The Long-Term Impact of Famine Experience on Harvest Losses

Abstract

Approximately one-third of the global food supply is lost or wasted each year. Given that the harvesting process is the initial stage following food production, minimizing losses in this crucial phase holds paramount significance in augmenting the food supply and ensuring food security. The 1959–1961 famine in China was one of the most catastrophic events in history and had long-term effects on human beings, particularly farmers. This paper aims to provide a new perspective on the variations in harvest losses across age cohorts by examining the impact of famine experiences. Using survey data from the 2016 Postproduction Food Loss and Waste Survey conducted by China Agricultural University and the Rural Economic Research Center, which involved 3538 farming households across 28 provinces, we construct a cohort difference-in-difference (DID) model to investigate the impact of famine experience on household harvest losses. The standard cohort DID estimation results indicate that in areas with severe famine, a 1% increase in excess mortality would reduce the rate of harvest loss by 3%, suggesting that farmers who have experienced extreme famine have a deeper memory of the event, which subsequently helps them reduce harvest losses. Moreover, the results of the heterogeneity test reveal that the more serious the famine that the household head experienced in early life was, the less harvest losses there were, particularly for those who were adolescents during the famine. The findings elucidate the importance of historical events in shaping current behaviors and contribute to a better understanding of the variation in harvest losses across age cohorts.

1. Introduction

Famine and food shortages continue to be pressing issues for governments and international organizations worldwide. Although the production of major cereals has increased steadily in recent years and surpassed 3 billion tonnes, the global demand for cereals has also risen, resulting in a basic balance between supply and demand [1]. Unfortunately, this has not been enough to prevent famine events from occurring in different parts of the world. The 2020 famine in Zimbabwe, the 2017 famine in South Sudan, and the 2011–2012 famine in Somalia are just a few examples of such events. In 2020, according to the Global Report on Food Crises [2], 155 million people across 55 countries and territories suffered from severe food insecurity. The problem of food loss and waste exacerbates the issue of global food security. The FAO estimates that approximately one-third of the world’s food is lost or wasted annually [3], which has a significant impact on food supply and demand. Reducing food losses and waste has become one of the most important ways to alleviate pressure on global food security.

The causes of famine are multifaceted, stemming from a combination of factors such as severe weather patterns, the presence of oversized rural communes, inadequate policy decisions, and the absence of the right of farming households to withdraw from collective organizations [4,5]. The primary focus of this article is not on the specific mechanisms that led to the occurrence of the famine but rather on the long-lasting impacts it had on those who survived it. Scholars have conducted extensive research on this subject, particularly on the physical health consequences. For example, one study found that children who survived the famine developed obesity that continued into their adult years [6,7]. Another study found that prenatal famine resulted in impaired glucose tolerance during adulthood [8]. Roseboom T et al. [9] found that poor maternal nutrition during gestation may contribute to restricted fetal growth, increasing the risk of coronary heart disease, atherosclerosis, abnormal blood clotting, heightened stress responses, and obesity later in life. In addition, fetuses and children deprived of nutrition during famine face more serious mental health issues, such as cognitive impairment, depression, and schizophrenia [10,11,12]. Research on decision making suggests that individuals tend to be excessively responsive to extreme events, which may significantly influence their subsequent choices [13,14]. For example, Callen [15] thought that exposure to the Indian Ocean Earthquake tsunami altered the time preferences of Sri Lankan wage workers, not only in terms of their rate of time preference but also in their beliefs about the future, background consumption, and intertemporal arbitrage opportunities. Griskevicius [16] proposed that individuals who grew up in harsh early-life environments were more impulsive, took more risks, and approached temptations more quickly, according to life-history theory. Feng and Johansson [17] believed that surviving a famine during one’s younger years is associated with more conservative financial, investment, and cash-holding policies; a lower likelihood of unethical behavior; and better firm performance during economic downturns.

Currently, there are two primary sources of food loss data. The first is the measurement of food losses at the macro level by international organizations and national governments. For example, the FAO estimates that in 2020, global food loss and waste accounted for approximately one-third of total food production [3]. The second is the measurement of food losses for microentities by segment, variety, and region through controlled experiments and sampling [18,19]. Regardless of the measurement method used, it is widely accepted among scholars that food loss rates are high and concentrated in developing countries during the harvesting, drying, and storage stages [20,21,22]. There are two main categories of reasons for harvest losses: objective conditions, such as climatic conditions, machinery quality, and planting scale [23,24]; and subjective reasons, such as grain-saving awareness, labor attitude, harvest timing, and field management [25,26]. However, most studies have focused on the objective causes of harvest losses, and there is a lack of research analyzing the influence of farmers’ characteristics on harvest losses. According to our survey, only approximately 30% of Chinese farmers engage in gleaning after harvest, which is in contrast to the social culture of valuing and preserving grains in China during the 1970s and 1980s. Therefore, what has led to this change?

In previous research, despite the extensive examination by numerous scholars into the influence of age on food loss and waste, a consistent consensus remains elusive. Some studies suggest that younger individuals are more likely to waste food than older consumers [27,28], while others have found evidence to the contrary. For instance, Hazuchova et al. [29] reported that food waste was highest among Czech residents over 65 years of age, and Koivupuro et al. [30] found no significant correlation between the age of the oldest person in a Finnish household and food waste. These divergent findings may be because individual behaviors do not show consistent changes with age and food loss and waste behaviors may stem from early life experiences [31]. Farmers who have experienced famine are more predisposed to cultivating a frugal mindset, making them less tolerant toward the incidence of food loss and waste [17,31]. Therefore, they may be more meticulous in harvesting and tend to choose harvesting methods with less loss, ultimately leading to a reduction in harvest losses [32,33]. The Chinese famine of 1959–1961 provided a quasi-natural experimental setting to test the hypothesis, as the famine could not have been foreseen and the strict household registration system in China prevented large-scale farmer migration across regions. To examine the impact of famine experience on harvest losses, we utilized data from the Postproduction Food Loss and Waste Survey and constructed a cohort difference-in-difference (DID) model.

This paper contributes to the literature in three ways. First, our study elucidates the underlying rationale behind the impact of early experiences on farmers’ behaviors, which was previously concealed by age characteristics. We explore the impact of famine experiences on farmers’ behaviors and broaden the scope of research into the long-term impact of famine experiences on individuals. Second, our study fills a gap in the literature by analyzing the factors influencing harvest losses from the subjective perspective of farmers. Third, this paper identifies age cohorts that are more susceptible to famine and estimates the impact of famine experience on harvest losses, providing empirical evidence for further research.

The rest of this paper is structured as follows. In Section 2, we conduct a literature review and propose a hypothesis. Section 3 introduces our data and models. Section 4 presents the results of the standard cohort DID, by-cohort DID, placebo test, robustness check and heterogeneity test. The findings are discussed in Section 5. Finally, Section 6 presents the research conclusions.

2. Background

2.1. The 1959–1961 Famine in China

From 1959 to 1961, China experienced an unprecedented three-year famine that caused significant loss of life (the mortality rates are presented in

Table A1. Mortality rates in 27 Chinese provinces (%), 1956–1961.

Province	1956	1957	1958	1959	1960	1961
Anhui	14.25	9.10	12.36	16.72	68.58	8.11
Beijing	7.73	8.19	8.08	9.66	9.14	10.80
Fujian	10.20	9.80	9.43	12.46	20.70	15.99
Hebei	11.30	11.30	10.90	12.30	15.80	13.60
Heilongjiang	10.08	10.45	9.17	12.76	10.52	11.13
Inner Mongolia	7.90	10.50	7.90	11.00	9.40	8.80
Jiangsu	13.02	10.26	9.40	14.55	18.41	13.35
Jiangxi	12.49	11.48	11.34	13.01	16.06	11.54
Jilin	7.53	9.05	9.12	13.43	10.13	12.05
Liaoning	6.60	9.40	8.80	11.80	11.50	17.50
Shandong	12.10	12.10	12.80	18.20	23.60	18.40
Shanxi	11.60	12.68	11.73	12.84	14.21	12.20
Tianjin	8.79	9.35	8.66	9.88	10.34	9.89
Zhejiang	9.46	9.32	9.15	10.81	11.88	9.84
Henan	14.00	11.80	12.69	14.10	39.56	10.20
Hubei	10.81	9.61	9.60	14.49	21.21	9.08
Hunan	11.51	10.41	11.65	12.99	29.42	17.48
Guangdong	11.08	8.36	9.17	11.10	15.24	10.82
Guangxi	12.46	12.35	11.74	17.49	29.46	19.50
Sichuan	10.40	12.10	25.20	47.00	54.00	29.40
Guizhou	13.01	12.35	15.26	20.28	52.33	23.27
Yunnan	15.21	16.29	21.62	17.95	26.26	11.84
Shaanxi	9.85	10.31	11.01	12.72	12.27	8.76
Gansu	10.78	11.33	21.11	17.36	41.32	11.48
Qinghai	9.43	10.40	12.99	16.58	40.73	11.68
Ningxia	10.58	11.06	14.98	15.82	13.90	10.71
Xinjiang	14.20	14.00	13.00	18.84	15.67	11.71

Source: China Compendium of Statistics 1949–2008 [38].

). Although there is no consensus on the precise number of famine deaths, many scholars estimate that between 15 million and 45 million people perished during this period [34,35,36,37]. During the Great Famine, all of China’s provinces experienced varying degrees of increased mortality. To illustrate the impact of famine on mortality, we calculated the average mortality rate in each province before (1956–1958) and during the famine (1959–1961) and measured the severity of the famine by the difference between the two rates (excess mortality).

The results in

Table 1. Mortality in 27 Chinese provinces before and during the 1959–1961 famine (%).

Province	1956–1958	1959–1961	Excess Mortality
Anhui	11.90	31.14	19.23
Beijing	8.00	9.87	1.87
Fujian	9.81	16.38	6.57
Hebei	11.17	13.90	2.73
Heilongjiang	9.90	11.47	1.57
Inner Mongolia	8.77	9.73	0.97
Jiangsu	10.89	15.44	4.54
Jiangxi	11.77	13.54	1.77
Jilin	8.57	11.87	3.30
Liaoning	8.27	13.60	5.33
Shandong	12.33	20.07	7.73
Shanxi	12.00	13.08	1.08
Tianjin	8.93	10.04	1.10
Zhejiang	9.31	10.84	1.53
Henan	12.83	21.29	8.46
Hubei	10.01	14.93	4.92
Hunan	11.19	19.96	8.77
Guangdong	9.54	12.39	2.85
Guangxi	12.18	22.15	9.97
Sichuan	15.90	43.47	27.57
Guizhou	13.54	31.96	18.42
Yunnan	17.71	18.68	0.98
Shaanxi	10.39	11.25	0.86
Gansu	14.41	23.39	8.98
Qinghai	10.94	23.00	12.06
Ningxia	12.21	13.48	1.27
Xinjiang	13.73	15.41	1.67

Source: China Compendium of Statistics 1949–2008 [38].

suggest that no province reported excess mortality rates below zero. Out of the 27 provinces, 11 had excess mortality rates above 5%. Among them, the mortality rates during the famine in Anhui, Sichuan, Guizhou, and Qinghai were two-to-three times higher than prefamine levels, with excess mortality rates surpassing 10%. The highest mortality and excess mortality rates were observed in Sichuan, with rates of 43.47% and 27.57%, respectively. Between 1959 and 1961, Sichuan experienced approximately 13.8 million excess deaths [38], which is only 1.1 million less than the number of global excess deaths caused by COVID-19 between 2020 and 2021 [39].

2.2. Food Harvesting Losses in China

In recent years, reducing food waste and loss has become a global priority, with many countries and international organizations devoting significant attention to this issue. For instance, in 2019, the FAO released a special report, “Moving Forward on Food Loss and Waste Reduction [40]”, calling for a global reduction in food loss and waste. This issue has also garnered increasing attention from other organizations, such as the United Nations Environment Programme (UNEP) [41] and the International Food Policy Research Institute (IFPRI) [42].

According to the FAO definition, food losses refer to the decrease in edible food mass throughout the part of the supply chain that specifically leads to edible food for human consumption [43]. Harvest losses are typically considered food losses that occur before harvested crops are transported, threshed, and cleared [18,33,44]. Furthermore, Arends-Kuenning et al. [18] found that harvesting is perceived as the most critical stage where food loss occurs. In the 2016 Post-production Food Loss and Waste Survey, the average harvest loss for sampled households was found to be approximately 51.256 kg (

Table 2. Average household harvest losses in 27 provinces in China (kg).

Province	Harvest Loss	Province	Harvest Loss
Anhui	56.689	Jiangxi	37.400
Beijing	5.133	Jilin	56.908
Fujian	14.350	Liaoning	44.473
Gansu	61.513	Ningxia	8.830
Guangdong	48.875	Qinghai	38.623
Guangxi	20.977	Shaanxi	46.477
Guizhou	14.449	Shandong	24.230
Hebei	42.096	Shanxi	40.173
Heilongjiang	101.120	Sichuan	16.803
Henan	30.739	Tianjin	136.786
Hubei	33.857	Xinjiang	41.229
Hunan	39.374	Yunnan	16.675
Inner Mongolia	146.911	Zhejiang	25.250
Jiangsu	52.755	Average	51.256

Note: The average harvest loss is the mean of the harvest loss of all sample households.

). According to the bulletin on the main data of the Third China National Agricultural Census [45], there are 207.430 million farming households in China, which means that harvest losses could be as high as 10.632 million tonnes.

3. Data and Methods

3.1. Data Source

This study used both macro- and microlevel data. The macrolevel mortality data for each province were obtained from the China Compendium of Statistics 1949–2008 [38]. The microdata were derived from the 2016 Postproduction Food Loss and Waste Survey conducted by China Agricultural University and the Rural Economic Research Center, Ministry of Agriculture and Rural Affairs of the PRC. The survey covered food losses of Chinese farming households during harvesting, storage, and transportation in 2015 and was conducted for farmers in 28 provinces, excluding Hong Kong, Macao, Taiwan, Tibet, and Shanghai.

The survey used stratified random sampling to obtain 6292 valid samples, with the sampling quantity of each grain variety determined by its output in China in 2015. The main survey provinces of each variety were selected based on their production proportion, and two counties were chosen in each surveyed province according to their geographical location and economic development level. Two villages were selected in each county, and 15–30 farmers were randomly surveyed in each village.

As Chongqing was not designated as a municipality until 1997, official statistics lack data on mortality rates during the Great Famine in this region. Therefore, the survey sample from this region was excluded from our analysis. Additionally, to ensure that the selected samples were engaged in agricultural production and possessed certain decision-making abilities, we also excluded farmers with zero arable land area, zero harvest yield, zero agricultural income, ages above 75 or below 35, and missing core variables. The actual sample size used was 3538.

3.2. Model Specification

The double difference model has been widely used by scholars as an important tool to study the effects of exogenous shocks on individuals [46,47,48,49]. It is based on a counterfactual framework to assess the changes in the explanatory variables in two conditions of shock occurrence and nonoccurrence. However, the shocks need to be completely exogenous; otherwise, there may be endogeneity problems due to sample self-selection.

The famine experience is a completely exogenous shock to the individual farmer, and no one can foresee the occurrence of famine, which satisfies the assumptions of the model. However, since data on farm household losses before the onset of famine are not available, we constructed the standard cohort specification DID model based on farm household birth cohorts and excess mortality rates during the famine in the farmers’ area [50]. The specific model used in this study is as follows:

$$ln{\mathrm{F}\mathrm{H}\mathrm{L}}_{\mathrm{i}\mathrm{j}\mathrm{p}\mathrm{a}}={\beta }_0+{\beta }_1{\mathrm{E}\mathrm{M}}_{\mathrm{p}}\times \mathrm{T}\left(1940\le j\le 1960\right)+{\beta }_2{\mathrm{X}}_{\mathrm{i}\mathrm{j}\mathrm{p}\mathrm{a}}+{\gamma }_j+{\delta }_p+{\mu }_a+{\epsilon }_{ijpa}$$

(1)

where i denotes the individual, j denotes the birth cohort, p denotes the province, and a denotes the crop type. ${\mathrm{F}\mathrm{H}\mathrm{L}}_{\mathrm{i}\mathrm{j}\mathrm{p}\mathrm{a}}$ represents the crop a harvest loss of farmer i in birth cohort j in province p. ${\mathrm{E}\mathrm{M}}_{\mathrm{p}}$ is the excess mortality rate during the famine in province p, calculated by the difference between the average mortality rate in 1959–1961 and the average mortality rate in 1956–1958, reflecting the severity of the famine. The dummy variable $\mathrm{T}$ is used to indicate the birth cohort, and we set farmers born in 1940–1960 as the treatment group with a value of 1 and farmers born in 1961–1980 as the control group with a value of 0. ${\beta }_1$ reflects the long-term effect of famine experience on the food loss of farm households.

The model also includes a series of individual-level control variables, denoted by ${\mathrm{X}}_{\mathrm{i}\mathrm{j}\mathrm{p}\mathrm{a}}$. These variables include arable land area, degree of fragmentation of arable land, harvest yield, farm household agricultural income, gender of household head, years of education of household head, labor affluence at harvest, and harvesting attitudes. The cohort fixed effect is represented by ${\gamma }_j$, which absorbs all changes across birth cohorts. The province fixed effect is denoted by ${\delta }_p$, and it controls all time-invariant province-level characteristics. Additionally, we control agrotype fixed effects (${\mu }_a$) to alleviate potential heterogeneity effects of crops. ${\epsilon }_{ijpa}$ is the error term.

The premise of obtaining an unbiased and consistent estimate of ${\beta }_1$ is that in the absence of famine, the harvest losses in each province should not be correlated with the excess mortality in that province. While it is not possible to directly observe counterfactual results, one can perform parallel trend tests using by-cohort differences-in-differences (DID) to test this hypothesis [51]. Thus, we set up the following model:

$$ln{\mathrm{F}\mathrm{H}\mathrm{L}}_{\mathrm{i}\mathrm{j}\mathrm{p}\mathrm{a}}={\beta }_0+\sum _{\tau =1940}^{1970}{\beta }_{1\tau }{\mathrm{E}\mathrm{M}}_{\mathrm{p}}\times \mathrm{T}\left(j=\tau \right)+{\beta }_2{\mathrm{X}}_{\mathrm{i}\mathrm{j}\mathrm{p}\mathrm{a}}+{\gamma }_j+{\delta }_p+{\mu }_a+{\epsilon }_{ijpa}$$

(2)

where $\mathrm{T}$ is a dummy variable indicating whether the head of household i was born in a certain cohort $\tau$ ($\tau$ = 1940, …, 1970). The 1971–1980 cohort serves as the baseline. All other settings of this model are identical to those of model (1).

3.3. Variables and Descriptive Statistics

Table 3. Variable definitions and descriptive statistics.

Variable	Definition	Mean	S.D.
FHL	Food harvest loss (kg)	51.256	82.262
AA	Arable land area per household (ha)	0.680	2.216
FI	Fragmentation of arable land (area of arable land/number of plots)	0.168	0.728
HY	Harvest yield (kg)	5638.455	84,305.555
Agrotype	Crop type (rice = 1, wheat = 2, corn = 3, soybeans = 4, canola = 5, peanuts = 6, sweet potato = 7, potato = 8, others = 9)	2.209	0.908
AI	Farming household agricultural income (yuan)	18,327.710	21,775.274
Age	Age of the head of the farming household	56.329	9.210
HS	Gender of the household head (male = 1, female = 0)	0.966	0.182
HE	Years of education of the household head	7.077	2.474
HP	labor availability during harvest (lack = 1, fair= 2, adequate = 3)	1.945	0.665
HA	operational attitudes during harvest (rough = 1, average = 2, fine = 3)	2.215	0.612

presents the variable definitions and descriptive statistics. The harvesting process includes food harvesting, field transportation, threshing and grain cleaning prior to warehousing. Food harvest loss (FHL) refers to the reduction in food quantity due to various human or natural causes during these stages. The surveyed households experienced an average harvest loss of approximately 51.256 kg, indicating significant potential for reduction. The average age of the head of the sample farming households is approximately 57 years old (birth in 1958), indicating that approximately half of the farming households were born before the Great Famine in China and suggesting an aging trend in the agricultural labor force in China and a relatively low quality of the workforce. Additionally, the arable land area per household is relatively small, with an average of 0.680 ha. The fragmentation of arable land is also high, with only 0.168 hectares per unit plot. Moreover, the total crop production amounts to 5638.455 kg, yielding an average of approximately 8728.259 kg/ha, which highlights a significant deficiency in production capacity. Despite recent advancements in mechanization and modernization, Chinese agriculture still primarily consists of small-scale, fragmented, and old farmers. The lack of uniform standards in the agricultural production process and significant variations in production behavior among farmers further contribute to losses in the harvesting process. This highlights that farmers may incur losses in the harvesting process not only due to objective conditions, such as eating by birds or rodents [22], mechanical damage [52], lodging [53], and abnormal weather [54,55], but also due to human factors, such as farmers’ education level [18,56], timely harvest [25,55,57], and operational meticulousness [18,57,58], which may be even more significant. Table 3 presents the means and variances of other control variables, such as farming household agricultural income, gender of the household head, years of education of the household head, labor availability during harvest, and operational attitudes during harvest.

Before conducting the empirical analysis, we classified farmers into a high mortality group and a low mortality group according to the excess mortality level in their area during the Great Famine and performed group-level statistics and T tests on food harvest loss, arable land area, and other control variables (

Table 4. Subgroup variable statistics.

	High Mortality	Low Mortality	Difference (H-L)
FHL	39.976 (1.559)	61.916 (2.223)	−21.941 *** (2.743)
AA	0.427 (0.012)	0. 918 (0.071)	−0.491 *** (0.074)
FI	0.106 (0.002)	0.226 (0.024)	−0.119 *** (0.025)
HY	2905.264 (96.789)	8221.388 (2754.262)	−5316.123 * (2834.825)
AI	14,840.387 (422.689)	21,623.317 (579.048)	−6782.931 *** (723.636)
HS	0.965 (0.004)	0.967 (0.004)	−0.002 (0.006)
HE	7.019 (0.060)	7.133 (0.058)	−0.114 (0.083)
HP	1.965 (0.015)	1.927 (0.016)	0.037 * (0.023)
HA	2.264 (0.014)	2.167 (0.015)	0.098 *** (0.021)
Observations	1719	1819

Note: * and *** represent significance at 10% and 1%, respectively. Standard errors clustered at the personal level are shown in parentheses.

). The average harvest loss was 39.976 kg and 61.916 kg in the high mortality group and low mortality group, respectively. Notably, the loss for farmers in the high mortality group was significantly lower than that for those in the low mortality group, and the T test rejected the original hypothesis that the data means of the two groups were equal.

4. Results

4.1. Standard Cohort DID

The estimation results for the overall sample are shown in Column (1) of

Table 5. The effect of EM on the food harvest losses of the standard cohort.

	(1)	(2)	(3)	(4)
Variables	All Samples	All Samples	High Mortality Group	Low Mortality Group
EM × T	−0.005	-	−0.030 ***	0.034
	(0.007)	-	(0.011)	(0.653)
HMG × T	-	−0.183 *	-	-
	-	(0.108)	-	-
AA	−0.010	−0.014	0.324 ***	−0.091
	(0.057)	(0.057)	(0.085)	(−1.035)
FI	0.109	0.119	0.549	0.332
	(0.174)	(0.174)	(0.480)	(1.244)
lnHY	0.688 ***	0.689 ***	0.616 ***	0.746 ***
	(0.031)	(0.031)	(0.047)	(16.590)
lnAI	0.092 ***	0.092 ***	0.050	0.113 ***
	(0.027)	(0.027)	(0.036)	(2.834)
HS	0.101	0.095	0.019	0.096
	(0.144)	(0.144)	(0.201)	(0.501)
HE	0.029 **	0.030 ***	0.015	0.049 ***
	(0.011)	(0.011)	(0.016)	(3.130)
HP	−0.089 **	−0.090 **	−0.043	−0.120 **
	(0.040)	(0.040)	(0.058)	(−2.137)
HA	−0.478 ***	−0.475 ***	−0.395 ***	−0.563 ***
	(0.046)	(0.046)	(0.065)	(−8.563)
Constant	−2.249 ***	−2.229 ***	−1.438 ***	−2.865 ***
	(0.316)	(0.313)	(0.465)	(−6.466)
Cohort Fixed Effects	Control	Control	Control	Control
Province Fixed Effects	Control	Control	Control	Control
Agrotype Fixed Effects	Control	Control	Control	Control
Observations	3538	3538	1719	1819
R2	0.401	0.402	0.384	0.448

Note: *, ** and *** represent significance at 10%, 5%, and 1%, respectively. Standard errors clustered at the personal level are shown in parentheses.

, where the coefficient of the interaction term between EM and the birth cohort dummy variable T is not statistically significant. The possible reason is that in areas with lower levels of famine, the impact of excess mortality on harvest losses is less pronounced even for those who experienced famine (i.e., those born before the famine). To address this issue, we re-estimated the model by replacing EM with the dummy variable HMG, which indicates whether a household belongs to a high mortality group (HMG = 1) or a low mortality group (HMG = 0). The results are presented in Column (2), which shows a significantly negative coefficient for the interaction term between HMG and T, indicating that famine experience can induce farmers to reduce harvest losses in high-mortality areas.

On this basis, we report the subgroup estimation results for the high mortality group (Column 3) and low mortality group (Column 4). The coefficient of the interaction term between EM and T in Column (3) is −0.030, which is statistically significant at the 1% level, indicating that a 1% increase in excess mortality in these areas will reduce the harvest loss by 3% for households with famine experience. However, in Column (4), although the estimated coefficient of EM × T is also negative, it is not statistically significant, suggesting that there is no significant effect of famine experience on harvest losses in areas with low excess mortality. In other words, even though farmers in these areas experienced famine in the past, the extent of the famine was not severe enough to create a lasting memory that would change their behavior.

4.2. By-Cohort DID

The estimation results of the by-cohort specification DID model are presented in Figure 1, where the 1971–1980 cohort serves as the baseline. Panel A of Figure 1 shows the estimation results of the interaction term coefficient ${\beta }_{1\tau }$ for each birth cohort with EM. ${\beta }_{1\tau }$ remains relatively close to 0 for farmers in birth cohorts after 1960, indicating that the excess mortality rate in their regions during the Great Famine is unrelated to harvest losses for farmers who did not experience the famine. This finding provides robust support for the parallel trend assumption.

4.3. Placebo Test

To ensure the accuracy of the estimates, we conducted two standard placebo tests on the sample of the high mortality group. First, we divided the famine-affected cohort (1940–1960) into a control cohort (1940–1945, T₁ = 1) and a placebo-treated cohort (1946–1960, T₁ = 0), and the tested results are presented in column (1) of

Table 6. Results of the placebo test.

	(1)	(2)
Variables	40–45 vs. 46–60	60–65 vs. 66–80
EM × T1	−0.017
	(0.021)
EM × T2		0.024
		(0.022)
AA	0.252	0.182
	(0.178)	(0.124)
FI	−0.118	1.460 **
	(0.723)	(0.690)
lnHY	0.667 ***	0.543 ***
	(0.071)	(0.074)
lnAI	0.025	0.170 ***
	(0.051)	(0.059)
HS	0.021	0.133
	(0.236)	(0.625)
HE	0.003	0.060 *
	(0.020)	(0.033)
HP	−0.062	0.025
	(0.077)	(0.115)
HA	−0.360 ***	−0.487 ***
	(0.084)	(0.122)
Constant	−1.525 **	−2.973 ***
	(0.635)	(0.917)
Cohort Fixed Effects	Control	Control
Province Fixed Effects	Control	Control
Agrotype Fixed Effects	Control	Control
Observations	996	481
R2	0.385	0.444

Note: *, ** and *** represent significance at 10%, 5%, and 1%, respectively. Standard errors clustered at the personal level are shown in parentheses.

. Next, we performed a similar test on the non-famine-affected cohort (1960–1980), dividing it into a control group (1960–1965, T₂ = 1) and a placebo-treated group (1966–1980, T₂ = 0), and the results are reported in column (2) of Table 6. The coefficients of the interaction terms for both tests were insignificant, and the estimates successfully passed the placebo test.

4.4. Robustness Test

Two robustness tests are performed in this paper. Test A replaces the core explanatory variables with the excess mortality rate EM₆₀ in 1960, which is calculated as the difference between the average mortality rate in the province in 1960 and the average mortality rate in 1956–1958. The results of the tests for the high and low mortality groups are presented in columns (1) and (2) of

Table 7. Results of the robustness tests.

	Test A. Change EM by 1960		Test B. Change Control Group with 1960–1970
Variables	(1) High Mortality Group	(2) Low Mortality Group	(3) High Mortality Group	(4) Low Mortality Group
EM60 × T	−0.011 **	−0.002
	(0.005)	(0.023)
EM × T3			−0.032 ***	0.043
			(0.012)	(0.055)
AA	0.328 ***	−0.090	0.372 ***	−0.086
	(0.085)	(0.087)	(0.098)	(0.100)
FI	0.520	0.333	0.456	0.323
	(0.480)	(0.266)	(0.513)	(0.302)
lnHY	0.615 ***	0.744 ***	0.633 ***	0.747 ***
	(0.047)	(0.045)	(0.050)	(0.048)
lnAI	0.047	0.113 ***	0.039	0.100 **
	(0.037)	(0.040)	(0.039)	(0.043)
HS	−0.001	0.105	0.007	0.119
	(0.200)	(0.189)	(0.212)	(0.206)
HE	0.015	0.048 ***	0.015	0.043 ***
	(0.016)	(0.015)	(0.017)	(0.016)
HP	−0.040	−0.123 **	−0.024	−0.176 ***
	(0.058)	(0.056)	(0.061)	(0.062)
HA	−0.393 ***	−0.562 ***	−0.388 ***	−0.525 ***
	(0.065)	(0.066)	(0.069)	(0.071)
Constant	−1.439 ***	−2.789 ***	−1.453 ***	−2.682 ***
	(0.473)	(0.435)	(0.504)	(0.458)
Cohort Fixed Effects	Control	Control	Control	Control
Province Fixed Effects	Control	Control	Control	Control
Agrotype Fixed Effects	Control	Control	Control	Control
Observations	1719	1819	1527	1612
R2	0.384	0.447	0.382	0.421

Note: ** and *** represent significance at 5% and 1%, respectively. Standard errors clustered at the personal level are shown in parentheses.

, respectively. Test B is conducted by replacing the control group, the 1960-1980 cohort, with the 1960–1970 cohort and estimating the high mortality group (column (3)) and low mortality group (column (4)) separately. The interaction term coefficients for the high mortality group are significantly negative in both Tests A and B, while the interaction term coefficients for the low mortality group are both insignificant. This indicates that in areas with severe famine, the higher the excess mortality rate is, the lower the harvest loss rate for farmers.

4.5. Heterogeneity Test

The impact of famine may vary across age cohorts, with farmers experiencing famine during adolescence more likely to be affected by future harvest losses. Therefore, we subdivided the age cohorts into four experimental groups of 40–45, 46–50, 51–55, and 56–60 (the 61–80 cohorts were set as the control group) and conducted regression analysis for the overall sample. The calculated results are shown in

Table 8. Impact of famine on different age cohorts: all samples.

	(1)	(2)	(3)	(4)
Variables	40–45 vs. 61–80	46–50 vs. 61–80	51–55 vs. 61–80	56–60 vs. 61–80
EM × T4	−0.010	-	-	-
	(0.014)	-	-	-
EM × T5	-	−0.019 *	-	-
	-	(0.010)	-	-
EM × T6	-	-	0.005	-
	-	-	(0.010)	-
EM × T7	-	-	-	−0.001
	-	-	-	(0.010)
AA	−0.003	−0.071	0.028	−0.066
	(0.071)	(0.073)	(0.051)	(0.071)
FI	0.089	0.294	−0.005	0.274
	(0.215)	(0.220)	(0.153)	(0.215)
lnHY	0.724 ***	0.706 ***	0.717 ***	0.680 ***
	(0.041)	(0.039)	(0.036)	(0.044)
lnAI	0.127 ***	0.110 ***	0.120 ***	0.093 ***
	(0.039)	(0.037)	(0.033)	(0.035)
HS	0.111	0.374 *	0.088	0.166
	(0.239)	(0.193)	(0.198)	(0.177)
HE	0.038 **	0.032 *	0.035 **	0.040 ***
	(0.018)	(0.017)	(0.015)	(0.015)
HP	−0.044	−0.045	−0.047	−0.062
	(0.055)	(0.053)	(0.051)	(0.051)
HA	−0.527 ***	−0.525 ***	−0.485 ***	−0.526 ***
	(0.064)	(0.062)	(0.058)	(0.060)
Constant	−3.032 ***	−2.864 ***	−2.967 ***	−2.408 ***
	(0.498)	(0.465)	(0.416)	(0.425)
Cohort Fixed Effects	Control	Control	Control	Control
Province Fixed Effects	Control	Control	Control	Control
Agrotype Fixed Effects	Control	Control	Control	Control
Observations	1737	1994	2225	2084
R2	0.443	0.436	0.421	0.409

Note: *, ** and *** represent significance at 10%, 5%, and 1%, respectively. Standard errors clustered at the personal level are shown in parentheses.

. Among the four cohorts, only the coefficient of the 46–50 cohort (10–15 years old at the time of famine) exhibited statistical significance. This indicates that farmers who experienced famine during adolescence contribute to the formation of enduring memories of famine, even in situations where the severity of the famine is not extremely pronounced.

On this basis, we employed this approach to examine the heterogeneity within the sample of farming households in high mortality areas and compared the results with the standard cohort DID findings. The results, as illustrated in

Table 9. Impact of famine on different age cohorts: high mortality samples.

	(1)	(2)	(3)	(4)
Variables	40–45 vs. 61–80	46–50 vs. 61–80	51–55 vs. 61–80	56–60 vs. 61–80
EM × T4	−0.048 **	-	-	-
	(0.022)	-	-	-
EM × T5	-	−0.047 ***	-	-
	-	(0.015)	-	-
EM × T6	-	-	−0.021 *	-
	-	-	(0.012)	-
EM × T7	-	-	-	−0.014
	-	-	-	(0.014)
AA	0.337 ***	0.320 ***	0.339 ***	0.301 ***
	(0.098)	(0.098)	(0.091)	(0.095)
FI	0.826	1.107 *	0.911 *	1.052 *
	(0.594)	(0.599)	(0.545)	(0.601)
lnHY	0.587 ***	0.573 ***	0.645 ***	0.500 ***
	(0.053)	(0.052)	(0.049)	(0.062)
lnAI	0.079	0.072	0.076 *	0.066
	(0.048)	(0.048)	(0.043)	(0.045)
HS	0.049	0.197	−0.140	0.175
	(0.338)	(0.323)	(0.262)	(0.249)
HE	0.019	0.011	0.028	0.028
	(0.024)	(0.024)	(0.022)	(0.021)
HP	−0.049	−0.017	0.006	−0.097
	(0.078)	(0.077)	(0.076)	(0.077)
HA	−0.418 ***	−0.432 ***	−0.348 ***	−0.490 ***
	(0.092)	(0.089)	(0.083)	(0.085)
Constant	−1.715 ***	−1.635 ***	−2.239 ***	−0.982 *
	(0.635)	(0.633)	(0.561)	(0.570)
Cohort Fixed Effects	Control	Control	Control	Control
Province Fixed Effects	Control	Control	Control	Control
Agrotype Fixed Effects	Control	Control	Control	Control
Observations	862	951	1093	981
R2	0.415	0.401	0.411	0.369

Note: *, ** and *** represent significance at 10%, 5%, and 1%, respectively. Standard errors clustered at the personal level are shown in parentheses.

, have revealed two intriguing findings. First, the notably expanded range of affected cohorts (from the 46–50 cohorts to the 40–55 cohorts) implies that the increase in famine severity drives more farmers to reduce their harvest losses, providing additional evidence in support of the hypothesis of this paper. Second, among the four cohorts, the regression coefficients for the 40–45 cohort (EM × T₄) and the 46–50 cohort (EM × T₅) are both statistically significant at the 1% level, exhibiting remarkably close values of −0.048 and −0.047, respectively. These coefficients are significantly higher than the regression coefficient (−0.021) for the 51–55 cohort (EM × T₆), while the regression coefficient for the 56–60 cohort (EM × T₇) decreases to −0.014 and does not demonstrate statistical significance. This further supports the notion that farmers who experienced famine during their adolescence exhibited a lower rate of future harvest loss. Farmers in the 0–5 age group at the time of famine could not form a long-lasting memory even after experiencing severe famine, and thus had no significant effect on future harvest loss.

5. Discussion

We used data from the 2016 Postproduction Food Loss and Waste Survey to investigate the long-term effects of famine experiences on farmers’ harvest losses. Our results indicate that famine experience reduced harvest losses of farm households, with the greatest reduction in household harvest loss rate observed among those who experienced famine during adolescence. Furthermore, in areas where severe famine occurred, farmers had a more vivid memory of the famine, leading to a greater reduction in harvest losses and a wider range of age groups affected.

Although previous studies provide evidence suggesting that older individuals tend to experience lower levels of food loss than younger individuals [27,28], these findings are typically based on cross-sectional observations rather than long-term follow-up monitoring. Consequently, they fail to account for the historical context in which these two age groups have lived. The underlying reason for the lower food loss among the elderly may lie in their prolonged exposure to poverty, hunger, and particularly extreme famine-like events, which have contributed to the formation of enduring habits in their lifestyles and work practices. This notion has been increasingly supported by existing studies [13,14,15,16], where farmers with famine experiences involuntarily develop a frugal mindset and a more conservative personality [17,59], which in turn prevents the recurrence of such events. Our research finds that in regions severely affected by famine, there is a significant correlation indicating that a 1% increase in excess mortality results in a 3% decrease in the rate of harvest loss. This observation suggests that farmers who have experienced extreme famine retain a profound and lasting memory of the event, leading to sustained reductions in harvest losses over the long term.

Notably, not all extreme events induce the same behavioral response in individuals, and different shocks may have varying effects on individuals. For instance, famine may induce farmers to be frugal [31], tsunamis may lead local households to change their settlement location [60], and train derailment may cause passengers to refuse the same mode of transportation in the future [61]. However, the commonality lies in the fact that affected individuals respond to the shock to avoid re-exposure to similar shocks, even though the strategies employed may differ. In contrast, after suffering a famine, farmers may reduce their harvest losses by adopting more meticulous production practices, selecting harvesting methods with lower loss rates, and collecting spilled grain after harvest, all in an effort to prevent the recurrence of starvation. Unlike short-term behaviors such as retaliatory consumption [62], these behaviors aimed at reducing harvest losses not only persist in the short term but are also driven by conservative dispositions, making farmers more reluctant to change their original production strategies in the long term [17,59]. Therefore, even though more than half a century had passed since the end of the Great Famine when the survey was conducted in 2016, we can still capture the impact of famine on farmers’ harvest losses.

Differences in the age at which farmers experience famine may lead to diverse responses to their famine experience. Farmers in adolescence are more susceptible to external influences [63,64], and this sensitive information perception leads them to subjectively magnify the negative impacts of certain events, thereby prompting corresponding behavioral feedback. Even if the famine was not severe in their area, it still made them more conscious of saving food in future agricultural production and affected their harvest losses. This finding is supported by the studies of Andersen and Teicher [65] and Cui et al. [66], who concluded that extreme shocks, such as hunger in adolescence, are especially harmful to mental health. Moreover, when famine is extremely severe, it will not only prompt frugal habits by affecting farmers’ risk perception [67] and personal experience [68] but also lead to post-traumatic stress disorder (PTSD) [69] and depression [10] in some farmers, making them intolerant of food losses.

Due to the constraints of data availability, this study primarily examines the specific impacts of famine experiences on farmers’ harvest losses. The exploration of underlying mechanisms is mainly based on theoretical analysis and lacks corresponding data validation. In addition, since the survey was conducted long after the famine, farmers aged 20 and above during the famine were over 75 years old in 2016. It is likely that their productivity had declined significantly, or they had withdrawn from agricultural production. As a result, the sample of farmers who had reached adulthood during the famine is relatively small. If we can overcome the data limitations, there is a possibility of obtaining more comprehensive and accurate research findings.

As the global economy continues to expand, there is a concerning rise in regional income disparity, accompanied by simultaneous challenges of excessive waste in developed countries [20] and famine in less-developed regions [2]. Our research findings indicate that the lack of first-hand experience with hunger may contribute to higher levels of food loss and waste among individuals in developed countries. Consequently, this implies that international organizations such as the Food and Agriculture Organization (FAO), International Food Policy Research Institute (IFPRI), and World Food Programme (WFP), in their efforts to promote food loss and waste reduction, have the potential to encourage individuals to minimize waste in their daily lives through innovative approaches, such as organizing hunger experience events.

6. Conclusions

Based on data from the 2016 Post-production Food Loss and Waste Survey, this paper investigates the impact of the Great Famine on Chinese farmers’ harvest losses using a cohort DID model. The findings demonstrate that the famine experience of the head of a farming household reduces harvest loss.

First, on average, Chinese farming households experience a harvest loss of approximately 51.256 kg, resulting in a total loss of approximately 10.632 million tonnes, which is a serious grain loss situation. Second, the harvest losses in the high mortality group and low mortality group are 39.976 kg and 61.916 kg, respectively, with farmers in the high mortality group experiencing significantly lower losses than those in the low mortality group. Third, the standard cohort DID results reveal that in areas with severe famine, the higher the excess mortality rate is, the lower the harvest loss rate of farming households. Furthermore, each 1% increase in the excess mortality rate results in a 3% reduction in the harvest loss rate. Finally, the results of the heterogeneity test indicate that the reduction in harvest loss rate is most pronounced for farmers experiencing severe famine during adolescence. Additionally, when excluding the low mortality group, the number of affected cohorts increases from five (1946–1950 cohort) to sixteen (1940–1955 cohort), suggesting that farmers experiencing extreme severity of famine developed a deeper memory of the event.

This study elucidates the influence of farmers’ early experiences on their behavior, which is hidden under age characteristics, and analyzes the reasons for the differences in harvest losses from the perspective of farmers’ famine experiences, expanding the boundaries of existing research. The findings confirm that famine experience has a significant negative effect on harvest losses. However, the mechanism of this effect remains unclear. Future studies should investigate whether famine experience can affect harvest losses by changing farmers’ awareness of grain savings, harvesting methods, and harvesting time. Additionally, it would be valuable to examine whether the famine experience had any effect on losses in other postproduction stages of the crop.