Famine, Fertility, and Fortune in China☆

Home , February 1958, The China Quarterly

China Economic Review 22 (2011) 244–259

Contents lists available at ScienceDirect

China Economic Review

Famine, fertility, and fortune in china☆

Xinzheng SHI ⁎

Economics Department, School of Economics and Management, Tsinghua University, Beijing 100084, China article info abstract

Article history: In this paper, I investigate the long term effects of China's Great Famine in 1959–1961 on Received 25 August 2010 cohorts affected by the famine in the first year of life. Using China's 2000 population census Received in revised form 20 December 2010 data and after controlling for positive fertility selections in the famine, I find that women Accepted 3 February 2011 exposed to the famine in the first year of life had a lower probability of completing high school Available online 25 February 2011 and lived in less wealthy households. I do not find any significant effects of the famine on men. In addition, I find that if positive fertility selections are not controlled for, the negative effects ﬁ JEL classi cation: become weaker. I21 © 2011 Elsevier Inc. All rights reserved. J13 J22 O12 O15 Q54

Keywords: Famine Long term effects Fertility selections Education Wealth China

1. Introduction

There were numerous famines during the twentieth century.11 The largest of these was China's 1959–1961 famine which resulted in about 30 million excess deaths.22 Previous research has focused on estimating excess mortality; however, there were undoubtedly important effects on the livings as well.33 In particular, children born during the famine may have suffered from malnutrition in the initial years of life, resulting in adverse long term health effects and then influencing economic and social attainments as adults. In this paper, I investigate the long term effects of China's 1959–1961 famine on the education, labor supply, and the wealth of rural Chinese women and men who were exposed to the famine in the first year of life. Children's health was damaged due to insufficient nutrition intake in the famine. This health shock may have had a long term effect simply because it persists over time. The health shock could also affect other outcomes such as educational attainments and labor market performance which help determine long-run well-being. On the other hand, estimates of famine impacts are

☆ I thank Albert Park and Dean Yang for their persistent guidance. And I thank participants of NEUDC 2006 in Cornell University, seminar participants in the University of Michigan, and the anonymous referee for helpful comments. The previous version of this paper was titled as “Does famine have long term effects? Evidence from China”. ⁎ Tel.: +86 10 62784920. E-mail address: [email protected]. 1 Sen, 1981; Ravallion, 1987. 2 Ashton et al., 1984. 3 Stein et al., 1975; Barker et al., 2005.

1043-951X/$ – see front matter © 2011 Elsevier Inc. All rights reserved. doi:10.1016/j.chieco.2011.02.001 X. Shi / China Economic Review 22 (2011) 244–259 245 complicated by potential selection biases. First, parents living through the famine could alter their fertility decisions. Parents unable to provide adequate support for children could choose to postpone childbearing or abort the fetus in the famine, so that only capable parents would still choose to have children. Second, because of the negative impacts of the famine on women's health, only those in good health status were able to conceive. These two led to positive selections of children born. I call these two selections as fertility selections in this paper since they are related to parents' fertility. Third, weaker children were more likely to die during the famine, leading to another positive selection with respect to children's health. The combination of negative shocks and positive selections complicates estimates of famine effects. In this paper, I take advantage of detailed information collected in China's 2000 Population Census to control for positive fertility selections and estimate the long term effects of the 1959–1961 famine. Because of lack of information, I cannot control the positive selection due to children's death in the famine. Therefore, my estimates can be considered as a low bound for the impacts of the famine. The data requirements for conducting such an investigation are considerable. Information is required on the extent of the famine experienced by individuals many years ago as well as detailed information on adult outcomes in the present days. I use provincial excess death rates in 1959–1961 to measure famine intensity. Outcome variables are from the China 2000 Population Census. Because the 2000 census data contain information of every individual's birth province as well as his or her year and month of birth, it is possible to link each individual's adult outcomes with the famine intensity experienced in the childhood. Additionally, the availability of the information of birth month makes it possible to control for fertility selections. The most recent research (Maccini & Yang, 2009) suggested that environmental shocks in the birth year had the significant effects on children's adult outcomes. Maccini and Yang (2009) also addressed that comparing with shocks experienced in infancy, there was no evidence to support that shocks in utero were importantly affecting the results (see Section 6.1 in Maccini and Yang (2009) for detailed discussion). Following Maccini and Yang (2009), I investigate the long term effects of the famine on individuals affected by the famine in the first year of life. The famine in China is usually called Three Year Disaster, meaning that the famine lasted three years, starting in 1959 and ending in 1961.4 Then, for cohorts born between February 1958 and October 1959, they were affected by the famine in the first year of life, but their parents' decision about whether to conceive must be made before the famine, therefore not affected by the famine. However, for cohorts born between July 1959 and October 1959, the famine had started when they were in the first five months in utero, during which the fetus could still be aborted and their parents' decisions about whether to abort the fetus might be affected by the famine, leading to a fertility selection. Therefore, for the sake of controlling fertility selections, I only include cohorts born between February 1958 and June 1959 in the treatment group. I estimate the long term effects of the famine by comparing cohorts conceived at least five months before the famine (therefore not affected by the fertility selections due to the famine) but affected by the famine in the first year of life, that is, cohorts born between February 1958 and June 1959, with those not affected by the famine in the first year of life, that is, cohorts born between January 1954 and January 1958, and comparing individuals born in different provinces. This identification strategy follows the idea of difference-in-difference. Using a sample of rural women from the China 2000 Population Census data, I find that with death rate higher than the normal level by 0.1 percentage point in the famine, women affected by the famine in the first year of life had a 0.09 percentage point lower probability of completing high school and lived in less wealthy households (measured by smaller houses: 0.006 fewer rooms per capita and 0.07 square meters smaller housing area per capita). With the average deviation of the death rates in the famine from the normal level equal to 0.57 percentage point (shown in Table 4), exposure to the famine in the first year totally reduced women's probability to complete high school by 0.5 percentage point and made women live in households having 0.034 fewer rooms per capita and 0.399 square meters smaller housing area per capita. However, I do not find any significant effects of the famine on men. In order to test whether positive fertility selections existed in the famine, I replace cohorts born between February 1958 and June 1959 with cohorts born between July 1959 and December 1961, that is, those affected by the famine in the first year of life and whose parents' fertility was also affected by the famine, and then repeat the analysis. I find that the famine effects become much weaker, providing evidence for the existence of positive fertility selections during the famine. Five other papers have focused on the long term effects of China's 1959–1961 famine. Chen and Zhou (2007) found that cohorts exposed to the famine had a lower average height, less labor supply, and less income. Luo, Mu, and Zhang (2006) found that women exposed to greater severity of famine were more likely to be overweight as adults. Almond, Edlund, Li, and Zhang (2007) found that cohorts exposed to higher exogenous mortality in utero were more likely to be poorer, disabled and illiterate; these cohorts were also more likely not to work, to have worse marriage market outcomes and have daughters. Mu and Zhang (2008) focused on the gender difference of the long term impacts of the famine. They found that the famine had a significant effect on women but not men in terms of disability rate, nonworking rate and illiteracy rate. Meng and Qian (2009) found that exposure to the famine reduced height, weight, weight-for-height, educational attainments and labor supply. Meng and Qian (2009) addressed the selection problem by estimating the impacts of the famine for individuals on different percentiles of the distribution of outcomes and found that the estimated effects were more adverse and larger for the individuals on the 90th percentile. But even if this pattern was found in the data, it is still hard to distinguish selection effects from heterogeneous effects of the famine on the individuals on the upper percentiles of the distribution. The distinction between my paper and their research lies in that I can use more precise information of individuals' birth years and months to control for positive fertility selections in the famine and provide evidence for the existence of the fertility selections. This paper is related to other research about the effects of famines in other countries such as 1944–1945 Dutch famine (Stein, Susser, Saenger, & Marolla, 1975; Barker et al., 2005). This paper is also related to a more general group of research using non-

4 Although the timing of the famine might vary in different provinces, it is hard to exactly identify the starting time and the ending time of the famine in each province. Therefore, I take January 1959 as the starting time and December 1961 as the ending time. It is consistent with other papers studying the long term impacts of the famine (Chen & Zhou, 2007; Luo et al., 2006; Almond et al., 2007; Mu & Zhang, 2008; Meng & Qian, 2009). 246 X. Shi / China Economic Review 22 (2011) 244–259

National Death Rate and Birth Rate 50 40 30 20 Death rate&Birth rate(0.1%) 10

1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 Year death_rate birth_rate

Source: China National Bureau of Statistics

Fig. 1. National death rate and birth rate (unit: 0.1%).

Chinese data on the impact of early life conditions on adult outcomes. Almond and Chay (2006) found that better conditions in early life for U.S. black women led to better health in adulthood and higher birthweight of the women's children. Behrman and Rosenzweig (2004) used the U.S. data to exploit intrauterine nutrient intake differences between monozygotic female twins. They found a strong impact of fetal growth on schooling attainment and height. Royer (2009) documented long-run and intergenerational effects of birthweight differences between twins. Almond (2006) found that U.S. cohorts who were in utero during the 1918 inﬂuenza pandemic had worse adult outcomes as they aged than cohorts born just before and after the pandemic in terms of educational attainment, physical disability, socioeconomic status and mortality. Lindeboom, Van den Berg, and Portrait (2006) showed that poor macroeconomic conditions in early life reduced longevity in the Netherlands. Alderman, Hodinott, and Kinsey (2006) found that rainfall shocks and exposure to war affected early-life nutrition and later height and schooling levels of young adults in Zimbabwe. Maccini and Yang (2009) found that higher early-life rainfall had a positive effect on the adult outcomes of women but not of men, using Indonesian data. The rest of this paper is organized as follows: Section 2 provides backgrounds of the famine. Section 3 describes the data sources, sample and variables used. Section 4 presents the empirical strategies and results. Section 5 discusses, and Section 6 concludes.

2. 1959–1961 famine

After the People's Republic of China was founded in 1949, the central government adopted a heavy-industry-oriented development strategy in order to quickly catch up to western countries. At the same time, it started agricultural collectivization to replace traditional family farms with collectively managed production teams. From 1952 to 1958, agricultural and industrial outputs increased continuously and dramatically. Prompted by this success and a desire to surpass developed countries, Chairman Mao launched the Great Leap Forward to accelerate economic growth. In rural areas, the People's Commune movement was launched on a full scale in the summer of 1958. The People's Commune created huge collectives and eliminated all private ownership; it also provided free food through large commune mess halls. The Communist Party told the people that China would soon enter the communism stage when people could get whatever they needed; however, the country was in crisis only one year later. Beginning in the winter of 1958, starvation was observed in Sichuan and Anhui provinces. By the spring of 1959, starvation became widespread. The estimated daily availability of food energy per capita during this period decreased considerably to about 1800 calories, reaching a low point of only 1500 calories in 1960.5 A study using demographic data released after the start of economic reforms concluded that this crisis resulted in about 30 million excess deaths.6 From Fig. 1, we can see that the crude death rate7 dramatically increased between 1959 and 1960. Facing such a severe famine, parents might decide not to have children or delay having children, and weaker women might not be able to conceive. According to Peng (1987), total fertility up to age 39 was about 5.6 births per woman in pre-famine years, but it dropped to its lowest historic level, 3.06, in 1961. From Fig. 1, we can

5 Ashton et al., 1984. 6 Ashton et al., 1984. 7 Crude death rate is deﬁned as number of deaths per 1000 people. I call it death rate for simplicity in this paper. The source and construction are discussed later. X. Shi / China Economic Review 22 (2011) 244–259 247

Population 750 700 650 Population(million) 600 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 Year

Source: China National Bureau of Statistics

Fig. 2. Population (unit: million). see that the crude birth rate8 dropped sharply between 1959 and 1961. Both the increase in the death rate and the decrease in the birth rate led to a dramatic decrease in the population between 1959 and 1961, shown in Fig. 2. Although the famine occurred nationally in 1959–1961, there was a large regional variation.9 Table 1 shows death rates in 1959–1961 across provinces. We can see from this table that the most severe famine in 1959 occurred in Sichuan province where the death rate was 4.7%; however, the lowest death rate (0.78%) was observed in Shanghai City. In 1960 when the famine was the most severe nationally, variation across provinces was still large. Anhui province became the province with the highest death rate (6.86%), while Shanghai City still had the lowest death rate (0.69%). In 1961 when the famine was nearing its end, Sichuan province had the highest death rate of 2.94% and Shanghai City still had the lowest death rate of 0.77%. In January 1962, the government abandoned its radical policies. The policy focus was shifted from steel production back to agriculture. As a result, grain output started to grow in 1962; in the same year, the famine ended. From Fig. 1, we can see that the death rate resumed its normal level immediately when the famine ended in 1962. However, the birth rate did not return to its normal level right away. There was a dramatic increase in the birth rate in 1962 and it reached the highest point in 1963. There are two possible reasons: parents postponing having children in the famine might have started to have children when the famine ended, and those women not able to conceive during the famine recovered their fertility when they had enough nutrition intakes after the famine ended. This dramatic increase after the famine can be thought of as compensatory birth. After 1963, the birth rate decreased and went back to its normal level.

3. Data

3.1. Population census data

The data used in this paper are from a 0.095% random sample drawn from the China's 5th National Population Census conducted by the China National Bureau of Statistics in 2000. This sub-sample includes 1,180,111 observations, covering all of China's 31 provinces.10 There are 604,050 (51%) men and 575,769 (49%) women in the sample. Because of the government's preferential treatment of urban residents, the effects of the famine on urban residents were not nearly as pronounced as on rural residents. In addition, children's educational attainments were disrupted by the Great Proletarian Cultural Revolution11 from 1966 to 1976 in urban regions.12 I therefore restrict the study to individuals born in rural areas.

8 Crude birth rate is deﬁned as number of births per 1000 people. I call it birth rate for simplicity in this paper. The source and construction are discussed later. 9 Due to the government's preferential treatment of urban residents through a grain rationing system and the maintenance of government-controlled stockpiles, the lack of food was much more devastating in rural than in urban areas. Because of the variation in the proportion of rural population, population density, exposure to natural disaster, and provincial response to food shortages, exposure to the famine also varied greatly across provinces. (Ashton et al., 1984; Lin & Yang, 2000). 10 The population census does not include Hongkong, Macao, and Taiwan. 11 The Great Proletarian Cultural Revolution in the People's Republic of China was a struggle for power within the Communist Party of China that manifested as wide-scale social, political, and economic chaos, and grew to include large sections of Chinese society; it eventually brought the entire country to the brink of civil war. (See http://en.wikipedia.org/wiki/Cultural\_Revolution). 12 Gregory & Meng, 2002a,b; Giles et al., 2008. 248 X. Shi / China Economic Review 22 (2011) 244–259

Table 1 Distribution of provincial death rates in different years. (Unit: 0.1%).

Province 1959 1960 1961

Beijing 9.7 9.1 10.8 Tianjin 9.9 10.3 9.9 Hebei 12.3 15.8 13.6 Shaanxi 12.8 14.2 12.2 Neimenggu 11 9.4 8.8 Liaoning 11.8 11.5 17.5 Jilin 13.4 10.1 12.1 Heilongjiang 12.8 10.5 11.1 Shanghai 7.8 6.9 7.7 Jiangsu 14.6 18.4 13.4 Zhejiang 10.8 11.9 9.8 Anhui 16.7 68.6 8.1 Fujian 12.5 20.7 16 Jiangxi 13 16.1 11.5 Shandong 18.2 23.6 18.4 Henan 14.1 39.6 10.2 Hubei 14.5 21.2 9.1 Hunan 13 29.4 17.5 Guangdong 11.1 15.2 10.8 Guangxi 17.5 29.5 19.5 Sichuan 47 54 29.4 Guizhou 20.3 52.3 23.3 Yunnan 18 26.3 11.8 Shanxi 12.7 12.3 8.8 Gansu 17.4 41.3 11.5 Qinghai 16.6 40.7 11.7 Ningxia 15.8 13.9 10.7 Xinjiang 18.8 15.7 11.7 National average 15.1 23.2 13.1

Source: China Compendium of Statistics: 1949–2004.

However, the census data only include information about whether individuals were living in rural or urban regions in 2000, but not record whether these individuals were born in rural or urban regions. I therefore restrict the sample to those individuals who lived in rural regions and had an agricultural hukou (household registration booklet) in 2000. Due to China's household registration system, the probabilities for these individuals to be born in rural regions are very high.13 Chongqing was an independent municipality directly under the jurisdiction of the central government in 2000, but it was a city of Sichuan province before 1997. In addition, Hainan was a province in 2000, but it was a part of Guangdong province before 1988. Therefore, I treat Chongqing as a part of Sichuan province but not as an independent municipality and treat Hainan as a part of Guangdong province but not as an independent province in this paper. Because of limitations in the death rate data, I removed people born in Tibet. Individuals born in Tibet only account for 0.2% of the total sample, so dropping them should not lead to a selection problem. Since I am investigating the effects of early life conditions on individuals' adult outcomes, so those born between January 1954 and December 1961, that is, 39 to 46 years old in 2000, are used for analysis. The rationale not to include cohorts born after the famine in the analysis is to avoid the potential negative selection after the famine. I will discuss the negative selection in more details in Section 4.1. The final sample used in this paper therefore includes individuals born between January 1954 and December 1961, living in rural regions and having an agricultural hukou in 2000, excluding individuals born in Tibet. Table 2 shows the distribution of this sample by birth provinces. Population census data include individual's birth province, birth year and birth month. I use the birth province to link adult's outcomes with provincial famine shocks in early life. The birth year and birth month can be used to more precisely identify the timing of when individuals were affected by the famine. In this paper, I investigate the effects of the famine on three types of outcome variables. The first type is about human capital, measured by years of schooling and an indicator about whether the individual completed high school; the second type is labor supply, measured by the number of work days from October 25 to October 31 in 2000; the third type is household wealth, measured by the average number of rooms per capita and the average housing areas per capita. For specific variable definitions, see the Data Appendix.

3.2. Famine intensity data

The measurement of famine intensity is generated from provincial death rates in different years. I obtain provincial death rates from China Compendium of Statistics: 1949–2004 which was compiled and published by the China National Bureau of Statistics. In

13 Chan & Zhang, 1999; Chan, 2009. X. Shi / China Economic Review 22 (2011) 244–259 249

Table 2 Distribution of sample in birth provinces.

Province Number Percentage (%)

Beijing 336 0.47 Tianjin 297 0.42 Hebei 5277 7.43 Shaanxi 2093 2.95 Neimenggu 1418 2 Liaoning 2305 3.25 Jilin 1438 2.03 Heilongjiang 1536 2.16 Shanghai 214 0.3 Jiangsu 4384 6.17 Zhejiang 2691 3.79 Anhui 2985 4.2 Fujian 1710 2.41 Jiangxi 2475 3.49 Shandong 6290 8.86 Henan 6234 8.78 Hubei 3249 4.58 Hunan 3803 5.36 Guangdong 3304 4.65 Guangxi 2563 3.61 Sichuan 6578 9.27 Guizhou 1948 2.74 Yunnan 2584 3.64 Shanxi 2543 3.58 Gansu 1531 2.16 Qinghai 232 0.33 Ningxia 263 0.37 Xinjiang 716 1.01 Total 70,997 100

this paper, I use excess death rate (EDR) to measure famine intensity in 1959–1961. In order to estimate excess death rate for each province in 1959–1961, I ﬁrst calculate for each province the average death rate in 1954–1958, death rate5458 , and the average death rate in 1962–1966, death rate6266 . I then calculate provincial predicted death rates for the three famine years in the following way:

death rate ; −death rate ; predicted death rate = death rate + 6266 j 5458 j ðÞz−1958 jz 5458;j 1962−1958

Here, j represents province j, z=1959, 1960 and 1961. The excess death rates in 1959–1961 are deﬁned as the deviation of the death rate in each province from provincial predicted death rate, as shown in the following14:

− EDRjz = death ratejz predicted death ratejz

EDRs in other years are zero. Table 3 shows the estimated excess death rate for each province in 1959–1961. The bottom row in Table 3 shows the average value of EDRs for the whole country. We can see that the highest average excess death rate appeared in 1960 (1.2%) and the lowest average excess death rate appeared in 1961(0.3%). The average excess death rate was 0.4% in 1959. This is consistent with the famine pattern, that is, the most severe famine happened in 1960 and the famine was coming to its end in 1961. Cohorts born in different months spent different time in the famine. For example, an individual born in October 1958 spent 9 months of the ﬁrst year of life in the famine, while an individual born in December 1958 spent 11 months of this period in the famine. I therefore calculate famine intensity experienced for cohorts born in province j in month m of year t and experiencing famine in different years using the following method, which is also consistent with the method used in Almond et al. (2007):

1961 ðÞ ∑ months spent in year z mt FIjmt = EDRjz z = 1959 12

14 In addition to this definition, I also construct another form of EDR which is defined as the deviation of death rates in the famine years from the average death rates in 1954–1958 in each province. All the results shown in this paper are robust to the change of EDR definition. These results are available under request. 250 X. Shi / China Economic Review 22 (2011) 244–259

Table 3 Estimated excess death rate in the famine (Unit: 0.1%).

Province Year

1959 1960 1961

Beijing 1.43 0.97 2.82 Tianjin 1.21 2.13 2.26 Hebei 1.29 5.22 3.45 Shaanxi 0.39 2.10 0.41 Neimenggu −0.13 −1.13 −1.14 Liaoning 3.43 3.32 9.51 Jilin 3.99 0.47 2.26 Heilongjiang 2.77 0.87 1.88 Shanghai 1.11 0.35 1.30 Jiangsu 3.74 8.02 3.50 Zhejiang 0.72 2.54 1.16 Anhui 5.12 58.28 −0.96 Fujian 2.63 11.10 6.67 Jiangxi 0.67 4.57 0.77 Shandong 6.03 11.73 6.83 Henan 2.33 28.77 0.31 Hubei 3.42 10.53 −1.16 Hunan 0.14 17.17 5.91 Guangdong 1.60 6.30 2.50 Guangxi 5.17 18.11 9.06 Sichuan 34.03 41.11 16.60 Guizhou 6.03 37.58 8.13 Yunnan 2.28 11.55 −1.98 Shanxi 1.72 0.88 −3.06 Gansu 4.49 28.82 −0.55 Qinghai 5.09 29.82 1.45 Ningxia 4.26 2.82 0.08 Xinjiang 5.14 2.86 −0.32 National average 3.93 12.39 2.77

ðÞ Here, FIjmt is the famine intensity experienced by cohorts born in province j in month m in year t. months spent in year z mt is the number of months in the first year of life spent in the famine year z by cohorts born in month m in year t. z equals to 1959, 1960 or 1961. 12 is the total number of months in the first year. Therefore, famine intensity assigned to each individual is actually a weighted average of different annual excess death rates, using the number of months in the first year of life spent by this individual in each famine year as weights. Table 4 reports selected summary statistic of the sample used for analysis. Since I restrict the sample to cohorts born between 1954 and 1961, the average ages of men and women were very close to each other, both about 43 in 2000. Women's education level was lower than men's. Only 5.23% of women completed high school and on average they finished 6.09 years of schooling. In contrast, 14.09% of men completed high school, with an average of 7.96 years of schooling. Men also performed slightly better than women in the labor market: men worked 5.67 days while women worked 5.10 days from October 25 to October 31 in 2000. However, women's living conditions, which are used to measure wealth, were slightly better. Women were on average living in households with 0.92 rooms per capita and 27.71 square meters housing area per capita, while on average men were living in households having 0.86 room per capita and 25.76 square meters housing area per capita. The last row reports the mean value of famine intensity experienced by individuals affected by the famine in the first year of life. We can see that the average values of famine intensity are 0.56% and 0.57% for men and women respectively. Since this variable is calculated based on excess death rates in provincial level, it is natural for them to be close to each other.

Table 4 Summary statistics.

Men Women

Mean SD OBS Mean SD OBS

Age 42.92 2.21 35649 42.90 2.24 35348 Proportion of completing high school 14.09% 0.35 35649 5.23% 0.22 35348 Years of schooling 7.96 2.62 35649 6.09 3.13 35348 Labor supply 5.67 1.94 35649 5.10 2.42 35348 Average number of rooms per capita 0.86 0.56 35649 0.92 0.63 35348 Average housing areas per capita 25.76 18.36 35649 27.71 20.72 35348 Famine intensity experienced in 1959–1961 0.56% 0.01 14261 0.57% 0.01 14310

Note: (1) Labor supply is measured by the number of work days in Oct. 25–Oct.31 in 2000. X. Shi / China Economic Review 22 (2011) 244–259 251

4. Empirical strategies and results

4.1. Empirical strategies

In examining the relationship between the famine experienced in the first year of life and adult outcomes, I seek to isolate effects of the famine on the cohorts experiencing the famine in the first year of life. I also seek to isolate deviation of adult outcomes from the mean in one's birth province as well as from the mean of the national birth cohort. Because particular provinces in China may be subject to slow-moving changes over long periods of time (reflecting, for example, different rates of economic development), I also try to isolate variations in a person's outcomes that diverge from the long-running trends in his or her birth province. In order to identify the impacts of the famine, I combine two variations: different famine intensities across provinces and comparison between cohorts experiencing and not experiencing the famine in the first year of life. The identification strategy essentially follows the idea of difference-in-difference. One concern about this strategy lies in the possibility of endogenous variation of famine intensities across provinces. In a recent study by Li and Yang (2005), they evaluated contributions of different factors to the dramatic drops in grain outputs which led to the famine. They found that Great Leap Forward policies, specifically diversion of resources from agricultural production and excessive procurement, contributed more than 50%. According to Yang (1996), provincial leaders' political attitudes in the Great Leap Forward determined how radically they implemented these policies. The OLS estimates could be biased if different provincial leaders' political attitudes affected the adult outcomes of children born in the famine in different provinces through channels other than the famine. In order to check this issue, I follow the method used in Duflo (2001). Specifically, I include in the regression interactions of provincial leaders' political attitudes in the Great Leap Forward with a dummy variable indicating whether the individual was affected by the famine in the first year of life. I use the ratio of Chinese Communist Party (CCP) members in the population in 1957 and the ratio of population accommodated by commune mess halls in 1959 to measure provincial leaders' political attitudes in the Great Leap Forward. As argued in Yang (1996), areas having lower ratio of CCP members in the population were more likely to be taken over later by CCP, their leaders therefore tended to be more radical than the leaders of old revolutionary areas in order to show their political loyalty in the Great Leap Forward; while having lower ratio of population accommodated by commune mess halls might represent lower economic development, therefore the provincial leaders tended to use more radical policies in order to catch up to other areas.

Finally, the regression I estimate is the following reduced-form linear relationship between adult outcomes Yijmt of an adult i born in province j in month m of year t and independent variables:

β β ς λ μ δ ε ð Þ Yijmt = 0 + 1 FIjmt + Politics Dummyijmt + jtTREND + j + t + ijmt 1

The coefficient of interest is β1 which shows the impact of the famine intensity experienced in the first year of life, FIjmt, on adult outcomes; Politics is a vector including two variables: the ratio of CCP members in the population in 1957 and the ratio of population accommodated by commune mess halls in 1959. Dummyijmt is a dummy variable for individual i born in province j in month m of year t. It equals 1 if individual i was affected by the famine in the first year of life. λjt TREND is a time trend specific to the province, which absorbs the long-run time trends in the outcomes that may vary depending on the province (TREND is a time trend, and the coefficient λjt allows the time trend to vary across provinces). I use two forms of time trends: one is a linear time trend and the other one is a quadratic time trend (including a linear term and its square). μj is birth province fixed effect, δt is birth year fixed effect, and εijmt is an individual-specific error term with mean equal to zero. Since individuals in the same province experienced the same or similar measured famine intensity, according to Moulton (1986) and Bertrand, Duflo, and Mullainahtanm (2004), serial and spatial correlations among the error terms lead to a downward bias in the OLS standard error estimates. Therefore, standard errors calculated in this paper allow for an arbitrary variance-covariance structure within birth provinces (standard errors are clustered by birth province). In this paper, special attention should be given to likely directions of any selections. A potential worry is that the famine might affect the likelihood of children's survival, and those whose survival was induced by the famine could have different initial characteristics from the overall population of births in a locality in a particular year. During the famine, parents might have postponed giving birth to a child or aborted fetus if they felt unable to provide enough support for children. Even they chose to give birth to a child, women might not be able to conceive because of “famine amenorrhea” due to insufficient nutritional intake and subsequent poor health during the famine.15 Therefore, children conceived and born in the famine should have better family backgrounds and healthier mothers, leading to positive selections. Fortunately, China 2000 Population Census collected information of individuals' birth months and birth years, using which I can identify a group of individuals who were affected in the first year of life by the famine but whose parents' fertility (including their decisions about whether to give birth to a child and the possibility for women to conceive) was not affected by the famine. Such individuals were born between February 1958 and June 1959. For example, if an individual was born in February 1959, then he or she should be conceived in April 1958 when the famine had not started, therefore their parents' fertility was not affected by the famine; but this individual was affected by the famine in the first year of life. One might notice that although individuals born between July 1959 and October 1959 were also conceived before the famine, I do not include them in this group. It is because when the famine started, these individuals were in the first five

15 See (Frisch & McArthur, 1974; Forster and Ranum, 1975; Frisch, 1978, 2002) for review of biological evidence. See (Ford et al., 1989; Langsten, 1985; Elias et al., 2007; Jowett, 1991) for description of “famine amenorrhea” in Bangladesh, China and the Netherlands. 252 X. Shi / China Economic Review 22 (2011) 244–259 months in utero when the fetus could still be aborted and their parents' decision about whether to abort the fetus might be affected by the famine. Therefore, these individuals could not be free from fertility selections. However, one caveat we need to bear in mind is that there were other selection effects caused by the famine. In the famine, weaker children would die because of insufficient nutrition intakes, leading to a positive selection. Since data used in this paper are from 2000 Population Census, people should be still alive in 2000 in order to be included in the survey. If cohorts affected by the famine in the first year of life are more likely to die, only those stronger people can be included in the data, leading to another positive selection. One other selection might come from migration during the famine. If people migrated from regions with more severe famine to regions with less severe famine, then the estimated famine effects would be biased toward zero. Because of data limitation, I cannot control for these selections. Therefore, the estimated effects of the famine in this paper can be thought of as a low bound of true effects. From Fig. 1, we can see that there is a dramatic increase in the birth rate right after the famine. One explanation could be that those parents postponing having children in the famine started to have children after the famine ended, leading to a negative selection of children conceived and born after the famine. For the sake of clean comparison, I do not include cohorts born after the famine in the sample for analysis. According to the above analysis, cohorts born between February 1958 and June 1959 are used as a treatment group in this paper. They were affected by the famine in the first year of life but not affected by the fertility selections. Cohorts born between January 1954 and January 1958 are used as a control group. Individuals in this group were not affected by the famine in the first year of life. In order to see how positive fertility selections affect the estimated famine impacts, I re-estimate the regression functions using a different sample. In the new sample, cohorts born between July 1959 and December 1961, therefore affected by the famine in the first year of life and fertility selections, are used as a treatment group; while the control group is kept the same. If the estimated impacts of the famine become weaker, then it provides evidence for the existence of positive fertility selections.

4.2. Effects of the famine controlling for fertility selections

Table 5 presents regression results from estimating Eq. (1) for a variety of outcome variables using the ﬁrst sample described above, that is, cohorts born between January 1954 and June 1959. The results presented in Table 5 are not contaminated by fertility

Table 5 Effects of the famine after controlling for fertility selections.

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)

Education Labor market Household wealth Education Labor market Household wealth performance performance

Years of Completing Labor supply Average Average Years of Completing Labor supply Average Average schooling high number of housing schooling high number of housing school=1 room per areas per school=1 room per areas per capita capita capita capita

Women Famine intensity 0.436 −0.089 0.430 −0.602 −7.053 0.127 −0.071 −0.207 −0.479 −7.337 experienced in the (0.433) (0.024)*** (0.258) (0.092)*** (3.405)** (0.591) (0.032)** (0.563) (0.178)** (3.952)* ﬁrst year of life Provincial linear time Yes Yes Yes Yes Yes No No No No No trend Provincial quadratic No No No No No Yes Yes Yes Yes Yes time trend Number of observations 26346 26346 26346 26293 26317 26346 26346 26346 26293 26317 R-squared 0.12 0.02 0.12 0.08 0.16 0.13 0.02 0.12 0.08 0.16

Men Famine intensity 0.032 −0.040 −0.225 0.101 −1.154 0.408 −0.008 −0.435 −0.114 −3.483 experienced in the (0.351) (0.040) (0.293) (0.089) (1.707) (0.754) (0.056) (0.611) (0.098) (3.013) ﬁrst year of life Provincial linear time Yes Yes Yes Yes Yes No No No No No trend Provincial quadratic time No No No No No Yes Yes Yes Yes Yes trend Number of observations 26904 26904 26904 26753 26742 26904 26904 26904 26753 26742 R-squared 0.09 0.03 0.07 0.08 0.14 0.09 0.03 0.07 0.09 0.14

Standard errors are calculated by clustering in province level; *significant at 10%; **significant at 5%; ***significant at 1%. Note:

(1) Labor supply is measured by the number of work days in Oct. 25–Oct.31 in 2000. (2) Birth year fixed effects, birth province fixed effects, and interactions of provincial political variables (including ratio of CCP member in population in 1957 and ratio of people accommodated by commune mess halls in population in 1959) with an indicator for being affected by the famine in the first year of life are included in all specifications. (3) Logit models are also fit for dependent variable in columns 2 and 7. Marginal effects calculated from the Logit models do not change much compared to those shown in this table. X. Shi / China Economic Review 22 (2011) 244–259 253 selections. This table is divided into upper and bottom panels. In the upper panel, estimates are from regressions using the sample of women. In the bottom panel, the estimates come from regressions using the sample of men. For each outcome in each panel, the coefficient of famine intensity experienced in the first year of life is presented. Standard errors are presented in parentheses. For brevity, regression coefficients for the constant term, the interactions of provincial variables (including the ratio of CCP members in the population in 1957 and the ratio of population accommodated by commune mess halls in 1959) with a dummy variable indicating whether the individual was affected by the famine in the first year of life, the large numbers of fixed effects and the provincial time trends are not shown. In columns 1 to 5, I control for provincial linear time trends, while I control for provincial quadratic time trends (including a linear term and its square) in columns 6 to 10. When discussing results, I focus on the coefficient of the famine intensity and discuss the impacts of per 0.1 percentage point increase of excess death rate (the unit of excess death rate is 10% in the data used for regressions), meaning 0.1 percentage point deviation of death rate in the famine from the predicted normal level. This format will be followed in subsequent tables. Columns 1 and 2 in the upper panel show the impacts of the famine on the educational attainments of women experiencing the famine in the first year of life. I use years of schooling (column 1) and the probability to complete high school (column 2) to measure educational attainments. The coefficient shown in the first column is 0.436, but not statistically significant. However, the coefficient in the second column is −0.089 and statistically significant at 1% level. Since the unit of excess death rate in the data is scaled to 10%; therefore, it means that with 0.1 percentage point increase of excess death rate, the probability for women to complete high school is 0.09 percentage point lower.16 Since the average value of famine intensity experienced by women in the first year of life is 0.57% (shown in Table 4), then we can know that on average exposure to the famine in the first year of life reduced women's probability to complete high school by 0.5 percentage point. Column 3 shows the impacts of the famine on the labor supply of women affected by the famine. Labor supply is measured by the number of work days from October 25 to October 31 in 2000. The coefficient is equal to 0.430 and not statistically significant. In addition to educational attainments and labor supply, I also investigate the effects of the famine on wealth. I use the average number of rooms per capita (column 4) and average housing areas per capita (column 5) to measure wealth. The coefficients shown in these two columns are both negative and statistically significant. The coefficient in column 4 (for the average number of rooms per capita) is equal to −0.602 and significant at the 1% level; and the coefficient in column 5 (for average housing areas per capita) is equal to −7.053, significant at the 5% level. With 0.1 percentage point increase of death rate in the famine from the normal level, the average number of rooms per capita decreases by 0.006 and average housing areas per capita decrease by 0.07 square meters, meaning on average exposing to the famine reduced the average number of rooms per capita by 0.034 and the average housing areas per capita by 0.399 square meters. There are two possible ways for the famine to affect household level variables; one works in a direct way: women affected by the famine had lower education, as shown in columns 1 and 2, which could reduce the income they were able to earn and contribute to the family; the other might go through an indirect way: women exposed to the famine in the first year of life were less educated, so they might marry to men having lower education level and thus lower income, reducing the household wealth as well. In columns 1 to 5, I control for linear provincial time trends. One might be concerned that provincial time trends might not be linear; therefore, I control for quadratic provincial time trends (including a linear term and its square) in columns 6 to 10. The coefficient in column 6 (for years of schooling) is still positive (0.127) but not statistically significant; and the coefficient in column 7 (for the probability to complete high school) is negative (−0.071) and statistically significant at 5% level. Then we can see from column 8 that the famine did not have significant effects on women's labor supply, and the coefficient is equal to −0.207. The coefficient in column 9 (for the average number of room per capita) is −0.479 and it is statistically significant at 5% level. And column 10 shows that the coefficient for average housing areas per capita is −7.337 and statistically significant at 10% level. From these results shown in columns 5 to 10, we can see that besides that the pattern of coefficients' significance level keeps the same, the magnitudes of the statistically significant coefficients do not change much, showing that the results are robust to different provincial time trends used. The bottom panel in Table 5 shows the regression results using the sample of men. In columns 1 to 5, I control for linear provincial time trends. We can see that none of the coefficients are statistically significant. Out of five outcome variables, there are two positive coefficients: 0.032 in column 1 (for years of schooling) and 0.101 in column 4 (for the average number of rooms per capita); while there are three negative coefficients: −0.040 in column 2 (for the probability to complete high school), −0.225 in column 3 (for labor supply), and −1.154 in column 5 (for average housing areas per capita). In columns 6 to 10, I control for quadratic provincial time trends. We can see the same pattern, that is, the coefficients of the famine intensity experienced by men in the early life are not statistically significantly different from zero. There is only one positive coefficient this time: 0.408 in column 6 (for years of schooling). And there are four negative coefficients: −0.008 in column 7 (for the probability to complete high school), −0.435 in column 8 (for labor supply), −0.114 in column 9 (for the average number of rooms per capita), and −3.483 in column 10 (for average housing areas per capita). From the bottom panel in this table, we can see that the famine effects on men are much weaker than on women. It is consistent with a gender bias model in which available resources are given to boys in bad times.17

16 Linear probability models are fit in this paper, but I also fit Logit models to calculate the marginal effects for binary dependent variables. Most of the marginal effects calculated using Logit models do not change much compared with those estimated using linear probability models. The results from Logit models are not shown in this paper, but available under request. 17 Dreze & Sen, 1989; Behrman, 1988; Rose, 1999; Alderman & Gertler, 1997; Duflo, 2003; Cameron & Worswick, 2001; Jayachandran, 2008; Maccini & Yang, 2009. 254 X. Shi / China Economic Review 22 (2011) 244–259

Table 6 Effects of the famine on the proportion of women in the sample.

Female = 1

(1) (2)

Famine intensity experienced in the ﬁrst year of life −0.044 −0.067 (0.061) (0.074) Provincial linear time trend Yes No Provincial quadratic time trend No Yes Number of observations 53250 53250 R-squared 0.00 0.00

Standard errors are calculated by clustering in province level; *significant at 10%; **significant at 5%; ***significant at 1%. Note:

(1) Birth year fixed effects, birth province fixed effects, and interactions of provincial political variables (including ratio of CCP member in population in 1957 and ratio of people accommodated by commune mess halls in population in 1959) with an indicator for being affected by the famine in the first year of life are included in all specifications. (2) Logit models are also fit for dependent variable in this table. Marginal effects calculated from the Logit models do not change much compared to those shown in this table.

One might be concerned that since the selection from the death of weaker children in the famine could not be controlled for in this paper, the insignificant effects of the famine on men might be driven by stronger selection effects on men if boys were more likely to die than girls during the famine. In order to test this, I regress a female indicator on famine intensity experienced in the first year of life after controlling for birth province fixed effects, birth year fixed effects, the interactions of provincial variables (including ratio of CCP member in population in 1957 and ratio of people accommodated by commune mess halls in population in 1959) with a dummy variable indicating whether the individual was affected by the famine in the first year of life, and provincial time trends. Table 6 shows the regression results. From this table, we can see that effects of the famine on the proportion of women in the sample are not significant, whether linear provincial time trends (column 1) or quadratic provincial time trends (column 2) are controlled for. This means that there is no systematic difference between the survival of boys and girls in the famine. Therefore, the insignificant effects of the famine on men should not be driven by the stronger selection effects of the famine on men. I also estimate the same specification as in Table 5 without controlling for the interactions of provincial variables (including the ratio of CCP members in the population in 1957 and the ratio of population accommodated by commune mess halls in 1959) with a dummy variable indicating whether the individual was affected by the famine in the first year of life. The coefficients of the famine intensity experienced in the first year of life do not change much.18 Although it does not necessarily mean that the famine intensities across provinces were not correlated with provincial leaders' political attitudes, it could provide some evidence suggesting that provincial leaders' political attitudes might not affect children's outcomes in adulthood through channels other than the famine, alleviating, if not eliminating, concerns about endogenous problem in OLS regressions.

4.3. Effects of the famine allowing for positive fertility selections

In Table 7, I investigate the effects of the famine on women and men, but not controlling for positive fertility selections. The sample used in Table 7 includes cohorts born between January 1954 and January 1958 and cohorts born between July 1959 and December 1961. The former group is not affected by the famine in the first year of life and therefore used as a control group; while the latter group is used as a treatment group, in which individuals were affected by the famine in the first year of life and fertility selections. If the estimated effects of the famine become weaker, then it provides evidence for the existence of positive fertility selections in the famine. We can see from Table 7 that the coefficients in column 2 (for the probability to complete high school), column 4 (for the average number of rooms per capita), and column 5 (for average housing areas per capita) are statistically insignificant now; while the corresponding coefficients in Table 5 are statistically significant. And compared with those in Table 5, the magnitude of these three becomes smaller: the coefficient in column 2 changes from −0.089 (in Table 5)to−0.003, the coefficient in column 4 changes from −0.602 (in Table 5)to−0.161, and the coefficient in column 5 changes from −7.053 (in Table 5)to−3.932. The coefficients in column 1 (for years of schooling) and column 3 (for labor supply) are still statistically insignificant, which is the same as shown in Table 5. We can see the same pattern after I control for quadratic provincial time trends. The results are shown in columns 6 to 10 in the upper panel in Table 7. No coefficients are statistically significantly different from zero. The coefficient in column 7 (for the probability to complete high school) is −0.004 (changing from −0.071, statistically significant at 5% level in Table 5), the coefficient in column 9 (for the average number of rooms per capita) is −0.167 (changing from −0.479, statistically significant at 5% level in Table 5), and the coefficient in column 10 (for average housing areas per capita) is −4.074 (changing from −7.337, statistically significant at 10% level in Table 5). We can see from the above that all these three coefficients become from statistically significant in Table 5 to insignificant in Table 7 and the magnitudes of the coefficients also become smaller. The coefficients in columns 6 and 8 keep statistically insignificant. Therefore, if we do not control for fertility selections in the famine,

18 I also do the same exercises using specifications shown in Tables 6–8, and find few changes in the estimated coefficients of the famine intensity. All these results are not shown in this paper but available under request. X. Shi / China Economic Review 22 (2011) 244–259 255

Table 7 Effects of the famine not controlling for fertility selections.

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)

Education Labor market Household wealth Education Labor market Household wealth performance performance

Women Famine intensity 0.565 −0.003 −0.166 −0.161 −3.932 0.573 −0.004 −0.151 −0.167 −4.074 experienced in the (0.385) (0.026) (0.279) (0.101) (3.567) (0.383) (0.026) (0.281) (0.100) (3.836) ﬁrst year of life Provincial linear time Yes Yes Yes Yes Yes No No No No No trend Provincial quadratic time No No No No No Yes Yes Yes Yes Yes trend Number of observations 33868 33868 33868 33786 33826 33868 33868 33868 33786 33826 R-squared 0.14 0.02 0.12 0.09 0.17 0.14 0.02 0.13 0.09 0.17

Men Famine intensity 0.071 0.020 0.039 0.083 3.164 0.078 0.014 0.102 0.074 2.884 experienced in the (0.300) (0.038) (0.198) (0.052) (2.710) (0.293) (0.037) (0.171) (0.051) (2.515) ﬁrst year of life Provincial linear time Yes Yes Yes Yes Yes No No No No No trend Provincial quadratic time No No No No No Yes Yes Yes Yes Yes trend Number of observations 34124 34124 34124 33826 33758 34124 34124 34124 33826 33758 R-squared 0.10 0.03 0.08 0.09 0.15 0.10 0.03 0.08 0.09 0.15

Standard errors are calculated by clustering in province level; *significant at 10%; **significant at 5%; ***significant at 1%. Note:

4.4. Comparison with results from other papers

Recently, there are many papers studying the same topic as this paper. In this section, I compare my estimates with estimates from two other papers: Chen and Zhou (2007), which was published, and Almond et al. (2007), which used a 1% sample from China 2000 Population Census, compared to a 0.095% sample from the same census used in this paper. Using CHNS (China Health and Nutrition Survey) data, Chen and Zhou (2007) found that the famine had significantly negative effects on labor supply of individuals born in 1959 and 1960. While in this paper, I do not find any significant effects of the famine on labor supply. One possible reason for the difference could lie in the fact that labor supply is measured differently in Chen and Zhou (2007) and in this paper. Chen and Zhou (2007) used annual total working hours to measure labor supply; while labor supply in this paper is measured by the number of work days during October 25 and October 31 in 2000. Measured in a much shorter period, labor supply in this paper is more likely to be affected by some accidental factors happening during this period, causing imprecise estimates of the famine impacts. This paper finds that the famine had significantly negative effects on women but not on men in terms of housing area per capita. While Chen and Zhou (2007) found that on average the famine had significantly negative effects on the housing area per capita of individuals affected by the famine, but they did not go further to check whether the famine effects were different for men and women. 256 X. Shi / China Economic Review 22 (2011) 244–259

The only common outcome variable investigated in Almond et al. (2007) and this paper is housing area.19 Both papers find that the famine reduced housing area of cohorts affected more by the famine in early years of life. However, Almond et al. (2007) found that the famine had significantly negative effects on both men and women, while this paper only finds significantly negative effects of the famine on women but not on men. One possible reason could lie in the fact that Almond et al. (2007) investigated the famine effects on cohorts experiencing the famine in utero while this paper examines the famine effects on cohorts experiencing the famine in the first year of life. Without knowing the gender of the fetus,20 parents could not treat boys and girls differently, leading to the similar effects of the famine on boys and girls. However, knowing the gender after the baby was born, parents favoring boys, which was very common in rural areas in China, might spend more resources on boys than on girls, causing different effects of the famine on boys and on girls. Therefore, finding different effects of the famine on men does not mean that there are any conflicts between results in Almond et al. (2007) and in this paper. On the contrary, since Almond et al. (2007) and this paper examine different periods in the early years of life affected by the famine, results found in these two papers could be considered as complementary.

5. Discussions

5.1. Are the results driven by post-famine shocks?

Since I am investigating the impacts of the famine on the individuals' outcomes about forty years later, one might be concerned that there was some other big shock such as the Great Proletarian Culture Revolution (1966–1976) which might be related with the famine and affect individuals' outcomes as well, leading to a bias in the estimation. If provinces having more severe famine also experienced more intense political movements in the Proletarian Culture Revolution, perhaps due to their leaders' more radical political attitudes, and if the Proletarian Culture Revolution had negative effects on individuals' outcomes like educational attainments,21 then our estimates of the famine effects would be upward biased. In this section, I do a pseudo test to show that it is not going to be the case. The logic of this pseudo test is as follows: if the estimated famine effects on individuals born in rural areas are not the true effects of the famine, but totally driven by the Cultural Revolution; then if we estimate the regression function using individuals born in the urban areas, we should also see the similar results as those estimated using rural sample since the Cultural Revolution affected the urban areas. If we do not see similar results, then it provides some evidence showing that the Cultural Revolution would not be the underlying factor driving the results. I therefore estimate Eq. (1) using an urban sample.22 Table 8 shows the estimated results. As before, I control for linear provincial time trends in columns 1 to 5 while I control for quadratic provincial time trends in columns 6 to 10. We can see results using the sample of women first. All the signs of the coefficients, except for the coefficients in column 1 and column 6 (both for years of schooling), are the opposites of those in Table 5. And the coefficient in column 9 (for the average number of rooms per capita) becomes significantly positive (0.209, at 5% level). For men, except for the signs of coefficients for years of schooling (columns 1 and 6), labor supply (columns 3), average number of room per capita (column 9) and average housing areas per capita (column 10), all other coefficients' signs are the opposites of those in Table 5. And none of them are statistically significantly different from zero either. Results shown in Table 8 imply that the estimated famine effects shown in Table 5 should not be driven by the Cultural Revolution; otherwise the estimates using the urban sample, especially the urban women sample, would be similar as those in Table 5. The estimates could still be biased if there were other post-famine programs targeting the specific population affected by the famine. For example, the government might invest more in provinces affected by the more severe famine. However, to the best of my knowledge, there were no such programs in China from the end of the famine to 2000.

5.2. Other potential biases in the estimates

Besides the biases due to the selection effects of children death in the famine and migration in the famine, which are discussed in Section 4.1, there are other potential biases in the estimates. I will discuss these biases in this section. One might be concerned that people who were conceived before the famine might still be subject to the selective fertility decision on younger siblings. For example, for the treated group of cohorts born in February 1958–June 1959, their parents were less likely to have additional births in the following years due to the famine. However, for the comparison group of cohorts born in January 1954–January 1958, their parents could have more births in the following years. More siblings could dilute family resources allocated to children, especially in the famine. It means that the treated group could have been affected more by the famine if their parents had the same likelihood to have births as the parents of the comparison group, leading to downward bias in the estimates and meaning that the famine effects could be larger than estimated.

19 Almond et al. (2007) used total housing area and this paper uses housing area per capita. 20 B-scan ultrasound devices, which could be used to detect fetus's gender, were not widely introduced to China until 1970 s (Chen, Li, & Meng, 2010). 21 Gregory & Meng, 2002a,b; Giles et al., 2008. 22 Urban sample includes cohorts born between January 1954 and June 1959; individuals in this sample lived in urban regions and had non-agricultural hukou in 2000. X. Shi / China Economic Review 22 (2011) 244–259 257

Table 8 Effects of the famine using the urban sample.

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)

Education Labor market Household wealth Education Labor market Household wealth performance performance

Years of Completing Labor supply Average Average Years of Completing Labor supply Average Average schooling high number of housing schooling high number of housing school = 1 room per areas per school = 1 room per areas per capita capita capita capita

Women Famine intensity 2.088 0.278 −0.405 0.088 0.863 1.185 0.196 1.239 0.209 4.904 experienced in the (2.023) (0.328) (0.668) (0.120) (6.631) (2.627) (0.378) (1.314) (0.101)** (7.188) ﬁrst year of life Provincial linear time Yes Yes Yes Yes Yes No No No No No trend Provincial quadratic time No No No No No Yes Yes Yes Yes Yes trend Number of observations 7681 7681 7681 7663 7663 7681 7681 7681 7663 7663 R-squared 0.04 0.05 0.03 0.05 0.05 0.04 0.06 0.03 0.05 0.05

Men Famine intensity 0.558 0.238 −0.559 -0.187 4.126 1.292 0.103 0.032 −0.438 -1.706 experienced in the (1.100) (0.238) (0.395) (0.237) (6.877) (1.244) (0.272) (0.694) (0.288) (8.559) ﬁrst year of life Provincial linear time Yes Yes Yes Yes Yes No No No No No trend Provincial quadratic time No No No No No Yes Yes Yes Yes Yes trend Number of observations 8179 8179 8179 8088 8087 8179 8179 8179 8088 8087 R−squared 0.03 0.05 0.01 0.03 0.04 0.03 0.05 0.02 0.03 0.04

Standard errors are calculated by clustering in province level; *significant at 10%; **significant at 5%; ***significant at 1%. Note:

The second concern comes from the possible measurement error in the famine intensity constructed. Because of the lack of precise information of the starting and ending time of the famine in different provinces, I assume that the famine started and ended at the same time in all provinces, and then construct famine intensity measurement accordingly. The existence of measurement error leads to attenuation bias in the estimation (Wooldridge, 2002), meaning that the true effects of the famine could be larger than the effects estimated after controlling for fertility selections in this paper. Another consequence of measurement error in the famine intensity variable lies in that the ﬁnding that the famine effects are not statistically signiﬁcantly different from zero if the fertility selections are not controlled for could be driven by the measurement error. However, in order to test the existence of positive fertility selections, what matters is the relative change from the estimates controlling for fertility selections to the estimates without controlling for fertility selections. Given that measurement error in the famine intensity variable leads both estimates (with or without controlling for fertility selections) toward to zero, the comparison of these two estimates should be less affected by, if not totally free from, the measurement error. It means that comparison between these two sets of estimates, as I do in this paper, can provide valid evidence for the existence of the positive fertility selections. The third concern comes from the possible selection due to the analysis sample based on household registration status in 2000. By using this sample, those changing from agricultural hukou to non-agricultural hukou before 2000 are excluded, leading to a potential selection bias. An important way to change hukou from agriculture to non-agriculture is to enter technical colleges or universities (Chan & Zhang, 1999; Chan, 2009). With the re-introduction of merit-based admissions to technical colleges or universities in the fall of 1979, the younger cohorts were more likely to have an opportunity to compete for college admissions than older cohorts. Therefore, the positive fertility selection cohorts (those born in July 1959–December 1961) might have the largest share of individuals completing college education and changing their hukou, then excluded from the sample, while the comparison cohorts (those born in January 1954–January 1958) might have the smallest share of such individuals. And the cohorts not affected by the fertility selections (those born in February 1958–June 1959) were in the midst. Then in the estimation by comparing positive fertility selection cohorts with comparison cohorts, the estimated famine effects could be upward biased, meaning that the estimated famine effects could have been weaker. It actually supports the existence of the positive fertility selections. However, in the estimation by comparing cohorts not affected by the positive fertility selection with comparison cohorts, the estimated famine effects might also be upward biased, meaning that the true famine effects might not be as strong as estimated. But without further information, we cannot say more about the degree of this bias. 258 X. Shi / China Economic Review 22 (2011) 244–259

6. Conclusion

The famine in 1959–1961 not only caused about 30 million excess deaths, but might also have long term impacts on those survivors. This paper uses the China 2000 population census data to study this issue. Using a rural sample from the China 2000 Population Census data and after controlling for positive fertility selections, I find that experiencing the famine in the first year of life had negative effects on women's adult outcomes. With death rate higher than the normal level by 0.1 percentage point in the famine, women affected by the famine in the early life had a 0.09 percentage point lower probability of completing high school and lived in less wealthy households (measured by smaller houses: 0.006 fewer rooms per capita and 0.07 square meters smaller housing area per capita). I do not find any significant effects of being exposed to the famine on men. In addition, I find that if fertility selections are not controlled for, the estimated effects of the famine become weaker by 70% on average. In addition to revealing the long-term effects of China's 1959–1961 famine, these results have important implications for policy. Our findings point to a group (infant girls), which is particularly vulnerable to fluctuations in exogenous environments. As such, findings in this paper provide additional justification for interventions, such as nutrition-enhancement programs, social insurance schemes, public health investments, or policies ensuring food security, which protect infants from temporary environmental and economic shocks.

Data Appendix

[1] Years of schooling: The number of years of schooling is created from respondent's self report of the highest level of education. In China, primary school is typically six years, followed by three years of junior high school, followed by three years of senior high school (or three years of professional school), followed by four years of university (BenKe) (or three years of college (DaZhuan)), followed by three years of graduate school study for MA degree, then followed by three years of graduate school study for Ph.D. degree. The respondents getting degrees from adult education programs are dropped. [2] High school indicator: Whether the highest level of education of the respondent is high school or above. [3] Labor supply: How many days did the respondent work during October 25 and October 31 in 2000? [4] Average number of rooms per capita: Total number of rooms divided by the total number of family members. [5] Average housing areas per capita: Total household construction areas divided by the total number of family members.

References

Alderman, Harold, & Gertler, Paul (1997). Family resources and gender differences in human capital investments: The demand for children's medical care in Pakistan. In Lawrence Haddad, John Hoddinott, & Harold Aldermand (Eds.), Intrahousehold resource allocation in developing countries: Models, methods, and policy (pp. 231−248). Baltimore and London: John Hopkins University Press. Alderman, Harold, Hodinott, John, & Kinsey, Bill (2006). Long term consequences of early childhood malnutrition. Oxford Economic Papers, 58, 450−474. Almond, Douglas (2006). Is the 1918 influenza pandemic over? long-term effects of in utero influenza exposure in the post-1940 U.S. population. Journal of Political Economy, 114, 672−712 (August). Almond, Douglas, & Chay, Kenneth (2006). The long run and intergenerational impact of poor infant health: evidence from cohorts born during civil rights era. New York: NY (Columbia University) and Berkeley, CA (University of California at Berkeley), Working Paper. Almond, Douglas, Edlund, Lena, Li, Hongbin, & Zhang, Junsen (2007). Long-term effects of the 1959–1961 China famine: Mainland China and Hong Kong. NBER Working Paper, 13384. Ashton, Basil, Kenneth, Hill, Alan, Piazza, & Robin, Zeitz (1984). Famine in China, 1958-61.Population Development Review, 10(4), 613−645 (December). Barker, David, J. P., Bleker, O. P., van Montfrans, G. A., Osmond, C., Ravelli, A. C. J., et al. (2005). Cardiovascular disease in survivors of the Dutch famine. The impact of maternal nutrition on the offspring. Nestle Nutrition workshop series. Pediatric Program, vol. 55. Behrman, Jere R. (1988). Nutrition, health, birth order and seasonality: Intrahousehold allocation among children in rural India. Journal of Development Economics, 28(1), 43−62. Behrman, Jere R., & Rosenzweig, Mark (2004). Returns to birthweight. The Review of Economics and Statistics, 86(2), 586−601. Bertrand, Marianne, Duflo, Esther, & Mullainahtanm, Sendhil (2004). How much should we trust difference-in-difference estimates? Quarterly Journal of Economics, 119(1), 249−275 (February). Cameron, Lisa A., & Worswick, Christopher (2001). Education expenditure responses to crop loss in Indonesia: A gender bias. Economic Development and Cultural Change, 49(2), 352−363. Chan, Kam Wing (2009). The Chinese hukou system at 50. Eurasian Geography and Economics, 2, 197−221. Chan, Kam Wing, & Zhang, Li (1999). The hukou system and rural-urban migration in China: Processes and changes. The China Quarterly, 160, 818−855. Chen, Yuyu, Li, Hongbin, & Meng, Lingsheng (2010). Prenatal sex selection and missing girls in China: Evidence from the diffusion of diagnostic ultrasound. Beijing, China (Beijing University), Beijing, China (Tsinghua University) and Beijing, China (Tsinghua University). Working paper. Chen, Yuyu, & Zhou, Li'an (2007). The long-term health and economic consequences of the 1959–1961 famine in China. Journal of Health Economics, 26, 659−681. Dreze, Jean, & Sen, Amartya (1989). Hunger and public action. Oxoford: Oxford University Press. Duflo, Esther (2001). Schooling and labor market consequences of school construction in indonesia: Evidence from an unusual policy experiment. The American Economic Review, 91(4), 795−813. Duflo, Esther (2003). Grandmothers and granddaughters: Old age pension and intrahousehold allocation in South Africa. The World Bank Economic Review, 17(1), 1−25. Elias, Sjoerd, G., Van Noord Paulus, A. H., Peelers Petra, H. M., Den Tonkelaar Isolde, Kaaks Rudolf, et al. (2007). Menstruation during and after caloric restriction: The 1944–1945 Dutch famine. Fertility and Sterility, 88(4), 1101−1107 (October-Supplement). Ford, Kathleen, Huffman, Sandra L., Chowdhury, A. K. M. A., Stan Becker, & Hubert Allen (1989). Birth-interval dynamics in rural Bangladesh and maternal weight. Demography, 26(3), 425−437 (August). Forster, Robert, & Ranum, Orest (Eds.). (1975). Biology of man in history. Baltimore and London: The Johns Hopkins University Press. Frisch, Rose E. (1978). Population, food intake, and fertility. Science, New Series, 199(4324), 22−30. Frisch, Rose E. (2002). Female fertility and the body fat connection. Chicago: University of Chicago Press. Frisch, Rose E., & McArthur, J. W. (1974). Menstrual cycles: fatness as a determinant of minimum weight for height necessary for their maintenance or onset. Science, 184, 949−951. X. Shi / China Economic Review 22 (2011) 244–259 259

Giles, John, Park, Albert, & Wang, Meiyan (2008). The Great Proletarian Cultural Revolution, disruptions to education, and returns to schooling in urban China. WB Policy Research Working Paper, 4729. Gregory, R. G., & Meng, Xin (2002a). The impact of interrupted education on subsequent educational attainment: A cost of the Chinese cultural revolution. Economic Development and Cultural Change, 50(4), 935−959. Gregory, R. G., & Meng, Xin (2002b). Exploring the impact of interrupted education on earnings: The educational cost of the Chinese cultural revolution.” Canberra, Australia (Australia National University) and Canberra, Australia (Australia National University). Working Paper. http://en.wikipedia.org/wiki/Cultural\_Revolution Jayachandran, Seema (2008). Air quality and early-life mortality: Evidence from Indonesia's wildﬁres. NBER Working Paper, 14011. Jowett, A. John (1991). The demographic response to famine: The case of China 1958–1961.GeoJournal, 23(2), 135−146 (February). Langsten, R. (1985). Determinants of natural fertility in rural Bangladesh reconsidered.Population Studies, 39(1), 153−161 (March). Li, Wei, & Yang, Dennis T. (2005). The Great Leap Forward: Anatomy of a central planning disaster. Journal of Political Economy, 113, 840−877. Lin, Justin Y., & Yang, Dennis T. (2000). Food availability, entitlements and the Chinese famine of 1959–61.The Economic Journal, 110, 136−158 (January). Lindeboom, Maarten, Van den Berg, Gerard J., & Portrait, France (2006). Economic conditions early in life and individual mortality. The American Economic Review, 96(1), 290−302 (March). Luo, Zhehui, Mu, Ren, & Zhang, Xiaobo (2006). Famine and overweight in China. Review of Agricultural Economics, 28(3), 296−304. Maccini, Sharon, & Yang, Dean (2009). Under the weather: Health, schooling, and socioeconomic consequences of early-life rainfall. The American Economic Review, 99(3), 1006−1026 (June). Meng, Xin, & Qian, Nancy (2009). The long run impact of childhood malnutrition: Evidence from China's great famine. NBER Working Paper, 14973. Moulton, Brent (1986). Random group effects and the precision of regression estimates.Journal of Econometrics, 32(3), 385−397 (August). Mu, Ren, & Zhang, Xiaobo (2008). Gender difference in the long-term impact of famine. IFPRI Discussion Paper, 00760. Peng, Xizhe (1987). Demographic consequences of the Great Leap Forward in China's provinces. Population and Development Review, 13(4), 639−670. Ravallion, Martin (1987). Markets and famines. Oxford: Oxford University Press. Rose, Eliana (1999). Consumption smoothing and excess female mortality in rural India.The Review of Economics and Statistics, 81(1), 41−49 (February). Royer, Heather (2009). Separated at birth: Estimating the long-run and intergenerational effects of birthweight using twins. American Economic Journal: Applied Economics, 1(1), 49−85. Sen, Amartya (1981). Poverty and famine: An essay on entitlement and deprivation. Oxford: Oxford University Press. Stein, Zena, Susser, Mervyn, Saenger, Gerhart, & Marolla, Francis (1975). Famine and human development: The Dutch hunger winter of 1944–1945. New York: London and Toronto: Oxford University Press. Wooldridge, Jeffrey M. (2002). Econometric analysis of cross section and panel data. The MIT Press. Yang, Dali L. (1996). Calamity and reform in China. Stanford University Press.