Area and data

Area

This study uses longitudinal micro-data from a rural area in north-eastern Spain. Located around 6–40 kilometres away from Zaragoza, the regional capital (see Fig. 1), this area is formed by a combination of plains and foothills near the Huerva river and comprises 13 small municipalities.Footnote ⁵ Most of the population lived in nuclear households and was essentially engaged in agriculture (mostly wheat and some wine) and sheep grazing. Throughout the study period, agricultural labourers made up approximately 80% of individuals with declared occupations in the census. About 10% were low-skilled artisans, while a little over 5% held upper-class occupations, primarily teachers, doctors, and veterinarians.

Figure 1.

Area of study: Middle Huerva (Aragón, Spain).

Note: Dark dots refer to the localities studied here (except Zaragoza, the provincial capital) and the corresponding shaded areas to their municipal boundaries. Apart from rivers (in grey) and main roads (dotted lines), the map also depicts neighbouring villages (white dots).

The predominance of nuclear families did not imply a detachment from extended family ties, as over 50% of children who survived to adulthood remained in their birthplace or nearby localities, allowing them to maintain familial and friendship support networks (Reher, Reference Reher, Dalla Zuanna and Micheli2004). This represents a classic demographic model of nuclear family structures within an agrarian setting, which would only come under strain with the demographic transition and subsequent socio-economic changes (Devolder, Reference Devolder1999). Inheritance systems were characterised by relatively equal distribution among sons and daughters, with marriage and the death of the first parent serving as key moments in the transmission of assets (Jarque and Salas Auséns, Reference Jarque Martínez and Salas Ausens2007). Households were officially headed by the husband/father, who maintained greater control over family property, except for assets received by the wife at the time of marriage.

In terms of land access, the study area was characterised by small plots that were divided among heirs, resulting in a predominant pattern of families owning some land yet unable to sustain themselves solely through that ownership. Approximately 15% of household heads were farmers able to subsist exclusively on their land, though none were large landowners. Large estate owners were rare but included, during the study period, the marquises who had retained manorial rights over these areas (and thus kept property ownership) and a few large proprietors from neighbouring localities who purchased land in the area to expand their holdings. It was only in the first half of the twentieth century (after the period studied) that the sale of extensive properties by heirs of the nobility enabled substantial groups of local owners to acquire sufficient land for self-sustenance (Marco-Gracia and González-Esteban, Reference Marco-Gracia and González-Esteban2024).

In relation to literacy acquisition, it was significantly influenced by parental education and the broader socio-economic context. Studies on the region indicate that a child’s likelihood of reporting literacy was closely tied to whether their parents were literate, reflecting a strong intergenerational correlation, especially during the 19th century when illiteracy was widespread (Luque de Haro and Marco-Gracia, Reference Luque de Haro and Marco-Gracia2024). Access to education was often constrained by the opportunity cost of children’s labour in agriculture and the limited value placed on basic education by many parents, who were themselves often illiterate. Although the Moyano Act of 1857 established compulsory schooling, aiming to provide free education to those unable to afford it, its impact was limited. Compliance with compulsory schooling laws was low, as local councils, responsible for funding schools, faced financial challenges and varied attitudes towards education. These constraints, coupled with broader social, cultural, and political attitudes towards literacy, and the region’s low level of development, restricted both the supply and demand for education (Martínez Galarraga and Beltrán Tapia, Reference Martinez-Galarraga and Beltrán Tapia2018). As a result, literacy rates improved only gradually in this rural area, where economic necessity frequently prioritised labour over schooling.

Demographically, prior to the onset of the demographic transition, most of the population in these villages practised natural fertility, meaning that married couples had children without actively using fertility control methods. This practice led to high fertility rates with no restrictions on the timing of childbirth within marriage. Consequently, families had large numbers of children, with births continuing until the mother reached menopause (Marco-Gracia, Reference Marco-Gracia2018). On average, Aragonese families had around seven children until the early twentieth century. However, this high number of births, combined with a greater survival rate of children, eventually contributed to population growth, thereby increasing demographic pressure over time. Infant and child mortality rates remained high, with only about half of infants surviving beyond age five. Similar to broader national trends, mortality rates began to decline toward the end of the nineteenth century, due to a decrease in epidemic crises and a reduction in water- and food-borne infectious diseases.

The demographic transition began during our study period in the study area. The peak of mortality occurred in the 1860s, and since then, childhood mortality began a slight decline, which continued to increase over the decades. Fertility control began in the early twentieth century (with women born in the 1870s and 1880s) and intensified particularly in the 1930s and 1940s, with the impact of the Civil War. The fertility transition was based on stopping, with very little use of spacing throughout the study period (Marco-Gracia, Reference Marco-Gracia2018). Regarding the evolution of gender preferences in the study area, previous research has shown that a preference for sons persisted into the first half of the twentieth century (Beltrán Tapia and Marco-Gracia, Reference Beltrán Tapia and Marco-Gracia2022; Marco-Gracia and Beltrán Tapia, Reference Marco-Gracia and Beltrán Tapia2022). However, it is possible that during the early decades of the twentieth century, these preferences shifted towards a preference for mixed-gender offspring, while still maintaining a greater interest in sons than in daughters (Marco-Gracia, Reference Marco-Gracia2021b). The inheritance systems, access to land, and the predominance of men as heads of household were not significantly altered during the study period.

In terms of living standards, the average height of Aragonese men during the nineteenth century was approximately 160 centimetres, according to historical measurements. These measurements indicate that biological well-being was low, as the height was considerably below the European average and even lower than that of other regions of Spain (Hatton and Bray, Reference Hatton and Bray2010). Even by 1924 (after the period studied here), the average household income was insufficient to support a nuclear family of two adults and two children with an adequate diet, even in the nearby city of Zaragoza, where wages were slightly higher than in rural areas (Marco-Gracia and Delgado, Reference Marco-Gracia and Delgado2024). This, along with additional qualitative evidence, such as records from archbishops’ delegates during pastoral visits, leads us to categorize the study area as a poor region, where most of the population lived close to subsistence levels. Under these conditions, economic shocks, like those caused by price crises, often jeopardise the survival and well-being of entire families.

Spanish women suffered a subordinate position that was not only visible in legal terms but also in the public and the private sphere (Borderías et al., Reference Borderías, Pérez-Fuentes, Sarasúa, Addabbo, Arrizabalaga and Owens2010; Borderías and Muñoz, Reference Borderías and Muñoz2018). As well as cultural and religious values, the status of women was linked to the importance of male labour in agricultural contexts (Alesina et al., Reference Alesina, Giuliano and Nunn2013; Giuliano, Reference Giuliano, Averett, Argys and Hoffman2018). As well as being less able to contribute monetary income to the household (Camps, Reference Camps1998; Germán, Reference Germán Zubero and Forcadell Alvárez2009), girls had to provide a dowry for their marriage and inheritance practices tended to favour male heirs. Economic considerations alone could therefore justify potential son preference. The gender literacy gap in the area of study was indeed remarkable, thus evidencing that parents invested differently in boys’ and girls’ education.Footnote ⁶ Recent research goes further and argues that discriminatory practices, including neglect, affected female mortality rates right after birth and during infancy and childhood (Marco-Gracia and Beltrán Tapia, Reference Marco-Gracia and Beltrán Tapia2021; Beltrán Tapia and Marco-Gracia, Reference Beltrán Tapia and Marco-Gracia2022).

In an area where poverty was widespread and son preference prevailed, the way families allocated their scarce resources mattered and discriminatory practices could have had lethal consequences. Periods of economic stress increased the need for rationing scarce resources, so studying these events can further test this hypothesis. In order to explore whether short-term economic difficulties affected boys and girls differently, this article links information on prices with the complete church registers of these villages. The latter provides high-quality information on all births, marriages, and deaths, thus enabling the reconstitution of their complete life histories.Footnote ⁷ In total, this article employs information on almost 34,000 individuals born between 1800 and 1910, including name, sex, place and date of birth, parents’ names and date of death, among others.

Data

This longitudinal dataset has also been complemented with the information on occupation and literacy contained in population lists (1747–1830), population censuses (1857, 1860), and electoral rolls (1890–1955). The population censuses contain data on occupation and literacy for all individuals in the study area, while the population lists include only the occupation of individuals over the age of 7, and the electoral rolls contain occupation and literacy data for all adults (with women appearing only after gaining the right to vote in 1931). The information was linked through an automated record linkage process using matching on the first name, both surnames (following the Spanish system), and year of birth (calculated from age) using Stata 14, with a matching threshold of 90%. When the match was not exact, the record linkage was manually reviewed on an individual basis to ensure the quality of the matching. Data on family characteristics (such as the number of children/siblings of each sex) were extracted from parish records once the family had been reconstituted (and therefore not from the census data). In cases where an individual changed profession throughout their life (or was recorded with different occupations in various censuses), the occupation closest to adulthood (typically around 40 years of age) was used as the reference.Footnote ⁸

We should nonetheless bear in mind that, since we rely on local records, we do not have all the information on those individuals who migrated to or out of our study area.Footnote ⁹ For the study period, we have no additional demographic events (such as marriage or death) recorded for 27% of children initially registered at baptism. Nearly 10% of these children came from families that did not record any further demographic events in the study area (suggesting that they likely left the area). Additionally, from occasional annotations (which were not mandatory) in the baptismal records – providing information on sacraments like confirmation, marriage, or death – we know that at least 15% of this 27% survived into youth, while only 2% were noted as having died before age 10 in another parish (although these data should be treated cautiously, as they were neither mandatory nor systematic). In any case, based on available information, we have several indications and safeguards suggesting that we are generally not introducing a bias by inaccurately capturing childhood mortality. However, small selection biases may be introduced that could influence the intensity of the coefficients, though they should not affect the sex distribution. In addition, given that we focus on children (young ages), this issue is somewhat limited.

This information is complemented with data on wheat prices from the city of Zaragoza (Peiró, Reference Peiró1987). Leaving aside self-consumption, the surplus of the wheat harvest from our area of study was sold, directly or through intermediaries, in Zaragoza, the regional capital. In order to ensure the quality of the demographic information, we focus on the period from 1800 to 1910. Likewise, instead of calendar years, we contemplate annual harvest years. We, therefore, consider prices from the summer harvest to the beginning of the following year’s harvest (from July to June) and associate each birth and death with the corresponding harvest season. For instance, a child born in August 1862 would be considered to have been born in the harvest year spanning from July 1862 to June 1863.Footnote ¹⁰ Figure 2, panel A, plots the raw data used here: the major economic crises took placed during the early nineteenth century (around 1802 and 1810) and price variations became more moderate over time. The evolution of wheat prices in Zaragoza is similar to that in the whole of inland Spain (Gallego Martínez, Reference Gallego Martínez2004).

Figure 2.

Annual evolution of the price of wheat in the city of Zaragoza.

Note: In Panel A, the unit of measure is pesetas per hectolitre. In the detrended price in Panel B, prices are measured in logs (average price = 0). High-price years, denoted by dashed lines, are those when the annual price is over 10% compared with average price.

Source: Peiró (Reference Peiró1987).

In this context, high wheat prices were generally a result of reduced production, leading to food scarcity rather than increased income for local producers. Throughout the 19th century, there was a gradual process of market integration, beginning before the construction of the railway linking Zaragoza with Barcelona in 1862 and intensifying afterwards (Pinilla Navarro, Reference Pinilla Navarro1995). Even so, studies confirm that years of high prices corresponded with declines in birth rates and increased mortality, indicating severe economic and nutritional stress for the community (Marco-Gracia, Reference Marco-Gracia2021a).

Following the pioneering studies that analyse the link between short-term economic stress (Bengtsson et al., Reference Bengtsson, Campbell and Lee2004; Bengtsson and Dribe, Reference Bengtsson and Dribe2006),Footnote ¹¹ the data have been detrended using a Hodrick-Prescott (Reference Hodrick and Prescott1997) filter and harsh years are subsequently identified as those showing an increase equal to or greater than 10 per cent from the average price.Footnote ¹² This figure serves to delimit the threshold at which the effects of an economic stress year were visible and affected the majority of the population. In this regard, while a 10-per cent variation in prices is serious enough to affect household consumption, it also allows prices to reach this threshold several times during the nineteenth century and therefore provides enough variation to identify its effects on demographic processes. In particular, following this method, we have identified 23 wheat-price crises between 1800 and 1910 (out of 110 years; see Fig. 2, Panel B).Footnote ¹³ In our area of study, these kinds of price variations disappeared at the end of the nineteenth century, so it is not necessary to extend the period of study. Once we have identified the crisis years, we can assess how economic crises affected demographic behaviour. While “Methodology” focuses on births and potential neglect right after birth, “Gender discrimination around birth in times of short-term economic stress” explores whether high-price years affected sex-specific mortality during infancy and childhood.

Methodology

The estimations included in the article are mainly based on the logit method and the Cox Proportional-Hazards (onwards COX PH) survival analysis model. We have used logit models to analyse events for which we have no observation period (e.g., probability of being baptised as a man or woman) or for which the observation period is very short (e.g., deaths during the first day of life, first week or first month). Logit models are a probability-based methodology that is useful when dealing with a binary dependent variable (such as being male or female) and various independent variables, both linear and categorical (Cramer, Reference Cramer2003; Hess and Train, Reference Hess and Train2017). The logit models can be expressed as follows:

$${P_i}\;\left( {male = 1} \right) = {1 \over {1 + {e^{ - \left( {\alpha + {\beta _1}*{X_{1i}} + {\beta _2}*{X_{2i}} + \ldots } \right)}}}} + \varepsilon $$

where P (left-wing=1) is the dependent variable (being male or female), for an individual i, βn, is the parameter, Xn denotes the independent variables used in each model, and ε is the error term. In addition, in order to analyse the joint significance of the independent variables, we have performed a likelihood ratio test and found that the test statistics follow a χ² distribution, which supports the explanatory power of the model.

In this case of analysing a long observation period, we have used COX PH regressions when studying the probability of death between 1 month and 10 years. The Cox regression method offers significant advantages in terms of flexibility, the ability to handle censored data and less restrictive assumptions compared to other analysis methods. Over the last few decades, the COX PH model has become widely used in studying disparities in mortality (Bengtsson and Lindström, Reference Bengtsson and Lindström2000; Edvinsson and Broström, Reference Edvinsson and Broström2012; Edvinsson and Lindkvist, Reference Edvinsson and Lindkvist2011; Schenk and van Poppel, Reference Schenk and van Poppel2011).

To prepare the underlying data for the COX PH models, we have applied the counting process format, enabling the use of multiple lines of data for a single individual. This approach subdivides the total at-risk follow-up time into smaller intervals. This format is recommended when using age-at-follow-up time as the outcome variable rather than time of follow-up. It also enables the inclusion of time-dependent variables, such as years of high prices, both at t and t−1 (Kleinbaum and Klein, Reference Kleinbaum and Klein2012). One of the reasons for this selection is that using a model based on age at the time of the event rather than one based on the observation time of the individuals may be preferable if age is a stronger determinant of the event of interest (in our case death) than the observation time. In any case, the control variables include the date of birth to consider the impact of economic improvements, advances in health and medical knowledge, as well as possible cultural changes on the probabilities of survival.

The risk function of the COX PH model for an individual with the characteristics of vector X can be formulated as follows:

$$h\left( {a,{\rm{\;}}X} \right) = {h_0}\left( {a{\rm{|}}{a_0}} \right)\exp [\sum {\beta _i}{X_i}]$$

where $h\left( {a,{\rm{\;}}X} \right)$ is the hazard of death; ${h_0}\left( {a{\rm{|}}{a_0}} \right)$ is the baseline hazard – that is, the hazard function for an individual having on all covariates the value 0.

$$i = [\rm Sex, year, village\ of\ residence, \ldots n], with\ n\ being\ the\ number\ of\ variables$$

${X_i} = \left[ {\matrix{ {{X_{Sex}}} \cr {{X_{year}}} \cr {\matrix{ \vdots \cr {{X_n}} \cr } } \cr } } \right]$ , where ${X_i}$ corresponds to the value of variable i.

${\beta _i}$ = [ ${{\beta _{Sex}}{\beta _{year}}\matrix{ { \ldots {\beta _n}}}}$ ], where ${\beta _i}$ corresponds to the value ‘beta’ of variable i.

The main output of this regression is the relative risk exp(β): If the variable is categorical, the relative risk reflects the ratio between two risk functions; one associated with the value of the variable and the other with the reference category. In case of a continuous variable, the relative risk indicates the impact on the outcome of increasing its value by one unit while keeping all other variables unchanged (Kleinbaum and Klein, Reference Kleinbaum and Klein2012).

Gender discrimination around birth in times of short-term economic stress

Panel A in Fig. 3 contrasts the evolution of baptisms with the occurrence of high-price years in our dataset. On average, harsh years experienced around 228 registered baptisms, 12 fewer than in the other years of the sample. Interestingly, the average sex ratios increased during years subject to high prices: while sex ratios during normal harvest years were around 108.3 males per hundred females, this figure increased to 112.2 in bad years (Fig. 3, Panel B).

Figure 3.

Economic crises, number of baptisms, and sex ratios, 1800–1910.

Note: The dashed vertical lines indicate the years experiencing high prices.

Observing more males being baptised during harsh years is somewhat surprising because, as mentioned in the introduction, males are biologically more vulnerable both in utero and right after birth (Waldron, Reference Waldron and Waldron1998; Drevenstedt et al., Reference Drevenstedt, Crimmins and Vasunilashorn2008; Di Renzo et al., Reference Di Renzo, Rosati, Donati Sarti, Cruciani and Cutuli2007; Dipietro and Voegtline, Reference Dipietro and Voegtline2017; Zarulli et al., Reference Zarulli, Barthold Jones, Oksuzyan, Lindahl-Jacobsen, Christensen and Vaupel2018). This feature was not only especially visible in the high-mortality environments that characterised pre-industrial Europe (Woods, Reference Woods and Woods2009; Beltrán Tapia, Reference Beltrán Tapia2019)Footnote ¹⁴ but should have been even more pronounced in the harsher circumstances characterising those years affected by economic crises. The female biological advantage around birth, therefore, implies that, if anything, harsher circumstances will result in (1) more male foetuses dying before birth (thus decreasing the probability of being born male) and (2) more males dying shortly after birth. This is particularly relevant for our study because, instead of registered births, we use baptismal records. Some infants could have, therefore, died before being baptised and this issue would especially affect boys due to their higher vulnerability.

Apart from less dramatic forms of fertility control, this evidence, therefore, suggests that economic crises fostered the neglect of female babies, thus accentuating discriminatory practices that were also visible in normal harvest years (Beltrán Tapia and Marco-Gracia, Reference Beltrán Tapia and Marco-Gracia2022). These patterns could, of course, have arisen from other causes, including the possibility of random variation, so they should be subject to a more sophisticated statistical analysis. In order to further explore whether economic crises resulted in female neglect around birth, we have analysed the individual-level information associated with all births occurring in our area of study between 1800 and 1910. Table 1 reports the results of estimating a logit model assessing whether the probability of being baptised male was different during bad years while controlling for individual and household characteristics. While column (1) shows the baseline specification that accounts for birth order and the number of children alive at the moment of birth (as well as a time trend, its square and village fixed effects),Footnote ¹⁵ column (2) adds a set of additional control variables: mother’s age, father’s occupation, and literacy. In addition, given that it is unclear whether our variable of interest would be affected by an economic crisis occurring during the same year (t) or the previous year (t−1), columns (3) and (4) replicate the analysis but focus on the latter. Finally, columns (5) and (6) consider these two variables simultaneously.

Table 1.

Economic crises and the probability of being baptised male

Robust standard errors in parentheses (clustered at the household level); ***p < 0.01, **p < 0.05, *p < 0.1. The basic controls include the number of siblings at birth, parity FE, village FE, year and year squared. The additional set of controls includes mother’s age, father’s occupation, and literacy.

The results clearly indicate that more boys were baptised during high-price years and the following year, even though we would expect the opposite results due to the higher male vulnerability. Although the coefficient is only statistically significant at the 90 per cent level, its impact is considerable: the probability of being baptised male increased from 51.4 per cent in normal harvest years to 52.7 per cent in high-price years (equivalent to going from a sex ratio of 105.8 to 111.4). The effect of the lagged variable is quite similar, and both are visible when considered simultaneously,Footnote ¹⁶ so the compound impact of an economic crisis on sex ratios at baptism was extremely high.

It can be argued that female under-registration of baptisms may affect the previous analysis.Footnote ¹⁷ Examining death registers therefore provides further evidence that baby girls were neglected because if under-registration systematically targeted girls, it would have affected both births and deaths around birth. The latter is actually more plausible because, unlike baptisms, registering deaths implies paying a fee. The bias would however now work in the opposite direction, and we would be observing fewer female deaths, thus making it more difficult to identify neglect against infant girls. Therefore, we will now explore whether high-price years affected the probability of dying during the first days of life differently for boys and girls.

Table 2 reports the results of estimating a logit model assessing whether high-price years (both in t and t−1) affected the probability of dying during the first days of life. We have distinguished between those deaths described in the sources as “nonatos” (stillbirths), those happening during the first day of life, those during the first week, and those occurring during the rest of the first month. All these specifications control for the same dimensions as before: parity, the number of previous children alive, year, year squared and village fixed effects; (odd columns), plus mother’s age and father’s occupation and literacy (even columns). In order to explore whether crises affected males and females differently, we have included a dummy variable identifying females as well as the interaction between a high-price year and being female. The latter variable therefore captures the distinct effect of bad years on female mortality right after birth.

Table 2.

Dep. variable: Dying during the period considered (1/0)

Robust standard errors in parentheses (clustered at the household level); ***p < 0.01, **p < 0.05, *p < 0.1. The basic controls include the number of siblings at birth, parity FE, village FE, year and year squared. The additional set of controls includes mother’s age, father’s occupation, and literacy.

Again, we should stress that the female biological advantage implies that more male infants were dying during this crucial period. This is apparent in the coefficient on the variable female, which clearly shows that the survival chances were significantly higher for baby girls.Footnote ¹⁸ The male vulnerability would be more accentuated in bad years, so we would expect that the coefficient capturing the interaction between high-price years and being female to be negative. Our results actually indicate the opposite: compared to males, females suffered higher mortality rates right after birth during the economic crises defined here. Although the effect is not statistically different from 0, we should bear in mind that we should be expecting a negative coefficient, so the benchmark for comparison is not 0 but a negative value. Discriminatory practices may have, therefore, offset the natural male vulnerability. These results hold regardless of whether we are exploring the contemporary effect (t) or the lagged effect (t−1). These findings support the previous evidence with respect to how high-price years increased the probability of being baptised male. In fact, these two sets of findings should be interpreted jointly: if female under-registration is an issue, our results either overestimate or underestimate female neglect depending on whether we are using baptismal or death registers. Taken together, and bearing in mind the female biological advantage, these results suggest that, in periods of economic hardship, behavioural decisions were acting against the survival of females right after birth.

Discrimination during infancy and childhood in times of short-term economic stress

Apart from parental behaviour right after birth, short-term economic stress may have also fostered discriminatory practices in older infants and children and therefore their sex-specific mortality rates. In order to explore this possibility, Table 3 analyses whether high-price years (both in t and t−1) affected the probability of dying during the first months and years of life. As explained previously, we use COX PH models for analysing long observation periods (however, in Table A1 in the Appendix, we have included logit regression models for individuals older than six months, similar to those presented in Table 2, confirming that both methodologies – logit models and COX PH regressions – yield similar and consistent results aligned with our initial hypothesis).

Table 3.

Cox proportional hazard models of mortality for children aged 1 month to 10 years

***p < 0.01, **p < 0.05, *p < 0.1; The basic controls include the number of siblings alive at birth, parity FE, village FE, year and year squared. The additional set of controls includes mother’s age, father’s occupation, and literacy. Note: These models have been replicated employing logit methodology as a robustness check (see Table A1 in the Appendix).

Table 3 presents the results of the survival model estimation to evaluate whether years of high prices influenced the probability of death in age groups separated into two panels: Panel A: 1–6 months, 6–12 months, and 1–2 years; and Panel B: 2–5 years and 5–10 years and 6 months to 10 years. All these exercises are repeated twice depending on whether they are accounting for the basic or for the extended set of controls described above (odd columns and even columns, respectively). In order to explore whether crises affected males and females in rural Spain differently, we have included a dummy variable identifying females as well as the interaction between a high-price year and being female. The latter variable, therefore, captures the distinct effect of bad years on female mortality risk. Again, we should bear in mind that the female biological advantage implies that boys would have suffered more under adverse circumstances, so not observing mortality differences between boys and girls would constitute indirect evidence of gender-discriminatory practices.

The exponent of the coefficient on the dummy variable female confirms the well-known female advantage, especially during the first six months of life. This advantage was reduced as children grew older and eventually reversed into a female penalty, suggesting that families prioritised their sons even in normal harvest years (Marco-Gracia and Beltrán Tapia, Reference Marco-Gracia and Beltrán Tapia2022). The exponent of the coefficient that captures the interaction between being female and the variable representing high-price years shows values greater than 1 in all the age groups analysed, although with a reduced degree of statistical significance. However, if, instead of looking at smaller age groups, we analyse the effect between six months and 10 years (columns 11–12), we can see how the value of this coefficient becomes statistically significant. These results are at odds with the male frailty already discussed, thus suggesting again that the distribution of resources and family care clearly affected the sex-specific survival chances and offset the male vulnerability. This effect is particularly telling because the introduction of solid food typically commenced at six months of age in the area of study (see Marco-Gracia and Beltrán Tapia, Reference Marco-Gracia and Beltrán Tapia2021) and was linked to an increased probability of acquiring gastrointestinal ailments owing to prevailing sanitary conditions (Guinnane and Ogilvie, Reference Guinnane and Ogilvie2014; Pérez Moreda et al., Reference Pérez Moreda, Reher and Sanz-Gimeno2015).

In order to strengthen our estimations and conduct a more detailed analysis of the individual effects of different variables, we have replicated the analyses by including the full set of variables but excluding the gender interaction with high-price years (Table 4). When the price increase surpasses 10% or 15%, there is an increase in the risk of death for the entire population; however, the statistical significance is low. The increase is more substantial (7.4%) and statistically significant when price increases exceed 20%. When we disaggregate the data between males and females, we can observe that the rise in the risk of death during high-price years is concentrated among women. The increase in female mortality risk during high-price years (more than 10%) compared to normal years is 6.3%, and 11.2% if it exceeds 20%. In the case of males, the variations are not statistically significant, and in the first two thresholds, they are even negative. The presence of high prices in year t−1 also results in an overall increase in mortality compared to normal years (1.5% at a 10% price increase, and 4.3% at 20%).

Table 4.

Cox proportional hazard models of mortality for children aged 6 months to 10 years. Wheat prices increase by 10 % (PANEL A), 15 % (PANEL B) and 20 % (PANEL C)

***p < 0.01, **p < 0.05, *p < 0.1. The basic controls include the number of siblings alive at birth, parity FE, village FE, year and year squared. The additional set of controls includes mother’s age, father’s occupation, and literacy.

Our results suggest that, in difficult times, families may have prioritised boys in terms of the quantity or quality of the food and perhaps also in the duration of the breastfeeding period. Alternatively, it is also plausible that parents devoted more attention and care to their sons than to their daughters when fighting against the diseases resulting from weaning and the subsequent deployment of the protective effect of breastfeeding.

In high-mortality contexts such as the one in our area of study, minor differences in how these children were treated were likely to have had lethal consequences, especially at sensitive ages. It is probable that other age groups also experienced similar discriminatory practices but their effect on mortality is less visible because they were not so high-risk: infants were protected by breastfeeding (a non-competitive resource) and older ones were subsequently stronger and had survived the dramatic weaning period. The expected male penalty, however, is not visible either. This strongly suggests that this kind of behaviour was also counterbalancing the female biological advantage at these ages.

The impact of socio-economic status and knowledge on mortality differences by sex during short-term economic stress crises

Previous studies on the effect of short-term economic stress have confirmed that socio-economic status has historically been related to changes in family demographic behaviour in response to these crises (Bengtsson, Reference Bengtsson, Bengtsson, Campbell and Lee2004; Dribe and Scalone, Reference Dribe and Scalone2010; Van Bavel and Kok, Reference Van Bavel and Kok2010; Jennings et al., Reference Jennings, Quaranta and Bengtsson2017). The poorest socio-economic groups were compelled to adjust their fertility and experienced the greatest impact on child mortality.

At the same time, in relation to literacy, having greater knowledge had a positive impact on the ability to access better-paying jobs (Vogl, Reference Vogl2014; Bargain and Zeidan, 2017; Bossavie et al., Reference Bossavie, Alderman, Giles and Mete2021) and being literate was related to having better knowledge about biological well-being and caregiving (Reynolds, Reference Reynolds2021). Advanced levels of education tend to be more concentrated in wealthier and more resourceful classes and are frequently used as an indicator of socio-economic status (Link and Phelan, Reference Link and Phelan2002; Link et al., Reference Link, Phelan, Hatzenbuehler, Major, Dovidio and Link2018). This factor has received significant attention in the literature as one of the key resources to explain the existence of social disparities in mortality (Clouston et al., Reference Clouston, Rubin, Phelan and Link2016; Luque de Haro, Reference Luque de Haro2024). In rural settings like ours, occupational categories are often broad (e.g., ‘farmer’), with limited variation. However, significant differences in wealth and resources may still exist within these categories. Literacy therefore serves as an additional, nuanced indicator of socio-economic status, capturing differences that occupation alone may not reveal. This is why in this section, we will explore the relationship between short-term economic stress and child mortality while controlling for parental literacy. Whereas in previous sections, socio-economic status was only used as a control variable (based on parental occupations and father´s literacy), at this point, we will examine the effect of socio-economic level by dividing parents into two groups based on whether either parent was literate or not.

Table 5 displays the models examining the impact of paternal literacy as a proxy for socio-economic status on female care discrimination. In columns 1 and 2, we have investigated models without interactions. The analysis reveals that the coefficient associated with the father’s literacy variable is significantly less than 1. This suggests that children aged between six months and ten years with literate parents had a reduced likelihood of mortality during childhood. When analysing the interaction of this variable with the occurrence of high-price years (columns 3 and 4), we can determine that children of literate parents had a noticeably reduced risk of mortality during those years and the subsequent year, compared to children without literate parents.

Table 5.

Cox proportional hazard models of mortality for children aged 6 months to 10 years stratifying by father´s literacy

***p < 0.01, **p < 0.05, *p < 0.1.

The basic controls include the number of siblings alive at birth, parity FE, village FE, year and year squared. The additional set of controls includes mother’s age.

Furthermore, we have conducted a stratification based on parents’ literacy levels. The results distinguish between children of parents possessing reading skills (columns 5 and 6) and those who did not (columns 7 and 8). It is evident from the analysis of models studying the mortality levels of the children of illiterate fathers that, in high-rice years, the mortality risk for females aged between six months and ten years is higher than that for males. However, the hazard ratio is almost one among the descendants of literate parents, implying the lack of gender disparities in mortality rate between six months and ten years during years of high prices. On the contrary, during ‘normal’ years, the risk is significantly higher for girls than for boys whose parents are literate, while among the children of illiterate parents, the differences in ‘normal’ years are not significant.

The findings suggest that discrimination against girls during childhood may have multifaceted causes. One of these factors is associated with the unequal distribution of resources, which appears to have a more pronounced impact on the daughters of families with lower socio-economic status during periods of short-term economic stress. However, the higher number of female fatalities during average harvest years among the general population, as well as among the daughters of literate parents, indicates the presence of an additional form of bias in caregiving. This cannot be solely attributed to financial limitations but is certainly linked to cultural factors and gender-specific role distinctions. It is plausible that in households with literate fathers, literacy was especially valued due to its economic and social benefits, potentially reinforcing son preference. Given the role of sons as carriers of literacy within the family, it is possible that these households directed resources preferentially to male children, further disadvantaging daughters.