Missing millions: the human cost of the Mexican Revolution
© 2001 Robert McCaa, University of Minnesota Population Center
There is no consensus among scholars regarding the demographic impact of the Mexican revolution or its components. Total losses range from 1.9 to 3.5 million. Table 1 summarizes the distinct scenarios proposed by nine specialists. Some discount the missing millions by emphasizing emigration and error, such as Loyo (1935), Collver (1965, Model "B"), Alba (1977), and Mier y Terán (1982). Others assign a major role to lost births and thereby reduce the scale of the horror (Gonzalez y Navarro (1974), Collver (1965), Greer (1966), and Mier y Terán (1982)). Still others blame massive mortality (Collver (1965, Model "A"), and Ordorica and Lezama (1993)), or exogenous mortality from the Spanish flu pandemic (Loyo 1935; Ibarra 1996). Each of these factors must be taken into account, but there is little agreement over their relative importance. Collver (1965)—the authority for subsequent demographic analyses by Greer (1966) and Ordorica and Lezama (1993)—encompasses the extremes, by offering two scenarios, both of which he characterizes as “implausible”. Collver’s model A maximizes the impact of mortality, with two million excess deaths. Model B minimizes mortality (“only” one-half million excess deaths) and balances the demographic equation by attributing the missing one and one-half million to census error. Collver's models are indeed implausible extremes. The best-case scenario falls between them, at around two-thirds the mortality maximum.
Since Collver’s research was published almost forty years ago, new information regarding the devastation of the revolution and new demographic methods have become available. The fatally flawed 1921 census is no longer a required benchmark, although it has been used by every previous effort to assay the human costs of the revolution, aside from the mathematical projection by Luna Mendez (1959). His graphical extrapolation shows 1921 to be wildly off the mark but nothing is made of it. Error in the 1921 enumeration is often used to explain away what otherwise was an unthinkably large number of missing. Error also discouraged researchers from attempting more refined analysis or proposing anything over than crude figures. Error can be reduced, on the one hand, by standard cohort analysis to 1930, when one of the best Mexican censuses of the twentieth century was conducted, and, on the other, by projecting females separately from, indeed prior to the male population, using inverse methods to model the 1930 age structure for each sex individually. I propose to ignore the 1921 census entirely, referring back to it only for the sake of completeness and comparability, to obtain an estimate of the undercount in that year and to facilitate comparison with other estimates of the human costs of the war.
Two-sex inverse projection offers significant advantages over one-sex conventional projections used by previous researchers. The most important advantage is that the goal of two-sex inverse projection is to balance the demographic equation not simply by population totals but by age and sex (McCaa, 1993). The inverse method requires minimal data, yet produces refined, surprisingly accurate demographic estimates, including life expectancies, gross reproduction ratios, and population age structures. Annual two-sex projections are the method’s greatest strength, even in the absence of good data. Since inverse projection demands little in the way of inputs, results may be compared against the most reliable data at hand. Various scenarios can be readily tested for plausibility. Even where the best model fails to fit the most reliable data, departures from empirical targets may be used to gain insight into greater than expected losses for some age groups and less for others. Finally, the method relies on data and demography, not mathematics. Instead of using mechanical methods favored by demographers, which tend to smooth away history (see, for example, Camposortega's summary of methods to resolve the age-heaping problems in Mexican censuses, 1992:19, 86), inverse projection encodes a chronology of births and deaths into the age and sex pyramid, year-by-year, cohort-by-cohort.
The power of the inverse projection method lies in its theoretical underpinnings, what demographers call the ergodicity theorem (Wachter 1986). Reduced to its simplest, this theorem states that the structure of a population is dependent upon demographic dynamics, not an age structure at some distant point in the past. Simply put, populations have no long-term memory when it comes to age structure. Moreover, contrary to common sense, instead of projecting backward, the inverse method is most powerful in projecting forward (Lee 1993). Thus, we can confidently begin a projection from, say 1895, without needing to know the demographics that came before, and if we successfully map the annual flow of births and deaths, the age structure will map underlying population dynamics. Estimating the annual number of births is most important, because if this figure is wrong for a specific year, results over the entire projected span will be distorted for the cohort born in that year. Accuracy in predicting the annual death flow is not as critical because deaths are distributed over all age groups, and error in one year may be compensated by errors in subsequent years. The same principle works for net migration. Given the fact that the annual volume of emigration is low (typically deaths greatly exceed migration), errors in measuring net migration are even less significant in this regard.
It is important to understand that the inverse projection method produces a model, which mimics reality to a greater or lesser degree. Deaths and emigration are apportioned out by age and sex for each year of the projection, based on the estimated total flow for that year and a pattern of change by age and sex derived from model mortality and migration tables. Where a five percent error rate is tolerable, the method yields amazingly accurate results. The strengths and limitations of the method are well documented, having been calibrated in a variety of trying historical and hypothetical conditions (Brunborg 1977; McCaa and Vaupel 1992; Lee 1993; McCaa 1993; Galloway 1994; Wrigley, Davies, Oeppen and Schofield, 1997). Nevertheless, it is important to realize that we can never know, in the case of Mexico for example, how many young boys died due to war in, say, 1915. The method helps predict how many might have died based on various scenarios. The only constraint is to match the age and sex structure of the 1930 census using conventional inverse projection methods and assumptions.
Inverse projection’s historical encoding is what sparked my interest in weighing the costs and consequences of the revolution in the first place. Some years ago as a classroom exercise, I did a simple inverse projection of the population of Mexico for the twentieth century, using Collver's crude birth and death rates. The results were surprising, particularly in mimicking corrected, as opposed to raw census age structures. For example, the census of 1960 reports children aged 0-4 years as constituting 16.59 percent of the total population. An authoritative correction by the Mexican demographers Raul Benitez and Gustavo Cabrera (1967) boosts the figure to 18.66 percent. Benitez and Cabrera, working in the heat of the moment to construct an accurate life table for 1960, sought to correct both the under-count and age-heaping. Their solution increased the proportion aged 0-4 by three percentage-points. A simple inverse projection yields the remarkably close adjustment of 3.4 points. I was further surprised to find that for the 1911-1920 birth cohort (aged 40-49 in 1960), mathematical smoothing yields a figure of 7.5 percent, while inverse projection points to a squeezed generation, amounting to only 6.8 percent of the total. Although the difference may seem slight, it is suggestive of how the historical origins of a cohort remain imprinted in an inverse projection—but may be erased by mechanical adjustments commonly used to smooth age data.
Inverse projection confirms that Collver's maximum models are implausible. The figures proposed here intersect his models at about two-thirds of the maximum for deaths and 90 percent for error (see Table 1). My estimates match closely the high, but plausible mortality findings of Ordorica and Lezama (1993). Nearly one-and-one-half million excess deaths occurred in the period 1910-1921, according to their analysis and mine. Our interpretations differ, however, on the matter of causes. While we agree that famine, disease, and epidemic were the proximate causes for the large number of excess deaths, I single out war as the root cause. Even in the case of the infamous Spanish Flu epidemic, which medical historians think was more devastating in Mexico than almost any other country in the world, its severity, in my view, is explained precisely by the disorder and weakened condition of the Mexican population vexed by years of unremitting violence, civil war and banditry. This assessment is shared by J. Gabriel Ibarra (1996), the author of the most up-to-date, comprehensive study of the epidemic in Mexico.
Lost births amounted to slightly less than 600,000, a surprisingly large number, but this is the smallest figure yet proposed by any demographic model (Table 1). At one-half million below the estimate by Ordorica and Lezama, our differences may be due to the fact that my accounting includes census error, and theirs does not. While their text accepts Gilberto Loyo’s rough estimate of undercount in the 1921 census as one-half million, the number is inexplicably omitted from their table (see Ordorica and Lezama, 1993, Table 9). My inverse projection ignores the flawed census of 1921 entirely, targeting 1930 totals by sex and age. Checking the results for 1921 from the best fitting inverse projection against the census yields a total undercount of 1.1 million in that year, only 100,000 short of the highest estimate by a demographer, Collver's self-described implausible error model.
The proposed emigration figure, 350,000 total net persisting emigrants over the 1910 decade, falls in the middle of earlier estimates. While Ordorica and Lezama favor four hundred thousand, the congruence between our figures is surprising given that they are derived by different data and methods. Emigration is the residual from their demographic balancing equation for the period 1910-1921, whereas my figures are based on foreign censuses, primarily of the United States. If we are in rough agreement on the number of Mexicans emigrating to the United States, our interpretations differ. In my view, economics, not politics, was the motive for many, probably most, Mexicans who emigrated during these years. As the economy of the Southwest boomed during the Great War, emigration from Europe to the United States slowed to a trickle, from 1.2 million in 1915 to less than 100,000 in 1917 (Haines 2000). This constellation of events lured many Mexicans “al norte”, often enticed by labor recruiters who readily advanced train tickets and even travel expenses, to work the rails, fields and factories in the United States (Martinez 1957, Cardoso 1980; Hall and Coerver 1990). In the 1920s economic slowdown and the renewed flood of Europeans dampened Mexican emigration. For the 1910 decade, sanctuary-seekers who remained permanently in the United States constituted only a fraction of the total emigration flow, almost certainly less than one-half. For this reason my figure in Table 1 is multiplied by 0.5. While the total comes from United States census data and is required to balance the demographic equation, a distinction should be made between emigration for political and economic reasons. One-half seems a not implausible fraction (see below). Emigration to Mexico during the decade was slight (González Navarro 1993-1994).
To sum up, in my view, the human cost of the Revolution was mainly internal, paid in Mexican blood. Of a total demographic cost of 2.1 million, excess deaths accounted for two-thirds, lost births one-fourth and emigration considerably less than one-tenth of the total. These fractions are not original, having been proposed by one or another researcher over the past three-quarters of a century. What makes them unique, in addition to weighing the cost for females and males separately, is how they combine and how they were derived. The remainder of the paper explains the data and methods in which this new combination of estimates is grounded.
Prior to 1930 vital statistics are unreliable for Mexico. Over thirty percent of births went unrecorded as recently as the 1920s, due in part to the disruption of revolution. At the beginning of the twentieth century, the record was better, although only by a couple of points, with an estimated 27 percent unrecorded. Death registration, in contrast, attained a remarkable level of completeness before the war. However, after 1910, as the intensity of fighting and looting increased with churches and government offices favored targets, the national registration system essentially collapsed (Collver 1965:138). While vital statistics may have been reliable for a few districts or even cities during the war years, there are none for the nation as a whole. Vital statistics for this period amount to crude quinquennial or even decennial estimates. Researchers derive crude numbers of births and deaths from basic demographic balancing equations suitable for projecting the total population forward from the census of 1910 to 1921 (see Collver 1965 and Mier y Teran 1982).
Censuses remain the best source of information on the Mexican population prior to 1930. Undercounting is typically the main shortcoming of this source, but in the case of pre-revolutionary Mexico, fraud, possibly for electoral purposes, may have been a more significant problem. Collver's maximum census-error model deflates the count for 1895 by 200,000, 1900 by 400-500,000, and 1910 by 200-500,000. In contrast, most researchers assume that fraud counter-balanced undercount in these early national efforts and accept the official figures for 1900 and 1910. For 1910, the last enumeration during the thirty-year rule of the aged dictator Porfirio Diaz, federal inspectors were sent into the field for the first time to monitor the taking of the census and the tabulation of returns, but counting and reporting remained in the hands of local officials (Greer 1966:28-38). I accept the official figures for 1910 as reported and leave aside those of 1895 and 1900, as not crucial to resolving the issues at hand.
The 1921 enumeration attempted to replicate 1910 procedures, but failed. General Alvaro Obregón's successful rebellion in May 1920 was followed by a purge of the National Statistical Office. The fourth national census of population, scheduled for October, was postponed for more than a year to November 30, 1921. Publication of results was delayed for many years by the failure of state officials to send in their tally sheets and by disorganization in the national office. Many of the more complex tables were never published, and many partial returns can be found today still in manuscript form, incomplete and unpublished in the Archivo General de la Nación (Gobernación, Fomento y Obras Publicas, Ramo de Censo y Estadística). Fraud is evident in the 1921 enumeration, with the returns for Colima appearing particularly suspicious (Greer 1966:7). Undercounting, however, was most severe, certainly worse than for any other Mexican census of the twentieth century. Loyo's estimate of a shortfall of 500,000 is widely accepted (Loyo 1960:4-5). Although this figure has never been examined critically, the adjustment is certainly in the right direction. On the other hand the 1921 numbers are so fundamentally flawed that ignoring them entirely in favor of those for 1930 seems a better approach.
In 1930, central processing of the original census returns was instituted. Thanks to this innovation, most of the sheets for the 1930 enumeration still survive and have been microfilmed. They may be consulted in the Archivo General de la Nación or on film at any branch library of the Genealogical Society of Utah. The 1930 effort is widely recognized as one of the best planned and competently executed of the century, although a complete enumeration remained an unattained ideal.
Error is the principal culprit for Collver (Model B), Mier y Terán, and Greer, with a million or more of the missing attributed to this factor. Collver's model B, or maximum census error model, would account for almost half the entire “loss” over the decade by this factor alone. Model B combines guesstimates of inflated counts in 1910 and suspected widespread undercount in 1921 to obtain a figure of 1.2 million total error. Greer's study of the demographic impact of the Mexican revolution endorses Collver's model B as the most likely scenario. Mier y Teran also relegates one of her missing millions to census enumeration problems.
Order of magnitude is the most that historians should expect given that adjustments for error of one-half million or more is required for every national census since the 1921 effort. The scale of adjustments for error, subtracting one-half million for 1910 and adding a similar amount for 1921 or 1930, shrinks estimates of real demographic loss by a substantial fraction. However, the evidence for widespread census fraud in 1910, as presented by Collver and Greer, is simply too scanty to be accepted. Neither author reports any systematic attempt to examine returns for individual states to detect inflated figures for specific census years. Arriaga (1968:163) and Mier y Terán (1982:353) place greater confidence in both the 1910 and 1930 enumerations than those of 1921 or 1895. Mier y Terán estimates an undercount of only 0.8 percent for 1930 (120-140,000), versus four percent for 1921 and two percent for 1940.
Alba, who pays little attention to early census data, places the undercount for 1930 at 510,000, less than for any other modern Mexican census (1977:18). To resolve the disagreement between Mier y Terán and Alba, splitting the difference at 300,000 seems reasonable. If Mier y Terán's estimate of error is closer to the mark for 1930, then I will have exaggerated census error by 150,000, but if Alba's is more accurate, then my figure is an underestimation of some 200,000. Whatever the total error, it must be parceled out to specific age groups. I assign one-half of the undercount to the age-group 0-4 (increasing the proportion from 15.2 to 15.8 percent), and the remainder, 75,000 for each sex, is spread proportionally over ages 5-85+. In other words, my adjustments are slight for all age groups except the youngest, which of course were born after the revolution and consequently are not as critical for answering the question at hand. Nevertheless, if a good-fitting inverse projection cannot be constructed for the decade after the revolution, then great expectations should not be held for modeling the decade of violence.
Prior attempts at assaying the costs of revolution focus on the decade of intense fighting. However, fighting did not end in 1917 or 1921. Indeed loss of life resulting from the Cristero rebellion in the late 1920s was extensive. The human cost of this event demands a study of its own, year-by-year and region-by-region. By examining the period 1910-1930, I base my assessment on more reliable data and make it possible to take into account destruction in later years as well. Finally, the estimate of undercount for the 1921 enumeration is derived simply by looking back from 1930, comparing the total population for 1921 from the inverse projection with the published totals.
Emigration, which accounts for as much as one-third of the total demographic loss for Gamio, Loyo and Mier y Teran, amounts to less than one-tenth for Gonzalez y Navarro and Collver's maximum mortality model. Ordorica and Lezama's figures for emigration show how slippery words can be when one tries to convert them into numbers, as in Table 1. While the authors give a numeric estimate of the emigration effect (400,000), the text attributes “a good part” of the missing millions to emigration. It seems to me that “a small part” might be more appropriate for describing a thirteen percent loss. For other authors, where no numbers are available (noted by brackets in Table 1), I have converted slippery narrative into concrete numbers.
Those who see emigration as a minor effect of the revolution can find comfort in new evidence from the original United States census manuscripts for 1920 which only became available to the public in 1992 (Gutmann, McCaa, Gutierrez-Montes and Gratton, 2000). Likewise, a re-working of rarely used published census figures also supports a low emigration scenario. The total Mexican born population in the United States increased from 221,915 in 1910 to 486,418 in 1920 and 641,462 in 1930, an increase of 260,000 in the first decade and 160,000 in the second (U.S. Census Bureau, 1933, vol. 2, p. 14, Table 17). These are net figures, not total flow. To arrive at the number of total persisting emigrants, an estimate is required of Mexicans who died in the U.S. in each decade and were replaced by new emigrants. Mortality of 10-20 percent per decade would encompass the true rate, as we shall see below. This would increase total net emigrants by 22-44,000 for the first decade and 50-100,000 for the second. Adding 33,000 and 75,000 yields mortality-adjusted numbers of 300,000 and 230,000 total persisting Mexican emigrants to the United States for the 1910 and 1920 decades, respectively.
Any complete assessment of the human impact of the revolution must take into account not only error, emigration and excess deaths, but also lost births—that is births that did not occur because of the disruption of normal family life. Here too, there is substantial disagreement among the authorities. Most demographers discern nearly a million lost births, equivalent to a decline of 10-15 percent in the crude birth rate over the entire decade. On the other hand, the father of Mexican demography, Gilberto Loyo, seems to have considered fertility effects as rather slight. Indeed his magnum opus on the population history of Mexico, although highly critical of the demographic attainments of the ancien regime of Porfirio Diaz as “no tan buena como debería haber sido”, omits any discussion of the impact of the revolution on fertility (1935:118). His assessment must be gleaned instead from musings toward the end of his career (1960:4). In contrast, Gonzalez Navarro's unpublished population history of Mexico in the twentieth century offers a figure of almost 1.5 million, which he describes as “crecimiento que no se afectuó [sic]”, but this number includes increased mortality due to disease, famine, and disorder.
Here, disagreement is greatest. On one extreme are the strict constructionists who would exclude epidemics as exogenous factors (Loyo and Gonzalez Navarro), and on the other are those who define mortality as the residual, after taking into account error, emigration and fertility (Collver, model A). Loyo, for example, blames the influenza epidemic of 1918 for much of the loss of the decade (1960:4):
Así, se puede estimar que la población de 1910 a 1921 perdió dos millones de personas. Una parte de estos dos millones, la menos, corresponde a las pérdidas de vidas en los años de las luchas armadas de la Revolución Mexicana, y la otra, la mayor, a la tremenda mortalidad por la epidemia de gripa llamada ‘influenza española’.
That the Spanish influenza epidemic was devastating in some regions of Mexico is beyond dispute, but its effects seem to have been most severe precisely where fighting had been most intense, and the population most weakened. For example, in Mexico City, deaths abruptly tripled in December 1918, reaching 4,329 (AHSS, Epidemiología, box 11, exp. 1-2). Nevertheless this number is well below the devastation of 1915, the year of hunger. 1915 was the deadliest year in the modern history of the city, with almost twenty-five thousand reported deaths, or more than five percent of the population. 1916 was nearly as bad with barely 500 fewer deaths. In the year of the flu epidemic over 23,000 deaths were recorded, but this fell short of 1915 by more than one thousand. Then too, after the epidemic passed, mortality declined by almost one-half in the following year to fewer than 12,000 deaths. The average for the biennium yields nearly normal levels of mortality. In contrast, recovery from war, hunger or typhus was not immediate, as mortality remained high (AHSS, Epidemiología, box 11, exp. 1-2). Similarly, in the historic city of Aguascalientes respiratory mortality in 1918 soared, but total deaths still fell considerably short of the 1916 record, where war, famine and other diseases tripled the pre-war average (González Esparza 1992:43). With respect to the Spanish flu pandemic, Ibarra's study (1996) offers a comprehensive review of the meager statistics available. Sánchez Rosales (2000:23), citing vague contemporary conjectures from the press, which placed total mortality for the Republic at 350-450,000 deaths (2-3 percent of the population), calls for a major study of the epidemic. The one more-or-less reliable statistic which he reports comes from the Mexican army. Of 125,000 men on the rolls, 1,862 died from the disease, a death rate of 15 per thousand. While the regional picture is far from complete, Allan Knight's detailed study of the revolution offers a succinct chronological and spatial description of the devastating effects of war on food supply, disease, and epidemics (see vol. 2:413-423).
Finally, it should be noted that public health efforts to contain epidemics did not completely collapse, even in the worst years of violence. Smallpox, for example, remained under control, notwithstanding isolated outbreaks. In 1910, as revolution erupted on the northern border, annual smallpox mortality in Mexico City fell to a half-century low of only 90 deaths for the entire year. Then the figure began to rise, to 390 in 1911 and 429 in 1912. As warring and banditry worsened, migration increased and the instinct for survival overpowered charitable inclinations of both parents and public officials. In 1915, the last great smallpox epidemic in the City's history erupted. Yet, it was a faint shadow of former bouts. From a weekly average of 10 cases (not fatalities) over much of 1914, 18 cases were recorded in the first week of 1915, rising to 50 per week in March and 70 in April. While total deaths for the year surged by nearly one-third, hunger and typhus accounted for much of the increase. Smallpox deaths probably numbered significantly less than one thousand, assuming a case fatality rate of one-in-four (AHSS, Epidemiología, box 11, exp. 1-2). When cause of death statistics become available again in 1918, smallpox mortality in the City had fallen to a mere 140 deaths for the entire year (AHSS, Estadística, box 10, exp. 27).
Unfortunately, what we lack is a general picture for the entire country. Nevertheless from the scattered statistics available it would seem, first, that Spanish influenza, while deadly, was not the biggest killer of the decade, particularly when averaged with the abnormally low number of deaths in the years following. Second, the general mortality level increased greatly due to violence, particularly in the four years when fighting was most intense, 1913-1916. Third, epidemic eruptions and famine—unlike any Mexicans had suffered since the end of Spanish colonial rule (McCaa 2000)—pushed death rates higher still. From the data on Mexico City and Aguascalientes (including its hinterland), an increase in the crude death rate of at least one-fourth above the norm occurred in each of four years over the decade, 1914, 1915, 1916 and 1918.
The best-fitting model for each sex will approximate 1930 census figures for each age group. Contrary to common sense, the inverse method most easily accomplishes this by means of annual projections, rather than simpler quinquennial or decennial computations (McCaa 1993). The preferred inverse projection maps annual fluctuations in crude birth, death and net emigration rates by sex. For the years of greatest crisis, higher mortality and lower fertility are postulated. Appendix I lists the computer instructions required to reproduce the various scenarios explained below using the inverse projection program Populate.
First, the female population is projected with the goal of matching the corrected female age structure in the 1930 census. The best-fitting annual series of births, deaths and net-migration for the female population provides a basis for projecting the male population. Then the male simulation, based on births obtained from the optimal female projection, adjusts deaths, but not births, to fit the 1930 male age structure. Extreme age-heaping in all Mexican censuses before 1960 (Camposortega 1992:94) foils attempts to compare cohorts by five or ten year age groups, so the age structures are presented graphically and slightly smoothed. Figure 1 reveals a surprisingly close fit between the observed and projected census age distributions for females. Indeed, based on the corrected census figures for 1930, for the 4.2 million females under 20 years of age, the absolute error is a trivial six thousand (0.1 percent). For females born in the first years of the revolution (1911-1915, aged 15-19 in 1930), the corrected census figure is 812.4 thousand and the projected is 811.6 while for ages 10-14 the respective figures are 901.6 and 899.6 thousand.
How sensitive is this model to mis-specification? Shifting the crude rates up or down by as little as one point increases error ten fold, to one percent ( ~50 thousand). If a high-pressure demographic regime is hypothesized, with crude rates, say, four points greater, error for ages 0-19 years balloons from -6 to +171 thousand. Decreasing the demographic pressure by four points creates a shortfall of 196 thousand at the same ages and a corresponding surplus at ages 20+. Note that the totals for females and males are matched in each instance, but the age distributions shift substantially with only slight adjustments in the crude rates. Because the favored model closely approximates the under-twenty age structure in 1930 it is superior to alternative models, and therefore, superior in approximating demographic conditions over the period 1910-1929.
For middle-aged and older women (35 years of age or more in 1930) the model is not as successful. In part this is due to the fact that the older groups continue to bear the imprint of the hypothesized age structure for 1895 used to initiate the projection. Beginning a projection prior to 1895 is an exercise in sheer invention because there are no national-level data prior to that year (Camposortega, 1992:12; McCaa 2000). Even for 1895, the published census age structures cannot be used because of odd groupings and pronounced age-heaping. A model age structure is necessary. In 1930, extreme age heaping remains a severe problem. Indeed as late as 1950, on the Whipple index scale of "exact" to "very bad," Mexican age structures for both sexes continued to score a "worst" rating. Only with the 1970 census was a rating of "bad" finally attained by each sex (Camposortega 1992:94).
The biggest disparities in the best fitting models are for females aged 35-44 (93 thousand too few in the model, -8.5 percent). This error is due to a combination of additional deaths and emigration at these ages over the 1895-1930 period, to age-heaping in the 1930 enumeration, and to errors in the hypothesized 1895 age distribution. Without additional data for the pre-1895 period, it is impossible to decompose these errors with much certainty. Since the overall totals match, excesses for one group must be counterbalanced by deficits for others. For females aged 55-59 (+75 thousand) and 65-69 (+35 thousand), age heaping, in this case age-avoidance in the census data is the likely problem, not the projection. The saw-toothed pattern of digit attraction is extreme at these ages with 293 thousand females aged 50-54, but only 164 at ages 55-59, surging to 201 at 60-64 and falling to 87 aged 65-69. For those 75 years of age and older, the number projected is half the census figure (42.5 vs. 87.3 thousand), a clear indication of the exaggerated ages common to the oldest old. For mature women aged 30-34 (588 thousand enumerated vs. 613 projected for females born between 1896 and 1900), the projection does not account for what may have been a more severe thinning of the cohort during the years of the revolution. While the difference of 25,000 amounts to one of the largest substantive errors in the model (at other ages digit attraction or age heaping is the greater problem), it illustrates the uncanny ability of the method to imprint cohorts with birth, death and even emigration histories as they are projected forward through time.
Once the female population is successfully modeled, the results are used to calculate the birth stream for males (1.05 times the estimated annual number of female births), year-by-year from 1895. It is hypothesized that male mortality will be higher than female, and this is borne out by the projection. Both before and after the decade of violence, the difference in crude death rates by sex is a matter of only a point or two, but during the years of greatest fighting and hardship, the best-fitting model requires an increase in the crude death rate above female levels of four points year-after-year. Even so, the projected male age structure does not fit the entire 1930 age distribution. For the youngest age groups (0-14), the model fits almost perfectly. The target is 3.417 million and the projected figure is 3.408 million. For 15-19 year olds (born 1911-15), the model projects 934 thousand, but the adjusted census figure is only 801. Excess deaths of perhaps 50-100 thousand seems the most likely explanation because by peeking ahead ten years to the 1940 census reveals 752 thousand reported for this group (versus 850 projected). For the following cohort (born 1906-10), peeking ahead also suggests heightened mortality. Compared with the model there is a deficit of 140 thousand, and the shortfall also persists in the 1940 enumeration. Redress comes at ages 35+, where the projection falls short of the enumerated population by 200 thousand. The model fails to capture the heightened mortality for teenage boys during the decade of revolution, and instead distributes deaths more evenly among younger and older males. Given the nature of this exercise, rather than tinkering with the model, I prefer to report its blemishes and use them to speculate on the effects of error, mortality, and emigration.
Given the general, often arbitrary character of the inputs used in these projections (Appendix I), it might be surprising to those unfamiliar with the inverse projection methodology that the simulated age structures match the enumerated population so closely, indeed almost perfectly for females below age 35. For males, inflating deaths in the late 1920s for the years of the Cristero rebellion (with a corresponding reduction in the most violent years of 1910 decade) would lead to a better fitting model. Instead, the projections are reported as they are with the proviso that male mortality is probably overstated by perhaps as much as 100,000 in the first decade and understated by a similar amount in the second.
In addition to the new inverse projection scenario, Table 2 summarizes four sets of vital rates postulated for the period—models by Collver (maximum mortality and maximum census error), Mier y Terán, and Greer. All scenarios balance the demographic equation for the period 1910 to 1921, but each does so somewhat differently.
Collver's maximum mortality model places Porfirian crude birth rates in the upper-40s, and for the decade of revolution allows them to decline by 3-6 points. Death rates surge from the low-30s to the high-40s, an increase of 10-15 points. Unfortunately, Collver’s models are not based on a well-founded understanding of the course of the revolution. This should not be surprising because Mexico is allotted a scant thirty pages, in a book devoted to developing base-line crude birth and death rates for all of Latin America from the earliest reliable vital statistics to 1960. Collver was not a student of the Mexican revolution, as evidenced by his statement that “most of the excess deaths probably occurred during two brief periods: the actual physical conflict which began in the fall of 1910, and the epidemic of Spanish influenza of 1918-19” (1965:38). Actual physical conflict and the deaths inflicted thereby were not confined to brief periods, certainly not the Fall of 1910.
In fact the overthrow of Diaz, which occurred in late spring 1911, was accomplished with little violence or destruction. The fighting scarcely began until 1911. The best example is the state of Morelos, where devastation was greatest (Holt Buttner 1962). There the dominant revolutionary chieftain for much of the decade was the legendary Emiliano Zapata. Zapata dallied some four months from November 1910 until March 1911, before finally answering Franciso Madero's call to revolt against Diaz. Three months later, Mexico's aged, six-term President resigned, not so much due to developments in Morelos as to the defeat of a 700 man-contingent of the federal army at the small, but strategic northern border town of Ciudad Juarez. In 1910, Ciudad Juarez numbered fewer than 20,000 inhabitants (Katz 1998:104). Victory at Ciudad Juarez came to the revolutionaries on May 10, 1911, after a siege lasting only a couple of days. Ten days later Zapata achieved his greatest success against Diaz, with the capture of Cuautla (pop. 11,169). Defended by a troop of 400 federales, the town was evacuated after a six-day siege. The classic account does not mention casualties (Womack 1969:84). The pact of Ciudad Juarez was signed on May 21, and Diaz resigned on the 25th.
The real fighting began, when the revolutionaries trained their weapons on one another over the course of the following six years. Indeed the first mutiny against Madero had already occurred, on May 13. Led by the victors of Ciudad Juarez, Pascual Orozco and Pancho Villa, Madero put it down without a single casualty by boldly leaping to the top of a car and appealing directly to the troop (Katz 1998:111).
While Zapata waited four months to rebel against the hated Diaz, not four weeks passed before he rebelled against the enormously popular Madero. In late November 1911, Zapata, tired of waiting to be appointed chieftain of the army of the South, denounced Mexico’s first democratically elected president by proclaiming the Plan of Ayala. While the Plan contained a fig-leaf statement on returning land to villagers whose holdings had been usurped (as well as the ever popular promise of land to revolutionaries and their heirs), most of the document’s 2,000 words was about power—and personality, not to mention perquisites—as a cursory reading of the complete document readily reveals (Womack 1969:394-397). Only in 1912 did serious fighting break out in Morelos. Elsewhere regional bands (and bandits), some with plans, others without, escalated the plundering of the countryside, hamlets and towns. As is well known, in less than two years after Diaz’s resignation the nation slid into chaos. With the assassination of Madero on February 21, 1913—probably at the order of the Madero-appointed commander-in-chief of the federal army, Victoriano Huerta—, civil war erupted in most regions of the country. Deprived of arms by a United States naval blockade, the usurper proved incapable of suppressing the many revolts. The battle of Zacatecas, June 23, 1914, where 6,000 federal troops died, sealed the fate of Huerta, who fled into exile two weeks later.
Now, an even bloodier phase of the revolution began, as, once again, the victors turned on one another. 1915 was the year of hunger. Marauding bands destroyed the few crops that were sown, many before they could be harvested. Destruction continued into 1916, although with the defeat of the northern chieftain Pancho Villa at the battle of Celaya in April 1915, the violence began to wane, however slowly.
Collver's maximum mortality model is based on a cursory understanding of the course of the decade of revolution. Nor does it take into account the not-inconsiderable amount of “robolución,” which plagued Mexicans for much of the decade. Collver's alternative model minimizes mortality by starting the decade with the total population discounted due to fraud and slightly lower birth rates and slightly higher death rates. With revolution, death rates in this model rise fewer than five points, but the fall in fertility is more dramatic—to account for the missing millions. Greer adopted Collver’s minimum mortality model as his own with some slight adjustments.
Mier y Terán offers a substantially different scenario. Porfirian vital rates are much higher, with births at 50 per thousand population and death rates approaching 40. The impact of the revolution is rather slight, with birth rates showing the greatest change. Death rates increase by only a few points.
The inverse projection scenario differs dramatically from others, because this model must balance two complex demographic equations—projecting the female and male population to 1930 using sex-specific crude birth and death rates. Counter-factual projections provide benchmarks to weigh effects of mortality and fertility, by simulating, on the one hand, what might have happened if there had been no revolution (“reform” in Table 2), and on the other, what the effects might have been if the revolution had only impacted mortality or fertility separately (“revolution” death rates and “reform” birth rates for the former and the reverse for the latter). For all projections crude rates for 1910 and 1920 provide anchors for interpolating intervening years. The reform model establishes a baseline—the course of vital rates without war. This model simply interpolates vital rates from 1910 to 1920, assuming no disturbances or change in the direction or pace of trend. The exception is 1918 where the Spanish influenza epidemic as an exogenous factor in the reform model adds four points to the crude death rate for that year (some 60,000 deaths). The revolution model, already discussed above, attempts to portray the likely course of vital rates during the war years. Two sex projections are made in each case.
If there had been no revolution and rates of natural increase had risen slowly over the decade from one percent to 1.4 percent per annum, the population of Mexico would have been 17.3 million in 1921, with normal emigration. The enumerated total was 3 million less, or 14.3 million. Adding 1.1 million to this figure for under-enumeration in 1921 (with a range of +/-200,000 depending upon whether one favors corrections for 1930 by Mier y Terán or Alba), reduces the net loss to 1.9 million (see Figure 2). If fertility had been unaffected by the war, there would have been 550,000 additional children born who would have survived to be enumerated in the 1921 census. This is equivalent to a four point loss in the annual crude birth rate over the decade. Under this scenario total population would have shrunk to 16.7 million. Excess war mortality accounts for 1.4 missing millions, pushing the total to 15.9. In relative terms, one-in-seven deaths over the decade was probably due to the violence and chaos of the revolution. Combining the two effects we arrive at the 15.4 adjusted figure for the 1921 census. In these scenarios the effect of the Spanish flu epidemic in 1918 is discernible, but clearly it was not the biggest killer in the larger picture. If the 400,000 flu deaths estimated by Ibarra (1996:65) is correct, the total is only slightly less than the United States's 550,000, but the population of the U.S. at 103 million was at least six times that of Mexico (Moyner and Garenne 2000). It would seem reasonable to ascribe at least half of Mexico's enormous loss from the flu, if true, to the disorder and weakened state of the population caused by a decade of war. The reform model adds “only” four percentage points to the crude death rate for that year or 60,000 deaths, as the more likely endogenous effect of the epidemic. To appease apologists of the Revolution who favor exogenous epidemics to explain the missing millions, we may, generously, sextuple the Spanish flu effect to 350,000, reducing deaths due to the revolution from 1.4 to 1.1 million. Apologists would be forced to conclude that on the whole endogenous epidemics probably accounted for much less than one-half of the excess mortality over the decade. War-related causes—hunger, violence, and the like—constituted the bulk of the 1.4 million lives lost. Persisting war-induced emigration, as noted above, probably totaled no more than 200,000 over the decade.
Alternatively, minimizing the number of excess deaths (and maximizing lost births to keep the population equations in balance) would halve the mortality cost but double the fertility effect. Unfortunately the high birth loss model cannot match the 1930 age structures. If we ignore this inconvenient fact and embrace the high fertility loss model anyway, excess deaths amounted to about eight hundred thousand (or half that if all Spanish flu deaths are considered exogenous). At the other extreme, the fertility effect could be minimized (and mortality maximized). This scenario, which also fails to produce a good fit with the 1930 population age structures, points to one-half million lost births and 1.7 million excess deaths. These simulations offer a range in which the true losses probably fall. Whichever scenario one chooses, the human cost was enormous: 0.8 – 1.7 million excess deaths and 0.5 – 1.6 million lost births. I favor mid-range estimates of 1.4 and 0.6 million, respectively, because this scenario most closely approximates the population age structures for females and males in 1930.
From the best fitting model, we learn that life expectancy probably fell 10-15 years during the periods of greatest violence, from 30-32 years in 1910 to 15-20 in 1913-1916 (Table 3). Infant mortality would have increased by one-fifth or more in those years, with more than one-fourth of all babies dying during the first year of life in 1915, 1916 and 1918. These basic demographic statistics fail to communicate the depths of the many personal tragedies, but they provide important new details for understanding broad national trends.
Table 3 near hear
For those not persuaded by the unconventional method of inverse projection, consider what a more conventional tool, cohort analysis, has to offer as a means of measuring the impact of the revolution. With this approach, data from the 1910 and 1930 censuses of both Mexico and the United States are combined to study Mexicans born before 1910 and enumerated in either country. To obtain the complete Mexican born population by sex in both years, we sum Mexicans residing in the United States to the population of Mexico.
As an aside, consider that very few Mexicans resided outside these two countries at either date. Mexicans supposedly sought refuge from revolutionary violence in Guatemala, Belize, Cuba and even Spain, but censuses show otherwise. The largest number of Mexicans in any of these countries, some 4,000, resided in Spain, where priests and nuns did indeed seek refuge. The figure is dwarfed by the several hundred thousand Mexicans resident in the United States. Then too, the number of refugees in the United States was never as great as often thought. If a half million “cultured” Mexicans sought refuge in the United States, as the often cited remark by the Mexican Ambassador to the United States asserts (Ordorica and Lezama, 1993, 46), they were not sufficiently cultured to appear on the United States census rolls for 1920. Meanwhile, their several hundred thousand illiterate compatriots experienced no such aversion or problem.
Gonzalez Navarro also questions the half million figure and the emphasis on sanctuary as a cause of emigration. According to Gonzalez Navarro (1994, vol. 3:194), “En la década 1910-1920 según el parecer más generalizado, el bracero emigraba en busca de un mayor salario, para huir de la servidumbre y por falta de garantías.” The fact that the 1920 U.S. census shows New Mexico attracting only a few thousand Mexican born, while California drew more than 50,000 and Texas double even that figure suggests that the giant sucking sound of the 1910s was the United States economy, not the storm of the Mexican revolution (Gutmann, McCaa, Gutierrez-Montes and Gratton, 2000; figures are from Haines, forthcoming). The scale of permanent war-related emigration has been greatly exaggerated, as the cohort analysis shows.
In this cohort analysis, fertility is not an issue, because we focus attention only on Mexicans already born before census day in 1910. In the enumeration of 1930, the survivors from this cohort were twenty years of age or older. Although census dates in the two countries differ by a few months, this distortion is minor given our interest in magnitudes. Likewise enumeration error, whether due to under-counting or misclassification of the Mexican born declaring themselves as citizens of the USA, is unlikely to have a significant effect on the proportions.
The purpose of Figure 3 is, first, to emphasize how small net emigration to the U.S.A was over the two decades, not only in absolute terms but also as a fraction of total population in 1910, and second, to stress how great the excess mortality was. Losses due to mortality were at least double and perhaps triple those due to emigration to the United States—without taking into account losses to those born after 1910. For females, the size of the cohort born before 1910 shrinks by forty-one percent from 7.7 million Mexican-born enumerated in both countries in that year to 4.4 million in 1930 (Figure 3). The figure for males, 44 percent, is significantly worse, with a decline from 7.6 to 4.0 million. As a comparison for the same period, Blacks in the U.S.A. lost 31 percent of their numbers over the two decades, with Black females enjoying a two-point advantage over males (Haines, forthcoming). Comparing U.S. whites is impossible due to the confounding effects of a substantial influx of European immigrants during the period. For Mexicans, a large fraction of the loss was due to higher mortality prevailing in Mexico. In 1910, life expectancy at birth hovered at thirty years, ten years lower than for Blacks in the U.S.A. Indeed, according to the best estimates, not until 1940 did life expectancy at birth reach forty years in Mexico, a figure attained three decades earlier by Blacks in the U.S.A (Arriaga 1968; Haines forthcoming). If, without the revolution, life expectancy had continued to improve only slightly over the decades, approximately 35 percent of Mexican females alive in 1910 and 36 percent of males would have died during the years 1910-1930. Excess mortality, then, amounts to roughly five percentage-points for females alive in 1910 and nine points for males, some 400 thousand excess female deaths and 700 thousand male. These figures do not include children born in the 1910 decade and who died due to revolutionary violence.
In contrast, emigration for the cohort born before the revolution was considerably less. The fraction residing in the United States more than doubled over the period 1910-1921, but the total remained small, some five percent of all female Mexicans born before census day in 1910 compared with 6.6 percent of males. The number of female emigrants born before 1910 and residing in the United States grew from 92 thousand in that year to 207 in 1930. For males the relative increase was similar, from 141 to 285,000. Discounting mortality from the 1910 cohort would point to 130-150,000 net female immigrants over two decades and 200-225,000 males, for a total of 350-400,000. However, emigration in the decade of revolution was, in fact, only slightly more than in the 1920s as noted above, so that halving these figures would be a rough approximation. Then the figures should be halved again to take into account the fact that many Mexicans were drawn to the United States, even during the 1910s, as much by the booming economy of the Southwest as by the search for sanctuary. Of course, both the push of revolutionary chaos and the pull of economic opportunity were at play. To assign half to each, placing war-induced emigration for the pre-1910 birth cohorts at 30-40,000 females and 50-60,000 males, would be less erroneous than to assign the entire sum to a single factor. Thus, the increase in permanently emigrating refugees among those already born in 1910 was probably less than 100,000, 0.7 percent of the population enumerated in 1910. The absolute maximum would be 1.5 percent, assuming that the entire net increase of the decade for the pre-1910 birth cohort resident in the U.S.A., 175-200,000, was due to war.
Cohort analysis can be something of a hammer in the demographer's toolkit. Comparing censuses two decades apart and combining results for two countries is difficult to do well. Conjecturing mortality rates is even more challenging. For these reasons, inverse projection offers an alternative, comprehensive means of assessing not only the total demographic losses over the decade, but also decomposing the losses into mortality, fertility, and migration components.
What inverse projection cannot do is assess the damage state-by-state ("entidad federal" in Mexican statistical parlance). Cohort analysis, on the other hand, can help distinguish where losses were greatest, although excess mortality is indistinguishable from losses due to migration because no table was published for migrants by age. In the worst case, that of Morelos, the total loss exceeded sixty percent for both males and females born before 1910 (Figure 4). Of 90,052 females counted in 1910, only 35,614 were enumerated in 1930. The greatest female losses, with less than 50% of the 1910 cohort enumerated two decades later, occurred in six of thirty-one entitidades--Morelos, Durango, Colima, Guerrero, San Luis Potosi, and Zacatecas. For males five additional entities must be added to the list: Campeche, Queretaro, Nayarit (Tepic), Mexico and Jalisco. In contrast, in-migration led to survival ratios greater than one for Baja California (males) and the Federal District (DF, females; 0.88 for males). Survival ratios of 0.6 or greater for both sexes were attained in only two other entities: Sinaloa and Tampico. Detailed studies such as that for Morelos (Holt Buttner 1962) are needed to assess losses state-by-state and settlement-by-settlement.
Given the magnitude of the human losses caused by the Mexican revolution, the silence of some scholars and disbelief by others is surprising. Indeed, the Mexican revolution is omitted from a recent historical summary of the human cost of modern warfare. Whether this is due to oversight, confusion, or selection criteria (the authors may not consider the Mexican Revolution as an “international” war), of twenty-five wars listed, the Mexican revolution would rank ninth, tied with the Spanish Civil War and surpassed by each of the two World Wars, the Russian Revolution, the Korean and Vietnamese Wars, the Napoleonic Wars (1803-1815), Sino-Japanese War (1937-1941), and the Soviet-Afghanistan war (Clemens and Singer 2000).
In the case of the Mexican Revolution, losses due to emigration was important, but this was not a major factor in the sharp fall of the population. Error, in the poorly enumerated census of 1921, created confusion for those trying to make an assessment, but the availability of high quality data from the 1930 enumeration resolved this problem many decades ago. Fertility effects can be discerned by examining the population below age ten in the 1921 census, or below twenty in 1930. Excess mortality can be estimated by focusing on Mexicans who were enumerated in the 1910 census, that is those who were born before the revolution began. Following this cohort in subsequent censuses provides an alternative means of assaying the demographic devastation of the revolution. This conventional method leads to the same conclusion as inverse projection: mortality costs of the revolution were massive, so great in fact as to be characterized as "implausible" by demographers.
The best two-sex inverse projection to 1930, taking into account the age and sex distribution of the population in that year, points to some three million missing as of 1921. Census error in the 1921 enumeration reduces this figure by one-million. Two-thirds of the remainder was due to one factor: excess mortality (1.4 million deaths), with 350,000 more male deaths than female. Lost births were substantially less at 550 thousand. Smaller still, at less than ten percent of the total loss, was emigration to the United States, with the persisting number of male refugees slightly more than 100,000, and females about three-fourths of this figure. Total persisting emigration was less than 400,000, of which half was probably due more to money than mayhem, the lure of better paying jobs than the flight for safety.
From the best-fitting inverse projection model then, excess mortality is the principal explanation for the missing millions. Ordorica and Lezama (1993) reached a similar conclusion some years ago, although attained by different methods. On the other hand, I estimate fertility and emigration effects to be considerably smaller than they. Our scenarios are members of the same high-mortality family—well-below Collver’s implausible maximum—and are unlike earlier estimates by historians and demographers. From a millennial perspective, the human cost of the Mexican Revolution was exceeded only by the devastation of Christian conquest, colonization, and accompanying epidemics, nearly four centuries earlier.
Alba, Francisco. La población de México: evolución y dilemas. México, D.F.: El Colegio de México, 1977.
Archivo General de la Nación (AGN), Gobernación, Fomento y Obras Publicas, Ramo de Censo y Estadística.
Archivo Historico de Salubridad Social (AHSS), Salubridad Pública. Arriaga, Eduardo A. New Life Tables for Latin American Populations in the Nineteenth and Twentieth Centuries. Berkeley: University of California Press, 1968.
Benítez Zenteno, Raul and Gustavo Cabrera. Tablas abreviadas de mortalidad de la población de México, 1930, 1940, 1950, 1960. México, DF: El Colegio de Mexico, 1967.
Brunborg, Helge. “The inverse projection method applied to Norway, 1735-1974,” unpublished manuscript of the Norwegian Central Bureau of Statistics, Oslo, 1977.
Camposortega Cruz, Sergio. Análisis demográfico de la mortalidad en México 1940-1980. México: El Colegio de México, 1992.
Cardoso, Lawrence A. Mexican Emigration to the United States 1895-1931. Socio-Economic Patterns. Tucson: Univ. of Arizona, 1980.
Clemens, Jr., Walter C. and J. David Singer. "The Human Cost of War: A Historical Perspective," Scientific American, 282:12 (June, 2000), 56-57.
Collver, O. Andrew. Birth Rates in Latin America: New Estimates of Historical Trends and Fluctuations. Berkeley, CA: Institute of International Studies, 1965.
Galloway, Patrick R. “A reconstruction of the population of north Italy from 1650 to 1881 using annual inverse projection with comparisons to England, France, and Sweden,” European Journal of Population/Revue Europeenne de Demographie, Vol. 10:3(1994), 223-74.
González Esparza, Victor Manuel. Jalones Modernizadores: Aguascalientes en el Siglo XX. Aguascalientes, AGS: Instituto Cultural de Aguascalientes, 1992.
González Navarro, Moisés. Población y sociedad en México (1900-1970). México, DF: Facultad de Ciencias Políticas y Sociales Universidad Nacional Autónoma de México, 2 vols., 1974.
-----. Los extranjeros en México y los mexicanos en el extranjero, 1821-1970. México, D.F.: Colegio de México, Centro de Estudios Históricos, 3 vols., 1993-1994.
-----. Historia demográfica de México contemporáneo. El Colegio de México, unpub. ms.
Greer, Robert Gordon. The Demographic Impact of the Mexican Revolution, 1910-1921. University of Texas, unpub. MA thesis, 1966.
Gutmann, Myron P., Robert McCaa, Rodolfo Gutierrez-Montes and Brian Gratton. “Los efectos demográficos de la revolución mexicana en Estados Unidos,” Historia Mexicana, 50:1(jul-sep 2000), 145-165.
Haines, Michael (ed.). Historical Statistics of the United States: Millennial Edition, forthcoming.
----. “The White Population of the United States, 1790-1920,” in Michael R. Haines and Richard H. Steckel (eds.), A Population History of North America, Cambridge, Eng.: Cambridge University Press, 2000, 305-370.
Hall, Linda B. and Don M. Coerver. Revolution on the Border. The United States and Mexico 1910-1920. Albuquerque NM: University of New Mexico Press, 1990.
Holt Buttner, Elizabeth. “Evolución de las localidades en el estado de Morelos según los censos de población,” Anuario de Geografía 2(1962), 9-126.
Ibarra, J. Gabriel. The Spanish Influenza in Mexico, 1918-1919. Washington State University, unpub. MA thesis, 1996.
Instituto Nacional de Estadística, Geografía e Informática (INEGI). Estadísticas históricas. Aguascalientes, 1996.
Katz, Friedrich. The Life and Times of Pancho Villa. Stanford: Stanford University Press, 1998.
Lee, Ronald L. “Estimating Series of Vital Rates and Age Structures from Baptisms and Burials: A New Technique with Applications to Pre-industrial England,” Population Studies, 28(1974) 495-512.
-----. “Inverse projection and demographic fluctuations: a critical assessment of new methods,” in R. Schofield and D. Reher (eds.), Old and New Methods in Historical Demography, Oxford, England: Claredon Press, 1993, 7-28.
Loyo, Gilberto. La políticia demográfica de México. México, D.F.: Instituto de Estudios Sociales, Politicos y Económicos del P.N.R. 1935.
-----. “La población de México: Estado Actual y Tendencias 1950-1980,” in Los Recursos Naturales renovables y el crecimiento demográfico de México, México, D.F.: Procedimientos del Instituto Mexicano de Recursos Naturales Renovables, 1960, 3-151.
Luna Méndez, G. “La natalidad y la mortalidad en el marco de la revolución mexicana,” Revista Mexicana de Sociología, 21(1959):103-126.
Martinez, John Ramon. Mexican Emigration to the U.S. 1910-1930. University of California, unpublished MA thesis, 1957.
McCaa, Robert. “Inverse Population Projection Benchmarks: England, Sweden, and a Modern Standard,” in R. Schofield and D. Reher (eds.), Old and New Methods in Historical Demography, Oxford, England: Claredon Press, 1993, 40-56.
-----. “The Peopling of Mexico from Origins to Revolution,” in Michael R. Haines and Richard H. Steckel (eds.), A Population History of North America, Cambridge, Eng.: Cambridge University Press, 2000, 305-370.
McCaa, Robert and Hector Perez-Brignoli. “Populate From Births and Deaths to the Demography of the Past, Present, and Future” (manual and micro-computer software). Minneapolis: Humphrey Institute Center for Population Analysis and Policy working papers series, 89-06-02, 1989.
McCaa, Robert and James W. Vaupel. “Comment la projection inverse se comporte-t-elle sur des données simulées?", in Alain Blum, Noël Bonneuil et Didier Blanchet (eds.) Modèles de la démographie historique, Paris: Institut National des Etudes Démographiques, Congrès et Colloques Nº 11, 1992, 129-146.
Mier y Terán, Marta. Évolution de la population mexicaine à partir des donnees des recensements: 1895-1970. Montreal, unpublished Ph.D. thesis, 1982.
Moyner, Andrew and Michel Garenne, "The 1918 Influenza Epidemic's Effects on Sex Differentials in Mortality in the United States," Population and Development Review 26:3 (Sept. 2000), 565-581.
Ordorica, Manuel and José Luis Lezama, “Consecuencias demográficas de la revolución mexicana,” in El poblamiento de México: Una visión histórico demográfica. México, DF: Consejo Nacional de Población, 1993, v. 4, 32-53.
Sánchez Rosales, Gabino. “La epidemia de influenza española de 1918; aproximación a una historia compartida México-Estados Unidos,” Boletín Mexicano de Historia y Filosofía de la Medicina, 3:2(2000), 17-25.
United States Bureau of the Census. 15th Census of the U.S. 1930. Population. vol. 2. General Reports Statistics by Subjects. Washington, 1933.
Wachter, K. W. “Ergodicity and Inverse Projection,” Population Studies, 40(1986) 275-287.
Womack, John Jr. Zapata y la revolución mexicana. México: Siglo Veintiuno Editores, 1969.
Wrigley, E. A., Davies, R.S., Oeppen, J. and R. S. Schofield. “Reconstitution and Inverse Projection,” in English Population History from Parish Reconstitutions, Cambridge, Eng: Cambridge University Press, 1997.