The population structure of the Amish, a rapidly growing ethnic religion in North America

Abstract

As Western populations face projected decline, understanding the demographics of high-fertility subpopulations becomes increasingly important. The Amish represent one rapidly growing North American subgroup, yet existing demographic studies are dated and narrowly focused. Here, we use a new population database of >50,000 households—the vast majority of Amish—to offer an up-to-date population-wide analysis that shows high fertility and low mortality and attrition. Specifically, women’s median age at marriage is 20.9, and 87.1 per cent marry by age 50; premarital conceptions are low (4.30 per cent of first births); spacing between marriage and first birth is short (mean 17.2 months); the total fertility rate is 6.1; infant mortality is 5.9; life expectancy at birth is 81.16 years; attrition is low (84.46 per cent retention for those aged 40+); and in-conversion is very low (154 individuals across nearly a century). These definitive population-wide figures open the way for testing predictors of population change and charting how growing subpopulations are shaping regions.

Introduction

Detailed demographic analyses of rapidly growing subpopulations are foundational for understanding the evolution of national population structures and informing tailored interventions across social, economic, and health systems. As many countries consider the potential impacts of projected population decline (Morland 2024; Spears et al. 2024), high-fertility populations within low-fertility settings warrant particular attention, as their cultural and demographic patterns will increasingly impact societies as a whole. For example, studies of ultra-Orthodox Jewish populations, who maintain exceptional fertility rates in both Israel and the United States (US) (Staetsky 2022; Stone 2023), have pointed to the need for specialized economic and educational strategies (Malach et al. 2016). Similarly, research on European Muslim populations has anticipated shifts in religious and cultural landscapes due to their higher fertility, despite it being only slightly higher than that of non-Muslim European populations (Coleman and Dubuc 2010; Hackett et al. 2019). By examining distinct high-fertility populations, demographers can provide the data and analyses needed to develop evidence-based policies that recognize changing demographics. Such studies also serve as the foundation for future work that can use predictive statistics to test theory-informed predictors of fertility change further within high-fertility populations.

In North America, the Amish represent one of the fastest-growing minority populations, with doubling times of just over two decades sustained over the past century. This remarkable growth has occurred within the context of low-fertility societies: the US and Canada, where total fertility rates (TFRs) have hovered around replacement level for half a century. The rapid expansion of the Amish population has significant implications for economic development, land use, cross-cultural relationships, political alignments, population health, disputes over the constitutional free exercise of religion, and rural public services. For example, with minimal formal education beyond eighth grade, Amish young people frequently enter manual jobs, for example, in construction, manufacturing, and agriculture: vital occupations in which highly educated, ageing societies often experience shortages. Their concentrated rural settlement patterns also provide economic vitality in otherwise depopulating regions (Anderson and Kenda 2015).

Using a new database that includes over 50,000 households—sourced from 71 unique Amish population registries (‘directories’) that cover 89.9 per cent of all Amish in North America—we provide an up-to-date population-wide examination of Amish demography that moves well beyond the sample sizes of previous studies. Our analysis of the Amish population structure is guided by the question: What population characteristics underlie Amish growth? As a framework for analysis, we use the balancing equation for population change (Preston et al. 2001):

$$P\left(t\right)=P\left(0\right)+B\left[0,t\right]-D\left[0,t\right]+I\left[0,t\right]-E\left[0,t\right],$$ (1)

where P(t) represents the population alive at time t; P(0) is the population at time 0; B denotes births between time 0 and t; D represents deaths between time 0 and t; and I is in-migration (or any type of population entry) and E outmigration (or any type of population exit) between time 0 and t. Around these four measures of population change, we present standard demographic indicators, including age at marriage, proportion ever married, age-specific fertility rates (ASFRs), TFRs, birth spacing, twinning rates, infant mortality, life expectancy, attrition/exit, and in-conversion. Our findings reveal that the Amish population sits in a middle, seemingly stalled, stage of the demographic transition (Caldwell 2006), where fertility remains high (with a low age at marriage, high proportion ever married, closely spaced births, and low proportion childless) but mortality is low and life expectancy high. Attrition is low, but so is in-conversion. Our calculations offer compelling evidence that the Amish population structure remains optimally poised for extremely rapid population growth due to high birth rates and high retention.

In describing the Amish population structure, we also acknowledge the importance of: (1) studying internal variations, historical trends, motivators for maintaining high fertility, and comparisons with other populations; and (2) using predictive, longitudinal, and inferred causal statistical methods that would allow theories to be tested. However, these areas are beyond the scope of this particular paper. In focusing on providing population-wide figures that articulate the Amish population structure, we open up avenues for such research. Future research will be better poised to address questions such as: What social structures and cultural processes predict this population structure? How can subpopulations significantly influence and reshape national demographics? And, how does population growth impact place, society, and public policy? Our contribution here is a foundational descriptive demographic study that uses up-to-date population-wide data to describe the population structure behind rapid growth (the descriptive ‘How?’) but does not posit explanations for observations or variations (the theoretical ‘Why?’).

Background

Defining the Amish population

To analyse a population’s growth, we must first define the population, determining inclusion criteria for individuals. We identify ‘Amish’ individuals as individuals who are listed in Amish directories and have not left the all-encompassing identity of Amishness. Such relative clarity of Amish identity is the product of a ‘thick culture’ that hybridizes ethnic and religious dimensions widely recognized in Amish theories of peoplehood (Anderson 2025) and is comparable in this respect to Orthodox Judaism (Pinker 2021).

Ethnically, Amishness represents a shared history tracing back to sixteenth-century Anabaptist dissenters in Swiss/German Europe, with an in-group language (Pennsylvania Dutch) and ethnic institutions (Enninger 1986). Most Amish migrated to North America in the eighteenth and early nineteenth centuries, and endogamy reinforced their ethnic distinctiveness. Religiously, Amish ‘adherence’ means adolescent or adult baptism (denoting church membership) in individual congregations, which are the locus of explicit religious behaviours, authority structures, and unbroken institutional succession. However, adherents also include members’ unbaptized offspring who still live at home (Petrovich 2022, 2023a). Offspring who decline baptism tend to leave home—and Amish life—from late adolescence to the mid-20s.

The Amish have experienced periods of significant attrition and schism, particularly in the nineteenth and early twentieth centuries. However, recent decades have seen lower rates of attrition and no seismic schisms; this has contributed to their current rapid population growth. The intertwining of ethnic and religious identity provides a clear demarcation for our study population, as those who leave the institutional setting of the church as adults are no longer viewed as fully Amish, even if they maintain some ethnic identity.

Recent Amish population growth rates

Inspired partly by Eaton and Mayer’s (1953a, 1953b, 1954) mid-century studies of the Hutterites (a religiously related Anabaptist tradition), scholarly efforts to document the structure of Amish populations began in earnest in the 1950s and ‘60s (Cross 2023). Fusing the interests of demographers and geneticists, these pioneering studies were citation network ‘main path’ publications that inaugurated the now predominant focus of population health in Amish studies (Anderson and Potts 2021). The pioneering researchers recorded Amish population growth rates close to peak-Hutterite levels, at a time when Hutterite fertility was beginning to decline (Sato et al. 1994). The Hutterite doubling time reached 17 years in the early twentieth century (Pratt 1969). Estimates of Amish population growth—calculated from various periods across the mid- to late twentieth century and in varying Amish settlements—have included:

• a doubling time of 23 years, representing an increase of 145 per cent each generation (Cross and McKusick 1970; Cross 1976);

• an annual growth rate of 2 per cent (Landing 1970a);

• a doubling time of less than 25 years, representing an annual growth rate of 4 per cent (Meyers 1991);

• doubling times for two different cohorts of 23 years (annual growth rate of 3.1 per cent) and 29 years (2.4 per cent) (Greksa 2002);

• doubling times of 22.5 and 20.5 years, respectively (Cooksey and Donnermeyer 2004; Donnermeyer 2015).

Historically slow population growth

This fast growth is unlikely to have occurred historically. Until around the mid-1900s, the Amish experienced higher rates of attrition and institutional schism. In the early years of the US, energetic pietistic groups, such as the German Baptists and Wesleyan Methodists, drew many adherents, causing Amish attrition (Cronk 1981). Then, as the mass appeal of evangelical Protestantism swept across North America in the nineteenth century, theological differences so fractured Amish churches that annual meetings of leaders—unprecedented in scope and frequency—were called from 1863 to 1878. Through these meetings, approximately two-thirds of Amish adherents opted to follow the trajectory of the change-oriented Amish Mennonites, while one-third remained in the Old Amish (the focus of this paper) (Yoder 1991; Reschly 2000). Amish-Mennonite borrowing of evangelical structures signalled an erosion of sectarian identity, which was more than complete by the mid-1900s (Bair 1952). The Old Amish experienced a series of further schisms in the 1910s–40s and then the 1950s, when sizeable minorities—perhaps 20–30 per cent—withdrew to join the Beachy Amish-Mennonites and Fellowship movements, respectively (Beachy 1955; Anderson 2011). Simultaneously, Old Amish migrants were starting new but tenuous settlements across the US, but the failure rate was high, at least during the 1940s. Individuals either returned to historical Old Amish settlements or switched to non-sectarian religions (Landing 1970b; Luthy 1986).

While switching out of Amish denominations continues, recent attrition rates are lower, and leaving is primarily an individual or household action (Meyers 1994; Greksa and Korbin 2002; Hurst and McConnell 2010). No major national schisms producing a non-Amish denomination have occurred since the 1950s, despite Amish-internal schisms persisting (Petrovich 2017). Furthermore, Amish mortality rates appear to have declined in parallel to those of non-Amish populations in North America (Smith 1960; Ericksen et al. 1979; Acheson 1994; Hewner 1997; Hewner 1998; Colyer et al. 2017). More recent Amish cohorts seem to match the life expectancies of Western non-Amish cohorts (Mitchell et al. 2012).

In summary, since the mid-twentieth century, no large-scale schisms have pushed subpopulations out of the Old Amish, individual/household attrition seems to have declined, and mortality rates appear to have dropped in parallel to those of non-Amish Western populations.

The driver of growth: High fertility

When researchers discuss Amish growth, they describe astonishingly high birth rates over a long period (Colyer et al. 2017; Lutz 2017). To underscore the sheer persistence of high fertility, we go back to two mid-century studies. In the 1950s, Elmer Lewis Smith conducted the first major study of the Amish population structure. He creatively compiled various public and field records to calculate demographic figures for the Amish in Lancaster County, Pennsylvania (PA). He found that: 1.4–5.1 per cent of couples were childless (depending on cohort); completed fertility ranged from 7.4 to 8.2 children per woman and actually increased in more recent cohorts; more than one-third of families had 10 or more children; 52.4 per cent of the population was under age 18; infant mortality rates (IMRs) had declined from 165 per 1,000 in 1890 to 45 per 1,000 in 1949; and the sex ratio (males per 100 females) ranged from 102.4 to 104.5 (Smith 1960). Shortly thereafter, Harold Cross and co-authors embarked on a groundbreaking population data collection effort in the Holmes County, Ohio (OH), Amish settlement (see Cross 2023). Their population data comprised 9,724 individuals in 1,933 families. For the Holmes County Amish, Cross found: a crude birth rate of 33.3 per 1,000 compared with 21.2 nationally; a population with 51.8 per cent aged under 20; a ratio of young children (aged 0–4) to women (20–44) that was 48.7 per cent higher among the Amish than among rural whites in the north-east; completed fertility for married women of 6.33 children, with 34.8 per cent having 10 or more pregnancies; a mean birth interval of 19.9 months; an almost balanced sex ratio (100.3), and a noticeably high twinning rate (Cross and McKusick 1970; Cross 1976). Studies since Smith’s and Cross’s—including the first and only multi-community analysis of the four largest Amish settlements (Ericksen et al. 1979)—have presented comparable figures, although their diverging methods and measures make cross-study comparisons difficult (Colyer et al. 2017).

Why have such high birth rates persisted? Several schools of thought exist. First, gender scholars—including historians and sociologists—have argued that Amish families maintain a diversified agrarian-style division of household labour, where Amish mothers make meaningful contributions (both on the farm and in the home) and manage children’s labour to generate value. This familial emphasis makes children desirable and also supports the existence of large families (Wright 1977; Olshan and Schmidt 1994; Jellison and Reschly 2020). Second, cultural anthropologists have argued that the Amish represent a holistic society, where individuality is subjected to shared/collective interests, notably the family and extended family, which are the primary social institutions of Amish life. Because holistic societies are characterized by thickly intertwined role relationships and because inheritance is determined through kin relationships, the social system rewards large families (Nagata 1968 [1989]; Hurd 1985a; Wenger 1991; Škender 2020). Finally, an economist has recently argued that Amish-specific religious rules latently channel the utility of children towards benefiting parents and older generations (Choy 2020).

Tests of these hypotheses—and others based on competing Amish peoplehood theorizations (Anderson 2025)—would greatly benefit from definitive population-wide Amish population structure figures, which currently do not exist. Existing Amish population studies are pilot studies because they are confined to particular locales and provide uneven analysis of the population structure. In this paper, we move beyond existing region-specific pilot studies by collectively synthesizing existing studies and identifying what we expect to find for today’s Amish, should past findings hold. We then use recent vital data that represent the vast majority of the population to present new figures for the Amish.

Literature review: Decomposing the Amish population structure

Following the balancing equation stated earlier (equation (1)), in this section we synthesize demographic research on Amish fertility, mortality, in-conversion, and attrition. Although existing demographic studies of the Amish are limited in geographic coverage and difficult to merge quantitatively into a single set of findings because of methodological diversity, they collectively provide a helpful foundation for identifying key components of Amish population growth. Thus, the purposes of this review are: (1) to identify existing calculations describing the Amish population structure, to inform this study’s calculations and optimize comparisons; and (2) to provide a benchmark for comparing this study’s comprehensive findings against past region-specific studies.

This section names numerous Amish settlements in North America, including the four largest (all in the US): the regions around Holmes County, OH; Lancaster County, PA; Northern Indiana (Elkhart and LaGrange Counties [referred to as Elkhart-LaGrange]); and Geauga County, OH. For profiles of these and other settlements, including maps, history, and social topography, see Anderson and Bacon (2024). Additionally, in reviewing these studies, we note that data collection generally occurs several years before publication unless otherwise specified, as Amish population analyses typically use the latest data.

High fertility

Maximum theoretical fertility exists when women begin childbearing soon after first menstruation and continue bearing children until menopause. This represents a span of over 30 years, from age 12–15 to 45–50, during which a child could be born approximately once a year (Henry 1961; Bongaarts 1978). While this scenario is hypothetical, it identifies the conditions under which populations can achieve relatively high birth rates. These include low age at nuptiality, childbirth across a woman’s full reproductive life lifespan, a high twinning rate, and short intervals between births.

Nuptiality structures.

Growing populations typically exhibit triangular-shaped population pyramids, representing age distributions skewed towards younger ages and balanced sex distributions. In monogamous cultures, marrying young maximizes the number of years a woman is at risk of pregnancy and minimizes generational turnover (i.e. the time until one generation’s children are able to start producing children themselves), thus contributing further to population growth. Several components underlie the Amish nuptiality structure, including age at marriage, premarital conception rates, sex ratios, and marriage stability.

Age at marriage.

In cultures that discourage marriage at very young ages due to psychological and physical development concerns (Schlegel and Barry 1991), the early 20s represents a reasonable marriage age for optimizing the full use of a woman’s reproductive lifespan. The Amish generally view the early 20s as the age appropriate for marriage; before age 18 is generally considered too early (Škender 2020). Previous studies have consistently found mean ages at marriage among the Amish to be in this range, although with some variation by time and settlement: 21.4–24.2 for males and 20.8–22.8 for females (Smith 1960; Markle and Pasco 1977; Ericksen et al. 1979; Hurd 1985b; Meyers 1991; Acheson 1994; Greksa 2002; Wasao et al. 2021).

Premarital conception.

Religious proscriptions among the Amish negatively sanction sexual intercourse outside marriage, especially for young people who do not keep themselves morally and ceremoniously pure before marriage (Škender 2020). Previous research has documented relatively low rates of premarital conception: 4.1 per cent in Lancaster in the early 1900s (Smith 1960) and 7 per cent in Holmes in the early to mid-1900s (Cross and McKusick 1970). Although permitting sexual intercourse outside marriage could potentially increase the birth rate, the net effect may work inversely. Because Amish value systems stigmatize premarital conception, Amish individuals pursue marriage in their young adult years, which results in most fecund women being routinely exposed to the risk of conception.

Sex ratio and marriage availability.

A balanced sex ratio optimizes monogamous marriage options. Studies have found generally balanced sex ratios among the Amish: 102.4–104.5 in early-twentieth-century Lancaster (Smith 1960); 100.3 in Holmes in the early to mid-twentieth century (Cross and McKusick 1970); and 100.0 in Geauga in the late twentieth century (102.0 for ages <15 and 84.0 for ages ≥50) (Greksa 2002).

Marriage stability and remarriage.

A married couple’s risk of conception is directly related to frequency of sexual intercourse and marital stability. While no strong research has documented Amish couples’ frequency of intercourse, marital separation is uncommon and divorce is rare; divorce is typically initiated by a spouse leaving the Amish. Remarriage while a former spouse is still alive is forbidden; however, widows and widowers may remarry. No research has documented the proportion remarried, but this practice potentially extends the fertile period, especially when widowers marry younger women.

Based on this review of nuptiality structures among the Amish, if findings from past regional studies hold for the entire Amish population in more recent years, we would expect to see the following:

• 1The sex ratio, specifically for those aged 15–50, will be nearly even, optimizing monogamous marriage options, and the age distribution will be disproportionately younger.
• 2The mean age at marriage for females and males will be in the early 20s, with low standard deviations.
• 3The percentage of first births within eight months of marriage will be ≤7 per cent.
• 4The population-wide proportion ever married will be high, say, >90 per cent.
• 5Couples will rarely divorce; widow(er)s will remarry quickly, where possible.

Birth quantities and timing.

High birth rates among the Amish have been well documented. Estimates of completed marital fertility from older studies include: 6.33 for Holmes in 1964 (Cross and McKusick 1970); 6.8 for Lancaster, Holmes, and Elkhart-LaGrange among <1928 birth cohorts (Ericksen et al. 1979); 6.92 and 7.67 for Elkhart-LaGrange in 1980 and 1988, respectively (Meyers 1991); 7.6 for Lancaster in 1988 (Dorsten 1994); 7.8 for Geauga in 1988 (Acheson 1994); and 7.7 for Geauga in 1993 (Greksa 2002). Other studies have reported the crude birth rate for the Holmes area Amish population at 33.3—compared with 21.2 for the US in the same year (Cross and McKusick 1970)—and the mean family size for couples (regardless of parental age) at 5.3 in 1988 (Wasao and Donnermeyer 1996) and 4.85 in 2015 (Wasao et al. 2021).

At the extreme ends of fertility ranges, couples without children remain rare, while very high fertility and childbirth at ages 40+ are common. Childlessness among women in the early to mid-twentieth century has been estimated at 1.4–5.1 per cent in Lancaster (Smith 1960), 3.8 per cent in Holmes (Cross and McKusick 1970), and 4.4–6.6 per cent in Holmes, Lancaster, and Northern Indiana combined (Ericksen et al. 1979); in mid-to-late-twentieth-century Geauga, estimated childlessness ranged from 3.0 per cent (Acheson 1994) to 3.2 per cent (Greksa 2002). Conversely, around 20–35 per cent of women in early-to-mid-twentieth-century cohorts with completed fertility had ≥10 births, specifically: close to one-third in Lancaster (Smith 1960); 34.8 per cent in Holmes (Cross 1976); and around one-fifth in Holmes, Lancaster, and Northern Indiana combined (Ericksen et al. 1979). Among mid-to-late-twentieth-century cohorts in Geauga, 27 per cent (Acheson 1994) or 28 per cent (Greksa 2002) had ≥10 births. Such high percentages suggest that childbirth at older ages—late 30s to early 50s—was common. Mean age at last birth was 37.57–39.40 in Northern Indiana (Markle and Pasco 1977). In Geauga, mean age at last birth was 37.5, 7.6 per cent of all births were to women aged 40+ (Acheson 1994), and 28.1–50.0 per cent of all women—depending on cohort—gave birth in their 40s or 50s (Greksa 2002).

Twinning rates contribute to population growth, both immediately by increasing the absolute number of births per pregnancy and by compounding growth across future generations. Multiple birth rates have been calculated for Holmes as 15.3 per 1,000 maternities (Cross and McKusick 1970) and for Geauga as 12.2 (Acheson 1994) or 13.9 (Greksa and Korbin 2002). For comparison, the global twinning rate has risen from 9.1 in the 1980s to a historical high of 12.0 in 2010–15, while the North American rate grew from 9.9 to 16.9 over the same period, due presumably to delayed childbearing and medically assisted reproduction (MAR) (Monden et al. 2021). By historical standards, Amish twinning has been high.

Following this review, if past Amish trends hold up to the present, we would expect to find that:

• 6The TFR will be between six and eight children per woman.
• 7Women will have their last child in their late 30s, on average.
• 8The percentage of childless couples will be low: between 1.4 and 6.6 per cent of all couples.
• 9Twinning rates will be between 12.2 and 15.3 per 1,000 maternities, which is higher than the US’s pre-MAR average but below current twinning rates.

Birth intervals.

To achieve high fertility, women must maintain short intervals between births. A minimum-length birth interval would consist of a nine-month period between conception and birth plus a period of postpartum infecundity ranging from three to 24+ months, depending on factors such as health and breastfeeding (McNeilly 2001). Among the Amish, prior research has found short intervals between marriage and first birth: 19.4 months in mid-to-latter-century Lancaster (Dorsten 1999); 19.9 months for women without premarital conceptions in early-to-mid-century Holmes; and 14.1–25.1 months across early-to-mid-century cohorts in Northern Indiana (Markle and Pasco 1977). The interval between first and second birth is also short: 18 months in Geauga (Greksa 2002) and 18.2 months in Lancaster (Dorsten 1999). Thereafter, spacing increases gradually for subsequent births (Greksa 2002) at least up to the sixth birth (the modal number of births in the Holmes study) (Cross and McKusick 1970). Typically, the longest interval is between the last two births (Cross and McKusick 1970).

Accordingly, we would expect that if past findings hold, Amish women will have short intervals between all births, although the interval will gradually increase after each birth. Specifically, we anticipate that:

• 10The mean interval between marriage and first birth for women who do not have a birth before eight months of marriage will be around 20 months.
• 11The mean interval between first and second birth will be around 18 months.
• 12Mean intervals will gradually increase up to the sixth birth, although all intervals will remain under, say, 36 months.
• 13The last birth interval will be the longest.

Mortality

Until recently in history, high infant mortality has limited life expectancy. Since then, recent advancements in nutrition and public management of disease, as well as numerous non-proximate influences of socio-economic development, have increased infant survival and life expectancy (McKeown and Record 1962; Preston 1975; Soares 2007). For the Amish, twentieth-century infant mortality has been relatively low. From 1890 to 1949, mortality for children aged under five declined from 165 to 45 per 1,000 among the Lancaster Amish (Smith 1960). More recent studies have calculated IMRs (for children under age one) at 20.9 per 1,000 (Ericksen et al. 1979) and 30 per 1,000 (Greksa 2002). In Geauga, the stillbirth rate ranged from 5.2 to 17.1 per 1,000 for cohorts born between 1948 and 1988, while the combined perinatal and infant mortality rate ranged from 15.6 to 32.8 per 1,000; around 25 per cent of women reported at least one prenatal, postnatal, or child death (Acheson 1994).

Life expectancy has also increased as older people live longer and deaths are more and more concentrated at the upper limits of human longevity (Wilmoth and Horiuchi 1999; Canudas-Romo 2010). Although few studies have addressed Amish longevity and mortality, several have suggested that Amish life expectancy is similar to or slightly higher than for the non-Amish. Analysing death certificates from the Lancaster, Holmes, and Northern Indiana settlements, Hamman, et al. (1981) found that Amish and non-Amish mortality rates by age group were very similar. Another study (Mitchell et al. 2012), using the geographically diverse Anabaptist Genealogy Database (Agarwala et al. 2001; Agarwala et al. 2003; Lee et al. 2010), found that among members of 1890–1921 birth cohorts who lived to 30+, Amish men were expected to live around three years longer than non-Amish men, while life expectancies among Amish and non-Amish women were similar. Finally, Hewner’s exhaustive study of the Amish population in Conewango Valley, New York State, produced an attractively low age-standardized death rate of 7.6 per 1,000, which was 19.6 per cent below the US rate (Hewner 1997).

Accordingly, we expect Amish mortality rates and life expectancy to be similar to or better than non-Amish figures. Specifically:

• 14The Amish IMR will be under 5.0 and at or below the non-Amish rate.
• 15Amish life expectancy will be at or above non-Amish life expectancy.

In-conversion

Both births and in-migration (i.e. religious in-switching) drive population growth. Scholars generally assume that in-conversion to the Amish is rare (Scott 2007; Anderson 2016), although no quantification exists. Multigenerational religions tend to experience their highest in-conversion in the first generation and then settle into patterns of growth via retention of offspring (Scheitle et al. 2011). That the Amish are also a non-proselytizing ethnic religion suggests in-conversion to be doubly improbable. Prospective converts must first take interest in the religion: interest in joining the plain people—who include cousin groups to the Amish—certainly exists but for many reasons may not translate into action (Anderson 2016). If seekers take action, they must move into a community and work to establish ties and trust. They must also overcome numerous cultural barriers to be accepted, including German language acquisition, adoption of a whole universe of subtle symbolic identifiers, and navigation of unending minutiae of how to think and behave (Enninger 1984). Finally, they must overcome the long-term fatigue associated with living in a different culture (Scott 2007).

For these reasons, we expect religious in-conversion to contribute negligibly to Amish growth. Specifically:

• 16The percentage of Amish who joined from non-Amish backgrounds will constitute a very small percentage of the total population: less than 0.5 per cent.

Retention/attrition

Exits from the Amish are concentrated in birthright adherents’ young adult years. This occurs largely prior to individuals being baptized into the Amish church (Mazie 2005; Nisly 2006; Faulkner and Dinger 2014), as leaving after baptism incurs formal social sanctions (Petrovich 2017; Petrovich 2023b). Relatively low rates of exit have been evident over a number of decades (Stoltzfus 1977; Ericksen et al. 1979), amplifying the impact of high fertility on population growth (Lutz 2017). In Northern Indiana, exit rates decreased from 21 per cent for the 1930s birth cohorts to 5 per cent for 1960s cohorts (Meyers 1994), whereas in Geauga they decreased from 30 to 5 per cent for the same cohorts (Greksa and Korbin 2002). Around 1970, in the three main settlements combined, fathers employed in farming retained 97.2 per cent of their grown children in the Amish, while non-farmers retained 87.4 per cent (Ericksen et al. 1979). More recently, Friedrich (2001), Hurst and McConnell (2010), and Choy (2020), respectively, calculated retention by denomination in the Holmes settlement as: New Order 53/67/64 per cent; Mainline Old Orders 85/>75/86 per cent; Dan Church 95/97/95 per cent; and Swartzentruber –/~90 per cent/– (ordered from least to most strict denomination). Despite variation across these studies, evidence suggests that exit rates have universally declined and are currently low. Accordingly, we expect that:

• 17The recent percentage of current Amish adherents’ adult children who have remained Amish will be >85 per cent.

Amish population data and method

The Amish develop and periodically update commercially available directories: population record books that provide extensive documentation of vital events and sociocultural information at the individual, household, and settlement levels (see Figure 1 for example entries). Directory information for each household generally includes: names of household members and immediate relations; dates of birth, marriage, childbirths, and death of first deceased spouse; geocodable street addresses; social variables (e.g. occupation of household head, denominational affiliation, standing in church, ministerial status, and date of ministerial ordination); and, for some, household migration history or complete community death lists.

Figure 1.

Examples of possible (not actual) entries from Amish directories

Source: Authors’ own.

Directories are impressively accurate. Directory data are collected by adherents, who provide households with surveys then compile the information into a book. Because fellow Amish request information for culturally relevant reasons, nearly all households respond. Strong group expectations for accuracy create social incentives for careful compilation. Additionally, the periodic publication of multiple editions allows for error correction over time. The internal consistency of vital events data provides further validation: birth dates, marriage dates, and birth intervals must align logically across thousands of households, making patterned errors or fabrication virtually impossible.

First appearing in the 1960s, these directories have become standard reference works for Amish adherents and have been widely used as reliable data sources in demographic research (Colyer et al. 2017; Cross 2023). The first author’s recently completed Cross-sectional Amish Population and Environment Database-2010s (CAPED-2010s) includes data from 71 commercially available unique current editions obtained through Amish stores and mail order. Directories are dated between 2009 and 2020 (median year 2015) and together cover approximately 89.9 per cent of all Amish in North America, as evidenced by comparing the CAPED-2010s US population count for 2010 with the count from the 2010 U.S. Religion Census (Grammich et al. 2012).

This paper’s first author conducted data entry and cleaning with 102 undergraduate research assistants across three years. Extensive internal validation procedures were implemented during compilation, including internal consistency checks and cross-referencing across multiple sources to identify and correct any discrepancies. For example, our extensive data validation procedures investigated spousal ages at marriage, birth spacing, and generational alignment, and we investigated and corrected cases that revealed inconsistencies, cross-referencing households and consulting older and newer editions, among other Amish and publicly available resources. External validity of the CAPED-2010s database has been demonstrated through cross-comparison with American Community Survey data: both sources produced remarkably consistent fertility statistics including TFRs, ASFRs, and birth timing patterns (Stone et al. 2025). This convergence between entirely independent data sources—one survey-based, one administrative—provides strong evidence for the reliability and representativeness of the CAPED-2010s.

Our population comprises all individuals listed as Amish at the time of directory compilation. As such, we have a naturally defined study population (current Amish adherents), where non-Amish are naturally excluded as an inherent characteristic of the data. Given the cross-sectional nature of our data, we employ period-specific measures rather than longitudinal or event history analysis methods that require sensitivity analyses for censoring biases.

CAPED-2010s units are households, not individuals, reflecting how directories communicate data. Before starting any analyses, we removed households with duplicate records (i.e. households appearing in two different directories) and those with extensively incomplete data (namely, from those few directories providing only geographic and/or relational data, no vital statistics). When individual-level units were needed—namely, to produce the age/sex structure displayed in our population pyramid—we transposed data to individual level by individually extracting household heads, their current wives, and children of current and previous spouses to a list. To determine an individual’s sex (information not explicitly provided in directories), we coded all married household heads as male and wives as female, reflecting the directory’s own structure. Given names among the Amish are almost universally specific to males or females (reflecting a religious position), so the sexes of unmarried household heads and a household’s children were determined by given name and/or, for married children, their spouse’s name. We coded the full list of given names as referring to either males or females. We were able to verify our method by analysing married children, where in this population a child’s sex must be the opposite of his or her spouse’s sex.

To construct a current population structure, the population in each directory was determined as at the time of publication of the directory minus two years, as directories provide data collected before the actual publication date. We excluded from the current population structure all individuals with a date of death (including stillbirths) and those coded as having left the Amish. Finally, to avoid cases of individual duplication (i.e. a person being counted as a child in their birth household and also as a head/wife in their independent household), we removed all cases of children in the birth household who were listed as married, as they will appear elsewhere in an independent household. Because a minority of households in CAPED-2010s are from directories without information on children’s attrition and marriage, we estimated these variables for those households, then conducted robustness checks, as described in the Results section. Our final data set contains 263,158 individuals from 50,065 Amish households in 67 directories.

For many calculations here, we reference a specific 2002–11 cohort whose members experience a certain event (e.g. marriage) during that period, and we present measures from that period. This period represents the most recent decade of data where the CAPED-2010s (which contains directories published at uneven intervals across the 2010s) has complete data. This cross-sectional, snapshot-in-time approach means that households that left the Amish prior to data collection—but belonged at some point during this period—are excluded entirely, while converts are included from their time of joining onwards. We acknowledge that this 2002–11 cohort approach has methodological limitations: ideally, rates would account for the timings of entry and exit to calculate accurate person-years of exposure. However, the directory data do not provide dates of conversion or departure, making proper censoring impossible to calculate. Our fertility and mortality rates therefore represent the demographic experience of the current Amish population, on the understanding that some bias may exist if the demographic patterns of leavers or converts are systematically different. We believe this represents the best possible analysis, given the available data structure, and our primary interest is in characterizing the current population’s demographic profile rather than tracking individuals longitudinally. Furthermore, the relatively low attrition and extremely low numbers in-converting (shown later) suggest that potential bias from our cross-sectional approach is minimal.

Our analysis of the CAPED-2010s uses standard demographic measures, including mean age at marriage, ASFRs, TFRs, birth intervals, and IMRs (Hinde and Demétrio 1998; Preston et al. 2001) and also life-table indicators, such as life expectancy at birth (Newell 1988). Some measures required two additional data compilation efforts beyond the CAPED-2010s.

First, directories do not allow life expectancy to be calculated directly. Even though the date of death is recorded for individuals within a household who have died, a household will disappear from a directory once both spouses are deceased; any rates calculated from the figures of a first spouse’s death alone will underestimate lifespan (i.e. the death date of the first deceased spouse is often recorded but not the second). Some directories provide an appendix of all deaths over time, which can be used as an alternative, and some communities have obituary books available. Obituary books are Amish-produced verbatim reprints of Amish-specific local newspaper obituaries. The rural, small-community settings and inclusion of even brief obituary notices for day-old deaths suggest extensive coverage of Amish deaths. From these sources, we compiled complete deaths data for four large Amish settlements—Holmes County (OH); Daviess County, Indiana (IN); Arthur, Illinois (IL); and Geauga County (OH). Our compiled deaths data include 1,213 deaths of Amish individuals that occurred from 2002 to 2011. To ensure that the numbers of deaths in different age categories do not fluctuate randomly, we calculated the average age-specific number of deaths across this period. Whereas these 1,213 data points comprise our information about deaths, the CAPED-2010s provides population data for calculating mortality rates and, ultimately, our life table. We describe robustness checks in the Results section.

Second, we used mixed methods to quantify in-conversion. Given the small number of converts, we decided to identify all conversions across the past century or so, to establish that in-converts contribute minimally to Amish population growth. We first identified non-Amish surnames in the database, for the household head and using wives’ maiden names. We then shared this list with two knowledgeable Amish converts who further helped us identify: (1) additional converts missing from the directory (e.g. those who were deceased, departed, or whose settlement was not represented in a directory); (2) first-generation converts as our individuals of interest, separate from subsequent generations that carry non-Amish surnames but are not considered converts; and (3) converts who happen to have an Amish surname (e.g. Miller). We opted not to include people converting from closely related plain Anabaptist groups, such as Low German Mennonites and Amish Mennonites, due to the occasional fluidity between such closely related groups and some difficulty in identifying them by surname, but we expect their numbers to be, at most, as low as for non-Anabaptist converts. We also excluded converts—largely rural peasants—to two Amish churches in Honduras (1970s) and five in Paraguay (1970s–90s) due to difficulties in counting them and their location outside the standard North American Amish areas. They were likely concentrated in two congregations (La Laguna, Honduras, and La Belleza, Paraguay) and certainly numbered <20 individuals/households combined. To summarize, our tally of in-converts is derived from a separate data effort/analysis from our fertility and mortality analyses, which use all Amish from the CAPED-2010s, regardless of population entry path.

Results

High fertility

Nuptiality structures.

We expected the Amish age–sex distribution to display a classic population pyramid shape: that is, a broad base that narrows with each increase in age group, illustrating a fairly symmetrical sex ratio and population sizes always larger in younger than older cohorts (with reference to expectation #1 from the Literature review section). Indeed, Figure 2 shows that the Amish population pyramid is this shape, confirming our expectation. The shape represents continuous population growth across several cohorts without noticeable generational change. Because the calculation of population pyramids requires the conversion of database households to individual level, and because knowing which household children have left the Amish or are now married is critical for accurate conversion, as described in our Methods section, we estimated children’s marriage and attrition for directories without this information. To validate our approach, we conducted a sensitivity analysis comparing the full data set with data from only those directories containing complete individual-level information (n = 114,324); both populations exhibited nearly identical age–sex structures, confirming the robustness of our population pyramid calculations. Of note, none of our subsequent analyses need to rely on this estimation method, except for estimating attrition itself.

Figure 2.

Amish population pyramid, 2002–11 snapshot

Source: Authors’ analysis of CAPED-2010s data.

For couples whose marriage date falls between 2002 and 2011 and where the wife married before age 50 (the end of reproductive age for women, after which the population structure is not affected by marriage), the mean age at first marriage was 21.6 years (Table 1), closely matching our expected range of the early 20s for marriage age (ref. #2). Between 2002 and 2011, the mean age at first marriage for women who married before age 50 (the end of reproductive age for women, after which the population structure is not affected by marriage) was 21.6 years (Table 1), closely matching our expected range of the early 20s for marriage age (ref. #2). The minimum age of Amish women at first marriage was 16.5 and the median was 20.9 years, indicating that half of all women married between these ages. For all US women in 2011, the median age at first marriage was 26.5 (U.S. Census Bureau 2011). With 75 per cent of Amish women marrying below age 22.3, marriage in the early twenties appears relatively universal among Amish women. A highly similar age schedule applies to Amish men, who were only roughly a year older at first marriage.

Table 1.

Age at first marriage for Amish women and men who married between 2002 and 2011 (where wife married before age 50)

Summary measure	Sex
	Females	Males
N	13,910	13,841
Minimum	16.5	16.3
25th percentile	20.0	21.1
Median	20.9	22.1
75th percentile	22.3	23.8
Maximum	49.8	49.5
Mean	21.6	22.9
Standard deviation	3.0	3.0

Source: Authors’ analysis of CAPED-2010s data.

For those who got married between 2002 and 2011 and had a first birth, 514 of their 11,945 first births occurred within eight months of marriage; accordingly, up to 4.30 per cent of births were conceived premaritally (what amount falls short are premature births), which falls within our expected range of at or below 7 per cent and indicates adherence to religious proscriptions against premarital sexual intercourse (ref. #3). This suggests that the Amish reserve sexual intercourse largely for marriage, although it is possible that unmarried individuals are using contraception more effectively in contrast to practices captured in earlier regional studies, although little is known about Amish contraceptive use and sexual practices (Anderson and Potts 2020), especially in premarital contexts (Anderson and Potts 2022).

An analysis of women aged 45–54 in 2002–11 reveals a consistent marriage pattern (Figure 3). While <1 per cent had married before age 18, approximately 75 per cent had wed between ages 18 and 24, with another 10 per cent marrying before age 30. A further <2 per cent had married by age 50, at which point 87.1 per cent had ever married. These findings approach our expectation of a population-wide proportion ever married above 90 per cent (ref. #4). Among those aged 45–54 during 2002–11, celibacy (defined as the percentage who never marry) was 11.6 per cent for women and 5.6 per cent for men. For other cohorts, celibacy ranged from 5 per cent to 15 per cent. Overall, marriage remains a normative life event for Amish women.

Figure 3.

Percentage of total Amish females ever married by a given age, 2002–11 snapshot

Source: As for Figure 2.

Amish individuals expect to be married for life. Separation is rare and divorce rarer still, as reflected in how Amish directories provide no formatting option clearly representing divorced households. In the event of a spouse’s death, remarriage is permissible, even desirable. Our analysis of all Amish remarriages (n = 1,098) reveals a median death-to-remarriage interval of 1.7 years (Table 2), supporting our expectation of rapid remarriage (ref. #5). However, the distribution is highly skewed; most remarriages occur within the first two years and many even within one year. Remarriage intervals longer than two years are relatively low in frequency, showing little variation between five and the maximum of 50 years (Figure 4). Men often remarry around one year after being widowed, whereas the gap between spousal death and remarriage for women is longer, with a larger standard deviation. Further, remarried men’s average age at death of first spouse is 52.8 compared with 39.2 for women, suggesting women are disproportionately more likely to remarry when younger (Table 2). The demographic consequence is that because men tend to remarry women much younger than themselves, more opportunities exist for unmarried women under 50 (who did not marry in their young adult years) to have children. Overall, the rapid timing of remarriage, especially for men, suggests how important family life is for Amish and how cultural norms privilege men’s initiation of courtship.

Table 2.

Timing of first spouse’s death and surviving spouse’s remarriage, as recorded throughout the CAPED-2010s database (n = 1,098)

Variable	Summary statistic	Overall	By sex
			Females	Males
Interval between first spouse’s death and remarriage	Min	0.2	0.3	0.2
	Median	1.7	3.9	1.4
	Mean	3.5	6.8	2.1
	Max	42.4	42.4	42.3
	Standard dev.	5.0	7.2	2.5
Surviving spouse’s age at death of first spouse	Min	19.9	19.9	20.6
	Median	48.8	39.2	52.8
	Mean	48.8	39.2	52.8
	Max	85.8	78.8	85.8
	Standard dev.	15.1	14.9	14.2

Source: As for Table 1.

Figure 4.

Interval between death of first spouse and remarriage for individuals remarrying between 2002–11

Note: Each interval is 0.5 years long; the values on the x-axis refer to the start of each interval.

Source: As for Figure 2.

Birth quantities and timing.

We use five-year age groups to measure birth quantities and timing. Our measures are calculated as follows (Hinde and Demétrio 1998):

$$TFR=5{\sum }_{i=15-19}^{45-49}{f}_i,\text{where}{f}_i\text{is the ASFR}.$$ (2)

$$f_i=\frac{\mathit{\text{Births in a year to women aged x years last birthday at the time of birth}}}{\mathit{\text{Mid}}-\mathit{\text{year population of women aged x years last birthday}}}\times \mathrm{1,000}.$$ (3)

The average TFR for all Amish women was 6.1 in the 2002–11 period (Table 3), meaning that an Amish woman would have, on average, just over six children across her reproductive lifespan if she experienced all the ASFRs within the given period. This TFR falls at the lower end of our expected range of six to eight children (ref. #6). Certainly, 6.1 is well above replacement level, whereas the US TFR in 2011 was 1.89 (Martin et al. 2012).

Table 3.

Total Fertility Rate and age-specific fertility rates, Amish 2002–11 snapshot

Age group	Women	Births	ASFR (live births per 1,000 women)
15–19	9,568	129	13.4
20–24	8,008	2,274	283.9
25–29	6,699	2,249	335.8
30–34	5,598	1,595	284.8
35–39	4,598	929	202.0
40–44	3,758	316	84.2
45–49	3,121	28	8.8
Total Fertility Rate			6.1

Source: As for Table 1.

The age schedule of fertility presented in Table 3 and visualized in Figure 5 indicates that fertility rises steeply between ages 15–19 and 20–24 and peaks around 25–29, with 336 children born each year per 1,000 women in this age group. Even for women in their 30s, the ASFR remains above 200 live births, only falling below 100 at ages 40–44. Women have their last child on average at age 36.52 (standard deviation = 4.86; not shown), closely matching our expectation of the late 30s for age at last birth (ref. #7). This fertility pattern is in sharp contrast to the current US age schedule of childbearing and is more similar to patterns seen in the mid-nineteenth century (Figure 5). The current US childbearing peak—around ages 25–29—is only slightly higher than the Amish level at ages 40–44.

Figure 5.

Age schedule of Amish fertility (2002–11) snapshot compared with historical and contemporary US fertility

Source: CAPED-2010s data; US 1844–49 data from Hacker (2003); US 2009–13 data from United Nations Department of Economic and Social Affairs Population Division (2019).

Figure 6 depicts completed cohort fertility for the 9,882 Amish women aged 45–54 in 2002–11, who can be assumed to have finished reproducing. The number of births averages 7.2 and ranges from 0 to 16. Only 4.0 per cent of women remained childless (ref. #8), confirming our expectation of low childlessness (1.4–6.6 per cent). Only 8.9 per cent had less than three children (i.e. a family size more typical of US households). Notably, 25 per cent of women had 10 or more children, underscoring the prevalence of large families among the Amish. Indeed, in the entire CAPED-2010s database, seven women had 19 children, four women had 20, and one woman had 21 children of her own (with two consecutive husbands).

Figure 6.

Amish completed cohort fertility: Women aged 45–54 in 2002–11 (n = 9,882)

Source: As for Figure 2.

Among all Amish births between 2002 and 2011, 12.55 per 1,000 were twin/triplet maternities (Table 4); this sits in the lower end of our expected range of 12.2–15.3 per 1,000 (ref. #9). The 967 twin/triplet maternities occurred within 929 families; accordingly, some families experienced more than one set of twin/triplet births. In this case, 30 families experienced two sets of twins, and four families even three sets of twins between 2002 and 2011. In the middle of the last century, the average US twinning rate was around 11. While Amish rates were considered a little high a few decades ago, the twinning rate has risen considerably in the general US population, to nearly 17 by 2015 (Monden et al. 2021), given the availability of MAR.

Table 4.

Twins and triplets born to the Amish throughout the CAPED-2010s database

Rates of twins/triplets among all maternities
Type of birth	Number of maternities	Children born	Twin/triplet maternities per 1,000 maternities
Singles	76,107	76,107	–
Twins	953	1,906	12.37
Triplets	14	42	0.18
Total	77,074	78,055	12.55
Frequency of twin/triplet births among families
Frequency of twin/triplet maternities		Number of families	Percentage
1		895	96.3
2		30	3.2
3		4	0.4
Total families with at least one twin/triplet maternity		929	100.0
Total twin/triplet maternities		967	–

Source: As for Table 1.

Birth intervals.

Relatively short intervals between marriage and first birth and between births are necessary to achieve high birth rates. The mean interval between marriage and first birth was 17.2 months for women in the database who gave birth to their first child between 2002 and 2011 (Figure 7). This is slightly shorter than our expectation of around 20 months, suggesting more rapid fertility initiation than previous regional studies have indicated (ref. #10). The average birth interval from first to second birth was 22.4 months, above our expectation of around 18 months (ref. #11). For comparison, among US women who married in 2006–10 without a premarital birth, the median interval between marriage and first birth was 2.7 years (32.4 months) (Hayford et al. 2014).

Figure 7.

Mean intervals in months between the first six births for any birth intervals completed between 2002–11 (n = 57,859)

Source: As for Figure 2.

Birth intervals for the subsequent births, up to the fifth to sixth interval, range from 25.4 to 28.3 months, gradually increasing with each subsequent interval (Figure 7). This confirms our expectation of gradual increases up to the sixth birth while remaining under 36 months (ref. #12). The last birth interval for women (not shown) averaged the longest of any other interval, at 3.13 years (37.6 months), confirming our expectation that the final birth interval would be the longest (ref. #13).

Mortality

Previous studies have suggested that Amish mortality patterns resemble those of the general US population. However, reliable data are usually presented only for IMRs; so far, other standard measures of mortality, such as life expectancy at birth, have been missing from the literature.

Historically, infant survival has improved due to non-proximate determinants (e.g. progress in public health provision, improved economic conditions, and changes in infant care), and the Amish have likely also benefited. In our analysis, we find that 77,073 live births occurred in 2002–11, and 456 infants did not survive until their first birthday. The IMR is all deaths under age one within a given year by all live births, multiplied by 1,000:

$$IMR=\frac{\mathit{\text{Deaths below age}}1\mathit{\text{year}}}{\mathit{\text{All live births within a year}}}\times \mathrm{1,000}.$$ (4)

Based on our data, the Amish IMR is 5.9, above our expectation of under 5.0 (ref. #14) but comparable with the US IMR of 6.1 in 2011 (Ely and Driscoll 2019).

We acknowledge one limitation: while directory entries explicitly labelled ‘stillbirth’ are excluded from IMR calculations, we cannot distinguish between late-term stillbirths and same-day infant deaths when both carry given names and identical birth and death dates. This may result in a slight overestimation of infant mortality. Notwithstanding, assuming that the Amish experienced similarly high IMRs to their non-Amish counterparts prior to the twentieth century, the Amish population seems to have benefited from contextual factors that reduce the IMR.

Even though studies have suggested that the Amish have long lifespans, directories provide insufficient mortality data to compare Amish life expectancy with that of other populations. Our alternative procedure described earlier—combining the CAPED-2010s with complete mortality data from four large Amish settlements—produced an annual population of 49,433, with an average of 121 annual deaths across the 2002–11 period. We then constructed a period life table (Preston et al. 2001), operationalizing current mortality conditions by inferring from a synthetic cohort subject to all age-specific mortality rates in that period. The age-specific mortality rate is defined by Hinde and Demétrio (1998) as:

$${{}_n{}^{}M}_X=\frac{\text{Deaths between ages x to x}+\text{n during the year}}{\text{Mid}-\text{year population in age interval x to x}+\text{n}}$$ (5)

Table 5 shows that the age-specific mortality rate remains below one death per 1,000 in the population until age 40. Death rates remain comparatively low up to age 65, when adult mortality surpasses infant mortality. From age 65, death rates increase substantially for each five-year age group. Low death rates across most ages result in high life expectancy at birth (e₀), specifically, 81.16 years in 2002–11 (Figure 8). This is slightly higher than the US population’s average of 78.7 years in 2011 (Arias 2015), aligning with our expectation that Amish life expectancy would be at or above the US value (ref. #15).

Table 5.

Life table for four Amish settlements: Holmes County (OH), Geauga County (OH), Daviess County (IN), and Arthur (IL), 2002–11 snapshot

Age (x)	Average population (2002–11)	Average deaths (2002–11)	nMx	nax	nqx	lx	ndx	nLx	5Px	Tx	ex
0	1,595	9	0.00603	0.068	0.00600	100,000	600	99,441	0.99349	8,115,952	81.16
1	6,136	2	0.00033	1.723	0.00130	99,400	129	397,305	0.99885	8,016,511	80.65
5	6,917	1	0.00014	2.500	0.00072	99,271	72	496,173	0.99880	7,619,206	76.75
10	5,951	2	0.00034	2.500	0.00168	99,199	167	495,577	0.99817	7,123,033	71.81
15	5,017	2	0.00040	2.500	0.00199	99,032	197	494,668	0.99841	6,627,456	66.92
20	4,200	1	0.00024	2.500	0.00119	98,835	118	493,881	0.99871	6,132,787	62.05
25	3,600	1	0.00028	2.500	0.00139	98,717	137	493,245	0.99849	5,638,906	57.12
30	3,065	1	0.00033	2.500	0.00163	98,580	161	492,500	0.99726	5,145,662	52.20
35	2,594	2	0.00077	2.500	0.00385	98,420	379	491,152	0.99470	4,653,161	47.28
40	2,209	3	0.00136	2.500	0.00677	98,041	663	488,547	0.99395	4,162,009	42.45
45	1,870	2	0.00107	2.500	0.00533	97,378	519	485,589	0.99423	3,673,463	37.72
50	1,604	2	0.00125	2.500	0.00622	96,858	602	482,786	0.98945	3,187,873	32.91
55	1,332	4	0.00300	2.500	0.01490	96,256	1,435	477,695	0.97853	2,705,087	28.10
60	1,051	6	0.00571	2.500	0.02814	94,822	2,669	467,437	0.96231	2,227,392	23.49
65	822	8	0.00973	2.500	0.04751	92,153	4,378	449,821	0.93785	1,759,955	19.10
70	620	10	0.01613	2.500	0.07752	87,775	6,804	421,866	0.88447	1,310,134	14.93
75	441	15	0.03401	2.500	0.15674	80,971	12,691	373,127	0.77997	888,268	10.97
80	260	18	0.06923	2.500	0.29508	68,280	20,148	291,028	0.43505	515,141	7.54
85+	149	32	0.21477	4.656	1.00000	48,132	48,132	224,113	–	224,113	4.66

Source: CAPED-2010s data plus additional deaths sample. Based on calculations for period life tables as in Preston et al. (2001).

Figure 8.

Remaining life years expected by age, Amish 2002–11 snapshot

Source: CAPED-2010s data plus additional deaths sample.

We took several approaches to validate our findings. First, to ensure we could reasonably infer mortality rates from our sample to the entire Amish population, we compared the IMR for the deaths sample (6.0) with the CAPED-2010s’ IMR (5.9), as complete data were available for both. The closeness of the two IMRs supports the strength of our sample for estimating population-wide mortality beyond infancy. Second, we acknowledge that the relatively small number of deaths might yield unstable mortality rates at older ages, particularly in a rapidly growing population (Scherbov and Ediev 2011). Hence, we compared expected deaths by applying 2010 US age-specific death rates (World Health Organization 2020) to our population of 49,433 Amish individuals. This revealed similar mortality patterns during infancy and childhood in the two populations. From age 15, US mortality was slightly higher than Amish mortality, with the gap widening after age 60. A similar comparison using 2010 rates from Israel (a very-low-mortality population) showed comparable adult mortality but lower infant and early childhood mortality than the Amish. Although Amish death rates are exceptionally high compared with both populations for ages 80+, our overall mortality patterns appear reliable despite the small numbers. Such post-transitional mortality patterns contribute to population growth through increased longevity and by enabling women (and men) to survive through their reproductive years, thus supporting sustained fertility throughout this period.

In-conversion

We identified 154 first-generation adult converts: 22 husband/wife pairs (44 people) plus 64 males and 46 females unmarried at conversion, all of whom had been accepted as Amish church members during a period of nearly a century (the earliest birth year included was 1934). This represents 0.146 per cent of the total Amish baptized membership alive in 2010. Many converts have left, but even if they had stayed, their children would have experienced a high risk of exit. In sum, as expected (ref. #16), we can safely conclude that in-conversion contributes only marginally to Amish growth, even when considering not just first generations but the limited compounded effect of multiple generations. Of note, a concentration of converts—24 of the 154—were in the Maine Amish churches, a small network that uses English rather than German in church services. This suggests that any seeker-receptive changes Amish churches make could increase the in-conversion rate.

Retention/attrition

Some directories include information about whether each couple’s offspring are: single and still living at home; baptized and married; no longer in the Amish; or single and living away from home (exact categories vary from directory to directory). As code meanings varied slightly across directories, we interpreted and standardized them to categorize individuals’ membership status. In total, just under half of all database households included codes, allowing us to analyse 117,259 individuals, including three of the four largest settlements: Holmes, Northern Indiana, and Geauga (but not Lancaster).

For comparison with past studies, we first took a cross-sectional approach, calculating the ratio of offspring with directory codes representing retention (e.g. child living at home or baptized married adult) vs disaffiliation. Across the directories’ various data collection time points, we found that 95.4 per cent of all offspring are coded as Amish and 4.6 per cent as having exited. This calculation—likely mirroring methods used in prior studies for generating Amish retention/attrition rates—is problematic because it will underestimate attrition. As younger people are inherently less likely to have left than older people, and because the population is skewed towards younger ages, this cross-sectional ratio disproportionately captures those retained so far but who may disaffiliate later in life.

To address this potential skew, we calculated retention for five separate age groups, allowing for a more nuanced understanding of disaffiliation across the life course (Figure 9). For those aged under 18, retention is nearly 100 per cent, with only 0.71 per cent disaffiliating. This is expected, as individuals rarely leave home until they are legally adults. In the 18–23 age group—the typical range for baptism and church membership decisions—retention remains high at 96.07 per cent, with 3.93 per cent disaffiliated. For those aged 24–29—representing the period just after baptism decisions—retention decreases to 90.63 per cent, with 9.37 per cent having disaffiliated. In the 30–39 age group, retention further decreases to 88.12 per cent, with 11.88 per cent disaffiliated, while for those aged 40+, retention is 84.46 per cent, with 15.54 per cent having disaffiliated. This final figure is the most representative of lifelong retention, suggesting that disaffiliations are concentrated in the 20s but continue through adulthood. The 84.46 per cent at ages 40+ is just below our expectation of retention above 85 per cent (ref. #17). Notwithstanding, overall retention remains exceptionally high for an American religious group. This high retention contributes significantly to the rapid population growth among the Amish, as it ensures that most individuals born into Amish families remain Amish.

Figure 9.

Cumulative Amish retention across the life course, 2002–11 snapshot

Source: As for Figure 2.

Two limitations should be noted. First, because not all directories include membership status codes, selection bias could exist. Yet, we would expect the inclusion of remaining Amish congregations that are without codes or unrepresented in the CAPED-2010s—mainly small congregations and strict groups—to raise retention. Second, when entire households leave the Amish (parents and children), their entry is removed and their children’s disaffiliation is not recorded. Failing to account for all members of disaffiliated families will underestimate attrition, especially for the <18 age group, where nearly 100 per cent retention is unrealistic, given that families with children do leave as a whole family. Despite these two countervailing limitations to our estimates, our analysis provides a robust analysis of Amish retention/attrition.

Discussion and conclusion

The Amish represent one of the fastest naturally growing populations in North America. This growth is occurring within the context of low-fertility societies: US and Canada, where TFRs are just below replacement level. In this study, analysis of 263,158 Amish individuals in 50,065 households reveals a demographic profile characterized by early marriage (mean age 21.6 years for women), high fertility (TFR 6.1), short birth intervals (mean 17.2 months between marriage and first birth and consistently below 29 months for the subsequent five births), and standard rates of twin/triplet births (12.55 per 1,000 maternities). Compared with findings from previous region-specific studies, today’s Amish women may have slightly fewer births, but they marry slightly younger and may concentrate births more heavily in their younger years rather than spreading births across the full reproductive period. Mortality patterns align more closely with those of the general US population, including an IMR of 5.9 per 1,000 live births and a life expectancy at birth of 81.16 years. In-conversion is extremely rare, with only 154 first-generation US/Canadian converts identified across just under a century. Retention is high, at 84.5 per cent for those aged 40+. In sum, the combination of high fertility, low mortality, and high retention explains the rapid growth of the Amish population.

Limitations of this study include potential under-representation of very small communities and the strictest Amish groups, possibly leading to underestimates of birth rates, overestimates of birth intervals, and underestimates of retention. While Amish directories provide nearly complete data for many measures, variations in directory publication dates and incomplete mortality data in some directories present calculation challenges.

The results from extensive population structure analyses of subgroups will have broad implications for public services, regional planning, healthcare provision, and cultural dynamics in areas with substantial Amish populations, although detailed analysis of these implications is beyond the scope of this paper. Given the cross-sectional nature of our database, future research should focus on longitudinal analyses and advanced statistical modelling to explain the ‘whys’ of this exceptional demographic profile. This could include exploring how specific cultural practices influence demographic outcomes, examining genetic predictors of demographic outcomes, analysing variations among different Amish affiliations, and projecting future population growth. Such studies would further enhance our understanding of Amish demography and its broader implications for population studies, testing theories that attempt to explain the persistence of large families in low-fertility contexts.

In conclusion, this comprehensive demographic analysis of North America’s Amish vastly improves our understanding of the Amish population structure. It also provides a robust foundation for understanding a population structure optimally poised for extremely rapid growth. It opens up new avenues for exploring the relationships between ethno-religious identity and demographic outcomes in contemporary societies, challenging assumptions about the universality of demographic transition, and inviting deeper investigation into the factors that shape population dynamics in diverse contexts.