Impacts of Income Gap on Migration Decision in China

Impacts of Income Gap on Migration
Decision in China:
A Verification of the Todaro Model
Nong ZHU
CERDI - IDREC
65 Boulevard François Mitterrand
63000 Clermont Ferrand - France
Tel. 04 73 43 12 36
Fax. 04 73 43 12 28
E-mail : [email protected]
Impacts of Income Gap on Migration Decision in China:
A Verification of the Todaro Model
Nong Zhu*
Abstract: Using survey data from China, this article examines the determinants of the ruralurban migrants' income. Specifically, it studies the effects of income gap on migration
decision and its sources. The empirical study demonstrates that income gap significantly
influences migration decision, which confirms the predictions of the Todaro model.
Moreover, our results show that migrants' income level depends greatly on education level
and profession, and that urban informal sectors are not always characterized by low salary.
By estimating urban-to-rural income gap, we find that the income differential between
migrants and non-migrants are larger for female than for male, which suggests that females
receive larger monetary returns as a result of migration. In terms of its decomposition, for
the males, their income gap is mainly determined by the differences in attributes; for the
females, both individual characteristics and the gain in returns to attributes play an
important role.
JEL Codes - E24, J31, O15, R23
Key words: China internal migration, Todaro model, sample selection, income gap
1 Introduction
Few models have been as dominant in their subject area within development
economics as that of migration introduced by Todaro (1969) and Harris and Todaro (1970).
This model provides a formal theory of rural to urban migration in developing countries.
According to this famous model, migration decision depends on the evaluation of expected
*
The author gratefully acknowledge the help from Elisabeth Sadoulet, Maréva Sabatier, Martin
Fournier, Mbolatiana Rambonilaza, Cécile Batisse, Sandra Poncet and Xubei Luo. Naturally, the
responsibility of any remaining errors lies with the author.
1
income of potential migrants. Two major factors determine this expected income, one is the
current urban salary, and the other is the subjective estimation of the probability of being
employed in urban formal sector, which is a function of unemployment rate. As suggested by
Todaro, the income gap (or expected income) constitutes the principal aspect of migration
motivation. The larger is this gap, the stronger is the migration propensity. Todaro theory is
the fundamental foundation of many empirical studies subject on migrations in developing
countries (Levy et alii, 1974; Todaro, 1976; Taylor, 1987; Lucas, 1988; Agesa et alii, 1999).
This famous theory suits also the case of China.
Prior to the recent tide of migration, China had for decades tightly restricted rural to
urban migration. This strict urban-rural segregation was mainly instituted following the
devastating famine that occurred between 1959 and 1961. The purpose of the policy was to
restrict the urban population size. The reason was that, among other things, the government
was responsible for feeding this population. Two fundamental methods were used to achieve
the segregation. One was to impose a high opportunity cost for leaving rural areas by tying
incomes to participation in daily collective farm work. The other was to make it difficult for
outsiders to live in urban areas through the denial of urban household registration (hukou), on
which employment, allocation of housing, food and other necessities were contingent. The
prolonged restriction on migration created large income gaps between urban and rural areas.
The income gap widened until 1978, declined between 1978 and 1984, and widened again
afterwards. The ratio of urban to rural per capita income was 2.57 in 1978, dropping to 1.86 in
1985, and then rising to 2.50 in 1994. From then on, it fluctuated around 2.50. (DCSNBS,1
1999: p.22).
The reforms began from the late 1970s. The people's commune came to an end and
the Household Responsibility System (HRS) was generalized throughout the rural areas,
which provided greater freedom of profession choice to farmers. The HRS had two farreaching and unintended effects on the control of migration. First, buying food in urban areas
without urban registration status become possible. The HRS increased the food supply
dramatically, which led to the availability of food on the free market in cities and eventually
led to the abandonment of food rationing. Second, the HRS returned personal freedom to rural
people. Rural labors could from there freely allocate their time. The large rural-urban income
gap incited farmers to leave agricultural activities for non-agricultural ones or migrate to
1
Department of Comprehensive Statistics of National Bureau of Statistics of China.
2
cities. Gradually, these spontaneous movements of the rural population broke the migration
constraints.
Given the great demographic pressure in rural areas, the income of farmers stayed at a
very low level. Motivated by the large rural-urban income gap, farmers were eager to leave
land. However, although the food shortage is no more a threat in nowadays, the government
continues to restrict migration by some direct or indirect regulation for three reasons. Firstly,
Urban residents are unwilling to share their higher living standards with rural people.
Secondly, the government does not want to increase its investment in urban infrastructures to
accommodate the rural migrants. In fact, in rural areas, it is the local population that takes the
burden of infrastructure investment. Thirdly, due to the State-Owned Enterprises reform,
unemployment becomes a serious problem in urban areas. The contradictions between rural
and urban areas, between official policies and farmers' aspirations, entail two major
consequences: first, the urban informal sector develops and the urban labor market
segmentation aggravates (Leila, 1999). Second, the rural non-agricultural sector prospers,
which provides another solution to the rural surplus labor (Aubert, 1995; Zhao, 1999). These
two rapidly growing sectors absorb a large amount of surplus agricultural labor.
Using the survey data, subject on "Migration and Regional Development", we try to
test the prediction of Todaro model, which suggests that migration decision depends on
urban-to-rural income gap. However, other than income gap, some other factors, such as
individual and family characteristics, rural-urban labor market conditions, etc., influence
migration decisions, too. The objective of this paper is to analyze the determinants of rural to
urban migrants' income and study impacts of income gap on migration decision and its
sources.
In the following section, we comment briefly on the principal theoretic model, Todaro
model, and its developments. Then, we introduce the analysis methods and data. In the forth
section, we present our econometric results. Finally, we conclude.
2 Todaro Model
Todaro and Harris’s fundamental premise is that potential migrants consider the
various labor-market opportunities available to them in the rural and urban sectors, and
choose the one that maximizes their expected wage gains. The key institutional assumptions
of the model include the followings (Bardhan and Udry, 1999: p.50):
a) The rural labor market is perfectly competitive.
3
b) Modern firms hire labor in the city, and the wage they pay is fixed above the marketclearing level, either by restrictive union activity or by governmental policy on wages.
c) Only urban residents can apply for jobs in modern firms, and if the job-applicants are
more than the available posts in modern firms, jobs will be allocated by drawing lots.
d) There is an “informal sector” in which urban residents not otherwise employed can
eke out a subsistence living using their labor power alone.
According to this model, the rural-urban migration decision is in response to the
difference of rural-urban expected remuneration.
The future expected net income from migration is given by:
∞
1
V0 = ∫ [ pwu − wr ]e − rt dt − C = [ pwu − wr ] − C
0
r
(2.1)
where wu and wr represent the average income of the formal sector and that of the
agricultural sector respectively; r, the migrant’s discount rate; p, the probability of finding a
job in formal sector, and C, the migration cost.
Migration takes place only if V0 is positive. In other words, migration takes place only
if there exists gains from moving, subject to certain employment prospects, i.e. only if
1
[ pwu − wr ] − C > 0 ⇔ pwu − wr > rC
r
(2.2)
The equilibrium migration condition is thus:
1
[ pwu − wr ] − C = 0 ⇔ pwu − wr = rC
r
(2.3)
The probability of finding a job in urban areas, p , is equal to the number of the total
available posts in urban areas, divided by the active urban population size after migration
takes place, namely Lu + MN r , where M represents the rural-urban migration rate and N r
represents the rural population size. If we suppose that Lu is determined by the exogenous
factors which are independent on migration and M is sufficiently small compared with N r ,
so that it does not influence rural population size, we have Lu = L u , N r = N r and:
p=
Lu
Lu + M N r
(2.4)
Substituting p by its value obtained from (2.4), we deduce the equilibrium migration rate:
 w − wr − rC  L u
M* =  u

 rC + wr  N r
(2.5)
4
If we take the existence of the urban informal sector into account, the rural expected
income will be equal to: pwu + (1 − p ) wb , where wb is the average income of informal sector.
In this case, the equilibrium migration rate is:
 ( w − wr ) − rC  L u
M b* =  u

 rC − ( wb − wr )  N r
(2.6)
Evidently, M b* > M * : the existence of informal sectors enlarges the absorption capacity of
urban labor market.
From (2.5) or (2.6), we get the familiar results:
∂M *
∂M *
∂M *
∂M *
>0 ;
< 0;
> 0;
<0
∂wu
∂wr
∂C
∂Lu
(2.7)
There are several policy implications of Todaro Model:
a) Any increase in urban marginal salary wu or decrease in rural marginal salary wr will
enhance rural-urban migration.
b) The more job-opportunities we create in urban areas, the stronger is the migration. In
consequence, the solution to urban unemployment problem does not only depend on
the expansions of industries.
c) Any decrease in migration costs will increase the rural depopulation. In other words,
to reduce the migration flux, it is necessary to increase the opportunity cost of
migration.
d) The most efficient ways to reduce migrations and alleviate the aggravation of urban
unemployment problem are to lighten the rural burdens and to develop the rural areas
in long term. (Guillaumont, 1985; Todaro, 1997).
Todaro model supplies a clear and simple explanation of migration phenomena and of
urban unemployment problems. It corresponds to the reality of labor market in most of the
developing countries. However, this model is based on some simplified hypotheses. Relaxing
some restrictions may lead to several extensions of it (Zenou 1995: p.300).
In Todaro model, the source of labor market imperfection, the rigidity of urban salary,
is exogenously determined. The salary level, which is imposed by the government, is
independent of the unemployment rate and the migration. Bardhan and Udry (1999: p.54-55)
develop an adverse selection model. It confirms the main conclusions of Todaro model
without basing on the hypothesis of exogenous urban salary fixation. They find that the salary
in manufacturing sector is higher than that in rural areas, but the salary in urban informal
sector is so low that it equalizes salaries in rural and urban areas.
5
Todaro reject the hypothesis of Lewis, which supposes that labor supply is unlimited
(Lewis, 1954) and there exists some agricultural labor whose productivity is so low that it
tends to zero. However, this hypothesis is true in some cases, for example, in the case of
China. During a long time, farmers were artificially blocked in the countryside. With the rapid
growth of rural population, the problem of the surplus agricultural labor became more and
more serious. With the difficulty in cultivable land enlargement, it is not easy to provide a
solution to this rural surplus labor problem, at least in the short run. The agricultural sector in
China is still characterized by technical stagnation and low productivity. In other words, it
stays in the traditional phase, which is defined by Schultz (1964). The agricultural income
thus stays at a very low level, namely, the subsistent level, except that it is influenced by some
exogenously factors, such as agricultural product price adjustment, land-allocation policies,
etc..
According to the Todaro model, increasing rural income will reduce the rural
depopulation incentive. However, its short-term marginal effect is ambiguous (Ghatak et alii,
1996: p.161,180). The rural development project may lead to a higher migration rate, at least
at its very beginning. The reason is that increasing rural income will eliminate the financial
constrains of potential migrants and enhance their capacity to overcome the migration
obstacles. (Schiff, 1996, p.7-9).
Some researchers, such as Bardhan and Udry (1999: p.51), suggest that the informal
sectors are not always characterized by low salary and they are not necessarily non-productive
sectors. Their findings are different to the descriptions in Todaro model. According to the
research results of informal sectors in Mexican (Roubaud, 1992), not all the posts in formal
sector are the ideal ones. As suggested by Gerdarme (1994: p.14), the informal sector may be
the source of certain dynamism and it is capable to initiate new techniques, which is not only
to ensure its survival but also to reply to the enlarging demand from urban population. This
marginal urban sector may constitute a transition to the more productive economic forms and
play a important role in auto-regulating employment and income within the cities of the third
world.
Contrary to the case in Mexican, in some developing countries, for example, in China,
there exists a serious urban labor market segmentation: rural migrants and local urban
residents participate in two different labor markets. Most of the non-qualified migrants arrive
in cities to take up marginal jobs that are characterized by long working hours, poor working
conditions, low and unstable pay, and no benefits – jobs which are unattractive to urban
residents. The possibility for them to enter the formal sector is very low. To these migrants,
6
the modern sector income is with sight but beyond reach. Moreover, the income gap between
formal and informal sectors is reflected not only by salary, but also by the advantages that the
formal sector workers received, such as housing subsidies, food provided, children education,
medical insurance and other social insurance. (Wang and Zuo, 1999: p.277)
3 Methodology and data
Using the data of a questionnaire survey in China, we will analyze the determinants of
the migrants' income, the effect of income gap on decision making and the sources of this gap.
3.1 Methodology
This study is composed of two steps. Firstly, we estimate simultaneously earnings
equations for the rural to urban migrants and non-migrants in rural areas, so that we study the
impact of income gap between these two groups on migration decision. Secondly, we study
the contribution of the various sources of income gap.
3.1.1
Effect of the income gap on the migration decision
Todaro suggested that the net difference between urban income and rural income plays
a dominant role in the migration behavior from countryside to cities. To analyze the impact of
income gap on migration, it is necessary to introduce the difference in urban to rural income
into the equation of migration decision. However, in our sample, urban income is observed
only if an individual has already migrated, which leads to the problem of sample selection
bias. We should take it into consideration and try to correct it by migration decision. So, we
should simultaneously estimate the income equation and the migration equation, which are
both endogenous. By using the two step procedure proposed by Heckman (1979), we can
solve this problem by the structural probit method (Nakosteen et alii, 1980; Perloff 1991;
Agesa et alii, 1999). This method includes three steps:
Firstly, we estimate a reduced form of the probit equation:
Pi * = α ' Z i + β ' X i + ε i'
(3.1)
where Pi* represents the migration decision; Z i and X i represent the independent variables
of the selection equation and those of the income equation respectively.
7
Secondly, we estimate the urban income equation and the rural income equation. In
order to correct the sample selection bias, we introduce the inverse Mills ratio issued from the
reduced form of the probit equation.
log Wui = β u X ui + γ u λui + µ ui
log Wri = β r X ri + γ r λ ri + µ ri
(3.2)
(3.3)
where Wui and Wri represent the migrant's income and the non-migrant's income respectively.
X ui and X ri are the vectors of individual characteristics. λ ri and λui are the inverse Mills
ratios.
Finally, from equations (3.2) and (3.3), we predict for each individual the value of
Ŵui , if he or she migrates to the cities; and that of Ŵri , if he or she stays in the countryside.
Then we introduce the income gap into the structural probit equation:
P * = αZ i + η (log Wˆ ui − log Wˆ ri ) + ε i
(3.4)
From this equation, we can study the impacts of different factors, the income factors and the
non-income factors, on migration decision.
The marginal effect of the regressors on the income in the observed sample consists of
two components. There is the direct effect on the mean of the income, which is β k . In
addition, for a particular independent variable, if it appears in the probability that Pi* is
positive, it will influence the income through its presence in the inverse Mills ratio λi
(Greene 1997: p.977-978). The full effect of changes in a regressor that appears in both X i
and Z i on the income is:
∂E (log W / P * > 0)
= β k − α k γ λ δ i where δ i = λi2 + (−αZ i )λi
∂x k
(3.5)
β k is a constant, while the total effect, noted as β ki' , varies. The value of β ki' is individually
dependent because of the existence of the term δ i . Any way, we can calculate some average
values of the total effect by calculating the arithmetic average of β ki' :
β k' =
∑β
i
n
'
ki
(3.6)
where n is the observation number.
8
3.1.2
Income gap decomposition
Various reasons, such as individual characteristics, discrimination between urban and
rural labor market and the other unobservable factors, will lead to the income gap between
migrants and non-migrants. Oaxaca (1973,1994) suggested a decomposition technique in
analyzing wage discrimination in labor market. Later, Reimers (1984) developed this method
to study the labor market discrimination, taking the adjusted wage into account to correct the
sample selection bias. Agesa et alii (1999) adopted this method to analyze the income
difference between migrants and non-migrants in Kenya (Agesa and al, 1999).
Based on equation (3.2) and (3.3), we estimate the urban income and rural income for
each individual:
log Wˆ ui = βˆ u X ui + γˆ u λui
(3.7)
log Wˆ ri = βˆ r X ri + γˆ r λ ri
(3.8)
So we can obtain:
~
~
log W u − log W r = log Wu − log Wr = βˆ u X u − βˆ r X r + γˆ u λ u − γˆ r X r
(3.9)
~
~
where Wu and Wr represent the geometric average value of the two groups respectively.
The income gap between these two groups can be decomposed as the following:
~
~
logWu − logWr = ( X u − X r )[ Dβˆ u + ( I − D) βˆ r ] +
[( I − D) X u + D X r ]( βˆ u − βˆ r ) +
(3.10)
γˆ u λ u − γˆ r λ r
where I is the identity matrix ; D is a diagonal matrix of weights. (3.10) decomposes the
difference between the geometric mean of observed wage rates for the two groups into three
parts : a) that due to differences in average characteristics of the groups, including differences
in local price levels where the groups members live ; b) that due to differences in the
parameters of the wage function, caused by labor market discrimination and other omitted
factors ; c) that due to differences in selectivity bias.
The measure of various sources of the gap depends on the choice of weights in matrix
D . If we assumed that discrimination penalizes the non-migrants by preventing them from
earning according to the migrants’ wage-offer function, then D equals I . If the
discrimination gives the rural-urban migrants an undeserved advantage, that they are paid
more than they would get in a non-discriminatory world, then D equals 0 (Reimers, 1984: p.
573; Oaxaca et alii, 1994: p.8).
9
We will base our studies on three hypotheses: 1) most of the rural-urban migrants
cannot enter the formal sector but only the informal one, where labor market is clean-up; 2) a
great number of agricultural surplus labor reduces rural income; 3) the differential between
industrial product price and agricultural product price also lowers the rural income. In
consequence, we take D = I in the following analysis:
~
~
log Wu − log Wr = ( X u − X r ) βˆ u + X r ( βˆ u − βˆ r ) + γˆ u λ u − γˆ r λ r
(3.11)
We define the ratio of the urban to rural income as R and the corrected value for
sample selection bias as R ' .
where
~
Wu
R = ~ = (1 + PX )(1 + Pβ )(1 + Pλ )
Wr
(3.12)
~
W
R' = ~ u
= (1 + PX )(1 + Pβ )
Wr (1 + Pλ )
(3.13)
PX = e ( X u − X r ) β u − 1 ,
ˆ
Pβ = e X r ( β u − β r ) − 1
ˆ
ˆ
and
Pλ = e γˆu λ u −γˆr λ r − 1
represent
the
contribution of the three sources of income gap between migrants and non-migrants
respectively.
3.2 Data description
The data we use in this article come from the research project "Migration and regional
development", which is financed by the China National Social Science Fund.
3.2.1
Sample restrictions
We did our survey in Hubei province in 1993. Our sample includes 2796 households,
which dispatches in 21 communities. In each household, we randomly choose one member of
at least 15 years old to answer the household questionnaire and the individual questionnaire.
After cylindering the data, we obtain 2573 valid observations. Here, we define migration as
the change of usual residency between towns or townships. The survey was executed in the
actual residency place. To the migrants, the survey holds on the destination place of their final
movement. Using a life history table, the survey registers the migration history from 15 years
old. However, in term of the other information about the household, only their actual situation
is registered because it is difficult to catch their evolution. To simplify our study, we consider
someone as a migrant if his or her birthplace is different from his or her actual residency
10
place. According to this definition, we divide our total sample into several groups, as
indicated in table 3.1.
Table 3.1 – Sample division
Actual residency place
Urban
Rural
Migrants
Urban to urban migrants
Rural to urban
Permanent
migrants
Temporary
Rural to rural Migrants
Urban to urban migrants
Non-migrants
Urban non-migrants
Rural non-migrants
The rural to urban migrants can be regrouped into two sub-groups: permanent
migrants and temporary migrants. The criterion is the change of the civil registration (hukou)
place. Generally speaking, we define the migrants whose usual residency place and the
registration place are different as the "temporary migrants"; on the contrary, we define the
migrants who have moved with their civil registration as the "permanent migrants". There are
two types of civil registrations: agricultural registration and non-agricultural registration. A
permanent rural-urban migration signifies a change from agricultural registration to nonagricultural registration. Many researches have already proved that there exist great
differences between permanent migrants and temporary migrants. (Goldstein et alii, 1993,
Chang, 1996; Wu and Zhou, 1997; Fan, 1999; Ma, 1999; Wang and Zuo, 1999, Yang et alii,
1999).
Firstly, the authorities immediately control the permanent migration so the permanent
migration decision depends greatly on government plan. The theory of push-pull does not
work in this case. On the contrary, the temporary migration is a consequence of the market
economy. It is spontaneous and settled by rural and urban labor market. Government policies
cannot influence it directly.
Secondly, the permanent migrants are rather the qualified migrants, who are generally
integrated in employment programs and government social protection program. Normally,
they can get permanent and stable posts in urban formal sector and receive the advantages
provided by the government. However, most of the temporary migrants are the non-qualified
farmers from the countryside. They can just earn a living in informal urban labor market and
work hard as manual workers. Generally, they just get the inferior posts, which are refused by
the citizens.
Thirdly, the permanent migration is generally a definitive movement. Permanent
migrants have not much relationship with their original place. On the contrary, temporary
11
migrants link closely to their departure place and often keep their plots of land for the sake of
security. In a certain circumstance, they can return to the countryside and recapture their
former professions. In fact, the temporary migrants include some seasonal migrants.
The objective of this paper is to analyze the impact of income gap on migration
behavior from countryside to city. Here, we are interested in temporary migration, whose
decision is made by the migrant himself or herself (or his or her own family) and depends
essentially on the labor market situation. In the following analysis, our sample restraints to the
temporary rural-urban migration. Among the temporary migrants, some of them have already
settled down in the cities for a rather long time. Though they keep their agricultural
registration, their living and employment styles almost match those of the urban inhabitants.
In consequence, we remove the migrants who have migrated to cities for more than five years
from our sample. In other words, we only keep the migrants who migrated to the destination
cities after 1988. We consider all that stay in the countryside, including the rural-rural
migrants, as the reference group, and we define them as the non-migrants. It is the reason why
there are only 1476 individuals in our sub-sample, including 474 rural-urban migrants and
1002 non-migrants.
3.2.2
Variables
Many researchers have shown that migration is strongly selective according to sex:
men and women face different labor market situations (Agesa et alii, 1999; Yang, 1999). So,
we estimate two separated models for each of the two sexes.
We have two types of equations: income equation and selection equation. In income
equation, the dependent variable is the natural logarithm of monthly income, measured by
Yuan (the Chinese money). In the questionnaire, the corresponding question was "what is
your average monthly income".
We introduce the following into income equation as independent variables: age,
squared age, education level, civil registration type (agricultural or non-agricultural), actual
profession and per capita GDP of the actual residency.
When it comes to the selection equation, the dependent variable is binary: it takes the
value of 1 when the individual has migrated and the value of 0 if the individual stays in
countryside.
The independent variables of the selection equation include: age, education level,
marriage situation before migration, family size, the number of brothers and sisters (a proxy
12
of the clan size), the eldest status, the size of cultivable land of the household and the per
capita GDP (in 1991) of the actual residency place.
Table 3.2 shows the average value of explanatory variables of the migrants and the
non-migrants according to their sexes.
Table 3.2 – Average value of explanatory variables
Age
Years of schooling
Education level (%)
Primary school a
Junior secondary school a
Senior secondary school a
Agricultural registration a (%)
Number of brothers and sisters
Eldest status (%)
Married (%)
Before migration a
Actual a
Type of ownership (%)
State-owned a
Collective-owned a
Self-employed individual a
Actual profession (%)
Worker a
Farmer a
Specialist and executive a
Retail trade and service personnel a
Monthly income (yuan)
Cultivable land size of the household (mu b)
Per capita GDP of the actual residency (yuan)
Males
Migrants
Nonmigrants
29.64
36.68
7.06
5.98
Females
Migrants
Nonmigrants
28.57
34.04
6.37
5.01
16.41
56.64
21.88
92.97
4.38
31.64
34.82
42.09
15.83
93.85
4.38
32.77
35.32
45.41
13.76
91.74
4.44
24.77
36.77
32.90
12.90
95.05
4.86
27.53
38.28
67.58
86.59
41.28
65.60
87.31
3.52
5.08
76.56
3.35
24.02
5.77
4.59
5.96
68.81
1.94
11.61
2.80
5.47
11.32
3.91
73.83
256.08
3.14
1847.77
14.53
63.69
8.38
6.89
150.22
4.61
1179.49
9.63
11.01
1.38
70.64
182.14
2.97
1675.26
6.88
81.29
2.80
4.30
107.27
4.27
1179.03
218
465
Number of observations
256
Note : a – Dummy variable. b – One mu is equal to 1/15 hectare.
537
From table 3.2, we find that, migrants are younger than non-migrants; migrants are
more educated than non-migrants and men are more educated than women. Among the male
migrants, the average schooling year is 7, which is equal to the junior secondary school level.
In terms of employment situation, most of the individuals are employed by individual
enterprises, among which there are many self-employed workers. In rural areas, many of the
non-migrants are employed by the collective enterprises, most of which are Township and
Village-ship Enterprises. Most of the migrants are employed in tertiary sector. Though we
cannot precisely identify the sectoral belonging of each individual from the type of the
enterprise they work for and their career, the large proportion employed by individual
13
enterprises and tertiary sector confirms the existence and the importance of informal sector to
a certain degree. The incomes of migrants are higher than those of non-migrants. The first
surpasses the latter in 70%. The men's income are higher than the women's.
4 Empirical results
In this section, we will analyze the determinants of the migrants' income and the effect
of income gap on migration decision, then we will study the sources of this gap.
4.1 Determinants of the migration decision
We firstly estimate a reduced form probit equation, which includes the independent
variables of the migration decision function and those of the income function (see regression
5 and regression 6 in table 4.3). Then, we estimate the urban income equation and the rural
income equation respectively, introducing the inverse Mills ratios which come from the
reduced form probit equation to correct the sample selection bias.
Table 4.1 shows the estimation results. Regression 1 shows the results of the male
migrants. The relation between income and age is inverse U form: at the beginning, income
increases with the increase of age, when age reaches its optimal level, income reaches its
maximum, then as age continues to increase, income decreases. This result corresponds to the
results of the other researches (Agesa et alii, 1999; Li 1997). On the one hand, age reflects the
accumulation of human capital, which includes the setting up of personnel connections and
the accumulation of experience (Li 1997: p.1012,1020). So, the increase of age favors the
increase of income. On the other hand, most of the rural to urban temporary migrants are the
non-qualified workers, they can only find the hard manual job. So, evidently, the old workers
will have some disadvantages.
The effect of agricultural registration is not significant. On the one hand, it may be
caused by the fact that not many temporary migrants are of non-agricultural registration; on
the other hand, to a certain degree, it reflects that the influence of the civil registration system
becomes less important as the market economy becomes more and more developed.
Education plays an important role in individual income determination. The results
above confirm that formal instruction level favors individual income. We find that as
education level increases, the value of its coefficient increases gradually.
14
Table 4.1 – Logarithmic income equation adjusting for sample selectivity
Dependent variable: Logarithmic income
Males
Migrants
Nonmigrants
Age
Age2 (/100)
Agricultural registration
Years of schooling
Females
Migrants
Nonmigrants
Regression 1
Regression 2
Regression 3
Regression 4
0.035**
0.046***
0.009
0.071***
(1.971)
(3.191)
(0.332)
(4.503)
-0.041*
-0.057***
-0.013
-0.079***
(-1.731)
(-3.357)
(-0.363)
(-4.290)
-0.157
-0.123
0.189
0.209
(-1.001)
(-0.824)
(0.958)
(1.288)
0.070***
0.105***
0.051**
0.061***
(4.107)
(6.889)
(2.452)
(4.096)
Actual Profession
Worker
Specialist and executive
Retail trade and service personnel
Per capita GDP of actual residency (/100)
Inverse Mills Ratio
Constant
R2
0.488**
0.162*
0.441*
0.301**
(2.530)
(1.618)
(1.869)
(2.003)
0.548**
0.218*
1.170**
0.778***
(2.574)
(1.667)
(2.412)
(3.810)
0.657***
0.573***
0.408**
0.625**
(4.791)
(2.657)
(1.988)
(2.129)
0.026***
0.083***
0.002
0.091***
(5.149)
(9.872)
(0.348)
(11.074)
0.023
-0.407*
-0.184
-0.074
(0.210)
(-2.313)
(-1.194)
(-0.425)
3.308***
2.351***
3.973***
1.318***
(7.532)
(6.621)
(6.303)
(3.274)
0.329
0.313
0.143
0.320
250
536
209
463
Number of observations
Note: The t-students are presented in parentheses.
*** indicates coefficient significant at 1% level; ** indicates coefficient significant at 5% level;
* indicates coefficient significant at 10% level.
When it comes to employment, the results are different from our prediction. Whether
an individual belongs to the category of "specialist and executive", such as professors,
engineers, doctors, directors, etc., which we consider as the formal sector posts, does not play
an important role in his income determination. Perhaps, it is due to the following reasons: 1)
There are not many individuals who belong to this type, so their influence is not sufficiently
represented in the sub-sample. 2) The income of the individual who is employed by formal
sector is not only reflected by the monetary salary, which is measured in our survey.
However, if the individual belongs to the category of "retail trade and service personnel", his
15
or her income is higher. In fact, many workers in informal sector belong to this type, such as
the small-scale property owners. It implies that informal sector is not always characterized by
low salary. From the results, we find that "workers" earn the lowest salary. Generally
speaking, in the secondary sector, the rural to urban migrants occupy only the inferior posts.
They held only the hard and dangerous jobs, for example, they work as porter, pit worker,
etc., which are refused by the citizens. On the contrary, the tertiary sector provides them an
opportunity to earn a better salary, because this sector is much more diversified.
Finally, we find that the per capita GDP of the destination, which is a proxy of the
regional development level, favors the migrants' income.
Regression 3 shows that the age of female migrants does not significantly influence
their income level. However, the effects of their schooling years remain significant. In terms
of the effects of professional situation, to the female migrants, being "specialists and
executives" does increase their income. However, the regional development level does not
influence their individual income level. It seems that income of female migrants depends
essentially on their human capital.
For the non-migrants, regression 2 and 4 show that the relation between age and
income is the same as that of the male migrants. The education level always plays an
important role. The non-agricultural income is higher than the agricultural income, which
indirectly demonstrates the important role of rural non-agricultural sector on household
income in countryside. These results coincide with those of the other researches (Rizwanul et
alii, 1994; Zhao, 1999).
In order to quantify the marginal effects of the independent variables, we calculate the
arithmetic average of the total individual effects for each variable, using formula (3.6). The
results are shown in table 4.2. The introduction of the variable "square of age" demonstrates
that income reaches its maximum when age is 42-45 for all the sub-groups. One more year of
schooling leads to a 7% increase of income for the male migrants, and 5% increase for the
female migrants. The effect of the education level on income is greater to non-migrants than
to migrants, which implies that human capital return is higher in rural areas than in urban
areas. One possible explication is that most of rural-urban migrants held the inferior posts and
work as manual workers, it reduced the effect of education level. On the contrary, for the
farmers, “a higher education level favors the acquiring and the usage of certain modern
factors” (Schultz, 1964: p.176), so it is more probable for them to participate in nonagricultural activities, which significantly increases their incomes. However, this conclusion
16
is valid only in the case of the comparison between rural-urban temporary migrants and those
stay in rural areas.
Table 4.2 – Total effect of the explanatory variables of income function
(Arithmetic average of the total individual effects)
Migrants
#
Males
Non-migrants
#
Females
Migrants
Non-migrants
#
0.034
0.054
0.028
0.074
Age
#
#
2
-0.039
-0.065
-0.033
-0.082#
Age (/100)
-0.163
-0.087
0.237
0.216
Agricultural registration
#
#
#
0.070
0.113
0.051
0.061#
Years of schooling
Actual profession
0.489#
0.156#
0.529#
0.313#
Worker
0.544#
0.237#
1.156#
0.776#
Specialist and executive
#
#
#
0.627
0.747
0.671
0.663#
Retail trade and service personnel
#
#
0.025
0.085
0.013
0.093
Per capita GDP of actual residency (/100)
Note : # indicates that the direct effect of this variable is significant at least at the 10% level in Table
4.1.
From income equations, we can predict urban and rural income, so we can predict the
income gap for a given individual. It enables us to study the impact of this gap on migration
decision by the structural probit equation estimation. The last two columns of table 4.3
(regressions 7 and 8) show the estimation results. We find that, for the two sexes, the relations
between age and migration probability are both inverse U forms. Education level plays a
positive role in migration decision only for the males, but not for the females. The per capita
GDP of the actual residency constitutes an attraction force to the migrants. On the contrary,
the lack of land drives as a repulsion force of the original place. Once married, the migration
probability is strongly reduced: the setting up of a family and the birth of children, mark the
start of a more stable live, which increases the migration cost. Finally, our results confirm the
famous prediction of Todaro: the income gap is an important determinant of migration
decision.
17
Table 4.3 – Probability of migration
Dependent variable: Migrant=1, Non-migrant=0
Age
2
Age (/100)
Agricultural registration
Years of schooling
Actual Profession
Worker
Specialist and executive
Retail trade and service personnel
Per capita GDP of actual residency (/100)
land size of the household
Married
Household size
Number of brothers and sisters
Eldest status
Reduced form equation
Males
Females
Structural equation
Males
Females
Regression 5 Regression 6
Regression 7 Regression 8
0.087**
0.166***
0.063**
(2.504)
(3.908)
(1.961)
(5.989)
-0.085**
-0.176***
-0.084**
-0.286***
(-2.070)
(-3.443)
(-2.199)
(-5.572)
0.367
0.422
(1.431)
(1.392)
0.079**
-0.434
0.108***
0.032
(2.389)
(-0.138)
(3.639)
(1.189)
-0.064
0.784***
(-0.277)
(3.058)
0.192
-0.120
(0.752)
(-0.240)
1.804***
2.311***
(10.931)
(12.031)
0.097***
0.094***
0.203***
0.415***
(6.692)
(5.615)
(12.044)
(8.609)
-0.112***
-0.116***
-0.066***
-0.053***
(-4.809)
(-6.077)
(-3.061)
(-2.802)
-1.783***
-1.912***
-0.816***
-0.662**
(-9.083)
(-8.715)
(-4.093)
(-2.526)
-0.017
0.086
-0.016
0.021
(-0.287)
(1.317)
(-0.289)
(0.393)
0.011
-0.027
-0.037
-0.023
(0.120)
(-0.274)
(-0.427)
(-0.271)
-0.037
0.212
-0.007
-0.107
(-0.231)
(1.168)
(-0.045)
(-0.695)
2.884***
(8.927)
4.116***
(6.493)
Income gap ( log Wˆ ui − log Wˆ ri )
Constant
0.265***
-3.408***
-4.743***
-4.556***
-10.438***
(-4.537)
(-4.685)
(-6.465)
(-7.127)
-210.109
-170.393
-244.613
-246.955
Log-likelihood
88.3
91.2
86.8
83.2
Percentage of correction predictions (%)
793
683
793
683
Number of observations
Note: The t-students are presented in parentheses.
*** indicates coefficient significant at 1% level; ** indicates coefficient significant at 5% level;
* indicates coefficient significant at 10% level.
18
In figure 4.1, we illustrate the relation between predicted migration and income gap.
Though several particular groups cause some perturbations, the result confirms the important
effect of income gap on migration probability in general.
Estimated
migration
probability
1
3.0e-07
-.672151
.504758
Income gap
 Wˆ
log10  ui
ˆ
 Wri




Figure 4.1 – Relation between probability of migration and income gap
4.2 Income gap decomposition
Having estimated urban and rural income, we can calculate the geometric average
value of the urban income and that of the rural income respectively. Then we can deduce the
ratio of urban to rural income and investigate in the decomposition of income gap. Agesa et
alii (1999) have used this technique to analyze the income gap between migrants and nonmigrants. They found that rural to urban wage gap is larger for males than for females, and
more importantly, the gain in returns to productivity-enhancing attributes is larger for males
than for females. The two above findings explain the fact that males have a higher incentive
to migrate to urban areas (in greater number than females) due to their larger gains in returns
to migration.
19
Table 4.4 shows our results: the geometric average of urban income and rural income
is respectively 204 yuans and 106 yuans for the male, 138 yuans and 77 yuans for the female.
So the ratio of urban to rural income is 1.929 for the male and 1.786 for the female. It
suggests that the rural to urban wage gap is larger for males than for females. However, the
correction of the selection bias changes the story: the ratio becomes 1.751 for the male and
1.899 for the female. It is thus contrast to the conclusion of Agesa et alii (1999, p.52). Our
results suggest that rural-urban income gap is larger for the female than for the male. In other
words, the female, compared to the male, will have greater profit to migrate. So the migration
propensity is higher for the female than for the male, which explains that there is no sensible
difference between the ratio of the male migration and that of the female in our case.
Table 4.4 – Income gap decomposition
Males
Before correction for sample selection bias
~
Geometric mean of urban income, Wu = e β u X u +γ u λ u (yuan)
ˆ
~
ˆ
Geometric mean of rural income, Wλ = e β r X r +γ r λ r (yuan)
ˆ
ˆ
~
Wu
Relative income, R = ~
Wr
Females
204
138
106
77
1.929
1.786
79.2
42.1
-2.3
33.6
10.2
-5.9
202
149
115
79
1.751
1.899
Contribution of the various components of income gap (%)
( X −X )β
Difference in attributes, PX = e u r u − 1
ˆ
Difference in returns to attributes, Pβ = e X r
( βˆu − βˆ r )
Sample selection bias, Pλ = e γ u λ u −γ r λ r − 1
ˆ
ˆ
After correction for sample selection bias
~
Wu
Geometric mean of urban income,
Geometric mean of rural income,
Relative income, R = ~
'
~
Wu
e γˆu λ u
~
Wr
e γˆr λ r
Wr (1 + Pλ )
(yuan)
(yuan)
−1
When it comes to the decomposition of income gap, we find that there is a great
divergence between the two sexes. For the male, the major composition is the portion due to
differences in attributes between urban and rural labors. It implies that a male farmer will not
earn a higher income in cities than in countryside except that he has some individual
attributes, which ensure that he can get a good job in cities. For the female, it is a different
20
case. Two factors, the differences in worker attributes in rural and urban areas, and the rural to
urban differences in returns to attributes both play an important role. Given the fact that, in
the countryside, quite many women participate only partially in the productive activities and
they are mainly in charge of housework, which does not create explicative income, and thus
their income is under-represented. However, in cities, most of the women have to go out to
work for the reasons of living: the salary of the men is generally not high enough to support
the whole family. So, the working rate is similar between females and males in urban areas.
The monetary yield to the attributes of the females is higher in cities than in countryside.
What's more, the females have further advantages in some informal sectors (Todaro,1997:
p.270, 276), so that the possibility that they can find a job is higher than the males, no matter
whether they have some individual attributes or not.
5 Conclusions
This paper provides some economic explanations for the actual rural-urban migration
decision in China. Our results confirm the prediction of the Todaro model: the migration
decision depends greatly on the rural-urban income gap. During a long period, the surplus
agricultural labor was artificially kept in rural areas and the disequilibrium of agricultural
labor and cultivable land aggravated. In consequence, the rural income staged in low level and
the rural-urban income gap stayed large. It is difficult to change this situation radically in a
short time, for there is the contradiction between the large rural population size and the
limited urban absorption capacity. The rural-urban income gap thus plays a dominant role in
the migration motivation of farmers for a long time.
As to the determinants of migrants’ income, we find that: 1) There exists an inverse U
form relation between the labor’s age and his income level. 2) The education level
significantly favors income as well as probability to migrate. 3) To a great degree, income
level is determined by type of propriety and that of profession. In particular, in rural areas,
income of the non-agricultural labor is significantly higher than that of the agricultural labor.
4) The urban informal sector is not necessarily characterized by low income, which is
different to Todaro theory.
Our results suggest that the urban to rural income gap is larger for females than for
males. This finding is consistent with the fact that females have an important incentive to
migrate to urban areas due to their larger gains in returns to migration. For males, their
21
income gap is mainly determined by the differences in attributes; for females, both individual
characteristics and the gain in returns to attributes play an important role.
Three policy implications may be drawn out: firstly, given the fact that there exists a
large urban-to-rural income gap in China, the number of migratory workers would be higher
without artificial barriers. It is a dilemma for the government to deal with. On the one hand, in
the absence of administrative control, there may be excessive expansion of urban population;
on the other hand, economic theory holds that migration can enhance efficiency by
reallocating labor from the activities of low productivity to those of high productivity. The
current system therefore reduces the overall productivity of labor, aggravates the urban-rural
distortion and causes a tremendous loss of social resources.
Secondly, the participation of non-agricultural activities significantly increases the
rural household income. Compared with the rural-urban migration, it entails much less costs
and risks. In consequence, we think that the development of the Township and Village-ship
Enterprises constitutes an important solution to the problem of the rural surplus labor.
Finally, the influence of education level (or human capital) on income is significantly
positive, to migrants as to non-migrants. According to some researchers as Yap (1976:
p.241)2, income levels seem to be a function of an individual’s human capital endowments
rather than of his migration status. Accordingly, strategies to alleviate poverty should place
more emphasis on raising the skill levels of the urban population than on restricting migration
to the cities. In short, it implies that it is important to invest in human capital in rural areas to
increase the productivity of rural and urban labor. Schultz (1964: p.176) pointed out: the
human capital is a major source of economic growth in agriculture.
However, this paper only tested the prediction of Todaro model. As suggested by
Stark (1991), some other factors predicted by “non-Todaro” theories play also an important
role in migration decision. We can study the migration behavior from the household aspect. In
addition, given the important role of the rural non-agricultural sector play on rural income, it
may be considered as an alternative solution to the problem of the agricultural labor surplus in
China. The impacts of the participation in non-agricultural activities on rural household
income will be an interesting subject of future studies.
2
Cited by Williamson (1988 : p.448),
22
References
AGESA, J. and AGESA, R.U., 1999, «Gender Differences in the Incidence of Rural to Urban
Migration: Evidence from Kenya », The Journal of Development Studies, Vol.35, No.6,
August 1999, p.36-58.
AUBERT C., 1995, « Exode rural, exode agricole en Chine, la grande mutation? », Espace
Populations Société, 1995-2, p.231-245.
BARDHAN, P. and UDRY, C., 1999, Development Microeconomics, Oxford, Oxford University
Press, 236 p.
CHANG S., 1996, « The Floating Population : An Informal Process of Urbanisation in China »,
International Journal of Population Geography, Vol. 2, 1996, p.197-214.
DCSNBS, 1999, Comprehensive Statistical Data and Materials on 50 Years of New China, Beijing,
China Statistics Press, 890 p.
FAN C.C., 1999, « Migration in a Socialist Transitional Economy: Heterogeneity, Socioeconomic and
Spatial Characteristics of Migrants in China and Guangdong », International Migration
Review, Vol.33, No.4, Winter 1999, p.0954-0987.
GHATAK, S., LEVINE, P. and PRICE, S.W., 1996, « Migration Theorie and Evidence : An
Assessment », Journal of Economic Surveys, Vol. 10, No. 2, p.159-197.
GOLDSTEIN, A. and GOLDSTEIN, S, 1993, « Determinants of permanent and temporary migration
in Hubei province, PRC », IUSSP Proceedings.
GREENE, W.H., 1997, Econometric Analysis, New Jersey, Prentice-Hall Tnternational, Inc., 1075 p.
GUILLAUMONT, P., 1985, Economie du développement (Tome 2), Paris, Presses Universitaires de
France, 605 p.
HECKMAN, J., 1979, « Sample selection bias as a specification error », Econometrica, Vol.47, p.15361.
HARRIS, J. and TODARO, M.P., 1970, « Migration, Unemployment and Development : a TwoSectors Analysis », American Economic Review, Vol. 60, p.126-142.
LEILA F., 1999, « Labor allocation of Chinese Rural Migrant Workers in Urban Areas: Job Election
or Job Enforcement? », Asian and Pacific Migration, Vol. 8, No. 3, 1999, p.329-40.
LEVY, M.B. and WADYCKI, W.J., 1974, « What is the opportunity cost of moving ?
Reconsideration of the effects of distance on migration », Economic Development and
Cultural Change, Vol.22, No.2, p.198-214.
LEWIS, W.A., 1954, « Economic Developement with Unlimited Supply of Labour », The Manchester
School of Economic and Social Studies, vol. 22, no. 2, p.139-91.
LI S., 1997, « Population Migration, Regional Economic Growth and Income Determination : A
Comparative Study of Dongguan and Meizhou, China », Urban Studies, Vol. 34, No. 7, 1997,
p.999-1026.
LUCAS, R.E.B., 1988, « Migration from Botswana », Economic Journal, Vol.95, p.358-82.
MA, Z., 1999, « Temporary migration and regional development in China », Environment and
Planning, A 1999, volume 31, p.783-802.
NAKOSTEEN, R.A. and ZIMMER, M.A., 1980, « Migration and Income : The Question of SelfSelection », Southern Economic Journal, Vol.46, 1980, p.840-51.
OAXACA, R., «Male-female wage differentials in urban labor markets », International Economic
Review, Vol.14, No.3, October, 1973, p.693-709.
23
OAXACA, R. and RANSOM M.R., 1994, «On discrimination and the decomposition of wage
differentials », Journal of Econometrics 61 (1994), p.5-21.
PERLOFF, J. M., 1991, « The Impact of Wage Differentials on Choosing to Work in Agriculture »,
American Journal of Agricultural Economics, Vol.73, No.3, August 1991, p.671-80.
REIMERS, C. W., 1983, « Labour Maket Discrimination Against Hispanic and Black Men », Review
of Economics and Statistics, Vol.65, No.4, p.570-79.
RIZWANUL, I. and JIN, H., 1994, « Rural Industrialization : An Engine of Prosperity in Postreforme
Rural China », World Development, Vol. 22, No. 11, p.1643-1662.
ROUBAUD, F. 1992, « La dynamique du secteur informel urbain au Mexique : le rôle de la mobilité
intersectorelle », Revue Tiers Monde, t. XXXIII, No.132, Octobre-Decembre 1992, p.893-924.
SCHIFF, M., 1996,«South-North Migration and Trade», The World Bank working paper 1696, 55 p.
SCHULTZ, T.W., 1964, Transforming Traditional Agriculture, New Haven and London, Yale
University Press, 212 p.
STARK, O., 1991, The Migration of Labor, Oxford, Basil Blackwell Inc., 406 p.
TAYLOR, J.E., 1987, « Undocumented Mexico-U.S. Migration and the Returns to Households in
Rural Mexico », American Journal of Agricultural Economics, Vol.69, No.3, p.626-38.
TODARO, M. P., 1969, « A Model of Labour Migration and Urban Unemployment in Less
Developed Countries », American Economic Review , vol 59, no. 1, p138-148.
TODARO, M.P., 1976, Internal migration in development countries : A review of theory, evidence,
methology and research priority, Geneve, BIT.
TODARO, M. P., 1997, Economic Developpement, London, Longman, 738 p.
WANG, F. and ZUO, X. (1999), « Inside China's Cities : Institutional Barriers and Opportunities for
Urban Migrants », AEA Papers and Proceedings, May 1999, p..
WILLAMSON, J.G., 1988, « Migration and Urbanization », in H. Chenery and T.N. Srinivasan (eds),
Handbook of Development Economics, Vol. I, Elsevier Science Publisher B.V., 848 p.
WU H. X. and ZHOU L., 1997, « Rural-to-Urban Migration in China », Asian-Pacific Economic
Literature, 1997, p.54-67.
YANG X., 1999, « Gender Differences in Determinants of Temporary Labor Migration in China: A
Multilevel Analysis », International Migration Review, Volume 33, Number 4, Winter 1999,
p.929-953.
YAP, L., 1976, « Rural-urban migration and urban underemployment in Brazil », Jounal of
Development Economics, Vol.4, p.227-243.
ZENOU, Y., 1995, « Chômage urbain et migration dans les pays en développement: une approche
théorique », Revue d'économique politique, Vol.105, No.2, Mars-Avril, 1995, p.293-313.
ZHAO, Y., 1999, « Labor Migration and Earnings Differences : The Case of Rural China », Economic
Development and Cultural Change, Volume 47, Number 4, July 1999, p.767-782.
24