Introduction

The sex ratio—the proportion of males relative to females in a population—exhibits remarkable heterogeneity across countries. In particular, in many countries around the world, there is a disproportionate number of males relative to females. This well-recognized fact led Amartya Sen to declare in 1990 that more than 100 million women were “missing” [1, 2]. Examining data on the global cross-country distribution of the average male-to-female sex ratio between 2000 and 2009, one finds that the sex ratio varies widely. For example, for children under age 5, it ranges from just under 1 for a number of countries (e.g., Rwanda, Angola, Togo, etc.) to 1.10 in China. S1 Fig shows the distribution of sex ratios for different age groups. As it is apparent from the figure, most countries in our sample have a sex ratio around 105. A non-trivial fraction of our sample (53 countries out of 153) has a sex ratio higher than 105. These countries belong to all continents (47% from Europe, 36% from Asia, 10% from Africa and 7% from North and South America). The sex ratio goes down with age but still shows a large amount of variation for all age groups. Although China is particularly well-known for having a male-biased sex-ratio, there are many other countries that also have sex ratios that are strongly male-biased, e.g., South Korea (1.08), India (1.08), Albania (1.07), Georgia (1.06), Singapore (1.06), Jordan (1.06), etc.

A range of factors underlying the noticeably larger proportion of males in many countries has been well studied. The correlates of the sex-ratio at birth include parental well-being [3], birth order [4], and fertility [5]. The correlates of the sex ratio after birth have also been widely studied and include economic opportunities for women [6, 7], patriarchy and kinship structures [8, 9], and disease burden [10]. For a review of the literature see [11, 12].

In this paper, we test the hypothesis that differences in the sex ratio are determined, in part, by differences in the agricultural technologies traditionally used by a society’s pre-industrial ancestors. Earlier research by Ester Boserup [13] has suggested that the use of plough agriculture generated a division of labor where men worked in the fields and women specialized in work within the home. This is because the use of the plough in agriculture requires significant upper body strength, grip strength, and bursts of power to pull the plough itself or control the animal that pulls it. This can be contrasted to shifting cultivation, where women tended to participate as actively in agriculture as men. This form of agriculture tended to use the hoe and digging stick, and, although it required very hard work, did not require the same level of physical strength as plough agriculture. In addition, shifting cultivation was compatible with child care, in contrast to plough agriculture, where the presence of large animals made child care dangerous.

According to Boserup, differences in agricultural practices led to the appearance of different norms about the role of women in society. Societies featuring plough agriculture exhibited weaker beliefs about the equality of men and women in society, which has persisted until today, affecting differences in female labor force participation [14]. An additional consequence of traditional plough agricultural is that girls came to be valued less than boys. In turn, this resulted in the emergence of the custom of the dowry, where girls’ parents are required to make a large payment to the newly formed family and/or the groom’s family. In societies without the plough, a bride price was more likely to arise, where a large payment is made from the groom’s family to the bride’s. These differences in marriage customs further reinforced societies preference for boys over girls in societies with plough agriculture. This male-preference is likely to result in male-biased sex ratios of children. Although this could occur through selective abortion or infanticide in more extreme cases, the most likely mechanism is through differential access to nutrition, health care, and similar resources, which result in differential mortality rates for boys and girls.

In Boserup’s own words: “In communities where girls live in seclusion, and a large dowry must be paid when they marry, parents naturally come to dread the burden of having daughters. In some of the farming communities in Northern India, where women do little work in agriculture and the parents know that a daughter will, in due course, cost them the payment of a dowry, it was customary in earlier times to limit the number of surviving daughters by infanticide.” (p. 37) Boserup continues to explain that “This practice has disappeared, in its outward forms, but nevertheless the ratio of female to male population in these districts continues to be abnormal compared to other regions of India. […] The only plausible hypothesis is that mortality among girls was higher than among boys. […] the persistence of socio-cultural factors are believed to be largely responsible for the excess of female mortality over the male. For example, one of these socio-cultural factors seems to be a widespread supposition that milk is not good for girls, but is good for boys. […] There is also a tendency to care more for sick boys than for sick girls.” (p. 37)

Recent research has shown that parents often treat boys and girls differently. For example, differential allocation of food to sons relative to daughters can explain the global distribution of stunting [15], and it is plausible that they also have effects on mortality rates, thus affecting the sex ratio. Evidence from India shows that boys are breastfed longer [16]. In the same country girls with congenital heart disease are less likely to have a surgery than boys [17]. More generally, parents may be more willing to spend money on healthcare on their son than their daughter. A son preference will also cause parents to have a sex-biased stopping rule [16]. For example, if a family that could only financially support four children had four daughters, a male gender bias would make this family more likely to have another child in an attempt to have at least one son. Thus, families that consist of more daughters tend to be larger and living beyond their means.

In this paper, we test whether a tradition of plough agriculture is associated with differences in the male-to-female sex ratio. Our focus on a specific historical determinant of the sex-ratio is not meant to imply that other factors are irrelevant in determining the sex ratio. The sex ratio can be affected by a host of factors at conception (primary sex ratio), during pregnancy (secondary sex ratio), and after birth (tertiary sex ratio) [18, 19]. A large literature in biology and medicine has studied the determinants of primary and secondary sex ratios [11]. Sociologists and economists have examined a wide range of explanations for differences in tertiary sex ratios, including income [20–24], intra-household allocation of resources across children [25, 6], different preferences about the gender of the children [7] and social norms ([26, 27, 28, 29] for the case of China). Our analysis accounts for those alternative determinants of the sex ratio through the use of an instrumental variables estimation strategy.

The hypothesis we test relies on the assumption that the historical adoption of the plough continues to affect the sex ratio today, decades after most societies have moved out of agriculture. Recent evidence in economics shows that historical events can have highly persistent effects on beliefs and values. For example, [30] show that areas that were more anti-semitic during the Black Death were also more anti-Semitic during the early 20^th century. It has been shown that the slave trade in Africa, which occurred roughly from 1500 until 1850, not only continues to affect economic outcomes like per capita income [31], but also a range of cultural characteristics, such as ethnic diversity [32], trust [33], polygamy [34], and gender roles [35]. Differences in geographical characteristics related to the suitability of crops with different caloric yields appear to be responsible for the evolution of more patient time preferences [36]. Evidence of such long-term persistence is also not confined to the economics literature. The classic work of [37] on the culture of honor traces the effects of ancestral herding on the culture of honor and violence that one observes in the U.S. South today, while, more recently, [38] document how historical specialization in wheat, rather than wet rice agriculture in China, led to a greater prevalence of individualism today.

Material and methods

Descriptive analysis

We begin our analysis by first examining the raw data. This is done visually in Fig 1, which reports the average sex ratio from 1960–2010 for two groups of countries. One is 79 countries for which more than 95% of the population has ancestors that used the plough traditionally and the other is 67 countries for which less than 5% of the population has ancestors that used the plough. From the figure, we see that for all age cohorts, the male-to-female sex ratio is greater among societies that traditionally engaged in plough agriculture. While this is true for the sex ratio at birth, it becomes stronger for older children. This is potentially explained by higher mortality rates among girls than boys.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Countries with ancestors who used the plough show a higher male-to-female sex ratio today.

The figure shows the average sex ratio (boys per 100 girls) in countries with ancestors that engaged in plough agriculture and in countries with ancestors that did not engage in plough agriculture for different age ranges. The sample includes 146 countries, 79 of which are countries that are categorized as traditionally engaging in plough agriculture and 67 as not traditionally engaging in plough agriculture. “Countries with plough agriculture” are countries for which 95% or more of its population has ancestors that belong to ethnic groups that traditionally engaged in plough agriculture. “Countries without plough agriculture” are countries for which 95% or more of its population has ancestors that belong to ethnic groups that traditionally did not engage in plough agriculture. The reported sex ratios are quinquennial averages from 1960–2000. The difference in the sex ratios between the two groups is significant at the 1% level for all age groups.

https://doi.org/10.1371/journal.pone.0190510.g001

The differences that we observe in the raw data are potentially explained by a host of other determinants other than traditional plough agriculture. Thus, we now turn to more formal multivariate OLS and IV estimates of the determinants of the sex ratio. In particular, the IV analysis addresses the possibility that societies with different gender norms historically may have been more likely to adopt technologies, like the plough, that reinforced the original biases. Before presenting the IV estimates, we first report, for comparisons, the multivariate OLS estimates of the relationship between ancestral plough agriculture and the sex ratio. These estimates account for a number of factors that may have their own effect on the sex ratio. The list of control variables included in the regression was described above.

The multivariate OLS estimates are reported in Panel A of Table 1. Each column of the table reports estimates for a different age cohort. The table reports coefficient estimates and robust standard errors, as well as Conley standard errors that adjust for non-independence of the observations [43, 44]. For our baseline estimates, we use the average geographical distance between a country’s ancestors to adjust for spatial dependence. As we discuss below, standard errors are similar to the use of a range of alternative methods to correct for non-independence using the Conley method.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Sex ratio and ancestral plough use.

https://doi.org/10.1371/journal.pone.0190510.t001

The estimates show that countries with more ancestral plough use tend to have a higher male-to-female sex ratio. In addition, the estimated effect is systematically larger for children of older age cohorts. Note that the finer age categories are not mutually exclusive. This should be taken into account when comparing the estimates across different dependent variables that measure the sex ratio of different age cohorts. The magnitudes of the estimated coefficients are sizeable. Based upon the estimates from column 1, a one-standard-deviation increase in ancestral plough use (0.44) is associated with an increase of 0.20 standard deviations of the sex ratio measure. For children aged 5–14, the estimated effect of a one-standard-deviation increase in ancestral plough use is 0.33 standard deviations of the sex ratio measure. Columns 5 and 6 of Table 1 report the principal component across the four different sex ratio measures (column 5) and across the sex ratio measures for the two non-overlapping cohorts, 0–4 and 5–14 (column 6). In Fig 2, we report the partial correlation plot for the specification in column 4, where the dependent variable is the sex ratio of children aged 5–14. As shown, the correlation is quite general and not driven by a small number of influential outliers. As part of Supporting Information, we also report the partial correlation plots for other age ranges. The coefficient increases with age (see S6 and S7 Figs).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Partial correlation plot: ancestral plough use and the sex ratio of children aged 5 to 14 (boys per 100 girls).

The sample includes 153 countries. The sex ratio is calculated as a quinquennial average from 1960–2000. The specification includes continent fixed effects, historical covariates (economic complexity, political hierarchies, the presence of large animals, agricultural suitability and a measure of tropical climate), and contemporaneous covariates (per capita GDP and its square, fertility, and infant mortality). Each country is labeled with its 3-digit iso code.

https://doi.org/10.1371/journal.pone.0190510.g002

As an additional measure of the magnitude of the historical plough use, we calculate the change in the R-squared when the historical plough use variable is added to the regression equation. For the specification where the dependent variable is the principal component of the sex ratio measures from all cohorts (column 5), the inclusion of the historical plough use variable increases the R-squared by 0.0226. (from 0.5733 to 0.5959). Therefore, a history of plough agriculture accounts for 2.3% of the total variation in the sex ratio’s principal component and 5.3% of the residual variation left unexplained by the control variables [(0.5959–0.5733)/(1–0.5733)]. For comparison, these magnitudes are similar to those from the inclusion of continent fixed effects.

Results and discussion

The IV estimates are reported in Panel B and C of Table 1, where we report the same set of specifications as for the OLS estimates. Panel C reports the first-stage estimates, where ancestral plough use is the dependent variable and plough-positive crop suitability and plough-negative crop suitability serve as instruments. The estimates show that while plough positive suitability is positively associated with the adoption of the plough, this is not true for plough-negative crop suitability. Thus, the estimates provide confirmation for Pryor’s hypothesis that crop type affected the adoption of the plough. In Tables D and E in S1 Supplementary Material, we examine the stability of these findings over the period of our analysis, 1960–2010. For simplicity, we report the results for the two non-overlapping sex ratios (age 0–4 and 5–14). The importance of the plough appears to be very stable and shows only a very small decline over time. They also provide confirmation about the relevance of our instruments for plough adoption.

The second-stage IV estimates are reported in Panel B of Table 1. They show that ancestral plough use is associated with higher male-to-female sex ratios. This provides evidence that the positive relationship between traditional plough use and the sex ratio, that was estimated using OLS, is not due to omitted variables bias or reverse causality and, thus, is likely causal. As with OLS, the IV estimates are sizeable in magnitude and show that the magnitude of the estimated effect is larger for older age cohorts. For example, based upon the estimates from column 4 and 5, a one-standard-deviation increase in traditional plough use (0.45) is associated with a 0.44 standard-deviation increase in the sex ratio of children aged 0–4 and a 0.58 standard-deviation increase in the sex ratio of children aged 5–14. Columns 5 and 6 of Table 1 report the principal component for the four different sex ratios (column 5) and for the 0–4 and 5–14 sex ratios (column 6). The results are qualitatively similar when we examine these aggregated measures.

Comparing Panels A and B of Table 1, we see that the magnitudes of the IV estimates are consistently larger than the OLS estimates. This is most likely due to a bias towards zero in the OLS estimates, which arises due to the fact that plough societies tend to be more economically advanced both in the past and today. If there is a positive relationship between economic development and equality of gender norms, and if our controls for historical and contemporary income are imperfect, then our OLS estimates will be biased towards zero, while the IV estimates will not suffer from this bias.

There are a number of potential limitations that should be kept in mind in interpreting our results. The primary is that although our procedure, which links historical ethnic groups to contemporary populations using languages, works well for many countries of the world, it works less well for countries where languages have been adopted by other groups within the country. An example is the United States, where all populations speaking English today are assumed to have ancestors that are English, even though we know that many ancestors of English-speakers within the U.S. were from non-English-speaking backgrounds, e.g., Italian, German, etc. To check the importance of this issue for our estimates, we have re-estimated our baseline regressions after omitting all countries from North and South America, as well as Australia, New Zealand and South Africa. The estimates, which are reported in Table F in S1 Supplementary Material, show that our estimates are very similar when we remove these countries from our sample.

A second potential concern relates to the non-independence of ethnicities in the Ethnographic Atlas, which has the potential to result in non-independence of observation in our country-level in our sample (i.e., Galton’s problem) [44]. We have addressed the issue of the non-independence of observations by also reporting Conley standard errors, which correct for non-independence of observations in our sample. The correction requires information on the distance between observations. For our baseline estimates, we have used the spatial distance, measured as the average distance between the traditional location of a country’s ancestors, which was calculated using the latitude and longitude of ethnic groups (as reported in the Ethnographic Atlas). An alternative strategy is to use genetic distance between countries [45]. Standard errors using this method are very similar to those calculated using geographical distance. These are reported in Table A in S1 Supplementary Material. The Conley correction also requires an assumption about the range of distances for which the correlation between observations is non-zero. Table A in S1 Supplementary Material reports calculated standard errors for different assumptions about this range. Overall, we find that the Conley standard errors, which correct for non-independence of observations, tend to be slightly smaller than conventional standard errors when we use geography and slightly larger when we use genetic distance. Overall, adjusting for non-independence does not alter our statistical inference nor our conclusions.

The third concern is related to the geographical variation in plough use. There is little variation within Europe and within sub-Saharan Africa. Therefore, our estimates could simply be driven by broad differences between continents. All of our estimates include continent fixed effects. To also rule out the possibility that we are not capturing differences between Sub-Saharan Africa and the rest of the world we also disaggregate the African continent indicator into an indicator for sub-Saharan Africa and one for North Africa. The estimates are robust to this additional specification (Table G in S1 Supplementary Material).

The fourth potential concern is that for our instrumental variable strategy, the difference between plough-positive and plough-negative environments may be correlated with geographic features that affect gender attitudes today through channels other than the plough. Although we have controlled for an extensive set of geographical and historical variables, which mitigates this concern, the violation of the exclusion restriction remains a possibility.

The final caveat is that our estimates only reflect the effects of having a tradition of plough agriculture versus not. Ploughs that were adopted varied significantly in size and complexity, and these may have had very different effects on gender roles and, thus, on the sex ratio today. Our analysis is only able to report the average effect of plough adoption across all of the types of ploughs that were adopted.

Conclusions

Our findings show that the variation in sex ratios that we observe today has deep historical roots. We find that descendants of societies that traditionally practiced plough agriculture have higher average male-to-female sex ratios today. The evidence is consistent with traditional plough use shaping the value placed on boys relative to girls, which continues to persist until today, affecting observed sex ratios.

Supporting information

Acknowledgments

We thank the editor and two anonymous referees for comments that greatly improved the paper.