Was China s rate of poverty reduction even faster than routinely assumed? Accounting for the effects of migration

Was China s rate of poverty reduction even faster than routinely assumed? Accounting for the effects of migration Klaus W. Deininger, John Giles, Songqing Jin and Hui Wang Abstract: Migration is recognized as a key mechanism of resce transfer and poverty reduction globally and in China. The way in which statistics are computed implies that, even after accounting for remittance inflows, migrant households welfare remains significantly below that of non-migrants. To explore whether this may be a result of data processing methodology, we use panel data from 8 provinces to compute alternative measures. Results suggest that adjusting for migrants absence increases per capita income and expenditure for migrant households (then estimated to be better off than non-migrants) and a significant reduction in poverty level. Implications for poverty estimates in China and elsewhere are explored. Keywords: Household definition, migration, poverty, inequality JEL: O15 R23 1. Introduction 1

Over the last two decades, the importance of international migration and the associated remittances has increased dramatically; in fact the latter have become a major determinant of global capital flows. Updating earlier figures (Ratha and Shaw 2007), the World Bank estimates that some 215 million individuals migrated across international boundaries in 2010. 1 The size of private migration remittances to developing countries, estimated at US$ 325 billion in 2009, exceeded that of portfolio investment and official development assistance, approaching the size of total foreign direct investment (Yang 2011). Migrant remittances not only exhibited enormous growth, increasing at 12.9% per annum in real terms in the 1999 to 2008 period, but also had other desirable characteristics. Most importantly, in contrast to FDI which is highly volatile, dropping by almost 40% in the wake of the 2008 financial crisis, they were very resilient to variations in the global economy. They have been shown to not only reduce poverty directly by transferring resces to poor areas and households but also to contribute to economic development by helping sending households and communities to cope with shocks (Yang 2008), invest in human and physical capital (Mora and Taylor 2006), start entrepreneurial activities, and help spread new ideas, values, and norms (Docquier et al. 2011). Internal migration in China, which was virtually non-existent even a few decades ago, has increased dramatically over the last decades to become a key driver of the country s economic expansion. In fact, internal migration in China is quantitatively more important than cross-border migration in the rest of the world combined: with 261.4 million, according to the 2010 Population Census (NBSC 2012), the number of China s internal migrants is larger than the 215 million cross-border migrants in the rest of the world (Sirkeci, Cohen, and Ratha. 2012). While international migration is either permanent or for long periods, China s migration is almost all seasonal and temporary 1 Detailed data by country are available at http://go.worldbank.org/jitc7nytt0. 2

(Whalley and Zhang 2007). It has been shown to have significant direct and indirect effects to reduce households exposure to agricultural production shocks and the need for low-return precautionary savings (Giles 2006). And by helping to reduce poverty in China at a fast pace, migration contributed to global achievement of the MDGs and poverty reduction. While there is thus little doubt that migration made a major contribution to China s progress in reducing poverty (de Brauw and Rozelle 2008, Du et al. 2005), official statistics suggesting that, even after migrants contributions are taken into account, households with migrants are significantly worse off than their peers, seems to be somewhat of a puzzle. A key reason for this may be the way in which migrants contribution to their household of origin is treated in national statistics. First, per capita consumption will be affected by whether or not migrants are considered part of the household of origin. In fact, contrary to what is practiced by most countries, China counts migrants as part of the sending household even if they are absent for periods longer than 6 months. By overestimating true household size, this may impart a systematical downward bias on migrant households estimated per capita income. Second, resident household members who respond to the standard household survey that underpins national estimates may either omit migrants income or consumption or report it with error. Any of these will affect estimates of sending households welfare, rates of rural poverty reduction, and rural-urban inequality. The importance of these issues has long been recognized; in fact China National Bureau of Statistics (NBS) has repeatedly alerted provincial offices of the need to ensure consistency in computing relevant figures. Still, empirical work is needed to assess whether such biases may be relevant in practice. To do so, we use individual micro-data for about 2,000 households in 8 provinces during the 5-years from 2005 to 2009 from 3

China s Rural Household Survey () applying a consistent methodology throughout. In line with discussion, two elements will be important to do so. First, we adjust household size by either dropping migrants who were absent for more than 3 months (as is done by most countries) or by pro-rating them depending on the number of months they actually were present in the household. Second, we net out migrants income or expenditure so that we only remittance-related transfers or items actually consumed by migrants during their presence in the household are counted. Results suggest that these methodological issues may indeed affect estimates of consumption and poverty. Our adjustments increase estimated per capita consumption or expenditure for migrant households by between 20 and 25%. While unadjusted figures would suggest migrant households being wore off than those without migrants, this is no longer the case with adjustments; in fact households with migrants are then estimated to be better off than non-migrant ones. Given the large number of migrants, this affects aggregate statistics; it points towards a much faster reduction of poverty than would be suggested by unadjusted figures with the poverty headcount decreasing by 2 to 5 percentage points. While it also points towards a slight increase in inequality within rural areas, this implies that, if data in line with international conventions are used, the drop in Chinese poverty has been even faster than otherwise. The paper is organized as follows: In Section 2, we draw upon existing studies of impact of migration on household welfare using different household definitions to motivate this paper. The data and alternative measures of household size are presented in Section 3. The empirical results are presented in Section 4, which is followed by conclusion and policy implication in Section 5. 2. Assessing the impacts of migration 4

Estimates of the economic impacts of migration in rural China may be biased for two reasons. First, migrants are counted as part of the sending household even if they stay away from a household for longer than 6 months, thus reducing estimated per capita income or expenditure. Second, if migrant income and consumption are recorded inconsistently across provinces or with error, outcome variables will be further affected even if otherwise data are of high quality. Data from China and elsewhere provide circumstantial evidence suggesting that this can indeed affect estimates of consumption, poverty, and distributional outcomes. While back-of the envelope efforts to correct for differences in household size are possible and indeed reported in the literature, they implicitly assume that the size and magnitude of bias introduced by potentially erroneous reporting of migrants income or expenditure is constant across reporting units. If this is not the case, proper adjustment will require use of micro data. 2.1 Importance and impact of migration in China With more than 250 million of migrants or 30% of the country s total labor force, the magnitude of internal migration in China is very large. Restrictions on migrants ability to gain residency at the destination imply that virtually all migration is temporary (Fleisher and Yang 2003), prevent equalization of income levels (Whalley and Zhang 2007), and contribute to persistent cross-regional imbalances (Au and Henderson 2006). Still, migration is credited with having made a major contribution to China s rapid progress in poverty reduction (Ravallion and Chen 2007). It also contributed to narrowing large rural-urban income gaps. With 41% of migration being between rural and urban and 18% between rural and rural areas, migrants originating in rural areas account for almost two thirds of total migration flows (Cai and Wang 2003).While high incidence of migration by the poor (de Brauw et al. 2002) could in principle reduce 5

poverty in sending communities (Zhao 2002), actual impacts are less clear (Du et al. 2005). Migration tends to benefit sending communities as migrants earn more at the destination than what they could have earned if staying at home. Reduced form crossnational estimates suggest that migrant remittances have a significant impact on reducing poverty; while one study finds that a 10% increase in remittances is estimated to lead to a 3.5% drop in the share of poverty (Adams and Page 2005), another one finds that a 2.5% increase in the remittance to GDP ratio is associated with a 0.5% decrease in poverty. Comparing the income loss due to migration (i.e. what migrants would have earned) to the gains from remittances generally points towards an equalizing effect of remittances but points towards wide variation across countries in the elasticity of poverty reduction to migration income (Acosta et al. 2008). For sending communities, key parts of the debate revolved around whether migration leads to a net loss of talent or whether emigrants remittances and returnees skills may compensate. Remittances can allow sending households manage risks (Rosenzweig and Stark 1989) encage human capital investment (Woodruff and Zenteno 2007, Yang 2008), improve educational outcomes (Mansuri 2006), and foster entrepreneurship in capital intensive sectors (Banerjee and Munshi 2000). Parents remittances and exposure to new ideas have been shown to improve migrant children s education (Cox Edwards and Ureta 2003, Glewwe and Jacoby 2004). These effects may be wealth-differentiated, for example due to entry costs associated with migration (Mendola 2008). The positive effects of migration may be tempered by other factors, e.g. young males dropping out of school forgoing secondary education to migrate (Acosta 2011) and in the process possibly unduly stretching females workload (McKenzie and Rapoport 2006). 2.2 Measurement issues that might affect poverty estimates 6

The main sce of information on poverty and household welfare in rural China is the rural household survey (), collected regularly by China s National Bureau of Statistics (NBS). This is a nationally representative panel survey. Samples are equally sized for each of China s provinces and the panel rotates every 5 years. Concerns about sampling error due to non-response of wealthy households (Benjamin et.al, 2005) seem to be limited to urban areas. Non-sampling error is reduced by requiring households to maintain a daily consumption diary, in addition to transaction books for cash and goods that serve as the basis for computing income and consumption (, 2004). Within the household, the relevant diaries are kept by the head or the working member with the highest level of education. Supervision is tight - a local supervisor - normally a village resident- visits households every fortnight and a control structure is built around county survey officers (Chen and Ravallion 1996). Compared to other panel surveys this results in high data quality (Jalan and Ravallion 1998) and low rates of attrition as (Alderman et al. 2001). While there is little doubt that the survey provides data of very high quality on household members present at the time of the interview, underlying definitional assumptions on household membership and attribution of migrants income or consumption may systematically affect information on households with migrants in ways that may differ from global practice. The internationally accepted household definition is based on sharing of food and habitation. 2 According to this definition, migrants who are absent from the household for longer than a given period, generally 3 months per year, are not considered members and excluded by most studies. 3 By contrast, the 2 The UN defines households based on the arrangements made by persons, individually or in groups, for providing themselves with food or other essentials for living (Handbook on rural households livelihood and well-being, UN, 2007).The definition by Demographic and Health Surveys (DHS) is a person or a group of persons, related or unrelated, who live together and share a common sce of food (DHS comparative studies No.14, 1994), and the definition in Living Standards Measurement Study (LSMS) bases membership on the actual number of months of residency in the house, except for household head (Benjamin et al. 2005). 3 Studies show that reported household size depends on questionnaire design. For example, in Mali inclusion of additional keywords in the questionnaire increased reported household size, significantly altering statistics on assets, consumption and production (Beaman and Dillon 2012). 7

definition, as laid out in the survey manual (Nong Cun Zhu Hu Diao Cha Fang An) is based on resce pooling. 4 Migrants are considered household as members long as all or part of their income is pooled with that of other household members even if they have been away for more than 3 months per year. If definitions are consistently applied this would affect international comparability but a unproblematic internally. The fact that resident members may not be able to provide accurate information on migrants income or expenditure can, however, lead to bias the magnitude of which may vary over time and space depending on a factors such as instructions given by the provincial NBS office and the relative importance of migration. To appreciate the potential for such bias, a closer look at the way in which data is collected is useful. Total expenditure is calculated from data recorded in two transaction books for each household as the sum of spending under the headings of food, clothing, housing, household facilities and articles, transport and communications, culture, education and recreation, healthcare and medical services and others. Household net income is defined as the sum of income from wages and salaries (including migration income), net income from family production, income from properties and income from transfer. Both data sces are is widely used for poverty measurement sces are is widely used for poverty measurement (Chen and Ravallion 1996, Chen and Ravallion 2008) Benjamin et.al, 2005; and to make inferences about household welfare (Brandt and Holz 2006). In addition, the obtains demographic and employment data from a quarterly interview that is administered by the resident supervisor. This includes information on household composition as well as detailed information on migrant workers such as destination, length of stay, income and expenditure during migration. 2.3 Empirical relevance 4 This definition of the households corresponds to those used in other widely used Chinese datasets such as the RCRE and CHIPS. 8

Evidence from other countries suggests that changes in household definition may indeed give rise to differences in estimated household welfare. In Vietnam s Household Living Standards Survey (VHLSS), for example, the definition of households was changed between 1998 and 2004. In the 1998 data, migrant households have lower expenditure than non-migrant ones (de Brauw and Harigaya 2007) whereas after the change, households with migrants who migrate more than 6 months but less than a year have higher expenditure than non-migrant ones (Nguyen and Winters 2011). Either the change in household definition between the two surveys (with migrant workers counted as household members in the 1998 but not in the 2004 VHLSS) or mis-reporting of income/consumption for migrants away from home could explain this. As failure to incorporate changes in household size may underestimate the impact of migration on income and poverty, a number of studies performed a rough adjustment for changes in effective size of migrant households by deducting the number of migrants from household size. In Ghana and Albania, two economies that depend on migration to a significant extent, adjusting for household size is shown to increase the estimated impact of migration to between 2.7 and 4.5 or 2.5 to 3.8 times the uncorrected impact, respectively. Impacts on poverty estimates are even bigger, between 2.6 and 4.4 times in Ghana and 1.1 to 2.5 times in Albania (Schiff 2008). In Mexico, a 2003 National Rural Household Survey obtained complete migration histories for more than 20 years from each household, making it possible to capture migration-induced changes in household size to obtain a consistent measure of per capita income/consumption (Mora and Taylor 2006). But imputing migrants home earnings from observations on non-migrants is not simple and involves a number of key decisions (Barham and Boucher 1998). In China, the issue has been discussed conceptually (Rozelle et al. 1999) and some studies aim to make empirical adjustments to address it. Household data from two 9

provinces illustrate that, although aggregate rural household income falls by 3%-12% as a result of migration, per capita income for those left behind increases by between 16%- 43% (Taylor et al. 2003). A sub of NBS 2008 data from 9 major migration sending or migration receiving provinces (Hebei, Jiangsu, Anhui, Zhejiang, Henan, Hubei, Guangdong, Chongqing and Sichuan) leads to similar conclusions: per capita income is lower for migrant households if migrants are counted as household members but the opposite holds if household size is adjusted to account for migrant members time or residence in the household (Yue and Luo, 2010). While all of these studies support the notion that adjustment is important, the focus on migration-related shifts in household size accounts only for one of the two potential sces of bias. If information on migrants income or consumption is systematically biased, such adjustment will be incomplete and there will be a need to complement it with evidence on actual incomes and spending from the micro-data. 3. Data and methodology We use individual micro-data from China s for 8 provinces to explore the extent to which migrants income may be underestimated. Our descriptive data suggest that, in line with definitions, migrants absent for more than 3 months are consistently considered as members. Data on income and consumption document great heterogeneity in these outcomes. We then discuss options to adjust household size and per capita income and expenditure. 3.1 Data and analytical issues To empirically assess how adjustments for household size or migrant expenditure may affect consumption and income estimates, we use a sub of panel data in 2005 to 2009 inclusive from 8 provinces (211 counties). In each county, one village was randomly selected for inclusion, yielding a total of 2,089 households with 7,498 10

individuals for whom data are available in every year of the 5-year period. As sizes are identical across provinces, provincial rural population in each of the years is used to construct weights. To make expenditure and income data comparable, we express them in 2005 prices using NBS provincial rural consumer price index (CPI) as deflator. Also, to reduce potential bias from cross-province variation of rural prices, we use RBSC s rural CPI at provincial level to construct a cross-province CPI deflator following the method suggested in the literature (Brandt and Holz 2006). To check consistency with aggregate statistics, we compare means from data to province-level information reported in statistical yearbooks. Results from doing so in table 1 highlight that aggregate statistics are close between from data and the entire. The data thus seem suitable for making inferences about the impact of different types of adjustment. Micro-data also allows us to check the extent to which household definitions were applied as intended and to explore the consistency with which expenditure and income is reported. Table 2 suggests that migrant workers are almost universally considered to be household members; in fact there are only very few non-permanent residents in the, most of them students. Regarding income and expenditure we note that, for 42% of migrant households, what is recorded as migration income in the households income module equals migrant remittances from the member level data in the Demographics and Employment module. This suggests that in many cases those in a migrants place of origin may not be fully aware of his or her income and that instructions differed across provinces so that simple back-of the envelope adjustments are unlikely to result in credible estimates. Beyond whether or not migrants are counted as household members or not, there may be biases in reporting of expenditure or income. In the diary of income and expenses filled by residents, consumption by migrants who are not physically present in the 11

household is often excluded. 5 On the income side, the fact that in many cases data report remittances as the only income -which is correct from the perspective of the sending household but not the migrant- would underestimate migration income. 6 The size of the difference will depend on the prevalence of migration; while differences are likely to have been minuscule in the 1980s when levels of migration were very low, they could be substantial for more recent periods and provinces with a large share of internal migrants. 3.2 Adjustments made To obtain consistent figures, we adjust for the number of household members and resce flows in line with the arguments made above. Two methods are used to adjust for variation of household membership. One is to rely on the accepted international definition and drop all members who lived in the household for less than three of the last twelve months. A second is to use information on the number of months a member resides at the place of origin to define effective membership in terms of the share of time spent at the home. 7 Table 3 illustrates the reduction in household size resulting from both adjustments; mean household size decreases from some 4.1 to about 3.5 or 3.6 adult equivalents for adjustments 1 and 2, respectively, a decrease of about 15% or 11%. For income and consumption, adjustment is potentially more complex. We follow Yue and Chu (2010) in assuming that (i) data on consumption recorded in the diary refers only to currently co-resident members, i.e. migrants consumption away from home is omitted; (ii) migrant remittances, representing an inflow for the reporting household, are recorded accurately; and (iii) income by migrants is either reported with 5 This was confirmed by NBSC staff. and is evident from the data: data, among 3571 observations with migrants migrating at least 6 months, 2084 of them have at least one heading migration consumption greater than household total consumption under that heading. 6 http://www.21cbh.com/html/2010-6-9/wmmdawmde4mtqwmq.html. Based on NBS staff, some provinces (e.g. Hunan and Guangdong in ) are much more careful than others in terms of instructing households to keep a consistent record of migration income. 7 In other words, if q is the number of months a member spends at home, his contribution to household size will be q/12. Since most migrant workers are long-term ones, the differences among two adjustments are not big. For distribution of migration periods, see Appendix Table 1. 12

error or inconsistently. This allows us to consumption and income for the core household members in 3 steps. First, we subtract reported migration income earned by all migrants (i.e., those away for more than 3 months) from reported net household income. Second, we add reported remittances by all migrants to what had been obtained in the first step. And finally, we add to what is obtained in the second step the reported migration income earned by non-migrants (i.e, those migrating for less than 3 months). Adjusted net per capita income or consumption is then calculated by dividing adjusted net household income by adjusted household size. 4. Results Using adjusted figures significantly increases migrant households per capita income (by between 12 and 26 percentage points) and consumption (by 23 to 28 points). With the adjustment, households with migrants are estimated to be better off than those without, contrary to what is suggested by standard analysis. Per capita consumption is estimated to increase by between 12 and 13 percentage points. While this leads to a modest increase in inequality within rural areas (with increases in the Gini of slightly less than one point for consumption and 3 points for income), it is likely to narrow the income gap between rural and urban areas and we demonstrate that estimated levels of poverty based on the adjusted measure are significantly below unadjusted ones. 4.1 Consumption and income Means and Gini coefficients for all households adjusted income and consumption per capita in different years are in table 4, with and without adjustment. We note that with adjustment, income and consumption increase, suggesting that NBS traditional household definition has underestimated rural households well-being. Under adjustment 1, which corresponds to international practice, per capita income increases by between 9.5 and 12.2 % and consumption by between 10.8 and 12.5 % (figures for adjustment 2 13

are 5.4-7.9 % and 11.3-13.3 %. In light of the similarly sized changes for income and consumption, the ratio of consumption over net income is thus little affected, in fact its mean and time trend are close for figures with and without adjustments. We do, however, note some time variation whereby, between years, it was highest in 2007, pointing to consumption smoothing during crisis. The right panel of Table 4 indicates that adjusted figures increases estimated inequality. While the size of the increase is marginal for the consumption-based measure (less than 1% in both cases), it is slightly larger for income-based measures. As adjustment will not affect estimated income for those without migrants, it will be of interest to see how it changes estimated welfare for migrant households. The first two columns of table 5 point towards marked differences between households with and without migrants in terms of per capita consumption with the former only consuming between 88% and 78% (in 2007) of the latter [note that differences are less pronounced for income]. This relationship is completely reversed with adjustment which increases migrant consumption by between 24 % and 27 % so that households with migrants enjoyed on average consumption that was higher by 8.8%-10.5% in all years except in 2007 when, due to the crisis, per capita consumption by both was almost equal. The same holds for income per capita which is between 8.8 and 11.7 % higher under adjustment 1or between 5.4 and 7.9 % under adjustment 2. In the case of consumption, results from the two adjustment methods are even more consistent with adjusted consumption expenditure per capita being 11-12 or 12-13 percentage points higher, respectively. As cdfs of per capita income and expenditure in appendix figure 1 illustrate, the adjusted measure dominates the unadjusted one fro consumption but not for income. 4.2 Poverty measures 14

To explore how adjustment affects measured poverty, we use the 1.25$ per day poverty line that would translate into an unadjusted Y 1579/year and Y 1152/year using the adjustment suggested by Chen and Ravallion (2008). 8 Results from computing the poverty head-count (P 0 ), the poverty gap (P 1 ), and the squared poverty gap (P 2 ) for this and the $1/day poverty line are reported in table 6. Beyond the marked reduction in poverty visible over the period, as a continuation of earlier drops in poverty, adjusting the data to correspond to international definitions further reduces estimated levels of poverty, with a drop of the headcount by between 2 and 5 percentage points. While adjustment also reduces the poverty gap, the decrease is less pronounced, suggesting that it is the poorest of the poor are less affected. As they use the same poverty line, results for 2005 in Table 6 are directly comparable with Chen and Ravallion (2007). Not surprisingly, as theirs is a national whereas s is restricted to rural areas, their headcount ratio is 26.4% and 15.6% for the unadjusted and adjusted poverty line, respectively, lower than figures of 37% and 16.7%. However, with adjustment, head count ratio (13.4% or 13.9%) is significantly below the national one reported by them. The same pattern holds for the poverty gap. The traditional household definition by NBS thus seems to result in an over-estimate of poverty. Adopting a definition in line with global practice would imply that China s progress in reducing poverty was even faster than traditionally believed. To demonstrate that this result holds independently of the specific poverty line applied, we report similar figures of the $1/day poverty line in the right half of panel 6. As can be easily verified, the substantive result is not affected. 5. Conclusion and policy implication 8 Using 2005 International Comparison Program (ICP) data, the consumption PPP for China in 2005 is 3.46 Yuan to the dollar so that we would obtain a consumption poverty line of 3.46 1.25 365 1579 Yuan per year. As 2005 ICP prices are not - representative of China s rural areas, it has been suggested to reduce them by 37% to account for price differences (Chen and Ravallion, 2008) and we follow this suggestion here. 15

The way in which household membership is defined and expenditure or income is accounted for has far-reaching impacts on estimated levels of poverty and, in light of China s global importance, achievement of the MDGs. Systematic differences in the way migration is accounted for between China and the rest of the world could thus have farreaching impacts on estimated levels and changes in income. Use of micro-data allows us to demonstrate that adjusting for these will affect estimated consumption growth and poverty reduction as well as the relative welfare of migrants vs. non-migrants. Making provisions in line with this will be important as migration is likely to become even more important in the future. 16

Table 1: Comparing income and consumption statistics between year book and data Net income per capita 2005 2006 2007 2008 2009 National 3255 3587 4140 4761 5153 Heilongjiang 3221 3671 3552 3933 4132 4620 4856 5149 5207 5639 Zhejiang 6660 6400 7335 7387 8265 8439 9258 8957 10007 9708 Anhui 2641 2963 2969 3138 3556 3642 4203 4616 4504 4679 Henan 2871 2925 3261 3644 3852 3866 4454 4920 4807 5258 Hunan 3118 3098 3390 3231 3904 3921 4513 4347 4909 4914 Guangdong 4691 4696 5080 4846 5624 5645 6400 6403 6907 6919 Sichuan 2803 3013 3002 3348 3547 4205 4121 4648 4462 5300 Guizhou 1877 1878 1985 2116 2374 2394 2797 2959 3005 3224 Total consumption per capita 2005 2006 2007 2008 2009 National 2555 2829 3224 3661 3994 Heilongjiang 2545 2816 2618 2865 3117 3362 3845 3849 4241 4660 Zhejiang 5433 5971 6057 6369 6802 7667 7534 7238 7732 7683 Anhui 2196 2333 2421 2346 2754 3083 3284 3689 3655 3465 Henan 1892 1913 2229 2264 2676 2832 3044 2962 3389 3534 Hunan 2756 2867 3013 3001 3377 3332 3805 4063 4021 4266 Guangdong 3708 3511 3886 3684 4202 4619 4873 5034 5020 5158 Sichuan 2274 2398 2395 2587 2747 3195 3128 3600 4141 4679 Guizhou 1552 1548 1627 1573 1914 2368 2166 2280 2422 2669 Food consumption per capita 2005 2006 2007 2008 2009 National 1162 1217 1389 1599 1636 Heilongjiang 924 986 924 959 1077 1140 1268 1301 1331 1383 Zhejiang 2061 2094 2219 2190 2431 2393 2779 2825 2812 2842 Anhui 1000 1014 1045 1054 1193 1253 1454 1564 1494 1581 Henan 859 893 912 947 1017 1043 1166 1146 1220 1230 Hunan 1433 1454 1463 1453 1675 1726 1948 2002 1968 2058 Guangdong 1789 1805 1887 1872 2088 2087 2389 2399 2426 2465 Sichuan 1244 1330 1216 1299 1436 1618 1628 1879 1741 2024 Guizhou 820 849 838 851 998 1050 1120 1193 1094 1098 Sce: Statistical yearbooks and own computation from NBS 2005-2009 panel 17

Table 2: Numbers of non permanent residents and migrants Year 2005 2006 2007 2008 2009 No.of households 2109 2097 2089 2090 2089 Number of migrants 1,178 1,250 1,259 1,204 1,272 --migration period >=3 months 1,113 1,201 1,197 1,157 1,205 --migration period >=6 months 974 1,072 1,089 1,024 1,085 No. of individuals 7,498 7,447 7,215 7,109 6,949 No. of non permanent residents 42 43 32 41 39 --enrolled students 25 24 24 26 23 --migrant workers 8 10 4 7 11 --others 9 9 4 8 5 Sce: Own computation from NBS 2005-2009 panel 18

Table 3: Mean household sizes under different definitions Traditional measure Adjustment 1 Adjustment 2 Persons Persons % of trad. Persons % of trad. 2005 4.14 3.59 86.58 3.71 89.65 2006 4.12 3.52 85.59 3.66 88.99 2007 4.09 3.47 84.94 3.61 88.35 2008 4.08 3.49 85.65 3.63 89.06 2009 4.04 3.43 84.89 3.56 88.28 Sce: Own computation from NBS 2005-2009 panel 19

Table 4 Mean of income and consumption under different definitions of household Mean Income per capita Gini coeff. Trad. Adj. 1 % incr. Adj. 2 % incr. Trad. Adj. 1 Adj. 2 2005 3088 3392 9.84 3278 6.15 0.333 0.366 0.358 2006 3448 3786 9.80 3654 5.97 0.342 0.381 0.373 2007 3854 4323 12.17 4158 7.89 0.345 0.383 0.374 2008 4442 4869 9.61 4681 5.38 0.348 0.385 0.376 2009 5159 5649 9.50 5448 5.60 0.353 0.386 0.381 Mean Cons. per capita Gini coeff. Trad. Adj. 1 % incr. Adj. 2 % incr. Trad. Adj. 1 Adj. 2 2005 2405 2666 10.85 2682 11.52 0.346 0.355 0.355 2006 2612 2907 11.29 2935 12.37 0.372 0.382 0.382 2007 3132 3528 12.64 3545 13.19 0.411 0.422 0.421 2008 3323 3700 11.35 3731 12.28 0.389 0.398 0.397 2009 3918 4409 12.53 4439 13.30 0.416 0.425 0.424 Note: Numbers are deflated by provincial rural CPI and weights are applied to adjust for differences in size of provinces. The traditional measure includes all permanent residents reported in NBS household roster. In all income measures, earnings by migrants at destination are excluded and remittances included. For consumption ***** Total household size either excludes those who had been outside their own county for 6 months or longer (Adjustment 1) or is computed in terms of effective membership, i.e. by computing the ratio of total months of migration by members outside their county (Adjustment 2). 20

Table 5: Comparing income and consumption for household with and without migrants Income per capita Year No migrant Traditional Adj. 1 % incr. Adj. 2 % incr. 2005 3109 3069 3750 22.19 3477 13.29 2006 3382 3509 4314 22.94 3986 13.59 2007 3877 3834 4903 27.88 4501 17.40 2008 4428 4455 5426 21.80 4985 11.90 2009 5318 5023 6063 20.70 5610 11.69 Consumption per capita Year No migrant Traditional Adj. 1 % incr. Adj. 2 % incr. 2005 2560 2262 2798 23.70 2824 24.85 2006 2778 2459 3071 24.89 3119 26.84 2007 3536 2769 3518 27.05 3556 28.42 2008 3558 3116 3872 24.26 3924 25.93 2009 4237 3647 4613 26.49 4659 27.75 Note: Figures are deflated by provincial rural CPI, weighted, and adjustments are as defined in the text. 21

Table 6: Levels of poverty under different household definitions Poverty line at $1 (2005 PPP value) Poverty line at $1.25 (2005 PPP value) Trad. Adj.1 Adj.2 Trad. Adj.1 Adj.2 Poverty 2005 0.37 0.319 0.314 0.167 0.136 0.133 headcount 2006 0.336 0.296 0.292 0.167 0.139 0.135 Poverty gap 2005 2007 0.268 0.223 0.219 0.124 0.106 0.104 2008 0.228 0.185 0.182 0.088 0.076 0.074 2009 0.179 0.144 0.142 0.075 0.059 0.058 0.0994 0.0822 0.0807 0.0390 0.0316 0.0309 2006 0.0972 0.0816 0.0798 0.0388 0.0323 0.0315 2007 0.0732 0.0624 0.0610 0.0281 0.0242 0.0235 2008 0.0557 0.0473 0.0462 0.0206 0.0177 0.0172 2009 0.0452 0.0375 0.0368 0.0184 0.0150 0.0147 Squared 2005 0.0395 0.0321 0.0314 0.0139 0.0111 0.0109 poverty gap 2006 0.0390 0.0326 0.0318 0.0137 0.0116 0.0113 2007 0.0290 0.0250 0.0244 0.0103 0.0090 0.0088 2008 0.0213 0.0184 0.0178 0.0072 0.0064 0.0062 2009 0.0183 0.0152 0.0148 0.0071 0.0059 0.0057 Note: We use $1.25 (in 2005 PPP value) as poverty line as recommended by Chen and Ravallion (2007) and Ravallion (2008). 22

Appendix table 1: Distribution of length of migration No of months of migration 2005 2006 2007 2008 2009 0-3 60 47 61 46 65 4.54 3-6 139 129 108 133 120 10.23 >=6 976 1,072 1,088 1,023 1,084 85.24 No. of migrants 1,175 1,248 1,257 1,202 1,269 100 % 23

Appendix table 2: Mean of food and non-durable goods consumption under different household definitions Per capita food consumption Trad. Adj 1 % incr. Adj 1 % incr. Gini coefficient Trad. Adj.1 Adj.2 2005 1086 1202 10.68 1212 11.60 0.264 0.278 0.279 2006 1115 1238 11.03 1253 12.38 0.273 0.29 0.291 2007 1207 1351 11.93 1367 13.26 0.279 0.297 0.298 2008 1351 1500 11.03 1517 12.29 0.295 0.313 0.314 2009 1608 1798 11.82 1821 13.25 0.3 0.319 0.321 Per capita food & non-food consumption Trad. Adj 1 % incr. Adj 1 % incr. Gini coefficient Trad. Adj.1 Adj.2 2005 1802 1896 5.22 1906 5.77 0.305 0.314 0.314 2006 1904 2027 6.46 2043 7.30 0.328 0.338 0.338 2007 2158 2239 3.75 2253 4.40 0.328 0.338 0.337 2008 2448 2486 1.55 2502 2.21 0.337 0.348 0.347 2009 2585 2835 9.67 2863 10.75 0.348 0.359 0.359 Note: Numbers are weighted by provinces, and deflated by provincial rural CPI. The traditional definition includes all permanent residents reported in the household roster. In adjusted definition 1, members who migrated outside of own county for 3 months or longer are excluded. In adjusted definition 2, members who migrated outside of own county for 6 months or longer are excluded. In adjusted definition 3, effective members (calculated as the ratio of total number of months of all members migrating outside of own county and 12 month) are excluded. In all the income measure, migration earnings at destination are excluded, while remittance is included. 24

Appendix table 3: Comparing food and non-durable consumption for household with and without migrants Non migrant Migrant Trad. Adj. 1 % incr. Adj. 1 % incr. 2005 1125 1050 1298 23.62 1311 24.86 2006 1143 1089 1357 24.61 1381 26.81 2007 1240 1178 1491 26.57 1513 28.44 2008 1393 1313 1629 24.07 1654 25.97 2009 1640 1580 1984 25.57 2018 27.72 Non migrant Migrant Trad. Adj. 1 % incr. Adj. 1 % incr. 2005 1834 1593 1973 23.85 1989 24.86 2006 1961 1686 2112 25.27 2139 26.87 2007 2179 1821 2315 27.13 2338 28.39 2008 2426 2054 2558 24.54 2587 25.95 2009 2737 2348 2952 25.72 3000 27.77 25

0.2.4.6.8 1 0 2000 4000 6000 8000 10000 cons cons_adj2 cons_adj1 0.2.4.6.8 1 0 2000 4000 6000 8000 10000 cons05 cons05_adj2 cons07_adj1 cons09 cons09_adj2 cons05_adj1 cons07 cons07_adj2 cons09_adj1 26

References: Acosta, P. 2011. "Labor supply, school attendance, and remittances from international migration: The case of El Salvador." Jnal of Development Studies 47 (6): 913-36. Acosta, P., C. Calderon, P. Fajnzylber and H. Lopez. 2008. "What Is the Impact of International Remittances on Poverty and Inequality in Latin America?" World Development 36 (1): 89-114. Adams, R. H. and J. Page. 2005. "Do International Migration and Remittances Reduce Poverty in Developing Countries?" World Development 33 (11): 1645-69. Alderman, H., J. R. Behrman, H. P. Kohler, J. Maluccio, and S. Cotts-Watkins. 2001. "Attrition in Longitudinal Household Survey Data: Some Tests for Three Developing-Country Samples." World Bank Washington, DC. Au, C. C. and V. Henderson. 2006. "How Migration Restrictions Limit Agglomeration and Productivity in China." Jnal of Development Economics 80 (2): 350-88. Banerjee, A. V. and K. D. Munshi. 2000. "Networks, Migration and Investment: Insiders and Outsiders in Tirupur's Production Cluster." Massachusetts Institute of Technology, Department of Economics Working Paper: 00/08. Barham, B. and S. Boucher. 1998. "Migration, Remittances, and Inequality: Estimating the Net Effects of Migration on Income Distribution." Jnal of Development Economics 55 (2): 307-31. Beaman, L. and A. Dillon. 2012. "Do Household Definitions Matter in Survey Design? Results from a Randomized Survey Experiment in Mali." Jnal of Development Economics 98 (1): 124-35. Benjamin, D., L. Brandt and J. Giles. 2005. "The Evolution of Income Inequality in Rural China." Economic Development and Cultural Change 53 (4): 769-824. Brandt, L. and C. A. Holz. 2006. "Spatial Price Differences in China: Estimates and Implications." Economic Development and Cultural Change 43-86. Cai, F. and D. Wang. 2003. "Migration as Marketization: What Can We Learn from China's 2000 Census Data?" China Review 3 (2): 73-93. Chen, S. and M. Ravallion. 1996. "Data in Transition: Assessing Rural Living Standards in Southern China." China Economic Review 7 (1): 23-56. Chen, S. and M. Ravallion. 2008. "China is poorer than we thought, but no less successful in the fight against poverty." Policy Research Working Paper 4621. Washington, DC: The World Bank. 27

Cox Edwards, A. and M. Ureta. 2003. "International Migration, Remittances, and Schooling: Evidence from El Salvador." Jnal of Development Economics 72 (2): 429-61. de Brauw, A. and S. Rozelle. 2008. "Migration and household investment in rural China." China Economic Review 19 (2): 320-35. de Brauw, A. and T. Harigaya. 2007. "Seasonal Migration and Improving Living Standards in Vietnam." American Jnal of Agricultural Economics 89 (2): 430-47. de Brauw, A., J. Huang, S. Rozelle, L. Zhang and Y. Zhang. 2002. "The Evolution of China's Rural Labor Markets during the Reforms." Jnal of Comparative Economics 30 (2): 329-53. Docquier, F., E. Lodigiani, H. Rapoport, and M. Schiff. 2011. "Emigration and Democracy." Policy Research Working Paper 5557. Washington, DC: World Bank. Du, Y., A. Park and S. Wang. 2005. "Migration and Rural Poverty in China." Jnal of Comparative Economics 33 (4): 688-709. Fleisher, B. M. and D. T. Yang. 2003. "Labor Laws and Regulations in China." China Economic Review 14 (4): 426-33. Giles, J. 2006. "Is Life More Risky in the Open? Household Risk-Coping and the Opening of China's Labor Markets." Jnal of Development Economics 81 (1): 25-60. Glewwe, P. and H. G. Jacoby. 2004. "Economic Growth and the Demand for Education: Is There a Wealth Effect?" Jnal of Development Economics 74 (1): 33-51. Jalan, J. and M. Ravallion. 1998. "Are There Dynamic Gains from a Poor-Area Development Program?" Jnal of Public Economics 67 (1): 65-85. Mansuri, G. 2006. "Migration, sex bias, and child growth in rural Pakistan." World Bank Policy Research Working Paper 3946. Washington, DC: World Bank. McKenzie, D. and H. Rapoport. 2006. "Can migration reduce educational attainment? Evidence from Mexico." World Band Policy Research Working Paper 3952. Washington DC: World Bank. Mendola, M. 2008. "Migration and Technological Change in Rural Households: Complements or Substitutes?" Jnal of Development Economics 85 (1-2): 150-75. Mora, J. and J. E. Taylor. 2006. "Does migration reshape expenditures in rural households? Evidence from Mexico." Policy Research Working Paper 3842. Washington, DC: The World Bank. Nguyen, M. C. and P. Winters. 2011. "The Impact of Migration on Food Consumption Patterns: The Case of Vietnam." Food Policy 36 (1): 71-87. 28

Ratha, D. and W. Shaw. 2007. "South-South migration and remittances." Working Paper 102. Washington DC: World Bank. Ravallion, M. and S. H. Chen. 2007. "China's (uneven) progress against poverty." Jnal of Development Economics 82 (1): 1-42. Rosenzweig, M. R. and O. Stark. 1989. "Consumption Smoothing, Migration, and Marriage - Evidence from Rural India." Jnal of Political Economy 97 (4): 905-26. Rozelle, S., G. Li, M. Shen, A. Hughart and J. Giles. 1999. "Leaving China's Farms: Survey Results of New Paths and Remaining Hurdles to Rural Migration." China Quarterly 158 367-93. Schiff, M. 2008. "On the Underestimation of Migration's Income and Poverty Impact." Review of Economics of the Household 6 (3): 267-84. Taylor, J. E., S. Rozelle and A. de Brauw. 2003. "Migration and Incomes in Sce Communities: A New Economics of Migration Perspective from China." Economic Development and Cultural Change 52 (1): 75-102. Whalley, J. and S. M. Zhang. 2007. "A numerical simulation analysis of (Hukou) lab mobility restrictions in China." Jnal of Development Economics 83 (2): 392-410. Woodruff, C. and R. Zenteno. 2007. "Migration networks and microenterprises in Mexico." Jnal of Development Economics 82 (2): 509-28. Yang, D. 2008. "International migration, remittances and household investment: Evidence from Philippine migrants' exchange rate shocks." Economic Jnal 118 (528): 591-630. Yang, D. 2011. "Migrant Remittances." Jnal of Economic Perspectives 25 (3): 129-52. Zhao, Y. 2002. "Causes and Consequences of Return Migration: Recent Evidence from China." Jnal of Comparative Economics 30 (2): 376-94. 29