The Role of Grandchildren in Old-age Support: Evidence from China (Preliminary)

Transcription

1 The Role of Grandchildren in Old-age Suort: Evidence from China (Preliminary) Xuezhu Shi 1 1 Deartment of Economics, STICERD, London School of Economics, United Kingdom. x.shi2@lse.ac.u Aril 3, 2017 Abstract Suort from children has always been one of the income sources for ageing arents in develoing countries. This aer resents the emirical evidence for indirect recirocities. The indirect recirocities address relationshis between three successive generations: grandarents, arents and grandchildren. This aer focuses on how the relationshis affect old-age suort received by the grandarents. Using a olicy change in China as the main identification strategy, the aer shows that sons with male offsring are more liely to suort their own arents comared to those with female offsring. For females, the grandchildren-gender effects exist but are less robust comared with males. However, the gender of grandchildren only affects old-age suort for grandarents when combining with the gender of arents. Keywords Old-age suort, Intergenerational transfers, Indirect recirocities I wish to than Nava Ashraf, Joan Costa-i-Font, Henri Kleven, Camille Landais, Dir Van de gaer and esecially Fran Cowell for their comments and suggestions for this aer in its early stages. Also, I want to than Sarah Clifford, Shiyu Bo, Jiajia Gu, Chao He, Yatang Lin, Panos Mavroonstantis, Arthur Seibold, Eddy Tam and other articiants in the Public Economics WIP Seminar in LSE. 1

2 1 Introduction In many countries, it is common for children to act as a suort scheme for the elderly. Esecially in develoing countries with underdeveloed ension marets and incomlete financial marets, it is a way to offset ossible riss for the elderly. This imortant role for children sometimes rovides a major motive for having children. 1 As reliable future income suorts, arents exect their children to rovide at least artial suort in their old-age and to imrove the rosect of the adequate rovision in the face of uncertainty. Many aers have focused on various characteristics of children that might affect old-age suort, esecially on the number and the gender of children. However, few studies examine the role of other family members, for examle, the offsring of their children. The existence of other family members, commonly the offsring, can affect savings, human caital investments and other ecuniary decisions (Barro and Becer, 1989; Wang et al., 1994 and Coeurdacier et al., 2015). So it is reasonable to believe that the offsring of children can affect old-age rovision. Grandchildren may affect the decisions of children on old-age suort for grandarents. 2 This aer fills a ga by examining the effects of the gender of grandchildren on old-age suort received by grandarents, jointly with the gender of children. It reveals a ossible mechanism behind the effects arising from the third-generation. The theoretical framewor underlying the effects of grandchildren on oldage suort received by grandarents is the Indirect Recirocity. This concet was firstly mentioned by a French anthroologist (Mauss, 1968), who used it to characterise a "gift-return-gift" relationshi between three successive generations. The relationshi can lead to the infinite chains of transfers (Arrondel and Masson, 2006). The characteristics of grandchildren can affect the interactions between arents and grandarents through this indirect recirocity. To reframe the indirect recirocity using an economic aroach and to mae it more alicable to the research question, I construct a model based on the indirect recirocity model of Cox and Star (1996). The model studies the connection between the gender of grandchildren and the old-age suort received by grandarents. It leads to the following conclusions: females will rovide more to their arents if they have female offsring, and males will rovide more with male offsring. The aer sets out to test emirically the redictions from the model. The reasons for using this Chinese data set is that first, there are strong family ties in China comared to Western countries. As a country embedded in Confucianism for more than a thousand years, roviding suort to ageing arents has become a ecuniary way of evaluating a erson s moral worth. 1 The motive is one of the underlying assumtions in macroeconomics literature focusing on fertility and economic growth and also examined in alied microeconomics literature. There are many aers discussing fertility and saving in Macroeconomics, such Barro and Becer (1989), Becer, Murhy and Tamura (1990) and Wang et al.(1994). Recently, Coeurdacier et al.(2015) study the One-Child Policy effects on savings. While in Microeconomics, there are Banerjee et al.(2010), Chouhmane et al. (2015), Cox and Star (1996). 2 Three generations: grandarents, arents and grandchildren 2

3 Also, suort by children is an imortant source of income for the elderly in China. The coverage of the ension system is relatively low, esecially in rural areas, not to mention the increasing burden on ublic ension system with the ageing oulation in China in recent years. 3 The emirical data set used in the aer is the China Health and Retirement Longitudinal Study (CHARLS). This data set rovides information on transfers among family members, family demograhics, income, consumtion, and other asects for survey resondents aged 45 or above. The information rovided by the data set is detailed enough to construct a samle with demograhic variables for three generations. The main emirical findings indicate that males with male offsring have a higher robability of roviding old-age suort to their arents than those males with female offsring, controlling for other variables. The results from the robustness checs show that females with sons are less liely to rovide comared with females with girls, but the results for females are less robust comared with the male results. The results match with the redictions from the model. The conclusion is that the robabilities of old-age suort rovision are closely lined to the gender of grandchildren and children. For non-ecuniary suorts, such as total visit days in a year, the gender effect of grandchildren is more significant for females than males. My emirical strategy is to use an interaction term of the gender of the arents and grandchildren to cature effects of the gender comosition on the rovision of old-age suort. There are otential biases from gender selections for the grandchildren. I tae advantage of a olicy change with constructed instrumental variables to alleviate the ossible selections in the gender of grandchildren. 4 The olicy change refers to a ban of gender selection technologies and consequential fines for those who conduct the gender selection by National Family Planning Commission (NFPC) in Also, other instrumental variables as the gender of the oldest child in a household. Detailed exlanations will be rovided in the later section. The aer roceeds as follows. Section 2 rovides the theoretical bacground in anthroology which leads the rediction of the gender of grandchildren affects the rovision of old-age suort. More bacground information on old-age suort and son reference in Section 3. Then it follows by Section 4 for the emirical findings. The bacground for the olicy change by NFPC is rovided also in Section 4. Section 5 concludes. 2 Indirect recirocity Altruism and exchange are two main motives in the standard model when analysing inter-generational transfers. However, the emirical results are not robust enough to suort these two motives in theoretical models. (Arrondel 3 Table.12 in the Aendix rovides more detailed information. 4 The arent generation in the survey are free from the gender selection bias because the technology of ultrasound B was not yet available nor common at the time of their births. 3

4 and Masson, 2006). 5 The theory of indirect recirocity acts as a reconciliation of the exchange and altruism motives. The concet of indirect recirocity is firstly mentioned by French anthroologist Mauss (1950, 1968). He exands the common "gift-return" recirocity relationshi between two generations to three generations and states that indirect recirocities involving three successive generations will lead to infinite chains of transfers. It also can be regard as a transmission of cultural traits and norms. The transfers by every generation mentioned by Mauss can be descending or ascending. Since this aer focuses on uward transfers from children to arents, I will only introduce relevant theories about ascending transfers. There are two existing theories: the "ension game" by Hammond in 1975 and the demonstration effect roosed by Cox and Star in In the ension game, layers anticiate the reactions of the next generation. To determine one generation s beliefs, it needs to determine only the next generation s strategy (Hammond, 1975). In this game, although the assumtions are based on the successive generation setting (an infinitely-reeated game), the actual beliefs or strategies determination rocess only involves two generations. The demonstration model roosed by Cox and Star includes that the characteristics of the third-generation. The model states that adult arents demonstrate to their children of roviding old-age suort to their own arents in order to shae their children s corresonding behaviours. The arents will tae the next generation s characteristics into consideration when roviding oldage suort. The model redicts with the resence of the grandchildren has a ositive effect on transfers from the arents to the grandarents. Cox and Star also test the rediction emirically in their 1996 and 2005 aers, using U.S. data sets. Combined the indirect recirocities and also the demonstration effect, I construct a model resented in Aendix. The model redicts males will rovide more to their arents if they have more sons and females rovide more with more daughters. There are other aers discussing the gender of grandchildren affecting the time sent with grandarents (Wolff, 2001; Mitrut and Wolff, 2009). None of the existing literature has been focusing on the gender effect of grandchildren on ecuniary rovisions for the elderly and also none of them studies the effectiveness of demonstration effect on different combinations of gender from arents and grandchildren. 3 Imortance of old-age suort in China Providing suort to ageing arents has been acting as one of the imortant sources for the elderly s economic well-being in China. Even though the ension and healthcare system for the elderly has been imroving, most of the ageing arents still receive regular or non-regular old-age suort from their children. In urban areas, 45.4% of the elderly had the ension as their rimary source 5 Most of literature and statements here are based on Chater 14 in Handboo of the Economics of Giving, Altruism and Recirocity, written by Arrondel and Masson. 4

5 of suort in 2005, while in 2010, this ercentage increased to 66.3%. 37% of them regarded family suort as the rimary source in 2005 and in 2010, it decreased to in urban areas. The ension system in rural areas is still not as well develoed as it is in urban areas. The same ercentage for the ension in rural was 4.6% in 2005 and unchanged in 2010; while the ercentage of the family suort in 2010 decreased by 7% comared to 2005, but nearly half of the elderly still regards the suort from family as the rimary source. 6 The statistics show that even with the develoment of the national ension system, old-age suort from children is still an imortant source for the elderly in China. Traditionally, males are resonsible for roviding old-age suort to arents. This is one of the reasons for son reference (Lee et al., 1994; Chan et al., 2002): "the more sons, the more blessings". It was common in China for households to have at least one son even right u to the imlementation of the "One- Child" Policy (OCP) (Milwertz, 1997; Ebenstein and Leung, 2010). So when the government introduced the OCP, son reference was a major obstacle. The gender ratio was balanced in the early years of OCP s imlementation, but it has been increasing since. The OCP imlementation was therefore followed by a camaign of gender equality in 1995, when the gender ratio of newborns had already increased to The government romoted the modern view on females: females should be the same as males, in terms of wor, education and also suorting the elderly. The government believed the camaign would wor and alleviate the foreseeable unbalanced gender ratio of males to females for the newborns in future. 8 It is a common belief that the advanced medical technology triggered the high gender ratio. The technology B-mode ultrasound allows eole to now the fetal sex and it has been commonly used in all areas after 1980 (White, 2001). Chen et al. (2013) conclude the increasing gender ratio is attributed to increased gender selection at birth. However, the technology is not the fundamental reason for abortion decisions. If there are no other underlying economic or social incentives, the technology itself will not be used as a tool for gender selection. Tae Jaan as an examle, the gender ratio of males to females newborns has been around 1.05 since Economic incentives that arents might obtain from sons are the ecuniary driver behind and one of the incentives might still be old-age suort rovided by offsring. Ebenstein and Leung (2010) study the effect having regulated 6 More detailed numbers are in Table.12 in the Aendix and the source of this data is from National Bureau of Statistics of China. 7 The China s government set "Gender Equity" as one of the basic state olicies in the Fourth World Conference on Women. Website htt:// The gender ratio comes from % samle oulation census. 8 The gender ratio of males to females for the newborns will be the gender ratio in the following art of this aer. 9 The data are available from the Statistics Bureau and the Director-General for Policy Planning of Jaan. A figure of Jaan s gender ratio for the newborns will be shown in Aendix. 5

6 ension system s effect on the sex ratio at birth in China. They conclude that old-age suort decreases the gender ratio of newborn children at county-level. Also, Qian discovered increasing income for females also increases female survival rate in her 2008 aer. Oliveira (2016) regards intergenerational suort as being largely affected by the demograhy of the offsring. Cox and Star (2005) also find that economic incentives affect reroduction of grandchildren. There are also researchers who view raising children as a rocess of self-enforcing recirocity norms, such as Guttman (2001). 10 Concluding the revious literature, old-age suort might be one of the economic reasons that affecting the demand for offsring. 4 Emirical findings 4.1 Data The data set that I am using is the China Health and Retirement Longitudinal Study (CHARLS) 2011 baseline survey. CHARLS is a longitudinal survey for 28 out of 34 rovinces for three waves in 2011, 2013 and 2014 u to now. 11 I use the 2011 data set in this aer as a cross-sectional data set. 12 The data set contains information on family, wor, retirement, wealth and also health for resondents. The main demograhic grou in the survey is eole aged 45 or above. In the 2011 samle, there are about 17,708 individuals in 10,257 households from two rovinces. The samle is randomly selected from four different level samling: county-level, neighborhood-level, household-level and resondent-level. 13 The ey information I need is the inter-generational transfers between the survey resondents, their revious generation and the next generation. Because the majority of the resondents have children at woring ages, most of them are receiving regular or non-regular suort from their children. Given the age of the resondent, there are fewer observations available if I limit the samle to the resondents roviding suort to their arents. So I choose the old-age suort that the resondents received from their next generation as the main transfer that used in this emirical section. The resondents in the survey are the assive old-age suort receiver in the model and the main regressions. I describe the resondent as G, their children as P and their grandchildren as K. The regressions are based on P while the survey is focused on G. I rearrange the data and construct a new data set with the main resondents are P with children. However, the original data set is designed for G, the demograhic information for P is not as detailed as G. The available demograhic variables about K are only their gender and how many children that a household from P 10 There are many aers discussing fertility and saving in Macroeconomics, such Coeurdacier et al.(2015) and Becer and Barro (1988). While in Microeconomics, there are Chouhmane et al. (2015) and Cox and Star (1996). 11 The detailed rovinces and counties distribution is resented in Fig Panel data for CHARLS will be used later 13 Detailed samling method at each level can be accessed at: htt://charls.u.edu.cn/en/age/about-samle

7 has. So, some assumtions need to be made when constructing the instrumental variables, which will be discussed later. The samle sized decrease to around 11,000-12,000 observations in the newly constructed samle. 4.2 Main regression The main regressions comare the robability and the amount of roviding different inds of suorts for households in P. The households are classified into different categories by the gender comosition of P and K. The main regression is y i = α + βmore_malek i + γp g i + δ(p g i More_maleK i ) + X iθ + ɛ i. (1) In the equations, i stands for household i in P. y i is outcome variables testing different asects of old-age suort. There are six different outcome variables used in this aer. Five of them are financial old-age suort: dummies equals 1 if P rovides any, regular and non-regular transfers and natural logarithms of the amount of regular and non-regular transfers rovided. Regular transfers are the transfers that P rovides transfers to his or her arents at fixed time. The non-regular transfers are the transfers that P rovides transfers to his or her arents at festivals, birthdays, weddings, funerals and also for other non-regular events. The dummy for any transfers equals 1 if P rovide regular or/and nonregular transfers. For the amount of the transfers, I unify the amount of the transfers to the amount of transfers er year. The one non-ecuniary outcome is the natural logarithm of the visit days er year. More_maleK i is a dummy that equals 1 if the gender ratio of K in a P household is larger or equal to 0.5. The gender ratio of K equals the number of boys in household i divided by the number of girls in the household if the household has more than one child. For households with one child, if the only child is a boy, then More_maleK = 1 and if it is a girl, then More_maleK = 0. P g i reresents the gender of P who is the child of the survey resondents and who reresent their household in the data. More_maleK i P g i is the interaction term and X i is the set of demograhic variables to be controlled for in the regressions. The error term is ɛ 1,i and it is clustered at household-level in generation G. The summary statistics for the ey deendent variables and also regressors are in Table 1. Table 2 shows the summary statistics for females and males searately. In the regression, I control for the household income of G and also the dummy variable that indicating whether on average G is enrolled with the ension as Ebenstein and Leung (2010) do in their aer. 4.3 The NFPC regulation and instrumental variables Emirical results from the regressing equation (1) might suffer from biases caused by some endogenous roblems. One of the concerns is the gender selection roblem for the variable More_maleK i. P g i is almost free from this 7

8 roblem because the average age of the arent generation in the samle is 39. At their time of birth, the gender selection technology was not yet available in China. To alleviate the gender selection bias for variable More_maleK i, I use a regulation imlemented in late 2002 together by the Ministry of Health, State Food and Drug Administration (SFDA) and National Family Planning Commission (NFPC) in China. The regulation bans the use of B-scan ultrasonograhy and other technologies for determining fetal sex from It states that any methods of gender selection should be banned. There are fines for different levels of violation of the regulation by individuals choose to select the fetal sex and the hositals that conduct these selections. This regulation initially was not as effective as the central government exected. The government found that it is difficult to eliminate the gender selection behaviours by an unified regulation that imlemented at rovincial-level. Two years after the initial imlementation of this NFPC regulation (2005), the Ministry of Health with NFPC started a year-long camaign targeting on illegal clinics and under-qualified doctors. The camaign esecially targeted on closing down clinics that conducting illegal gender selective abortions and also illegal gender tests. The camaign was imlemented by refectural governments, so the local governments can have better controls on the actual imlementation of the camaign. Different refectural-level cities have different ways of starting this camaign. Some of the cities have been imlementing this camaign every year since 2005 and others might just have the camaign only for 2005 and sometimes even started the camaign later than the NFPC required year. How many years cities have been enforcing the camaign and also the starting year of the camaign can be an indicator of the strictness of the regulation on gender selection behaviours at refectural-level. I collect the relevant information from different government websites and also the newsaer about when and how many year refecturallevel cities has been enforcing this camaign and generate an index indicating how strict the cities are about the gender selection behaviours. These camaigns and regulations of course cannot eliminate all the gender selection behaviours, but they can alleviate some of the gender selection behaviours. 15. I also have to use the index that I create before together with another instrumental variable that maes the gender of K more exogenous. This IV is the gender of the oldest child in household i. The OCP does not strictly limit all households to have only "one child", esecially in rural areas. Usually, the first-born children have the lowest gender ratio comared with the gender ratio for other birth sequences children. Figure 1 shows the statistical data from National Poulation Census in China, the gender ratio for the newborns as the first child increased from 1.05 to around While for children with other birth sequences, the gender ratios are all larger than the gender ratio for the first-born children. These two variables will be used as instrumental variables 14 Website: htt:// 15 The yearly national-level gender ratio is in Figure 2 and the increasing trend of the gender ratio had been decreased after 2006 and started to increase around This shows that the camaign did have some effects. 8

9 for More_maleK i in the regression equation (1). There are some assumtions that I have to mae when constructing the instrumental variables. As discussed before, I restructure the data set from a data set with main resondents from G to a data set with main resondents from P. But in the original data, there is information on the birth year for art of the grandchild generation. The year of birth is available if grandchildren are living with G, but there is much data still unnown. However, the data does rovide information on the gender and number of the K that is above age 16. I can use this information to wor out the gender of the first child for most of the households. But some estimations are still needed in this rocess. 4.4 Regression results The current literature finds that the gender of P does not affect the rovision of old-age suort (Oliveira, 2016). It is a rather surrising emirical result given the gender inequality in the labour maret and also the rooted son references in China. I want to verify Oliveira s results by running ure correlation regressions, only to observe what the data set shows about the gender differences of P in terms of old-age suort. The regression equation for this art is simler than equation (1): y i = α 0 + β 0 P g i + X i θ 0 + ɛ 0,i. (2) Table 3 shows the correlations for the gender of P and the suorts rovided. There are no gender differences for any transfers rovided, but if I loo into detailed tyes of transfers, the gender differences aear. For regular transfers, male P have higher robabilities of rovision, while for females, they are more liely to rovide non-regular transfers instead. For the transfer amount, males significantly transfer more in terms of the regular transfer amount and the gender difference is not significant for the non-regular transfer amount. The gender difference for the non-ecuniary suort is also significant: males visit their arents more than female. Summing u the results, it is difficult to conclude that male P now always rovide more suorts to G than female P. However, males still rovide more regular suorts to their arents than females Gender effects of K and old-age suort There might be some other gender-related factors affect old-age suort. From the theory of indirect recirocity, the characteristics of grandchildren might also affect old-age suort received by G. The effect of numbers of grandchildren is studied by Cox and Star (1994) and the conclusion they have is that the number of grandchildren increases the uward suort from arents to grandarents. One of the characteristics that I want to focus on is the gender of grandchildren. Because the indirect recirocities usually wor through arents demonstrating to their children, and this mechanism wors better for the same-sex arents and children (Bussey and Bandura, 1999). Using the regression equation (1), the results for OLS regressions are dislayed in Table 4. 9

10 There are four different tyes of households in P : a male in P with more male offsring in K (MM), a male in P with more female offsring in K (MF), a female in P with more male offsring in K (FM) and a female in P with more male offsring in K (FF). The redictions are MM (FF) households should on average rovide more transfers and ay more visits than MF (FM) households. In Table 4, the baseline grou is FF households, "0 1" reresents for FM households, and "1 0" and "1 1" reresent MF households and MM households resectively. From the corresonding regression results, the comarison between FF and FM grou is only significant for the ln (visit days). It is also difficult to conclude the comarison between MF and MM households from Table 4. I run the regression searately for the different gender of P. The searated regression results are in Table 5. For regular transfers, non-regular transfers and any tyes of transfers, male P are more liely to rovide with more sons in their family. While the results in Table 3 show that males are less liely to rovide than females on average for non-regular transfer, but more for regular transfers. Table 5 also shows that for the absolute amount of financial suorts, the effects of gender comosition of K are not significant. Also for the days of visiting arents, the gender effect of K is insignificant for male P. For females, the gender effects of their offsring are not significant for any ecuniary outcomes. But for visit days, females with more daughters are more liely to visit their arents than females with more sons. Visit days for male P are not affected by the grandchildren gender comosition. Female P on average visit their arents less than male P. 16 Interreting the results of Table 3 and 5 together, they show a very interesting henomenon. For robabilities of roviding transfers and the natural log of visit days, the effects of the gender comosition of K are differently for male P and female P. Males are more liely to be affected by the gender comosition of his offsring for ecuniary transfers and females want to demonstrate more to their daughters through non-ecuniary values. This is the conclusion from the OLS results. Due to the gender selection roblems for the newborns in China and also the increasing gender ratio, the gender comosition of K in each household could be endogenous with references, incomes or other variables at household-level. Using the instrumental variables discussed in Section 4.3, I can alleviate these ossible endogeneity roblems by the gender selection behaviours. The regression with IVs are dislayed in Table 6. Comared with the OLS results, using instrumental variables boosts the third-generation gender effects for any-transfer, nonregular and ln (visit days) for male P, but also maes the gender effect less significant for regular transfers for males and ln (visit days) for females. The IV results show male P are more roense to rovide old-age suort to arents if they have more sons in their family. 16 This is might because more grandarents are living together with male P s. 10

11 4.5 Robustness chec Family comositions of P One of the ossible reasons for why some of the results cannot fully match with the anticiation model is that sometimes eole need to or are resonsible for suorting their arents. They have to suort their ageing arents, regardless of the gender of their offsring. It is true for most of the males in China and also true for female P, if they do not have older brothers. To test this assumtion, I searate the samle into four different grous: female P with or without older brothers and male P with or without older brothers. I use the same regression equation and the identification methods that I used in the revious section and the results are showed in Table 7. The results indicate that the gender difference in P are less significant for those with older brothers than it is for those without, excet for the amount of the regular transfer The results imly that for males without older brothers, the gender comosition of K ositively affects the visits to arents, the robability of roviding non-regular suorts, but negatively for the amount of regular transfer. There is no effect from the gender of the third generation on any financial suorts for them comaring with males with older brothers. For females, the results still imly that for those with or without older brothers, the gender of their offsring does not have significant effects on ecuniary and non-ecuniary outcomes. Most of the results here corresond to the results in Table 6. In Chinese culture, male P who are the oldest son are usually considered to be the one to tae care of their ageing arents comared to those with older brothers. If the oldest sons want to ass this culture onward to their offsring, they should suort arents more if they have more sons in their own household. So, the results are showing that the coefficients for males without older brothers are larger and more significant than those for males with older brother. There is only one inconsistent result for the outcome variable. The amount of regular transfers for males without older brothers is negatively affected by the thirdgeneration gender comosition. One of the ossible exlanations for this result could be that the oldest sons in a family also suorts their arents from different asects other than regular transfers, so they might decrease the corresonding amount Urban-rural differences Passing the gender role to the offsring, esecially the traditional one, exists more commonly in rural areas than urban areas in China. The exected results are that the third-generation gender effects should be larger or more significant in rural areas than the effects in urban areas. Table 8 resents the regression results for urban and rural using the instrumental variables. In the urban areas, the gender effects of the grandchildren are insignificant for ecuniary outcomes for both male P and female P. However, for the total days of visiting arents, females visit their arents more with more daughters with their households, while males visit arents more with more sons. 11

12 For the old-age suort behaviours in rural areas, the third-generation gender effects are significant only for males in terms of the robability of roviding and the amount of non-regular transfers to arents. The effects are also larger. For examle, a male with more sons in his family can increase around 33% of the amount of the non-regular suort transferred to his own arents. But the gender effects are insignificant for others outcomes and also for female P. Analysing the results for urban and rural areas, it seems that the demonstration effects exist in both areas. However, the differences between the coefficients of More malek for male P and female P in urban areas are smaller than the differences in rural areas. This can be the indirect evidence for the statement before: the third-generation gender effects are larger in rural areas than urban areas. Pecuniary and non-ecuniary resources are sent differently on the ageing arents in urban and rural areas. Comared to those living in rural areas, eole living in the urban areas have less time and more money. Peole might want to demonstrate to their offsring to suort the ageing arents through the scarce resources in their areas. Also, the gender effects are insignificant for ecuniary values in urban areas but significant in rural areas could also be exlained by the larger ension coverage for the elderly in urban areas Single child or non-single child households Aart from the behavioural differences of old-age suort for urban and rural areas, the effects of the third-generation gender effects might also be different in household tyes. Parents with the single child should behave differently from arents with more than one child. Again, I run the regression searately for single-child households and non-single child households using the instrumental variable method and the results are in Table 9. Table 9 shows a clear contrariety for single and non-single child families. The third-generation gender effects are showing their significance only for the singlechild households, although the significance levels have droed a bit comared to the main regression results in Table 6. In the relevant literature, the effect of the number of children is ositive on the old-age suort received by the arents (Oliveira, 2016). This existing emirical result can also exlain why the results are more significant for the single-child households: to ensure that they can receive more suort from their single child, arents need to rovide more to their own arents for have a ositive demonstration effects on their child. From the relevant literature in Psychology, the gender secification for young children is also affected by their siblings (Rosenberg and Sutton-Smith, 1971). If a child is the single child in a family, the gender role secification should mostly come from the arents; hence the arents should demonstrate more to their sons or daughters in the single-child family. The emirical results in Table 9 also match with the relevant literature in Economics and Psychology. 12

13 4.5.4 Uward-transfer from G The revious results only imly that P is trying to demonstrate to their offsring, but they do not show what is the benefits of doing such demonstrative behaviours. Utilising the data set, I obtain the transfers information of G roviding to their arents, which are the grandarents of P. I use the gender of P, uward-transfers of male and female G and the interaction term of these two variables to construct the new regressions. The two new OLS regressions for checing the return of the demonstrative behaviours are: and y i = α 2 + β 2 mgt i + γ 1 cg i + δ 2 (gg i mgt i ) + X iθ 2 + ɛ 2,i (3) y i = α 3 + β 3 fgt i + γ 1 cg i + δ 3 (gg i fgt i ) + X iθ 3 + ɛ 3,i. (4) In the regression equation (2) and (3), excet mgt i and fgt i, other variables have the same definitions as they are in the regression equation (1). So i still reresents individual i in generation P. suort i in this section has three different outcomes: dummies for any, regular and non-regular transfers. mgt i is a dummy stating whether male G of individual i rovides any/regular/non-regular transfers to their arents and fgt i is the corresonding variables for female G. The control variables are the same as the controls in Table 1. One extra control that I have for the articular regressions is the average self-reorted health. Health roblems of the grandarents of P, might affect suorts rovided, so I also add for the health for grandarents of P in the controls. The results for the regression equation (2) and (3) are combined in Table 10. The ey regressor in Panel A is the cross term of the gender of P and whether the corresonding male or female G rovides transfers to their own arents. The three coefficients for the interaction term are all significant and ositive for all different tyes transfers, showing that male G s transfers to their arents have ositive effects on the transfers received from their offsring. Using the results for any-transf er as an examle to interret, if the interaction term is significant and equals 0.036, it means male G s transfers have a larger imact on the transfers from their sons than on the ones from their daughters. For males, the demonstration effect seems taing into effects from the same gender channel: sons are more affected than daughters by their father. With the same interretation, the interaction term in Panel B is exected to be negative and significant if the results also want to show that mothers have more imacts on their daughters than sons. However, the results turn out as unexected. All three coefficients are ositive but not as significant as the results in Panel A. However, it coincides with the revious results, for financial suorts, only the results for males can show ossible demonstration effects. There are some otential roblems using the uward transfers for two different generations in the same year. More transfers rovided to G might increases the rovision of uward transfers to G s arents. However, given the fact that the average age of G is already around 65, the grandarents of P are 13

14 mostly in the very last eriod of their life. Usually, the transfers to these greatgrandarents are relatively constant over years, because the consumtion level of elderly in China is relatively low and constant over The low-level consumtion for the very old oulation in China is the reason that I am only focusing on dummies that indicating whether G transferred or not. I also control for the health status of arents of G. Because this health status also affects the rovision of transfers from G to their arents. I also use the original data set and test whether the revious generations (G and their arents) has similar demonstration effects as the current generations (P and G). I advance the inter-generational transfer relationshis by one generation. As I stated in the data section because G s average age is around 65. Given this average age, the G with arents alive do not consist a large enough samle as the reconstructed one. However, it is reasonable to use the results as a iece of suorting evidence and also as a chec to see whether the transfer attern changes from one generation to another. The regressions results are resented in Table 11. To mae sure the results more comarable with the results in Table 5, I mae the controls as similar as ossible to the controls used before. The samle size is droed to around 2,000. The outcomes are the dummies for P and G roviding uward financial suorts. The results for the G generation imly similar conclusions as for the P one. The coefficients of gender comosition for grandchildren are ositive for males and all negative for females. Although in terms of the significant level, only the coefficients for any transfers for males is significant. Given the samle size, the standard deviations of the coefficients are relatively large comared to the revious results. It is ossible with larger samle size, the results might be more significant. This new results also show some behavioural differences for two different generations, G and P. I will just focus on the three ecuniary outcomes in Table 11. Comaring two tables, I observe for females, the gender effect from the third generation changes from an insignificant negative imact to an insignificant ositive one. Also for males, the gender effect decreases in terms of the magnitude. Tae the coefficient of any transfers as an examle, the gender effect of the third generation decreased from around 10% for G to 1.6% for P. The gender effect of the third generation seems decreased. This could due to the OCP imlementation and also the increasing emowerment of females in social status and labour marets now in China comared to the revious generation. Summarising all the robustness checs, the results imly that there are ossible demonstration effects for males for ecuniary and non ecuniary suorts while for females, the demonstration effects are only artly significant in terms of visiting arents. 5 Conclusion Children are an imortant source of old-age suort in develoing countries. This aer finds that different combinations of gender effect of arents and 14

15 grandchildren affect the old-age suort received by the grandarents. On average, the robabilities of suorting arents are about the similar for male and female children, but, deending on the gender of their offsring, the robability varies. Males with daughters are less liely to rovide transfers than males with sons, controlling for other variables. For aying visits to arents, females with daughters visits their arents more than females with sons and for males, the gender of the third generation does not have significant effects. However, on average, males visit their arents more than females. The aer also rovides a model to exlain this transfer behaviour. The model shows that, for those with the traditional view of females, the gender ratio of offsring affects males and females with this trait differently. The redictions are that old-age suort rovided by males increases with the gender ratio and with the reverse for females. 17 Alying this model to the data, the gender ratio for the children at family-level should match its effect redicted by the model. But the emirical results are artly consistent with the model s rediction. Males and females demonstrate to their in different asects of old-age suort. The inconsistency of the model s rediction and the emirical results might be because of different social exectations on the resonsibilities of males and females. Males are resonsible for more regular financial suorts and more visits, but they have more demonstrative behaviours in non-regular suorts. Females overall are more liely to rovide non-regular transfers, but the gender of their children have more effects for days of visits. In summary, the aer finds the emirical evidence for the gender of grandchildren affecting robabilities and the amount old-age suort rovided by arents, financially and non-financially in China. The story behind it is more comlicated than a ure gender effect of the grandchildren. The combination of the gender of children and grandchildren is the factor that affecting the suort that ageing arents received. Given the ageing oulation and the rising ension burden in China, 18 it aears that old-age suort is again becoming imortant. The gender ratio might ee increasing in future. The government needs to romote Gender Equity not only by having legislation rotecting women s rights in bequest, roerty right and labour maret, and also having strong enforcement of the legislation. 19 Once the gender difference in terms of roviding old-age suort is reduced, the gender selection roblem may become less severe. The combined gender effect of children and grandchildren may also have a very imortant imlication for China. The revious literature states that the economic incentives, esecially old-age suort, are one of the reasons behind fetal sex selections (Qian, 2008; Ebenstein and Leung, 2010). It might be artly due to the imlementation of One Child Policy. Before the OCP, arents could ee having more children until they had a son in their family. However, with 17 Again, the gender ratio here is the nationwide males to females ratio for newborns. 18 Rising ension burden in China. Reorted in China Daily. The website address for this is: htt://m.chinadaily.com.cn/en/ /12/content_ htm 19 The gender ratio in Korea has decreased artly due to the establishment of legislation suorting females legal right. 15

16 the OCP s limitation on the number of children, gender selection technologies become the ey method to ensure a boy in a family. From the results, males are more een to rovide regular suort and they are more liely to demonstrate to their children for ecuniary suort. While arents usually generate less monetary benefits from daughters when they are old. However, I can only conclude some ossible demonstration effects from the emirical results, its imlications on the increasing gender ratio in China are still unclear. 16

17 References [1] J. D. Angrist, W. N. Evans, Children and their arents labor suly: Evidence from exogenous variation in family size, The American Economic Review 88 (3) (1988) [2] L. Arrondel, A. Masson, Altruism, exchange or indirect recirocity: what do the data on family transfers show, in: S. C. Kolm, J. M. Ythier (Eds.), Handboo of the Economics of Giving, Altruism and Recirocity, North Holland., 2006, [3] J. Baer, The history of sonograhers, Journal of Ultrasound in Medicine 24 (1) (2005) [4] G. S. Becer, R. J. Barro, A reformulation of the economic theory of fertility, The Quarterly Journal of Economics 103 (1) (1988) [5] A. Bisin, T. Verdier, "beyond the melting ot": Cultural transmission, marriage, and the evolution of ethnic and religious traits, The Quarterly Journal of Economics 115 (3) (2000) [6] K. Bussey, A. Bandura, Social cognitive theory of gender develoment and differentiation, Psychological Review 106 (4) (1999) [7] C. Chan, P. H. Paul Yi Ernest Ng, C. Chan, J. Au, Gender selection in China: Its meanings and imlications, Journal of Assisted Reroduction and Genetics 19 (9) (2002) [8] Y. Chen, H. Li, L. Meng, Prenatal sex selection and missing girls in China: Evidence from the diffusion of diagnostic ultrasound, The Journal of Human Resources 48 (1) (2013) [9] T. Chouhmane, N. Coeurdacier, K. Jin, The One Child Policy and household saving (2015). [10] N. Coeurdacier, S. Guibaud, K. Jin, Credit constraints and growth in a global economy, American Economic Review 105 (9) (2015) [11] D. Cox, O. Star, Intergenerational transfers and the demonstration effect, Tech. re., Boston College Woring Paers in Economics, 329 (1996). [12] D. Cox, O. Star, On the demand for grandchildren: tied transfers and the demonstration effect, Journal of Public Economics 89 (9) (2005) [13] A. Ebenstein, S. Leung, The missing girls of China and the unintended consequences of the One Child Policy, The Journal of Human Resources 45 (1) (2010) [14] A. Ebenstein, S. Leung, Son reference and access to social insurance: Evidence from China s Rural Pension Program, Poulation and Develoment Review 36 (1) (2010)

18 [15] P. Hammond, Charity: altruism or cooerative egoism?, in: E. S. Phels (Ed.), Altruism, Morality, and Economic Theory, Russell Sage Foundation, 1975, [16] Y. J. Lee, W. L. Parish, R. J. Willis, Sons, daughters, and intergenerational suort in Taiwan, American Journal of Sociology 99 (4) (1994) [17] H. Li, J. Zhang, Testing the external effect of household behavior: the case of the demand for children, The Journal of Human Resources 44 (4) (2009) [18] I.-F. Lin, N. Goldman, M. Weinstein, Y.-H. Lin, T. Gorrindo, T. Seeman, Sons, daughters, and intergenerational suort in Taiwan, Journal of Marriage and Family 65 (1) (2003) [19] J. M. Guttman, Self-enforcing recirocity norms and intergenerational transfers: theory and evidence, Journal of Public Economics 81 (1) (2001) [20] M. Mauss, Sociologie et anthroologie, PUF, Paris., [21] M. Mauss, Essais de sociologie, Editions de Minuit, Paris., [22] C. N. Milwertz, Acceting Poulation Control: Urban Chinese Women and the One-Child Family Policy, Richmond Surrey: Curzon Press., 1997, Ch. Demograhic Change in the Peole s Reublic of China, [23] A. Mitrut, F. C. Wolff, A causal test of the demonstration effect theory, Economics Letters 103 (1) (2009) [24] J. Oliveira, The value of children: Inter-generational suort, fertility, and human caital, Journal of Develoment Economics 120 (2016) [25] N. Qian, Missing women and the rice of tea in China: The effect of sexsecific earnings on sex imbalance, The Quarterly Journal of Economics 123 (3) (2008) [26] B. Rosenberg, B. Sutton-Smith, Sex-role identity and sibling comosition, The Journal of Genetic Psychology 118 (1) (1971) [27] T. White, Domination, resistance, and accommodation in China s One- Child Camaign, in: E. J. Perry, M. Selden (Eds.), Chinese Society: Change, Conflict and Resistance (Asia s Transformations), Routledge, 2001, [28] F. C. Wolff, Private intergenerational contact in France and the demonstration effect, Alied Economics 33 (2001) [29] Y. Xie, H. Zhu, Do sons or daughters give more money to arents in urban China?, Journal of Marriage and Family 71 (1) (2009)

19 6 Tables 19

20 Table 1: Summary Statistics VARIABLES Mean Std. Dev. Minimum Maximum whether P rovides any transfers whether P rovides a regular transfer whether P rovides a non-regular transfer ln amount of regular transfer ln amount of non-regular transfer ln visit days gender of P gender comosition of K No. of K age of P income level of P education of P whether P has a rural huou whether P is married P living in rural areas No. of siblings of P P s raning in siblings No. of brothers whether P is living with G rofessional title of P distance from G household head of G average age of G average woring status of G average ension of G average education level of G (dummy) average education level of G who should suort G have G retired whether G have deosit household income of G e+8 hours of G taing care of grandchildren any transfers from G regular transfers from G non-regular transfers from G Notes: The first six variables are deendent variables. P stands for an individual from the Parent Generation and G is the arents of P. K is the children of P. Other variables are demograhic variables for P and their arents G. Because the ey resondents for the data set are not P, so there is no information on very detailed P s income, but their income levels. For variables education level of G, rural huou and age, the information comes from the arent of P who is the resondent in the survey. For variables whether G has retirement ensions, their household income, deosit, woring status and education, the information is the average value for the arents of P. Other variables for G are the corresonding summation. For the variable "who should suort G", it is a categorical variable and 1 in this variable indicates G believe that they will be suorted by their children, P, when they are old. 20

21 Table 2: Summary Statistics: females and males differences VARIABLES Females Males Mean Std. Dev. Mean Std. Dev. whether P rovides any transfers whether P rovides a regular transfer whether P rovides a non-regular transfer ln amount of regular transfer ln amount of non-regular transfer ln visit days gender comosition of K No. of K age of P income level of P education of P whether P has a rural huou whether P is married P living in rural areas No. of siblings of P P s raning in siblings No. of brothers whether P is living with G rofessional title of P distance from G household head of G average age of G average woring status of G average ension of G average education level of G (dummy) average education level of G who should suort G have G retired whether G have deosit household income of G hours of G taing care of grandchildren any transfers from G regular transfers from G non-regular transfers from G Notes: The first six variables are deendent variables. P stands for an individual from the Parent Generation and G is the arents of P. K is the children of P. Other variables are demograhic variables for P and their arents G. Because the ey resondents for the data set are not P, so there is no information on very detailed P s income, but their income levels. For variables education level of G, rural huou and age, the information comes from the arent of P who is the resondent in the survey. For variables whether G has retirement ensions, their household income, deosit, woring status and education, the information is the average value for the arents of P. Other variables for G are the corresonding summation. For the variable "who should suort G", it is a categorical variable and 1 in this variable indicates G believe that they will be suorted by their children, P, when they are old. 21

22 Figure 1: Gender ratios for the newborns in China: birth sequence Data source: National Poulation Census. 1990, 1995, 2000, 2005 and

23 Figure 2: Gender ratios for the newborns in China: yearly trend Data source: China Poulation and Emloyment Statistics Yearboo

24 Table 3: Gender effect of P on the old-age suort rovision VARIABLES any-transf er regular nonregular ln regular ln non-regular ln (visit days) P g ** *** 0.253*** *** (0.005) (0.005) (0.005) (0.043) (0.065) (0.039) P demograhics Yes Yes Yes Yes Yes Yes G demograhics Yes Yes Yes Yes Yes Yes Observations 12,524 12,524 12,521 11,607 11,607 12,524 R-squared Notes: Robust standard errors in arentheses, clustered at household-level. Stars indicate statistical significance. *** <0.01, ** <0.05, * <0.1. P stands for individuals from the Parent Generation and G is the arents of P. K is the children of P. P g is the gender of arents. The controlling variables for P are age, marriage status, rural huou, region, education, rofessional title, income level, whether P lives with arents and the distant to arents lace, the number of siblings, the number of children. The controlling variables for G, age, education level, rural huou, ension status, household wage and income of G. If P is a male, then the gender of P equals 1 and 0 otherwise. If P only has boys in the household, then gender comosition of K equals 1 and equals 0 if P has all female K. The coefficient resented here for OLS coefficients. The deendent variable any-transfer, regular and nonregular equal 1 if P rovides any, regular, and non-regular transfer to G corresondingly. ln regular and ln non-regular are natural log of the amount of regular and non-regular transfers. ln (visitdays) is the natural log of visit days in last year. 24

25 Table 4: Gender effect of K and P on rovision of old-age suort VARIABLES any-transf er regular nonregular ln regular ln nonregular ln (visit days) P g#more malek ** (0.007) (0.005) (0.005) (0.047) (0.089) (0.046) * *** 0.200*** (0.009) (0.008) (0.010) (0.064) (0.103) (0.058) *** *** ** (0.008) (0.007) (0.008) (0.057) (0.095) (0.054) P demograhics Yes Yes Yes Yes Yes Yes G demograhics Yes Yes Yes Yes Yes Yes Observations 11,101 11,101 11,098 10,191 10,191 11,101 R-squared Notes: Robust standard errors in arentheses, clustered at household-level. Stars indicate statistical significance. *** <0.01, ** <0.05, * <0.1. P stands for individuals from the Parent Generation and G is the arents of P. K is the children of P. P g is the gender of arents and Kg is the gender ratio of K The controlling variables for P are age, marriage status, rural huou, region, education, rofessional title, income level, whether P lives with arents and the distant to arents lace, the number of siblings, the number of children. The controlling variables for G, age, education level, rural huou, ension status, household wage and income of G. If P is a male, then the gender of P equals 1 and 0 otherwise. More malek equals 1 when the gender ratio for K is equal or larger than 0.5 in P s household. The coefficient resented here for OLS coefficients. The deendent variable any-transfer, regular and nonregular equal 1 if P rovides any, regular, and non-regular transfer to G corresondingly. ln regular and ln non-regular are natural log of the amount of regular and non-regular transfers. ln (visitdays) is the natural log of visit days in last year. 25

26 Table 5: Gender effect of K on rovision suort for different gender of P OLS VARIABLES any-transf er regular nonregular ln regular ln non-regular ln (visit days) Male P M ore malek 0.023*** 0.018** 0.022** (0.009) (0.008) (0.009) (0.060) (0.087) (0.046) R-squared Observations 5,901 5,901 5,899 5,092 5,092 5,901 Female P M ore malek * (0.008) (0.005) (0.008) (0.047) (0.090) (0.046) R-squared Observations 5,200 5,200 5,199 5,099 5,099 5,200 P demograhics Yes Yes Yes Yes Yes Yes G demograhics Yes Yes Yes Yes Yes Yes Notes: Robust standard errors in arentheses, clustered at household-level. Stars indicate statistical significance. *** <0.01, ** <0.05, * <0.1. P stands for individuals from the Parent Generation and G is the arents of P. K is the children of P. P g is the gender of arents and Kg is the gender ratio of K The controlling variables for P are age, marriage status, rural huou, region, education, rofessional title, income level, whether P lives with arents and the distant to arents lace, the number of siblings, the number of children. The controlling variables for G, age, education level, rural huou, ension status, household wage and income of G. If P is a male, then the gender of P equals 1 and 0 otherwise. More malek equals 1 when the gender ratio for K is equal or larger than 0.5 in P s household. The coefficient resented here for OLS coefficients. The deendent variable any-transfer, regular and nonregular equal 1 if P rovides any, regular, and non-regular transfer to G corresondingly. ln regular and ln non-regular are natural log of the amount of regular and non-regular transfers. ln (visitdays) is the natural log of visit days in last year. 26

27 Table 6: Gender effect of K on rovision suort for different gender of P with IV IV VARIABLES any-transf er regular nonregular ln regular ln nonregular ln (visit days) Male P M ore malek 0.031** ** ** (0.014) (0.013) (0.015) (0.098) (0.137) (0.069) R-squared Observations 5,901 5,901 5,899 5,092 5,651 5,901 Female P M ore malek (0.010) (0.007) (0.010) (0.077) (0.126) (0.064) R-squared Observations 5,200 5,200 5,199 5,099 5,099 5,200 P demograhics Yes Yes Yes Yes Yes Yes G demograhics Yes Yes Yes Yes Yes Yes Notes: Robust standard errors in arentheses, clustered at household-level. Stars indicate statistical significance. *** <0.01, ** <0.05, * <0.1. P stands for individuals from the Parent Generation and G is the arents of P. K is the children of P. P g is the gender of arents and Kg is the gender ratio of K The controlling variables for P are age, marriage status, rural huou, region, education, rofessional title, income level, whether P lives with arents and the distant to arents lace, the number of siblings, the number of children. The controlling variables for G, age, education level, rural huou, ension status, household wage and income of G. If P is a male, then the gender of P equals 1 and 0 otherwise. More malek equals 1 when the gender ratio for K is equal or larger than 0.5 in P s household. The coefficient resented here for OLS coefficients. The deendent variable any-transfer, regular and nonregular equal 1 if P rovides any, regular, and non-regular transfer to G corresondingly. ln regular and ln non-regular are natural log of the amount of regular and non-regular transfers. ln (visitdays) is the natural log of visit days in last year. The instrumental variables are the gender of the first-born child in the family together with the index that indicating how strict the cities on the gender selection behaviours at refectural-level. 27

28 Table 7: Gender effect of K on the old-age suort rovision: P with or without brothers IV VARIABLES any-transf er regular nonregular ln regular ln nonregular ln (visit days) With older brothers M ore malek (for male P ) (0.021) (0.020) (0.022) (0.155) (0.209) (0.107) M ore malek (for female P ) (0.016) (0.010) (0.016) (0.118) (0.201) (0.098) No older brothers M ore malek ** * *** (for male P ) (0.019) (0.018) (0.019) (0.118) (0.171) (0.085) M ore malek (for female P ) (0.013) (0.009) (0.013) (0.105) (0.161) (0.083) P demograhics Yes Yes Yes Yes Yes Yes G demograhics Yes Yes Yes Yes Yes Yes Notes: Robust standard errors in arentheses, clustered at household-level. Stars indicate statistical significance. *** <0.01, ** <0.05, * <0.1. P stands for individuals from the Parent Generation and G is the arents of P. K is the children of P. P g is the gender of arents and Kg is the gender ratio of K The controlling variables for P are age, marriage status, rural huou, region, education, rofessional title, income level, whether P lives with arents and the distant to arents lace, the number of siblings, the number of children. The controlling variables for G, age, education level, rural huou, ension status, household wage and income of G. If P is a male, then the gender of P equals 1 and 0 otherwise. If P only has boys in the household, then gender comosition of K equals 1 and equals 0 if P has all female K. The coefficient resented here for OLS coefficients. The deendent variable any-transfer, regular and nonregular equal 1 if P rovides any, regular, and non-regular transfer to G corresondingly. ln regular and ln non-regular are natural log of the amount of regular and non-regular transfers. ln (visitdays) is the natural log of visit days in last year. The instrumental variables are the gender of the first-born child in the family together with the index that indicating how strict the cities on the gender selection behaviours at refectural-level. 28

29 Table 8: Gender effect of K on the old-age suort rovision: Urban-rural differences IV VARIABLES any-transf er regular nonregular ln regular ln nonregular ln (visit days) Urban M ore malek * (for male P ) (0.021) (0.019) (0.022) (0.136) (0.204) (0.101) M ore malek ** (for female P ) (0.016) (0.011) (0.017) (0.126) (0.198) (0.094) Rural M ore malek ** * (for male P ) (0.019) (0.018) (0.020) (0.135) (0.180) (0.090) M ore malek (for female P ) (0.013) (0.008) (0.013) (0.092) (0.161) (0.082) P demograhics Yes Yes Yes Yes Yes Yes G demograhics Yes Yes Yes Yes Yes Yes Notes: Robust standard errors in arentheses, clustered at household-level. Stars indicate statistical significance. *** <0.01, ** <0.05, * <0.1. P stands for individuals from the Parent Generation and G is the arents of P. K is the children of P. P g is the gender of arents and Kg is the gender ratio of K The controlling variables for P are age, marriage status, rural huou, region, education, rofessional title, income level, whether P lives with arents and the distant to arents lace, the number of siblings, the number of children. The controlling variables for G, age, education level, rural huou, ension status, household wage and income of G. If P is a male, then the gender of P equals 1 and 0 otherwise. If P only has boys in the household, then gender comosition of K equals 1 and equals 0 if P has all female K. The coefficient resented here for OLS coefficients. The deendent variable any-transfer, regular and nonregular equal 1 if P rovides any, regular, and non-regular transfer to G corresondingly. ln regular and ln non-regular are natural log of the amount of regular and non-regular transfers. ln (visitdays) is the natural log of visit days in last year. The instrumental variables are the gender of the first-born child in the family together with the index that indicating how strict the cities on the gender selection behaviours at refectural-level. 29

30 Table 9: Gender effect of K on the old-age suort rovision: Single-K family IV VARIABLES any-transf er regular nonregular ln regular ln nonregular ln (visit days) Single-child M ore malek 0.023* * *** (for male P ) (0.014) (0.013) (0.014) (0.100) (0.137) (0.068) M ore malek (for female P ) (0.011) (0.007) (0.011) (0.083) (0.140) (0.068) Non-single-child M ore malek (for male P ) (0.037) (0.035) (0.038) (0.247) (0.326) (0.168) M ore malek (for female P ) (0.023) (0.017) (0.024) (0.172) (0.276) (0.152) P demograhics Yes Yes Yes Yes Yes Yes G demograhics Yes Yes Yes Yes Yes Yes Notes: Robust standard errors in arentheses, clustered at household-level. Stars indicate statistical significance. *** <0.01, ** <0.05, * <0.1. P stands for individuals from the Parent Generation and G is the arents of P. K is the children of P. P g is the gender of arents and Kg is the gender ratio of K The controlling variables for P are age, marriage status, rural huou, region, education, rofessional title, income level, whether P lives with arents and the distant to arents lace, the number of siblings, the number of children. The controlling variables for G, age, education level, rural huou, ension status, household wage and income of G. If P is a male, then the gender of P equals 1 and 0 otherwise. If P only has boys in the household, then gender comosition of K equals 1 and equals 0 if P has all female K. The coefficient resented here for OLS coefficients. The deendent variable any-transfer, regular and nonregular equal 1 if P rovides any, regular, and non-regular transfer to G corresondingly. ln regular and ln non-regular are natural log of the amount of regular and non-regular transfers. ln (visitdays) is the natural log of visit days in last year. The instrumental variables are the gender of the first-born child in the family together with the index that indicating how strict the cities on the gender selection behaviours at refectural-level. 30

31 Table 10: Effects of the revious generation transfers on the current one OLS VARIABLES any-transf er regular nonregular Panel A P g ** (0.015) (0.014) (0.015) U ward transf er of male G 0.031* ** (0.019) (0.010) (0.018) Cross term 0.036* 0.046** 0.035* (0.021) (0.018) (0.021) Observations 2,718 2,718 2,718 R-squared Panel B P g (0.014) (0.012) (0.014) ) U ward transf er of f emale G 0.089*** 0.019* 0.083*** (0.010) (0.010) (0.017) Cross term * (0.020) (0.018) (0.021) Observations 3,156 3,156 3,156 R-squared P demograhics Yes Yes Yes G demograhics Yes Yes Yes Notes: Robust standard errors in arentheses, clustered at household-level. Stars indicate statistical significance. *** <0.01, ** <0.05, * <0.1. P stands for individuals from the Parent Generation and G is the arents of P. The cross term is the interaction term between the gender of P and the uward transfer of male or female G to their arents corresondingly in Panel A and Panel B. The controlling variables for P are age, marriage status, rural huou, region, education, rofessional title, income level, whether P lives with arents and the distant to arents lace, the number of siblings, the number of children. The controlling variables for G, age, education level, rural huou, ension status, household wage and income of G. If P is a male, then the gender of P equals 1 and 0 otherwise. If P only has boys in the household, then gender comosition of K equals 1 and equals 0 if P has all female K. The coefficient resented here for OLS coefficients. The deendent variable any-transfer, regular and nonregular equal 1 if P rovides any, regular, and non-regular transfer to G corresondingly. 31

32 Table 11: Gender effect of the third generation: evidence from G OLS VARIABLES any-transf er regular nonregular Male G P g 0.093** (0.045) (0.035) (0.045) R-squared Observations Female G P g (0.037) (0.021) (0.038) R-squared Observations 1,280 1,280 1,280 G demograhics Yes Yes Yes P demograhics Yes Yes Yes Notes: Robust standard errors in arentheses, clustered at household-level. Stars indicate statistical significance. *** <0.01, ** <0.05, * <0.1. P stands for individuals from the Parent Generation and G is the arents of P. The cross term is the interaction term between the gender of P and the uward transfer of male or female G to their arents corresondingly in Panel A and Panel B. The controlling variables for P are age, marriage status, rural huou, region, education, rofessional title, income level, whether P lives with arents and the distant to arents lace, the number of siblings, the number of children. The controlling variables for G, age, education level, rural huou, ension status, household wage and income of G. If P is a male, then the gender of P equals 1 and 0 otherwise. If P only has boys in the household, then gender comosition of K equals 1 and equals 0 if P has all female K. The coefficient resented here for OLS coefficients. The deendent variable any-transfer, regular and nonregular equal 1 if P rovides any, regular, and non-regular transfer to G corresondingly. 32

33 Aendix 33

34 Figure.3: Distribution of CHARLS samle counties and districts Data source: Official reort by CCER. Website: htt://charls.u.edu.cn/uloads/document/ublic_documents/alication/challengesof-poulation-aging-in-china-final.df 34

35 Table.12: Primary Source of Suort of China s Elderly, 2005 and Urban Rural Source of suort Average Male Female Average Male Female Labour income Pensions Dibao Insurnace and subsidy Proerty income Family suort Other Source: NBS, Most significant share of suort reorted Urban Rural Source of suort Average Male Female Average Male Female Labour income Pensions Dibao Insurnace and subsidy Proerty income Family suort Other Source: NBS, Most significant share of suort reorted. 35

36 Figure.4: Jaan s male roortion for the newborns from 1949 to 1999 Data source: The Statistics Bureau and the Director-General for Policy Planning of Jaan. The technology, model B ultrasound, aeared before 1980s in Jaan, and the gender ratio has been decreasing since. 36