Great Expectations? The Subjective Well-being of Rural–Urban Migrants in China

World Development Vol. 38, No. 1, pp. 113–124, 2010� 2009 Elsevier Ltd. All rights reserved

0305-750X/$ - see front matter

www.elsevier.com/locate/worlddevdoi:10.1016/j.worlddev.2009.03.002

Great Expectations? The Subjective Well-being of Rural–Urban

Migrants in China

JOHN KNIGHTUniversity of Oxford, UK

and

RAMANI GUNATILAKA *

Monash University, Melbourne, Australia

Summary. — This paper is among the first to link the literatures on migration and on subjective well-being in developing countries. Itposes the question: why do rural–urban migrant households settled in urban China have an average happiness score lower than ruralhouseholds? Three basic hypotheses are examined: migrants had false expectations about their future urban conditions, or about theirfuture urban aspirations, or about their future selves. Estimated happiness functions and decomposition analyses, based on a 2002 na-tional household survey, indicate that certain features of migrant conditions make for unhappiness, and that their high aspirations inrelation to achievement, influenced by their new reference groups, also make for unhappiness. Although the possibility of selection biasamong migrants cannot be ruled out, it is apparently difficult for migrants to form unbiased expectations about life in a new and differentworld.� 2009 Elsevier Ltd. All rights reserved.

Key words — aspirations, China, happiness, relative deprivation, rural–urban migration, subjective well-being

1. INTRODUCTION

This paper contributes to the voluminous literature on rur-al–urban migration in developing countries. It does so from anew angle—by examining the subjective well-being of respon-dents living in migrant households in China. It raises an inter-esting puzzle. The normal assumption of migration theory isthat rural people migrate in order to raise their utility, at leastin the long run. Yet our sample of migrants has a mean hap-piness score of 2.3, well below the mean score of the rural sam-ple (2.7) and also below that of the urban sample (2.5). Ofcourse, initial hardship is to be expected—and indeed it is pre-dicted by the models—but these are migrants who have estab-lished urban households and whose average urban stay is noless than 7.5 years. Did the migrants come with excessivelygreat expectations?

Section 2 briefly explains the theory of the decision to mi-grate, and the implications of models based on the theory. Sec-tion 3 provides background on migrants in China anddescribes the data to be analyzed, and Section 4 provides thehypotheses to be examined. Section 5 reports the results fromour estimates of happiness functions. Section 6 considerswhether the relatively low happiness of migrants is due toself-selection. Section 7 summarizes and concludes.

* We are grateful to three referees for helpful comments, to the Nuffield

Foundation for supporting the research from its Small Grants Scheme,

and to the Global Poverty Research Programme of the UK ESRC for a

grant which helped to fund the collection of data on subjective well-being

in the survey on which the research is based. Final revision accepted:March 27, 2009.

2. MODELS OF MIGRATION

The economic literature on internal migration in developingcountries is surveyed in Lucas (1997): there is emphasis on util-ity as providing the motive for migration but there is no men-tion of measured happiness. The original probabilistic modelof rural–urban migration (Todaro, 1967) has spawned a pleth-ora of models, but almost all of them have the following

113

features in common. First, the model is based on the assump-tion either of utility maximization or of income maximizationwhere income serves as a proxy for utility. Second, the workeris assumed to migrate if the discounted present value (in termsof utility or income) of migrating to the urban location (givenan expected length of urban stay) exceeds the discounted pres-ent value of remaining in the rural location. Third, the migranttakes into account the probability of obtaining a desired urbanjob in any period, and the need to remain unemployed or in alow-income activity until such a job is found. Thus, fourth, therelevant urban discounted present value is an ‘‘expectedvalue,” represented by the wage in a desired urban job multi-plied by the probability of obtaining one, this product beingdiscounted by the degree of risk aversion.

Consider the implications for migration equilibrium. Withunrestricted migration and no transaction costs, migrationtakes place until the rural supply price, as defined above, isequal to this urban ‘‘expected wage.” Marginal migrants areindifferent as to whether they migrate or stay, expecting thesame utility from both actions, but intra-marginal migrantsbenefit from their migration in the form of economic rents. Gi-ven that the expected wage incorporates an initial period ofsearch and some probability of failure, migrants who haveundergone initial hardship and secured a desired job shouldalso derive a net benefit. Thus, migration models predict thatmigrants who have made the transition into urban employ-ment and living have more utility than they would have

http://dx.doi.org/016/j.worlddev.2009.03.002

114 WORLD DEVELOPMENT

received had they remained at home. Can this prediction bereconciled with the relatively low happiness that we observe?

An alternative theoretical approach to rural–urban migra-tion is to view it as income diversification in order to reducethe risks faced by households in a poor rural environment(Stark, 1991, chap. 5). Here the migrant is assumed to remaina part of the rural household. In principle, urban income,being imperfectly correlated with rural income, should reducethe variability of household income. Insofar as householdmembers place a value on income stability, the subjectivewell-being of the household as a whole should be raised evenif household income is not. However, it is possible that migra-tion redistributes happiness within the household, with the mi-grant becoming less happy and non-migrants becoming morehappy as a result of migrant remittances and of the greatersecurity, even if not greater income, that these provide. Canthis interpretation be reconciled with the settled nature ofthe migrant households and with their income net of remit-tances?

Another view of rural–urban migration is that it stems fromthe ‘‘push” of rural poverty rather than from urban ‘‘pull” fac-tors (for instance, Sabates-Wheeler, Sabates, & Castaldo,2008). There is a conceptual difficulty in distinguishing pushand pull, as theory predicts that it is the expected rural–urbandifference which provides the incentive for migration. We canstandardize for the observed characteristics of migrants incomparing rural and urban incomes and happiness. However,we cannot standardize for unmeasured characteristics, such asown or family misfortune, unhappy personality, or bad rela-tionships. Could unobserved idiosyncracies explain the lowhappiness of migrants that we observe?

3. THE BACKGROUND AND THE DATA

The phenomenon of rural–urban migration in China hasbeen different from that in most other poor countries (Cai,Park, & Zhao, 2008; Knight & Song, 1999, chaps. 8 and 9).During the period of central planning the movement of peo-ple, and especially movement from the communes to the cities,was strictly controlled and restricted. Even after the com-mencement of economic reform in 1978, migration was verylimited although temporary migration was permitted when ur-ban demand for labor exceeded the resident supply. The sys-tem of residential registration (hukou) initially prevented andlater hindered rural people from settling in the cities. Thehardships and disadvantages that temporary migrants facedin the cities caused many to prefer local non-farm jobs when-ever they were available (Zhao, 1999). When, increasingly, mi-grants began to settle in the cities with their families, they weresubject to discrimination in access to jobs, housing, education,and health care. City governments favor their own residents,and migrants are generally treated as second class citizens(Knight & Song, 1999, chap. 9; Knight & Song, 2005, chaps.5 and 6; Solinger, 1999). For instance, they are allowed onlyinto the least attractive jobs that urban hukou residents shun;many enter self-employment, which is less regulated. Althoughthe labor markets for urban and rural hukou workers have be-come less segmented over time, the degree of competition be-tween them remained very limited in 2002 (Knight & Yueh,2008).

Despite these drawbacks, rural–urban migration has bur-geoned as the controls on movement have been eased andthe demand for urban labor has increased. Two official esti-mates of the stock of rural hukou migrants in the cities placethe number at 81 and 84 million people in 2002 (Cai et al.,

2008, p. 192). It is very likely the case that we are observing‘‘the greatest migration in human history.” Although a largepercentage of migrants have chosen to come temporarily tothe cities with the intention of returning home, an increasingpercentage wish to settle in the cities, and are establishing ur-ban households. Thus, many are revealing their preferencesfor urban living.

In this study we examine a sample of rural–urban migrantsliving in households. As we shall see, these are migrants whoare settled in the cities. The sample was collected as part ofa national household survey, organized by the Institute ofEconomics, Chinese Academy of Social Sciences, and designedby Chinese and foreign scholars. The survey was conducted bythe National Bureau of Statistics early in 2003 and its informa-tion generally relates to 2002. It is described in Gustafsson, Li,and Sicular (2008, pp. 331–353). There is very little panel (re-call) element in the survey, and none that we can use. The ur-ban and rural samples are sub-samples of the official annualnational household survey. However, because the official ur-ban survey does not yet cover rural hukou households, the rur-al–urban migrant sample was based on a sampling ofhouseholds in migrant neighborhoods in the selected cities.

The migrant survey contains a great deal of informationabout the household and each of its members, including in-come, consumption, assets, housing, employment, labor mar-ket history, health, education, and rural links. Less commonly,various migrant attitudes and perceptions were explored. Thesame question was asked of one of the adults in each sampledhousehold: ‘‘Generally speaking, how happy do you feel?.”The six possible answers were: very happy, happy, so–so,not happy, not happy at all, and do not know. This is a keyvariable in our analysis.

The analysis must be qualified at the outset. We have to relyon a single cross-section dataset, albeit rich in relevant infor-mation. The data lack a panel element, including previous in-come and happiness, in particular a measure of happinessprior to migration. These lacunae mean that some of our testswill be weak and inconclusive and some possible explanationswill be left unexplored. Nevertheless, the topic and questionsare sufficiently important and original to deserve a first at-tempt at answers. The analysis can in turn suggest directionsfor future research and provide pointers for improving fieldsurveys and hypothesis tests.

It is helpful first to provide descriptive information aboutthe migrants before presenting the happiness functions thatwill explain the determinants of subjective well-being. Con-sider the characteristics of those household members—77%of whom were the household heads—who responded to theattitudinal questions. Sixty-one percent were men, 90% weremarried, 93% were employed, and 88% were living with theirfamily. These respondents were generally not pessimistic aboutthe future: 7% expected a big increase in real income over thenext five years, 55% a small increase, 28% no change, and only10% a decrease. Rural links were commonly retained: 53% hadfamily members who still farmed in the village, 51% remittedincome to the village, and 32% had one child or more childrenstill living in the village.

Even though the mean happiness score of migrants is lowerthan that of rural people, their mean income is not. The aver-age income per capita of migrant households is 2.39 times thatof rural households. Even allowing for the smaller number ofdependents in migrant households by comparing total insteadof per capita household incomes, the ratio is still 1.54. The cor-responding ratios of household income per worker and ofwage income per employee are 2.01 and 3.02, respectively.Whichever concept is considered most relevant, migrants are

GREAT EXPECTATIONS? THE SUBJECTIVE WELL-BEING OF RURAL–URBAN MIGRANTS IN CHINA 115

at a considerable income advantage. This compounds the puz-zle: higher income appears not to raise happiness.

Table 1 shows the percentage distribution of migrantsamong the five categories of happiness. We see that over43% are happy or very happy, and that fewer than 12% are un-happy or not at all happy. On the scale of 4 for very happydown to zero for not at all happy, the mean score is 2.37. Withmigrant households divided into income per capita quintiles,the happiness score increases monotonically from 2.13 forrespondents in the lowest quintile to 2.56 for those in the high-est quintile.

The respondents in the categories unhappy and not at allhappy were asked the reason for their unhappiness. The pre-dominant reason, offered by over two-thirds of the respon-dents, is that income is too low. The next most importantreason, reported by over 11%, is uncertainty about the future,suggesting that insecurity is a problem. Discrimination againstmigrants is mentioned only by 2% of the respondents. The evi-dence suggests that income will be an important determinantof migrant happiness. Migrants were asked what they consid-ered to be the most important social problem. Lack of socialsecurity is the most common response, mentioned by 24% ofrespondents. Environmental pollution comes second (20%),corruption third (18%), followed by social polarization(11%), discrimination against migrants (10%), and crime (8%).

Migrants were asked: ‘‘Compared with your experience ofliving in the rural areas, are you happier living in the city?.”No fewer than 56% felt that urban living gave them greaterhappiness, 41% felt that it gave the same, and only 3% re-ported greater rural happiness. When asked what they woulddo if forced to leave the city, more migrants replied that wouldgo to another city (54%) compared to those who replied thatthey would go back to their village (39%). These results addto the list of questions. If most migrants view urban livingas yielding them greater happiness, and most wish to remainin an urban area, why are their mean happiness scores lowerthan those of rural residents?

4. HYPOTHESES

There are several possible explanations for our puzzle, giv-ing rise to hypotheses that we wish to test, or at least to ex-plore. Our first hypothesis is that migrants, when theydecided to migrate from the village, had excessively highexpectations of the conditions that they would experience inthe city. We shall look for evidence that this might be the caseby considering the characteristics of their urban life that re-duce their welfare.

Table 1. Percentage distribution of happiness,

Happiness category Overall

1st 2nd

Very happy 6.35 3.5 6.0Happy 37.33 27.8 35.0So–so 45.09 50.0 47.1Not very happy 9.47 15.4 10.3Not happy at all 1.76 3.3 1.5Total (number) 1985 396 397Mean happiness 2.37 2.13 2.34

Notes: Data for this table and for all subsequent tables are derived from the RThe data relate to the household member—normally the household head—whoquintile into which that household falls.The happiness score is based on cardinal values assigned to the qualitative assesand not at all happy = 0.

Second, the puzzle might be solved by recourse to the adap-tation theory that has been developed by Easterlin (2001). Hisargument is that happiness is a function of both income andaspirations, the former having a positive effect and the lattera negative effect. Moreover, as income rises over time, aspira-tions adapt to income, so giving rise to a ‘‘hedonic treadmill.”This account is consistent with the finding (e.g., by Easterlin,1974) that happiness rises with income in cross-section butdoes not do so in time-series datasets. Easterlin (2001), usingsuccessive cross-section surveys to create a synthetic panel,finds that the income of a cohort rises over the working lifeand then falls in retirement, but that its average happinessscore remains remarkably constant. His explanation drawson the psychological literature to make the distinction between‘‘decision utility” and ‘‘experienced utility”: the utility ex-pected at the time of making a choice and the utility subse-quently experienced from that choice. When respondents areasked to assess their happiness in the past, when their incomewas lower, they tend to judge it by their current aspirations forincome and therefore tend to report that their happiness waslower. Similarly, when they are asked to assess their happinessin the future, when they expect to have higher income, they donot realize that their aspirations will rise along with their in-come and therefore report that their happiness will be higher.Rabin (1998, p. 12) summarized the findings from social psy-chology thus: ‘‘we don’t always predict our own future prefer-ences, nor even accurately assess our experienced well-beingfrom past choices.” Easterlin (2001) marshals this evidenceas support for his argument that aspirations are a functionof income and tend to rise in proportion with income. 1

If current judgements about subjective well-being, whetherin the past, the present, or the future, are based only on cur-rent aspirations, this might explain why the mean happinessof migrants is lower than that of rural people: aspirationscould have risen after having made the decision to migrate.Aspirations might not be quantifiable, but the predictions ofthe theory can be tested. Similarly, we might also find anexplanation for why it is that migrants nevertheless generallyreport that happiness is higher, or at least no lower, in urbanareas than in rural areas. The second hypothesis, like the firstone, involves false expectations, but in this case the expecta-tions are about subjective aspirations rather than about objec-tive outcomes.

The notion that aspirations depend on income relative tothat of the relevant reference group, coming from the sociolog-ical literature (for instance Runciman, 1966) has been devel-oped for China in companion papers on subjective well-being (Knight & Gunatilaka, in press; Knight, Song, & Guna-tilaka, 2009). Other studies for developing countries which

overall and by income per capita quintile

Income quintile

3rd 4th 5th 5th–1st

5.8 6.5 9.8 6.339.3 42.1 42.5 14.743.6 42.8 42.0 �8.08.6 7.8 5.3 �10.12.8 0.8 0.5 �2.8397 397 3982.37 2.46 2.56 0.43

ural–urban Migrant Household Survey, 2002.responded to the question about happiness, and to the income per capita

sments as follows: very happy = 4; happy = 3; so–so = 2; not happy = 1;


show the importance of reference groups include Graham andPettinato (2002) (shifts in reference norms in Peru and Russia),Kingdon and Knight (2007) (comparison with close neighborsin South Africa), and Fafchamps and Shilpi (2008) (rural–ur-ban migrants retaining a village reference group in Nepal). Ifthe group with which the migrants compare themselveschanges as a result of rural–urban migration and urban settle-ment, this might explain why their aspirations change. We cantest whether migrants show relative deprivation in relation tourban society.

Our third hypothesis draws on theories of migration that arebased on decision-making by the rural family. It is that thepresence of members left behind in the village can place a bur-den on the urban members of the two-location family. Insofaras migrants remit part of their income, their own happinessscore might fall and that of their rural family might rise.Equivalently, our measure of the income per capita of the ur-

Table 2. Happiness functions of rural

Mean or proportion

(1)

Log of per capita householdincome 2002

8.55 0.197186***

Expect big increase in incomeover next 5 years

0.07 0.323740***

Expect small increase in incomeover next 5 years

0.55 0.047957

Expect decrease in income overnext 5 years

0.10 �0.399027***

Male 0.61 �0.275356**

Married 0.90 �0.00971Male and married 0.55 0.333382***

Education (years) 7.99 �0.00145Net financial assets (‘000 Yuan) 16.51 �0.00008Unemployed 0.01 �0.06881Working hours (‘00 per year) 31.94 �0.00138In good health 0.90 0.112137*

Duration of urban residence(years)

7.51 0.018281**

Duration of urban residence,squared

84.83 �0.000657**

Log of average per capita urbanincome in city of currentresidence

8.97

Log of average rural income inprovince of origin

7.81

Living with family members 0.88Number of relatives and friendsin city

7.19

Child still in village 0.32Ln remittances per capita 2.76House area per capita 11.39Living in own house 0.11No heating 0.65Constant 0.541127**

R2 0.077Number of observations 1930

Notes: Dependent variable: Score of happiness based on cardinal values assignso = 2; not happy = 1; and not at all happy = 0.Eqns. (1)–(3) are for the full sample. Eqns. (4) and (5) are based on sub-sampThe omitted categories in the dummy variable analyses are: single female; empnormal mood; and no change in income expected in the next five years.In this and subsequent tables, ***, **, and * denote statistical significance at thEqns. (2)–(5) have been clustered at city level for robust standard errors.

ban migrant household might overstate its disposable incomeper capita.

Fourth, our results might be explained by selection bias. Thelower mean happiness score of migrants may be the result oftheir, or of their households, having characteristics differentfrom those of the rural population as a whole. If this werethe case, they would indeed have been less happy on averagehad they remained in the village. Such happiness-reducingcharacteristics might be available in the dataset—and thuscapable of being conditioned on—or they might be unobserv-able to the researcher. For instance, it is possible that thoserural dwellers who by nature are melancholy or have highand unfulfilled aspirations hold their rural life to be responsi-ble and expect that migration will provide a cure. They mightthus be more prone to leave the village for the city. If the self-selected migrants are intrinsically less happy, this might ex-plain why the sample of rural–urban migrants has a lower

–urban migrants: OLS estimation

Full sample Below medianduration

Above medianduration

(2) (3) (4) (5)

0.208102*** 0.200896*** 0.129454*** 0.276563***

0.298398*** 0.295402*** 0.267288** 0.337325**

0.026176 0.0244 0.050754 �0.003458

�0.403299*** �0.402897*** �0.322104** �0.450602***

�0.268374** �0.270976** �0.361654*** 0.044724�0.05981 �0.060922 �0.267322* 0.292311

0.349128*** 0.351775*** 0.504235*** �0.020951�0.009834 �0.010387 �0.019562 �0.003055�0.000247 �0.000256 �0.000415*** 0.000648*

�0.056733 �0.058675 �0.314096 0.2937620.000093 0.000097 0.000159 �0.000113

0.123086** 0.124074** 0.026615 0.169086**

0.013580* 0.013154*

�0.000547* �0.000529*

�0.120432 �0.12766 0.005306 �0.280012**

0.070021 0.078382 0.124506 0.051889

0.134726 0.14282 0.207858** 0.1283470.003869* 0.003822* 0.007609 0.001566

�0.124977** �0.143516** �0.125387** �0.1131130.006364

0.00262 0.002745 0.000607 0.0033810.034593 0.036657 0.057436 0.026951�0.149865** �0.150708** �0.204220*** �0.116598*

1.024808 1.073999 0.465825 1.670172

0.100 0.101 0.091 0.1341850 1850 925 926

ed to qualitative assessments as follows: very happy = 4; happy = 3; so–

les selected according to the length of stay in urban areas.loyed or labor force non-participant; not healthy; in normal or worse than

e one percent, five percent and ten percent levels, respectively.


mean happiness score than the sample representative of therural population of which they were previously a part. Thatis a third form of false expectations, now based on self-misdi-agnosis. Its implications can be tested.

5. THE DETERMINANTS OF HAPPINESS

We estimate happiness functions in order to discover thedeterminants of happiness among rural–urban migrants. Thisenables us to test Hypotheses 1, 2, and 3. We proceed instages: first, we estimate OLS estimates of the happiness score,both with a basic specification and with a full set of explana-tory variables; second, we confine the sample to employed mi-grants, as this enables us to introduce a series of work-relatedvariables; and third, we estimate the same equations with theincome variable instrumented.

Table 2 reports, for the full sample, the basic model and theextended model with the full set of explanatory variables avail-able. Eqns. (1) and (2) show the coefficients of the basic equa-tion and the full equation, respectively. The asterisks showlevels of statistical significance. The coefficient on ln incomeper capita is significantly positive, and its values (averaging0.20) indicate that a doubling of income raises the happinessscore by about 0.14 points. Income is relevant, as predicted,but its effect does not appear powerful by comparison witheither the presumptions of economic theory or the effects ofsome other variables in the equations. However, expectationsof income over the next five years enter powerfully and signif-icantly: those expecting a ‘‘big increase” have a higher happi-ness score than those who expect income to remain the same,by 0.32 and 0.30, respectively, and those expecting a decreasehave a lower score, by �0.40 in both cases.

It appears that peoples’ current happiness depends partly ontheir expectations of future income. This result deserves inves-tigation because of its implications for economic theory. It isconsistent with the notion that people are efficient inter-tem-poral utility maximizers on the basis of their ‘‘permanent in-come,” that is, people derive their happiness from theircurrent consumption, and current consumption in turn isdetermined by their expectation of permanent income. How-ever, when Eqns. (1) and (2) of Table 2 were re-estimated withconsumption per capita replacing income per capita, the coef-ficients on the expected income variables were barely changed

Table 3. Happiness functions of employed rural–urb

Mean or proport

Satisfaction with job 1.98Index of discrimination 5.35Permanent or long-term contract work 0.05Temporary work 0.24Can find another job in two weeks 0.11Can find another job in a month 0.23Can find another job in 2 months 0.10Can find another job in 6 months 0.13Need more than 6 months to find another job 0.17

R2

N

Notes: Dependent variable: Score of happiness based on cardinal values assignso = 2; not happy = 1; and not at all happy = 0.The specifications of Eqns. (1) and (2) are identical to those of Eqns. (1) and (reported.The omitted categories in the dummy variable analyses reported are: self-empEqn. (2) has been clustered at city level for robust standard errors.

(equations not shown). There is the further possibility that thevariable denoting the expected change in income is endoge-nous, with naturally happy people being more optimisticabout the future. Since we were unable to find good instru-ments for expected income change, we adopted a different ap-proach. We estimated an equation with expected incomechange as the dependent variable, and city dummies (to reflecteconomic conditions) and a proxy for natural disposition to-ward happiness as the explanatory variables. As our proxywe used the individual residuals in the happiness function(based on Table 3, Eqn. (2)) because they are likely to be pos-itively correlated with unobserved personal disposition.Whether an ordered probit equation or a regression equationusing the cardinal measure (2, 1, 0, �1) of the four expecta-tion categories was used, the coefficient on the residual termwas never significantly positive. There is no evidence for theendogeneity explanation. We are left with an explanation interms of adaptation theory. The result is consistent with thefindings of the psychological literature as interpreted by Eas-terlin (2001), that is, people evaluate their future income onthe basis of their current aspirations, on the assumption thattheir future aspirations will not adjust to their future income.

We see that men have lower happiness, ceteris paribus (theircoefficients being �0.28 and �0.27 in Eqns. (1) and (2), respec-tively), and that marriage has a negligible effect for women buta positive and significant effect (0.33 and 0.35, respectively) formen. Surprisingly, none of years of education, net financial as-sets, unemployment, and hours worked has a significant effect.However, being in good health raises happiness.

We expect migrants to adjust over time to urban life in var-ious ways. On the one hand, as they overcome initial difficul-ties and become more settled, we expect their happiness to rise.On the other hand, their reference groups might change, fromthe, poorer, village society to the, richer, urban society, andthis fall in perceived comparative status might reduce happi-ness. The length of time spent in the urban area is introducedas an explanatory variable, and also its square so as to allowfor non-linearity in the relationship. Both the variable andits square are significant, the former positively and the latternegatively, although only at the 10% level in the full equation.The coefficients imply that the happiness score rises to a peakafter 14 years and then declines in the basic equation, and thatthe peak is reached after 12 years in the full equation. How-ever, it is possible that there is selection bias on the duration

an migrants: OLS estimation, selected variables

ion (1) (2)

0.080024*** 0.073527*

�0.039273*** �0.032196***

0.128294 0.133763�0.016973 0.007874�0.10626 �0.099676�0.133214** �0.121339**

�0.145655** �0.147820*

�0.183292*** �0.191704**

�0.237879*** �0.214012***

0.111 0.1291784 1715

ed to qualitative assessments as follows: very happy = 4; happy = 3; so–

2), respectively, of Table 2, but the variables contained in Table 2 are not

loyed and can find a job immediately.


variable: happier migrants are more likely to stay on in thecity. This would tend to bias upwards the returns to durationof urban residence. In summary, it would appear that mi-grants’ happiness tends to rise over several years of urban liv-ing, but the evidence is weak.

We turn to the variables that enter only in the full equation.In order to pursue the notion that reference groups can beimportant, we investigated the effect of relative income. Draw-ing on the urban and rural samples of the 2002 national house-hold survey, we introduced the average urban income percapita in the destination city and (lacking information onthe origin county) the average rural income per capita in theorigin province of the migrant, the hypothesis being that bothhave a negative coefficient, reflecting relative deprivation. Thecoefficient on destination income is indeed large and negativebut not significant; and that on origin income is trivial and notsignificantly different from zero. If the migrant is living withfamily, or has relatives in the city who can be turned to forhelp, the effect on happiness is positive, but not significantlyso in the former case. Having a child still in the village has asignificant depressing impact. Of the housing variables onlylack of heating is significant: the effect is predictably negative.

Eqns. (4) and (5) of Table 2 reproduce the full equation fortwo sub-samples: those who had less than the median urbanstay (7.5 years) and those who had more, respectively. Wemention only the notable determinants for which there is a sig-nificant difference in coefficients. The long-stayers have a high-er coefficient on the income variable (0.28 compared with0.13). Even when ln income per capita is replaced by absoluteincome per capita in the equations (not shown), so providingan estimate of the marginal utility of income, the coefficientfor the long-stayers is twice as high. This might be the resultof self-selection. However, the result is also consistent with mi-grants learning to enjoy the costly opportunities of urban lifeand thus becoming more materialistic as they get more in-volved in urban society. The long-stayers are more sensitiveto average urban income per capita in the destination city (asignificant �0.28 compared with a non-significant 0.005.). Thissuggests that over time urban residents increasingly becomethe reference group for migrants. Moreover, their resultantfeeling of relative deprivation might explain why additional in-come becomes more important for their happiness.

Long-staying men cease to be at a significant disadvantagecompared with women and, whereas among short-stayersmarriage is bad for women’s happiness but good for men’s,among long-stayers these effects are weakened. This changemight reflect the evolution of power relationships or socialnorms as migrants become more urbanized.

Table 3 is based on estimates of the basic and the full equa-tions of Table 2 but only for employed respondents, the reasonbeing that it is then possible to add various employment-re-lated explanatory variables as well. Fortunately, we lose veryfew observations (1.3% of the sample). The first column pro-vides mean values, Eqn. (1) shows the basic equation, andEqn. (2) the full equation. Because the coefficients of the vari-ables in common are barely changed, Table 3 reports only theadditional variables. The hypothesis is that the unpleasantnessand insecurity of urban work contribute to the unhappiness ofmigrants.

Where satisfaction with the current job is rated 4 for ‘‘verysatisfied” down to 0 for ‘‘not at all satisfied,” this cardinal var-iable has the expected positive and significant coefficient inboth specifications. Respondents were asked whether ruralworkers enjoyed the same treatment as urban workers in sevendifferent aspects of the employment relationship. The negativeanswers were added to form a cardinal index of discrimination

(ranging from 0 to 7).The coefficient is negative and significantin both equations, indicating that perceptions of discrimina-tion contribute to unhappiness. Compared with being self-em-ployed (the predominant activity), having permanent work orlong-term contract work raises happiness but this result couldarise by chance. Another aspect of the insecurity of urbanemployment can also be incorporated. Respondents wereasked how long it would take for them to find another job withequivalent pay if they lost their current job. Compared with‘‘within one week”—the reference category—the coefficientsare generally significantly negative and increase monotonicallyin size in both equations. The evidence is consistent with ourhypothesis: migrant employment can be unpleasant and inse-cure, and this depresses migrant happiness.

The third hypothesis emerges from theories of rural–urbanmigration expressed in terms of decision-making by the ruralfamily, of which the migrant remains a part. The inference isthat the mean happiness score of migrants is low because theysupport their rural family members through remittances, andtherefore that our dependent variable cannot reflect the fullgain in happiness of the two-location family. In principle theargument is weak. First, it is less plausible for settled migrantsthan for temporary migrants. Second, utility-maximizing eco-nomic agents are assumed to allocate their income optimally,that is, at the margin gifts yield as much utility for the giveras does consumption. Happiness thus need not fall if incomeis remitted. It is nevertheless true that migrant household dis-posable income per capita is reduced by the presence of familymembers elsewhere.

Roughly half (51%) of migrant households made remit-tances; remittances represented 9% of household income forthe sample as a whole and 17% for the remitting households.Our hypothesis is that remittances reduce the happiness ofrespondents in migrant urban households, and so contributeto the low mean happiness score. Our test is whether the var-iable ln household remittance per capita (with zero remittanceset equal to one yuan) is significantly negative in the happinessfunction. Whether we add this term to the basic or the fullequation of Table 2, and whether for the full sample or forthe sub-sample of remitters, the coefficient on the remittancevariable is in no case different from zero. To illustrate, Eqn.(3) of Table 2 corresponds to Eqn. (2) but with remittancesadded; the other three experiments produced very similar re-sults. We can find no evidence in support of Hypothesis 3.

In a methodological paper, Ferrer-i-Carbonell and Frijters(2004) examined the sensitivity of results to the influence ofthe unobserved determinants of happiness. They did so bycomparing the coefficients on the income variable derivedfrom the cross-section data with those derived from the paneldata, so identifying the effect of unobserved time-invariantfixed effects. Use of the panel element in the German nationalhousehold survey reduces the size of the positive coefficient onincome, suggesting that the cross-section results are biased up-wards. For instance, unobserved characteristics such as per-sonal energy might raise both income and happiness, orhappiness itself might improve motivation and so might raiseincome. Accordingly, the basic and full specifications of Table2 were estimated with the potentially endogenous variable thatis most relevant to our tests, ln income per capita, now instru-mented. The exclusion restrictions were mother’s years of edu-cation, spouse’s years of education, and the income that themigrant earned in the village before migrating. It is plausiblethat these variables do not directly influence current happiness(not even own education has a significant positive effect in Ta-ble 2). The instruments passed the conventional statistical testsfor good instruments: they were neither weak nor endogenous,


and they were needed. The single notable difference in the re-sults (not shown) lay in the income variable itself. Contrary toexpectations, this was now higher, being more than doubled insize, to 0.69 and 0.63 in the equations corresponding to Eqns.(1) and (2) of Table 2, respectively. 2 However, this effect wasstill modest, for example, a doubling of income raised thescore by between 0.48 and 0.43, that is, less than half theequivalent of moving from being so-so (a score of 2) to beinghappy (3). One explanation for the rise is the possibility thatthe hidden relationships have the opposite sign, for example,higher aspirations raise income but lower happiness, or happi-ness discourages effort, or farsightedness means accepting cur-rent hardship in the expectation of future happinessAlternatively, instrumenting might reduce the attenuation biascaused by error in the measurement of income.

Ferrer-i-Carbonell and Frijters (2004) also tested whetherhappiness levels should be treated as cardinal (as psychologistsgenerally do) or ordinal (as economists generally do). Their re-sults were not at all sensitive to the choice of dependent variableand thus to the choice between OLS and latent variable meth-ods. We too examined the robustness of our estimates to differ-ent forms of the dependent variable: happiness as a binaryvariable estimated by means of a probit model, and as a multi-nomial variable using an ordered probit model. The pattern ofresults was very similar to that of Table 2. Thus, our results wererobust to alternative versions of the happiness variable.

6. ARE MIGRANTS SELF-SELECTED?

Hypothesis 4 is that migrants have lower mean happinessthan rural people because they are self-selected. Thus, theirlower happiness might be the result of differences in character-istics. A different testing methodology is required. We wish tocompare the migrants with both rural and urban residents,employing the standard Blinder–Oaxaca decomposition tech-nique, based on identical happiness regression equations forthe groups being compared. The choice of explanatory vari-ables used is governed by the availability of the same variablein the two datasets, and by its success in the happiness func-tions. The objective is to pinpoint the reasons for the differ-ences in happiness.

Table 4. Decomposition of the difference in mean happiness score between rural–

Using the rural happiness function

Due to characteristics Due to co

Ln income per capita �55.51 1.1Health �26.39 114.Income expectations 14.71 32.9Age 13.97 �138Education �2.55 22.6Male �4.70 �23Marital status 2.49 �1.Ethnicity 1.10 2.1CP membership 5.01 1.3Unemployment 0.09 0.0Working hours 16.65 �23Net financial assets �13.43 21.2Constant term 0.00 140.Sum (percentage) �48.56 148.Sum (score) �0.1485 0.45

Notes: The mean happiness scores are 2.6764 in the case of rural residents andequal to +100%) to be explained by the decomposition. Thus the combined cThe composite variables are age and age squared for age, married, single, divodecrease for income expectations.

We began by conducting a decomposition analysis of thedifference in household mean income per capita (not shown),in order to throw some light on the representativeness, andthe motivation, of the migrants. This involved estimating iden-tical rural and migrant household income functions. Thosemigrating from rural China are indeed a selective and unrep-resentative group. Migrant households, had they remained inthe rural areas, would on average earn 10% less income thanrural resident households. There is also a considerable incomeadvantage to their migration. The mean income that migranthouseholds actually earn is 2.64 times what they would earnin the rural areas. By contrast, if they were to migrate, averagerural households would earn 2.19 times more than they actu-ally earn. Rural households possess productive characteristicsthat are relatively valuable in the countryside whereas migranthouseholds possess productive characteristics that are rela-tively valuable in the city.

The mean happiness score of rural people was 2.68 and thatof migrants was 2.37, implying a migrant shortfall of 0.31. InTable 4 we decompose this gap into the parts which can be ex-plained by differences in the mean values of the characteristicsof the two groups and by differences in the coefficients of thetwo happiness functions. The results differ only a little accord-ing to whether we pose the counterfactual question ‘‘whatwould be the effect on the mean happiness of migrants if theyhad the same happiness function as rural people?” or the ques-tion ‘‘what would be the effect on the mean happiness of ruralpeople if they had the same happiness function as migrants?.”The former question involves decomposition according to

H r � Hm ¼ X mðar � amÞ þ arðX r � X mÞ ð6:1Þand the latter according to

H r � Hm ¼ X rðar � amÞ þ amðX r � X mÞ ð6:2Þwhere Hr and Hm are the mean happiness scores in the ruraland migrant samples, respectively, Xr and Xm are the vectorsof rural and migrant mean characteristics, and ar and am arethe vectors of rural and migrant coefficients. To illustrate,the entry �55.51 in row 1, column 1 of Table 4 is obtained,according to Eqn. (6.1), by multiplying the difference in meanln income per capita by the rural coefficient of ln income percapita, and the entry 1.13 in row 1, column 2 is obtained by

urban migrants and rural residents: percentage contribution to the difference

Using the migrants’ happiness function

efficients Due to characteristics Due to coefficients

3 �55.39 1.0199 �5.81 94.418 11.34 36.36.82 6.69 �131.541 �0.13 20.18

.87 0.74 �29.3082 0.89 �0.222 0.13 3.108 0.40 5.992 0.10 0.02.94 5.53 �12.818 0.29 7.5648 0.00 140.4856 �35.23 135.2344 �0.1078 0.4137

2.3703 in the case of migrants, creating a migrant shortfall of 0.3061 (setontributions of characteristics and coefficients sum up to 100%.rced and widowed for marital status, and big increase, small increase, and


multiplying the mean migrant ln income per capita by the dif-ference in coefficients, and then by expressing these productsas percentages of the gross mean difference in happiness.The entries show the percentage contributions of the differencein income and of the difference in the returns to income,respectively.

The effect of characteristics is actually an increase in the dif-ference in the mean happiness scores. This is mainly due to thevariable ln income per capita: its coefficients are the same(0.194) in the two samples but migrants have higher mean in-come. The reason why migrants have lower mean happiness isthus to be found in the different happiness functions. The con-stant term, health, and income expectations are the main con-tributors. The importance of the constant term implies thatthere are unobserved characteristics that reduce migrant hap-piness relative to rural happiness. For instance, we are unableto standardize for the various social disadvantages that mi-grants encounter in the cities because the same variables arenot available in the rural dataset. Perhaps because rural peopleare on average less healthy than migrants—poor health beinga deterrent to migration—they place a higher value on goodhealth.

In both samples happiness is highly sensitive to expectationsabout future income in five years’ time. It appears that expec-tations of future income can influence current happiness. Withthe expectation of no change in income as the omitted cate-gory in the dummy variable analysis, the coefficients in the mi-grant sample vary from 0.31, if a large increase is expected, to0.05, if a small increase is expected, and to �0.39, if a decreaseis expected; the corresponding estimates for the rural sampleare 0.41, 0.19, and �0.19, respectively. The fact that in the mi-grant sample the coefficients are uniformly lower, in relation tothe expectation of static income, suggests that migrants havehigher aspirations relative to current income. This can be ex-pected if aspirations depend on the income of the relevantcomparator group. Whereas the rural respondents are repre-sentative of rural society, and so their mean income is closeto the mean income of their likely comparator group, the mi-grant sub-sample is unrepresentative of urban society: mi-

Table 5. Decomposition of the difference in mean happiness score between rural–

Using the urban h

Due to characteristics

Ln income per capita 43.20Income expectations �47.03Living standard in second highest quarter in city �16.81Living standard in third highest quarter in city �8.19Living standard in lowest quarter in city 194.35Age 1.52Male 11.53Education �8.65Marital status 0.18Ethnicity �2.12CP membership 15.69Unemployment �6.68Health �54.21Working hours �1.08Net financial assets 1.69Constant term 0.00Sum (percentage) 123.41Sum (score) 0.1372

Notes: The mean happiness scores are 2.4845 in the case of urban residents andequal to +100%) to be explained by the decomposition. Thus the combined cThe composite variables are age and age squared for age, married, single, divodecrease for income expectations.

grants tend to occupy the lower ranges of the urban incomedistribution. If migrants make comparisons with urban-bornresidents as well as with other migrants, their aspirations willbe high in relation to their current income. However, there isanother possible interpretation of this result: people mighthave migrated with pre-existing high aspirations.

Is the low mean happiness of migrants a general character-istic of city life? We pursue our inquiry further by comparingmigrants with ‘‘urban residents,” that is, persons who are ur-ban-born or in other ways have acquired urban hukou status,with the rights and privileges that accompany it. The meanhappiness score of urban residents is 2.48 and that of migrantsis 2.37, implying a migrant shortfall of 0.11. Table 5 provides adecomposition similar to that provided by Table 4 but with adifferent set of explanatory variables—those that are commonto the two datasets.

In this case the differences in coefficients add slightly to themigrant shortfall in mean happiness score. The coefficient onthe income variable is higher for urban residents (0.173) thanfor migrants (0.111), so raising urban happiness relative torural happiness, and the effect of income expectations is alsostronger for urban residents. The positive effect of incomeexpectations reflects the lower coefficients in the migrant sam-ple: with static expectations as the reference category, for mi-grants an expected big increase in income has a coefficient of0.21, a small increase 0.00, and a decrease �0.37, whereasfor urban residents the corresponding estimates are 0.34,0.10, and �0.29, respectively. Again, migrants appear to havehigher aspirations relative to their current income. The contri-bution of the income variables to the explanation of the differ-ence in mean happiness is offset by the negative effects ofvariables such as age, gender, and the constant term. Note thatposition in the city income distribution has a powerful effecton happiness. With the highest quarter of households beingthe omitted category, happiness falls monotonically, to lowerthan �0.80 in the lowest quarter. As this is true of both sam-ples, it does not affect relative happiness.

The migrant shortfall thus has to be explained in terms ofdifferences in mean characteristics. Two variables stand out:

urban migrants and urban residents: percentage contribution to the difference

appiness function Using the migrants’ happiness function

Due to coefficients Due to characteristics Due to coefficients

457.57 28.15 472.6266.43 �39.92 59.329.40 �33.68 26.2874.32 �11.71 77.84�26.79 175.93 �8.37�562.72 32.85 �594.05�62.39 �4.08 �46.78

8.22 �11.54 11.112.63 �1.96 4.773.19 �0.34 1.401.00 7.63 9.06�2.01 �0.68 �8.0178.08 �28.01 51.8922.20 10.50 10.623.85 �2.46 8.01�96.38 0.00 �96.38�23.41 120.67 �20.67�0.0260 0.1342 �0.0230

2.3703 in the case of migrants, creating a migrant shortfall of 0.1143 (setontributions of characteristics and coefficients sum up to 100%.rced and widowed for marital status, and big increase, small increase, and


the higher mean income of urban residents improves their rel-ative happiness, and their superior position in the city incomedistribution has the same effect. A far higher proportion of mi-grants than of urban residents report that they fall in the low-est quarter of city households in terms of living standard (35%compared with 11%). This fact alone can explain more thanthe entire migrant deficit. If the income of the relevant com-parator group influences aspirations, the inferior position ofmigrants in the city income distribution can also explainwhy they appear to have higher aspirations in relation to theircurrent income.

We see that the constant term in the decomposition pre-sented in Table 4 explains more than the entire difference inthe mean happiness scores of migrants and rural dwellers. Thissuggests that differences in unobserved characteristics areimportant. Migrants may be less happy on average simply be-cause inherently unhappy people tend to be the ones who mi-grate. Support comes from answers to the question as towhether urban living had yielded greater happiness than rural

Table 6. Determinants of urban living happ

Log of per capita household income 2002Duration of urban residence (years)Duration of urban residence, squaredExpect big increase in income over next 5 yearsExpect small increase in income over next 5 yearsExpect decrease in income over next 5 yearsDifference between actual and predicted happiness scoreMaleMarriedMale and marriedEducation (years)Working hours (‘00 per year)Net financial assets (‘000 Yuan)Log of average per capita urban income in city of current residenceLog of average rural income in province of originLiving with family membersChild still in villageNumber of relatives and friends in cityHouse area per capita

Living in own houseNo heatingPermanent or long-term contract workTemporary workSatisfaction with jobIndex of discriminationCan find another job in two weeksCan find another job in a monthCan find another job in 2 monthsCan find another job in 6 monthsNeed more than 6 months to find another jobUnemployed

Pseudo R2

Number of observations

Notes: The dependent variable is the probability of being happier in urban arediscrete change of dummy variable from 0 to 1.The variable, difference between actual and predicted happiness scores, has been3.The omitted categories in the dummy variable analyses are: single female; empnormal mood; and no change in income expected in the next five years.Eqns. (2) and (3) have been clustered at city level for robust standard errors.

living. Despite the mean happiness score being lower for mi-grants than for rural people, 56% of migrants answered affir-matively and only 3% negatively. This is the picture thatcould emerge if migrants are intrinsically unhappy peoplewhose happiness remains low despite improving after migra-tion.

Migrants might be unhappy people because by nature theyare melancholy or they have high but unfulfilled aspirations.However, the latter reason fits better with the stereotype of mi-grants as relatively self-confident, optimistic, and risk-lovingindividuals. Consider the implications of assuming both thatmigrants are naturally unhappy people and that migrationdoes indeed generally raise happiness. Insofar as those mi-grants with a relatively unhappy disposition become abso-lutely happier albeit still relatively unhappy after migration,we might expect as high a proportion of unhappy migrantsas of happy migrants to report that their life is more satisfac-tory in urban areas than in rural areas. In fact the proportionfalls, from 67% in the highest happiness category to 34% in the

ier than rural living: probit estimation

Marginal effects

(1) (2) (3)Full sample Employed sample Employed sample

0.053172** 0.046613* 0.050583*

0.017410*** 0.019026*** 0.017368***

�0.000342* �0.000351 �0.0003120.152552** 0.176613*** 0.165676**

0.077823** 0.094063*** 0.086907**

�0.102598 �0.055929 �0.0556620.173596***

0.082351 0.080844 0.0798510.041908 0.02991 0.021325�0.084901 �0.072965 �0.068252�0.025580*** �0.026373*** �0.026748***

0.002915*** 0.004453*** 0.004369***

�0.000021 0.000235 0.0002120.147553 0.173367* 0.144629�0.211506*** �0.242126*** �0.222568***

0.120859** 0.107023* 0.128551**

�0.040104 �0.031336 �0.0512380.001886 0.00136 0.0017580.000238 �0.000068 0.000289

0.143046** 0.128608** 0.130402**

0.056151 0.089106 0.061625�0.018118 �0.0122610.042303 0.045310*

0.076827*** 0.071872***

�0.006276 �0.00548�0.026297 �0.025783�0.079845 �0.0727580.014598 0.011422�0.011039 �0.010450.005574 0.009036

�0.084094

0.0638 0.1256 0.07991850 1715 1715

as. For the dummy variables the marginal effects are denoted by dy/dx for

derived by obtaining the predicted happiness score from Eqn. (1) in Table

loyed or labor force non-participant; not healthy; in normal or worse than


lowest happiness category, suggesting that this sort of self-selection can at best be only a partial explanation for thelow mean happiness of migrants.

The argument can be tested more rigorously as follows. Ifwe estimate the predicted happiness score for each respondent(from Eqn. (1) of Table 3), the residual (actual minus pre-dicted) score is the part of happiness that cannot be explainedby our variables. The residual is made up of measurement er-ror and two sorts of unobserved characteristics of the respon-dent: those which were present before migration and thosewhich came after migration. A disposition to be happy or un-happy is of the former sort. Assume that migration had a sim-ilar effect on the happiness of all respondents whoseunobserved characteristics did not change pre- and post-migration. In that case, we can test whether the residual helpsto explain whether the respondents reported that their happi-ness was higher in the city than in the village.

Table 6 reports the determinants of a logit equation predict-ing an affirmative answer. Its three columns, presenting themarginal effects of each explanatory variable, indicate whetherthe full sample or the employed sample is used. The object is toidentify the characteristics which have raised happiness. Com-paring Tables 2 and 3 with Table 6, we see that some of thesame variables that determine happiness also correspondinglydetermine an increase in happiness: the income variable, dura-tion of urban stay, expectation of future income change, andjob satisfaction. When the residual is introduced into the equa-tion corresponding to column 1 of Table 6, we expect that itwill not be significantly different from zero if inherent andunchanging personality is the cause of unhappiness. However,we find that the coefficient is positive and significantly so at the1% level (column 2), and the marginal implies that a residualof +1.0 raises the probability of an affirmative answer by17% points. This positive effect suggests that migration chan-ged the unobserved characteristics of migrants, in which caseinherent disposition cannot solve our puzzle.

Instead, migration may be subject to selection bias on thebasis of unobserved characteristics that are different or havedifferent effects in the two locations. Several examples cometo mind (beyond the case discussed under Hypothesis 2, i.e.,migrants’ aspirations rise). If people who are dissatisfied withlife in general but with village life in particular have a highpropensity to migrate, migrants might have low average hap-piness in both locations but particularly in the village. For in-stance, own or family misfortune or bad family or villagerelationships could reduce a person’s happiness but more soif he/she remained in the village. If migrants have high pre-existing aspirations which cannot be fulfilled in the villagebut have the potential to be fulfilled in the city, this might havethe same effect. In each of these cases the migrants would belikely to report that their urban life is better than their rurallife had been, despite their low average urban happiness. A testof this type of hypothesis would require a survey which couldreveal the happiness score, and the reasons given for unhappi-ness, before migrating.

7. CONCLUSION

This research is among the first to link the literatures on rur-al–urban migration and on subjective well-being in developingcountries. Indirectly, Stark (1991, chap. 6); Stark and Taylor(1991) introduced relative deprivation based on income rankwithin the origin village as a determinant of internal migra-tion, and adduced supporting evidence for Mexico. Faf-champs and Shilpi (2008) found that the perceived adequacy

of consumption of rural–urban migrants in Nepal was inver-sely related to mean consumption in their home village. Mostdirectly, De Jong, Chamrarrirthirong, and Tran (2002) calcu-lated that a somewhat higher proportion of the permanent mi-grants in their sample for Thailand experienced an increase inlife satisfaction after migration than a decrease.

We have posed the question: why do rural–urban migranthouseholds which have settled in urban China report lowerhappiness than rural households? Migrants had lower meanhappiness despite their higher mean income: the income differ-ence merely adds to the puzzle. It is a question that cannot beanswered easily in terms of the conventional models of rural–urban migration based on utility maximization. Four hypoth-eses were examined. We found no evidence for Hypothesis 3,in terms of migrant support for family members in the village.Each of the other three involves false expectations, but for dif-ferent reasons: prospective migrants have false expectationsabout their urban conditions, or about their urban aspirations,or about themselves. What they have in common is that rural–urban migrants are unable to form ‘‘rational expectations” be-cause they lack the necessary information as they enter a newand different world. In each case there are reasons why theirexpectations are too favorable.

Consider the light that the happiness functions throw onHypothesis 1, that is, the low mean happiness score of migrantsis due to false expectations about their conditions in the city.The fact that happiness appears to rise over several years sug-gests that migrants are able to overcome the early hardships ofarriving, finding work, and settling in the city. However, somehardships remain, relating to accommodation, family, andwork. The unsatisfactory conditions in which migrants liveand the unpleasant and insecure nature of their employmentdepress migrant happiness. That adverse urban conditionsare important can be illustrated by the following simulations,based on Table 3, Eqn. (2). The mean migrant happiness scorewould rise by 0.04 if no migrants had children remaining in thevillage, by 0.02 if all lived with their family, by 0.09 if all hadheating, and by 0.17 if all reported zero discrimination. In com-bination, these counterfactual assumptions would correspondto the entire difference in mean scores between rural and mi-grant households. Provided that unbiased information wasavailable to prospective migrants, we expect them to haveinternalized the effects of adverse conditions on their happinesswhen deciding to migrate: expectations would not have beenfalse. Why should migrants have overestimated the conditionsof their urban life and work? It is possible that, whereas ex-pected income is quantifiable and understandable, other as-pects of urban life have to be experienced to be understood.Moreover, expectations of conditions might be based on theimages of the lives of urban residents rather than on those ofthe lives of rural–urban migrants. The migrants, when theymade their decisions to move, may have been realistic abouttheir urban income prospects, whereas their expectations of liv-ing and working conditions could have been biased upwards.However, there is a caveat: the better the information flowsto the villages, the weaker is the case for this hypothesis.

Hypothesis 2 is that migrants had false expectations, formedat the time of migration, that their aspirations would not alterin the city. Consider the reasons why migrants’ aspirations mayhave risen and now exceed their actual achievements. When weconducted a decomposition analysis to discover why migrantshave a lower mean happiness score than both rural residentsand urban hukou residents, in each case a major contributioncame from the higher aspirations of migrants in relation to cur-rent income. This is consistent with the fact that over two-thirds of migrants who were unhappy or not at all happy gave


low income as the predominant reason for their unhappiness.The relatively high aspirations might be explained by the lowlyposition of most migrants in the city income distribution: hav-ing relatively low income was shown to reduce their happiness.The evidence suggests that migrants draw their referencegroups from their new surroundings, and for that reason havefeelings of relative deprivation. It is plausible that migrants,when they took their decisions to move, could predict that theirincomes would rise but not how their aspirations would rise asthey became part of the very different urban society.

Consider Hypothesis 4, that is, people with unobserved andinvariant characteristics that reduce happiness have a higherpropensity to migrate, in the false expectation that migrationwill provide a cure, and their continuing unhappiness pullsdown the mean happiness score. It was found that the residual,unexplained component of the happiness score is itself anargument in the function that determines the migrant’s viewthat urban living provides more happiness than rural livinghad done. This suggests that the residual reflects either chang-ing unobserved characteristics or location-specific effects ofunchanging unobserved characteristics. Provided that natu-rally unhappy or aspirational people experience the samechange in happiness as a result of migration as do happier peo-ple, inherent disposition itself cannot be a predominant expla-nation for the low mean happiness score of migrants.

There are other hypotheses which cannot be adequatelytested by means of our dataset. The one mentioned above isthat migration is subject to selection bias on the basis of unob-served characteristics which are different or have different ef-fects in the two locations. Another is that rural–urbanmigrants, once they settle in the city, are induced by urban cul-tural norms to use a different scale for measuring happiness,and thus to report happiness scores lower than those reportedby rural residents. We would expect the reported happiness ofmigrants to be higher before they have time to adjust their hap-piness scale. However, the average happiness score of migrantswho have been in the city for less than three years is 0.08 pointslower than the mean for all migrants, and the regression resultsin Tables 2 and 3 suggest that the standardized happiness scorerises for more than a decade after arrival. Although it is notpossible to refute the rescaling hypothesis, this evidence failsto confirm it. Yet another possibility is that migrants are will-ing to sacrifice current happiness for future happiness—plausi-ble in a country with an overall household saving rate of no lessthan 24% 3. We saw that the income that is expected in fiveyears’ time is important to current happiness, implying thatits effect is already at least partly internalized (Table 2); andthat expected future income increases the probability thatrespondents report that their urban happiness exceeds theirrural happiness (Table 6). Migrants might thus be willing toput up with unhappiness because they feel that life will eventu-ally get better for them or their children. The absence of testsfor these alternative explanations means that our results canmerely suggest solutions that are consistent with the availableevidence. Further research based on better datasets is required

to explain the puzzle in China and, if it is found to be a generalphenomenon, in other poor urbanizing societies.

Whatever be the explanation, the obvious question arises:why do unhappy migrants not return to their rural origins?One reason is that the majority do perceive urban living tohave yielded them more happiness than rural living. This re-sult was found to be sensitive to expected income, and themajority of migrants did indeed expect that their incomeswould rise over the next five years. Migrants were also morelikely to favor urban living the longer they stayed in thecity—possibly because they increasingly valued aspects of ur-ban living that were not to be found in rural areas. Social psy-chology might again be relevant: migrants do not take intoaccount how their aspirations will adjust if they return to vil-lage life. Thus there might be symmetry in the way they viewleaving their rural residence and not leaving their urban one.On the other hand, aspirations might not adjust downwardsif horizons, once broadened, cannot be narrowed (Knightet al., 2009). Another possible reason why unhappy migrantsdo not return to their origins—unfortunately not pursued inthe survey—is that the cost might be prohibitive. This is plau-sible if their households have forgone the tenurial rights to vil-lage farm land and housing land that they previously held.

The study has broader implications. Should the social wel-fare function used by policy-makers reflect measured happi-ness? The argument against is examined in Clark, Frijters,and Shields (2008) and found wanting. The distinction madeabove between expected utility (which economic agents are as-sumed to maximize) and experienced utility (which happinessscores are assumed to measure) is relevant. Insofar as there is asystematic difference between the two, this can arise because ofan unpredicted change in aspirations (for instance, owing to achange in reference group) or because the happiness measuredoes not fully capture ‘‘eudaimonia,” that is, functional as-pects of well-being such as autonomy, competence, self-accep-tance, and good relationships. In our judgement, changes inaspirations should be taken into account in assessing welfare(argued in Kingdon and Knight, 2006), and the happinessequations do appear to be sensitive to some functional aspectsof well-being: measured happiness is indeed relevant to policy.

In many developing countries rapid rural–urban migrationgives rise to various social ills—such as urban poverty, slums,pressure on infrastructure, unemployment, and crime—whichadversely affect the welfare of all urban residents. In contrast,by attempting to restrict migration the Chinese governmenthas curbed these outcomes. For instance, in the 2002 nationalhousehold survey few urban hukou residents reported that thepresence of migrants constituted the greatest social problem—well behind corruption, lack of social security, and environ-mental pollution. The fact that rural–urban migrants are theleast happy group suggests that they themselves might fomentunrest. However, because social instability probably requiresnot only unhappiness but also a perception that it is man-made and capable of being remedied, no such conclusioncan be safely drawn.

NOTES

1. In posing the ‘‘commuting paradox” (compensating variation does notprevent longer commuting distance from reducing subjective well-being inGermany), Stutzer and Frey (2008), after tests, similarly favour anexplanation based on wrongly predicted adaptation.

2. Our result is not unique: Powdthavee (2007), using the BritishHousehold Panel Survey, also found that instrumenting raised thecoefficient on the income variable.

3. Calculated from China Statistical Yearbook 2003, Tables 4.1 and 10.1.


REFERENCES

Cai, F., Park, A., & Zhao, Y. (2008). The Chinese labor market in thereform era. In L. Brandt, & T. Rawsi (Eds.), China’s great economictransformation. New York and Cambridge: Cambridge UniversityPress.

Clark, A., Frijters, P., & Shields, M. (2008). Relative income, happinessand utility: An explanation of the Easterlin Paradox and other puzzles.Journal of Economic Literature, 46(1), 95–144.

De Jong, G., Chamrarrirthirong, A., & Tran, Q. (2002). For better, forworse: Life satisfaction consequences of migration. InternationalMigration Review, 36(3), 838–863.

Easterlin, R. A. (1974). Does economic growth improve the human lot? InP. A. David, & M. W. Reder (Eds.), Nations and households ineconomic growth: Essays in honor of Moses Abramovitz. New York:Academic Press, Inc..

Easterlin, R. A. (2001). Income and happiness: Towards a unified theory.Economic Journal, 111(July), 465–484.

Fafchamps, M., & Shilpi, F. (2008). Subjective welfare, isolation andrelative consumption. Journal of Development Economics, 86(1), 43–60.

Ferrer-i-Carbonell, A., & Frijters, P. (2004). How important is method-ology or estimates of the determinants of happiness?. The EconomicJournal, 114(497), 640–659.

Graham, C., & Pettinato, S. (2002). Frustrated achievers: winners, losersand subjective well-being in new market economies. Journal ofDevelopment Studies, 38(4), 100–140.

Gustaffson, B., Li, S., & Sicular, T. (2008). Inequality and public policy inChina. New York and Cambridge: Cambridge University Press.

Kingdon, G., & Knight, J. (2006). Subjective well-being poverty versusincome poverty and capabilities poverty?. Journal of DevelopmentStudies, 42(7), 1199–1224.

Kingdon, G., & Knight, J. (2007). Community, comparisons andsubjective well-being in a divided society. Journal of Economic Behaviorand Organization, 64, 69–90.

Knight, J., and Gunatilaka, R. (in press). The rural–urban divide inChina: Income but not happiness? Journal of Development Studies.

Knight, J., and Song, L. (1999). The rural–urban divide. Economicdisparities and interactions in China. Oxford: Oxford University Press.

Knight, J., and Song, L. (2005). Towards a labour market in China.Oxford: Oxford University Press.

Knight, J., Song, L., & Gunatilaka, R. (2009). The determinants ofsubjective well-being in rural China. China Economic Review, 20(2), inpress.

Knight, J., & Yueh, L. (2008). Segmentation or competition in China’surban labour market?. Cambridge Journal of Economics, 33(1), 79–94.

Lucas, R. E. B. (1997). Internal migration in developing countries. InMark. Rosenzweig, & Oded. Stark (Eds.). Handbook of population andfamily economics (Vol. 1B, pp. 721–798). Amsterdam: North Holland.

Rabin, M. (1998). Psychology and economics. Journal of EconomicLiterature, 36(March), 11–46.

Powdthavee, N. (2007). How much does money really matter? Usingpayslip and lagged regional house prices to instrument for income inlife satisfaction equations. Mimeo: Institute of Education, Universityof London (December).

Runciman, W. G. (1966). Relative deprivation and social justice. Berkeley:University of California Press.

Sabates-Wheeler, R., Sabates, R., & Castaldo, A. (2008). Tacklingpoverty-migration linkages: Evidence from Ghana and Egypt. SocialIndicators Research, 87(2), 307–328.

Solinger, D. (1999). Contesting citizenship in urban China: Peasantmigrants, the state and the logic of the market. Berkeley: Universityof California Press.

Stark, O. (1991). The migration of labour. Cambridge, Mass and Oxford:Basil Blackwell.

Stark, O., & Taylor, J. E. (1991). Migration incentives, migration types:The role of relative deprivation. Economic Journal, 101, 1163–1178.

Stutzer, A., & Frey, B. (2008). Stress that doesn’t pay: The commutingparadox. Scandinavian Journal of Economics, 110(2), 339–366.

Todaro, M. (1967). A model of labor migration and urban unemploymentin less developed countries. American Economic Review, 59(2),138–148.

Zhao, Y. (1999). Labor migration and earnings differences: The case ofrural China. Economic Development and Cultural Change, (47),766–782.

.sciencedirect.com
Available online at www

Date post:	13-Sep-2016
Category:	Documents
Upload:	john-knight
View:	215 times
Download:	2 times

Great Expectations? The Subjective Well-being of Rural–Urban Migrants in China

Documents