard deviation of earnings, schooling and experience for the wh~le sample
Statistics
Mean Standard deviation No. of cases
Log earnings
7.41 .58
413
Years of schooling
8.99 3.75
413
Years of labour market expenence
5.20 3.89
413
The final earnings equation for the whole sample explains total earnings in terms of
'segment of employment' of the worker, his/her 'fathers' occupation', 'religion', 'marital
status', 'parents' education' 'first job' 'on-the-job-training', and 'labour market experience'.
lbe equation is significant, and our final correlation matrix does not show existence of
multicollinearity among explanatory variables (see Table 6.12). This earnings equation
explains nearly 74 percent of earnings variations in the sample (see Table 6.13). As has
already been discussed earlier, 'labour market structure' or 'segment of employment' is the
most powerful explanatory variable in the earnings equation, and it alone explains nearly
35 percent of earnings variation in the whole sample (R2==.34715). This finding supports
our argument that in the small scale manufacturing sector in Delhi structural factors in the
labour market are very important in determining wages of a worker. \\'ben 'labour market
experience' is considered along with the 'structure of the labour market', the equation
explains nearly 50 percent of earnings variations (R2==.50100). Here, R-square change is
.15385, implying that the contribution of labour market experience alone to the explanatory
power of the equation is about 15 percent. When ·fathers' occupation' divided into 6
dummy variables is included in the equation, its explanatory power increases by around 10
percent and the equation explains 60 percent of earnings variations (R2==.60382). When
'religion' of the worker is brought in the equation, the equation explains nearly 66 percent
of earnings variations, the increase here is by 6 percent.
227
Table 6.12 Final correlation matrix for the aggregate equation for the whole sample
LOG EAR E8 E57 PAEDU V6 LMEX LMEXQ F6 R2 J\
LOG EAR 1.000 .2070** .4817** .5780** .5891 ** .491 o• • .3920** .5406** . \589* -.5381**
E8 1.000 .0861 -.1836** -.0695 .3779** .2383** .678 .0282 .0009
E57 1.0000 .3130** .3106** .3012** .3145** .4809** -.0563 -.1909**
PAEDU 1.0000 .7176** .0757 .0693 .3844** -.0506 -.4821**
V6 1.0000 .1782** .1575* .3044 •• -.1408* -.5282**
LME 1.0000 .9094** .2041 ** -.0001 -.1436*
LMEXQ 1.0000 .2151 ,.. .0006 -.1048
F6 1.0000 -.0600 -.2037**
R2 1.0000 -.2306**
11 1.0000
Number of cases : 413 2 -tailed significance : • - .0 I ** - .00 I
22R
Table 6.13 : Regression results for the whole sample (Dependent variable : log earnings)
Vairable
Segment of employment Labour market experience Father's occupation
Administrative or Professional Religion (Muslim) Parents' education Labour market experience squared First job unskilled manual worker On-the-job training Marital status Constant Multiple R R- square Adjusted R - square Standard Error F= Regression Residual No. of cases
B-coefficient
.277726
.084892
.743984
.399225
.016425 -.002071 -.173946 .374189 .112043
6.815261 .85975 .73917 .73334 .29736
126.89459 9
403 413
Standard Error ~-coefficient T-Value of B-coefficient
.051434 .217017 5.400
.010027 .572848 8.466
.088333 .256984 8.422
.059434 .185578 6.717
.003013 .215663 5.451 4.33721E-04 -.307955 -4.774
.037676 -.151196 -4.617
.093826 .121716 3.988
.038806 .085608 2.887
·----·-------·-----··
229
Significance ofT
.0000
.0000
.0000
.0000
.0000
.0000
.0000
.0001
.0041
In the next step, 'parents' education' along with the earlier 4 variables raises the
explanatory power of the equation to about 69 percent. In other words, other things
remaining the same, 'parents education' explains nearly 2 percent of earnings differentials
in the whole sample, R2-change is .02388. Inclusion of' labour market experience squared'
in the equation increases the explanatory power of the equation by 2 percent (R2-change =
.02252), and the equation explains nearly 71 percent of earnings variations in the sample.
'First job' of the worker divided into 6 dummy variables along with the earlier variables
explains about 72 percent of earnings variations among workers in the whole sample. Here,
the individual contribution of the ·first job' variable in explaining earnings differentials in
the sample is only one percent. When 'on-the-job training' is entered in the equation, the
R2 value increases to .73377, and finally, 'marital status' of the worker along with the
earlier 8 variables raises the R2-value to .73917. The contribution of 'marital status' in
explaining earnings variations in the whole sample is relatively very low, i.e. only 0.5
percent.
The final earnings equation for the whole sample suggests that structural factors in the
labour market and socio-economic backgrounds are most important determinants of
earnings of the sample workers, and the ·segment of employment' explains a little less than
half of the earnings variations in the sample. Besides, 'labour market experience' also
emerges as the next best determinant of earnings, and the effect of 'religion of the worker'
on his/her earnings level is significant but very low. Social background of the worker also
plays an important role in determining the earnings of an individual worker in the urban
manufacturing labour market in Delhi.
6.4 Some Findings
We ran several alternative regressions separately for each individual labour market
segment and for the whole sample. As can be seen in Table 6.14, finally we have selected
3 regressions, one each for the secondary segment, primary segment and for the whole
sample, as our 'best fits'. In this section, our discussions of the findings will focus on these
three regressions. All the regressions discussed here are highly significant and confirm to
230
the partial F-test, with 95 percent of confidence limit. Also Students' T-test confinn that
all the B-coefficients of these regressions are significant at 5 percent level, though some
are even significant at one percent level.
Our final regression equation for the secondary segment explains total earnings in
terms of five original variables. such as 'marital status', 'parents' education', 'average
annual family income'. 'labour market experience', 'labour market experience squared' and
two dummy variables such as 'religion' of the worker (i.e. Hindu) and first job (i.e.
unskilled manual worker). And, the equation explains about 52 percent of earnings
variations among individuals in the secondary segment (R2 = .51895). The results ofthis
regression suggest that the 'first job' of the worker is one of the important predictors of
earnings in the secondary segment. The ·first job' is divided into 6 dummy variables, and in
the secondary segment equation, the dummy variable representing the 'unskilled manual
worker' has a negative and large coefficient (see Table 6.14). So far as the manufacturing
urban labour market in Delhi is concerned, this finding suggests that workers who get into
low paid unskilled manual first jobs continue to earn 21 percent less in their current jobs
compared to the workers who get into relatively high paid first jobs. We have already
discussed in chapter 5 that job changes in the secondary segment of the manufacturing
labour market is quite frequent compared to the frequency of job changes by workers in the
primary segment of the labour market. The above finding therefore suggests that workers
having low paid first jobs are still caught up in the secondary segment of the labour market.
Thus, job history of the worker is one of the important determinants of earnings in the
manufacturing labour market in Delhi.
It can be seen in Table 6.14 that the coefficient of the religion dummy (i,e. Hindu) is
negative and large (i.e. B- coefficient = -.36946). This means that being a Hindu rather
than Muslim or Christian depresses the earnings by 36.9 percent in the secondary labour
market. This is perhaps a surprising finding. However, if we examine the characteristics of
the secondary labour market in our sample, we will find that most of the Hindus are
working in unskilled manual jobs which are lower paid jobs, and most of the workers
belonging to other religions such as Islam or Christianity are working in the skilled
231
Table 6. I 4: Regression result~ for ~econdary 11e2ment, primary segmenl nnd whole snmple (l>ependent variable: log earnin2s)
Vnrinble
Sex Religion
Hindi Muslim
Marital status Parents' education Father's occupation
Administrative or professional Average annual family income Years of schooling squared Y car~ of vocntional schooling squared First job
Unskilled manual worker On-the-job training Labour market experience Labour market experience squared Segment of employment or labour market structure Constant Multiple R R-square Adjusted R-square Standard Error F= Regression Residual No. of Cases
Secondary Segment
B-cocfficient
-369436
.I i3702
.014717
2.637586E-06
-.209691
.78168 -.00 I 947
7.210028 .72038 .51895 .50730 .28480
44.53858 7
289 297
T-Valuc -.6.140
-6.140
2.496 3.259
2.200
-5.409
.0000 -2.145
Note: All the 8-coefficicnts arc significant at 5 percent level.
232
Primary Segment Whole Sample (Aggregate/Equation)
B-coe fficient -.232952
.700172
.0026 7(1
.022613
.2R8124
.092363 -.002426
6.852817 .86407 .74661 .73019 .30812
45.45971 7
108 116
T-Value B-coefficient T-Value -·--2.641"·----·-·--------------- ---
.399225 6.717
.112043 2.887
.016425 5.451
6.829 .743984 8.422
·1.:10.\ 3.106
. 17.:19-16 -4.617 2.722 .374189 3.988
.000 .84892 .0000 -4.175 -.002071 --4.774
.277726 5.400 6.815261
.85975
.73917
.73334
.29736 ! 26.89459
Q
403 -113
occupations. This is perhaps one of the reasons explaining disadvantaged position of Hindu
workers in the secondary segment.
'Marital status' of the worker also emerges as one of the important predictors of
earnings in the secondary segment. Given the coefficient of .113702, it implies that being
married raises one's earnings by 11.4 percent in the secondary segment. This finding
somehow suggests that perhaps employers consider married persons as relatively stable
workers in the secondary segment, and it is this characteristics of this category of workers
which yields them relatively a higher market premium than the unmarried workers. Among
the family-background variables. ·parents' education' and ·average annual family income'
are two important determinants of earnings in the secondary segment. Given the
statistically significant and positive coefficients of these two predictors, it means that an
individuals earnings are higher, higher the educational attainments of his/her parents, and
also higher the average annual income of his/her family. Thus, family- background of an
individual influences, to a large extent, the level of earnings in the secondary segment. This
finding do support the claims of the segmented labour market theory that socio-economic
backgrounds of an individual rather than his/her productivity related characteristics are
important in determining earnings in a highly structured labour market, like the one we
have in our sample.
'Total labour market experience' of the worker is the most important predictor of
earnings in the secondary segment, the size of its B-coefficient is .078168. What it means
is that, given the semi-logarithmic form of the earnings equation, an additional year of
labour market experience gives a return of about 7.8 percent in the secondary segment of
the manufacturing labour market. Moreover, a negative coefficient of 'labour market
experience squared', suggests that, after a certain initial years of labour market experience,
an additional year of labour market experience exerts a declining effect on the level of
earnings of the secondary segment worker.
Thus, the earnings equation for the secondary segment suggests that socio-economic
backgrounds, such as religion, family income, marital status, parents' education', labour
233
market experience, a.'1d tirst job (unskilled manual) of the worker are important
determinants of personal earnings. One of the objective measures of the importance of an
independent variable in the final regression equation is the P-coefficient. If we examine
the P-coefficients of the regression equation for the secondary segment, we will find that
'labour market experience' contributes most to the explanation of the variations in earnings
in the secondary segment. In terms of relative importance of the explanatory variables,
'labour market experience' is followed by 'parents' education', 'marital status', 'average
annual family income', 'first job', and ·reiigiOn'. However, what is striking here is that
'years of schooling' of the secondary segment worker does not yield any returns in the
urban manufacturing sector in Delhi, and therefore it does not figure in the final earnings
equation. This finding of our regression analysis supports the proposition of the segmented
labour market theory that there are no returns to schooling, and if there is any, it is too low
in the secondary segment of the labour market. This finding also suggests that wage
setting mechanism in the secondary segment does not take into consideration the variations
in productivity related characteristics across individual workers. Moreover, our finding
also indicates that secondary segment workers do not. receive any on-the-job training
facility, which further supports the proposition of the segmented labour market theory.
Besides, our final earnings equation for the secondary segment explains only 52 percent
of the variations in personal earnings, and nearly 48 percent of variations in earnings can
not be explained by this equation. Perhaps, what it indicates is that the wage determination
process in the secondary segment is highly influenced by market forces such as demand for
and supply of labour in the manufacturing sector. Given the high rate of rural-urban
migration in the country, employers find surplus labour supply in the secondary segment of
the manufacturing labour market. Existence of this surplus labour supply depresses the
'reservation wage' level of the secondary segment workers in Delhi, and also employers
find it easy to obtain labour by offering low wages. It should also be noted here that
workers in the small scale manufacturing sector in Delhi, especially in our sample. are not
unionised, which indicates that these workers have no bargaining power in the market.
234
Regression analysis for the primary segment of the labour market reveals that 'sex',
'fathers' occupation, 'labour market experience', 'education', and 'on-the-job training' of
the worker are important predictors of personal earnings, and the earnings equation
explains about 75 pe~cent of variations in earnings. It should be noted here that human
capital variables such as education, experience and on-the-job training are valued in the
primary segment of the labour market. and education finds a place in the final earnings
equation. Among all the predictors in the primary segment earnings equation. the size of
the B-coefficient of 'labour market experience' is .092363, and it means that an additional
year of labour market experience in the primary segment raises the earnings of the worker
by around 9.2 percent. If we compare the coefficients of thjs variable in the earnings
equations for primary segment and secondary segment, we will find that returns to an
additional year of labour market experience in the primary segment is higher than the
returns to an additional year of labour market experience in the secondary segment of the
manufacturing labour market in Delhi. This indicates that workers in the secondary
segment have a relatively flat experience- earnings profile compared to the experience -
earnings profile of the primary segment workers. This finding also suggests that 'labour
market experience' has a relatively high market premium in the primary segment than in
the secondary segment.
Among the social background variables, 'fathers' occupation' (administrative or
professional) is an important predictor of personal earnings in the primary segment, and the
B-coefficient of this variable is . 700172. This implies that if the father of an worker in the
primary segment is in administrative or professional job, his/her earnings increases by
nearly 70 percent. The B-coefficient of 'sex' variable is negative and significant. It can be
seen in Table 6.14 that, in the primary segment, being a female rather than a male
depresses one's earnings by 23.3 percent, all other things remaining the same. This finding
perhaps indicates that in the primary segment employers prefer males to females as their
employees.
235
One of the important findings here is that both general and vocational education of the
worker do have market premium in the primary segment of the manufacturing labour
market. In our final equation for the primary segment, positive coefficients of 'years of
schooling squared' and 'years of vocational schooling squared' that the level of general and
vocational education of the worker positively influences the worker's earnings, and
perhaps, these are used as 'screening devices' in this segment. It also implies that earnings
in the primary segment are a parabolic function of level of general and vocational
education, and the parabolic effects of years of general and vocational schooling are
positive. In this segment, those workers who enjoy on-the-job training facility on the
current job earn nearly 28.8 percent more than those workers who do not enjoy this facility.
The above findings support one of the claims of the segmentation theorists that human
capital variables in general and education in particular yields a positive return in the
primary segment of the labour market. Another important finding here is that if we
compare the B-coefficients of 'labour market experience squared' in the primary segment
equation and in the secondary segment equation, we will see that both the coefficients are
negative and significant. But the size of the B-coefficient of this variable in the secondary
segment equation is smaller than that of the B-coefficient of the· primary segment equation.
This means that, in the secondary segment, declining returns to an additional year of labour
market experience begins at an early stage compared to the years of labour market
experience after which every additional·year of experience yields a declining return In the
primary segment.
A comparison of regression results for primary and secondary segments of the labour
market reveals that the earnings functions that 'best fits' the individual segments are
different, and the size of the coefficients of the common variables in these earnings
functions are also different, generally large in case of primary segment equation. This
finding strongly supports our conclusion that the urban small scale manufacturing labour
market in Delhi is divided into distinct segments, where workers face different earnings
functions. And, even if workers face the same earnings function in both primary and
secondary segments, the size of the coefficients of the common explanatory variables in the
236
equations differ between the individual segments of the labour market. We, therefore,
argue that a distinct low wage secondary labour market exists in the manufacturing sector
in Delhi where wage setting mechanism is greatly influenced by the market forces.
In the earnings equation for the primary segment, if we rank the seven determinants of
earnings in terms of their relative contributions in explaining total earnings (on the basis of
size of p-coefficients of these variables), we will find that ·labour market experience' heads
the list and it is followed by 'fathers' occupation' (administrative or professional), 'years of
general schooling squared', 'years of vocational schooling squared', 'on- the-job training',
'sex' and 'labour market experience squared'.
As can be seen in Table 6.14, the aggregate equation for the entire sample explains total
earnings in terms of: (I) ·segment of employment' (i.e. structure of the labour market); (2)
'labour market experience'; (3) 'fathers' occupation' (administrative or professional); (4)
'religion' (Muslim); (5) 'parents' education'; (6) 'labour market experience squared'; (7)
'first job' (unskilled manual); (8) 'on-the-job training'; and (9) 'marital status' of the
worker. The 'segment of employment' (i.e. whether a primary or secondary segment
employee) emerges ac; one of the powerful predictors of personal earnings in the sample
small scale manufacturing labour market in Delhi. The B-coefficient of this variable is
.277726, which implies that being employed in the primary segment rather than in the
secondary segment increases the earnings of a worker by 27.8 percent. This finding draws
our attention to the nature of jobs found in these two labour market segments. In the
primary segment, jobs are protected in terms of some form of written agreement of
employment, and the secondary jobs are unprotected. This very fact influences the level of
earnings of worker to a great extent, indicating the importance of institutional factors in the
process of earnings determination in the sample manufacturing labour market in Delhi.
Another important finding is that 'religion' of the worker also proves to be an important
predictor of personal earnings in the sample, the coefficient of which is .399255. This
means that being a Muslim can raise ones earnings by 39.9 percent. This finding is
somewhat not surprising because of the fact that most of the low paid secondary segment
237
workers in our sample are Hindus, and Muslims and Christians are either working in
skilled or technical jobs which are relatively higher paid.
The 'first job' (unskilled manual) as an important predictor explains substantially the
earnings variations ~ong individuals in the sample. This dummy variable has a negative
and significant B-coefficient, i.e. -.173946. This means that, in the sample, those who have
their first job as unskilled manual worker earn about 17.4 percent less. This finding
indicates the importance of ones initial job history in wage determination process in the
sample manufacturing labour market. Among the family-background variables, 'parents'
education' and 'father's occupation' (administrative or professional) prove to be important
determinants of earnings in the entire sample, An additional year of parents' schooling
raises ones earnings by 1.6 percent in the entire sample. Similarly, if one's father is
employed in administrative or professional occupation, he/she earns around 74.4 percent
more in the sample labour market. These findings do support the proposition of the
segmented labour market theory that social background plays an important role in the
earnings determination process in a highly segmented labour market. Moreover, being
married also increases ones earnings by 11.2 percent in the entire sample, which indicates
that in the sample labour market, the stability related characteristics of a worker are valued
by the employers. 'On-the-job training' as a predictor also plays an important role in
explaining total earnings in the entire sample. The B-coefficient ofthis variable is .374189,
and it means that, in the sample population of workers, those who enjoy the on-the-job
training facility on current job earn nearly 3 7.4 percent more than those who do not. Apart
from this, 'labour market experience' also exerts a positive influence on the level of
personal earnings of the worker in the sample manufacturing sector. The B-coefficient of
this variable is positive and significant, i.e .. 084892. This means that an additional year of
labour market experience raises ones earnings by 8.5 percent. We also find that the
parabolic effect of labour market experience on earnings is negative, thereby implying that
after a certain years of work experience, returns to every additional year of work
experience starts declining.
238
Ranking of the independent variables on the basis of size of the P-coefficients in the
aggregate earnings equation show that 'years of labour market experience' is the most
important predictor of personal earnings in the entire sample, and this is followed by
'father's occupation' (administrative or professional), 'segment of employment', 'parents'
education', 'religion' (Muslim), 'on-the-job training', 'marital status', 'first job' (unskilled
manual), and 'labour market experience squared'.
Our regression analysis in this chapter strongly indicates the existence of
compartmentalised labour market segments in the sample urban manufacturing sector in
Delhi. Moreover, our findings suggest that structural factors and socio-economic
backgrounds are important in determining earnings in the urban manufacturing labour
market, and the overall labour market outcomes of this sector reflects the existing socio
economic inequality in our society. The functioning of this small scale manufacturing
labour market in Delhi also reproduces the existing socio- economic inequality. This is
demonstrated by the fact that socio-economically disadvantaged groups are found
concentrated in the vulnerable low labour market segments in the urban manufacturing
sector in Delhi. In other words, results of our analysis show that the relatively less
privileged groups in our society are a less fortunate lot in the urban manufacturing labour
market in India in general, and in Delhi in particular.
239
l