Spatiotemporal distribution of the North American Indigenous population prior to European contact

Significance

The decline in the Americas’ Indigenous population after 1500 CE is well known. Less understood is its precontact demography. Radiocarbon dates confirm a continental peak US Indigenous population ~1150 CE, which declined and stabilized by ~1450 CE, declining again after European arrival. Analyzed by watersheds, radiocarbon dates show the decline occurred at different times in different places before 1500 CE and, in some areas, only after contact. Precontact declines were most likely due to climate, especially drought, endogenous disease in large settlements, emigration, and warfare. Had Europeans arrived a few hundred years earlier, they would have encountered a larger Indigenous population organized in places into large, cohesive polities. Further, the subsequent history of North America might have been very different.

Abstract

We examine spatiotemporal trends in the pre-European-contact Indigenous population of North America using radiocarbon (¹⁴C) dates of the past 2000 y. At a continental scale, the Indigenous population of the past ~14,000 y peaked at ~1150 CE and then declined until a brief recovery shortly before 1500 CE, after which ¹⁴C probability declines precipitously. After testing, we reject the hypothesis that the 1150 CE peak and decline is a result of ¹⁴C sampling issues. We then examine the ¹⁴C record of the past 2000 y in each of 18 watersheds where we find peaks ranging from ~800 to 770 CE to after European contact, with the majority, in the interior of the continent, declining ~1080 to 1300 CE. Although all Indigenous populations declined after European contact, that of a large portion of the country (the Great Lakes, New England, the Mid-Atlantic, the Central Plains, the Northwest, and California) did not decline until after contact.

Researchers have long sought to estimate the size of the North American Indigenous population on the eve of European colonization (e.g., refs. 1 and 2) with an eye toward assessing the impact of colonization on the continent’s Indigenous population. Although estimates vary, the Indigenous population declined severely after 1500 CE due to disease, genocide, warfare, export slavery, and environmental degradation (3–6). However, the spatiotemporal distribution of the Indigenous population prior to European arrival is critical to estimates of the precontact population and to understanding the demographic impact of European contact in different areas.

Using radiocarbon (¹⁴C) date frequencies in the Canadian Archaeological Radiocarbon Database (CARD), Peros et al. (7) discovered a continental precontact population decline in North America ~1150 CE. They attributed this decline to increased mortality due to disease in large, dense settlements and a decline in researchers’ use of ¹⁴C dating for post-1500 CE sites. At the time, CARD was smaller than the currently available database, incorporated poor dates to increase sample size, and, for the continental United States, was biased toward the northern states. Here, we use a larger, updated CARD (8) to examine spatiotemporal patterns in the last 2,000 y of Indigenous population history in the continental United States. We replicate the previous conclusion and find that population declines initiated at different times in different regions over the past 2,000 y, possibly beginning as early as ~770 to 800 CE in some places and not until after contact in others.

Archaeologists have previously documented precontact population declines in local areas of North America, e.g., the Savannah River valley in the late 14th to mid 15th centuries (9–11), Cahokia and the American Bottom in the mid-12th to 15th centuries (12), and the northern plains after ~1450 CE (13). In the Southwest, population peaked ~1250 to 1300 CE, leaving many large pueblos abandoned before the arrival of the Spanish ~1540 CE (14–17). The population of the Mesa Verde/four-corners region declined significantly after 1300 CE (14), and only a small population inhabited southern Arizona’s once densely inhabited Hohokam region by ~1450 CE (17). The populations of southwest Texas, southeast New Mexico, and the Mimbres area in central/southern New Mexico also declined ~1400 to 1480 CE (15, 18).

In many cases, local population declines were associated with emigration. The huge settlement of Cahokia and its vicinity was depopulated as people moved up the Ohio river as well as into Tennessee (5, 6, 19, 20). In northwest Wyoming, population declined severely after ~1150 CE, with some emigration north and west (21), and maize-dependent inhabitants of the Mesa Verde region emigrated south, to, e.g., the upper Rio Grande valley, with the vacated region reoccupied by the Ute, a lower-density foraging population (14). And yet, emigration must have been accompanied by actual population loss to produce the results presented here. Although devastated by European contact, the Indigenous population did not simply increase to a peak at 1500 CE and then decline (5). Here, we offer a bird’s-eye view of Indigenous demography of the past 2,000 y to help frame future research aimed at testing the cause(s) and precise timing of regional declines both before and after European contact.

Analysis

We previously replicated the Peros et al. (7) study through a continental (lower 48 states) summed probability distribution (SPD) using >24,000 AMS ¹⁴C dates covering the last ~15,000 y (8, Fig. 1). This revealed a peak at ~1150 CE, followed by a significant decline, with evidence for a slight recovery or plateau just prior to 1500 CE, and then a steady decline to the effective limits of ¹⁴C, ~1750 CE. The peak is not a function of the calibration curve at ~1150 CE as the peaks of individual watersheds lie at different times (see below). The decline after 1500 CE reflects population loss due to European emigration, but also the replacement of ¹⁴C dating with other methods, primarily dendrochronology and contact era artifacts.

Fig. 1.

Summed probability distribution of the continental United States, generated from 23,726 archaeological AMS dates (excluding shell dates) covering the last ~15,000 y. The uptick in the SPD for the very youngest age-range is due to calibration curve effects. Dashed line is a 200 y smoothing.

To determine whether the peak at ~1150 CE was uniform across the continent, we examined spatial variation using the 18 USGS 2-digit (highest level) hydrologic units (HU) that cover the continental United States (https://nas.er.usgs.gov/HUs.aspx; 22, Fig. 2). We used HUs since these roughly mirror regions of both similar ecology and precontact Indigenous culture areas (e.g., California, Southern Plains) in which we might expect human behavior to be consistent at a gross level of adaptive response. Although an analysis using precise site locations is preferable (e.g., ref. 23), the National Historic Preservation Act and the Archaeological Resources Protection Act prohibit revealing site locations (24); in CARD, US site locations are recorded at the county level only. Site county designation was used to assign individual dates to HUs based on the USGS Watershed Boundary Dataset. Sites in Arizona are not identified to county but to 1 × 1° maps; since >90% of Arizona falls within HU 15, we assigned the entire state to it.

Fig. 2.

Map of the continental United States with HU regions outlined in black. Labels are the 2-digit HU designation, sample size, and median peak age in years CE (Wyoming included in HU10).

We created SPDs for each HU covering 2,500 to 100 cal BP (error ≤ 100) by drawing data from a cleaned set of radiocarbon dates (n = 61,002) compiled from (8) plus a few thousand dates collected since 2022 (SI Appendix, Part I and Figs. S1 and S2 and Dataset S4). We created these SPDs using three, increasingly larger datasets for each HU (excluding shell dates; see below) to check for effects of sample size/quality: a) only AMS dates, b) AMS and all normalized non-AMS dates, and c) AMS, normalized, and any measured dates on charcoal normalized to an assumed −25‰ [many of the normalized dates of (b) were, in fact, normalized this way in the original reference].

We ran each of these three samples for each HU using a permutation test (25; SI Appendix, Part II) in the “rcarbon” package (26). These permutation tests control for interregional sampling biases by developing a population model based on the permutation of SPDs from each HU region. Statistically significant peaks and troughs for each HU are determined by comparison with the population model permuted from all HU SPDs. The sample size of the null model depended on which of the three samples was analyzed: a) all AMS (n = 16,773), b) all AMS and normalized (n = 34,038), and c) all AMS, normalized and assumed normalized (n = 45,298). In brief, this effort found no major differences among the three samples for each HU, except HU 9, where the sample size is small (SI Appendix, Part III). Although this gives confidence in (c), the largest sample, to be cautious we used data from (b) for the peak analysis described below.

The initial decline date for each dataset is its peak, and we use a unique, albeit simple and straightforward method to estimate the peak in each HU (SI Appendix, Part V and Figs. S5 and S6http://www.pnas.org/lookup/doi/10.1073/pnas.2419454122#supplementary-materials and Dataset S2) for the last 2,000 y. Calibration can produce artificial peaks and plateaus where the curve is “steep” or “flat,” aggregating or smearing ¹⁴C probability onto a shorter or longer range of calendar ages. To alleviate these effects, researchers commonly smooth their data. However, calibration effects can alter a peak’s age estimate with different levels of smoothing, and the “correct” level of smoothing (200, 500, and 800 y?) is unknown. To cope with this uncertainty, we smoothed the data at different intervals, from 1 (no smoothing) to 500 y (in 1-y intervals) and calculated the peak for each run. From the resulting sample of 500 peak ages for each HU, we calculated descriptive statistics (Table 1 and Fig. 3). Herein, the discussion of “peak dates” refers to these descriptive statistical medians. Tests using hypothetical date samples (SI Appendix, Part V) show that the method produces accurate results if it employs dates with SE ≤100 and a sample size of such dates that provides a date density of at least 0.2, or, for a 2,000-y interval, ≥400 dates. In all but HU 9, we exceed this sample minimum, often considerably (Table 1).

Table 1.

Descriptive statistics of radiocarbon date peak analysis by HU, in cal BP with sample size (N)

HU	Min	1st	Cal BP median	Mean	3rd	Max	N
01	399	499	540	545	602	637	470
02	338	411	455	454	490	554	1,596
03	741	766	781	789	819	845	1,716
04	343	403	450	449	497	555	871
05	567	615	655	649	682	709	1,828
06	704	737	783	800	863	927	354
07	808	828	843	850	869	925	1,892
08	707	737	773	775	816	886	828
09	980	1,028	1,123	1,094	1,139	1,163	149
10	1,022	1,074	1,093	1,089	1,105	1,173	3,422
11	408	436	467	465	484	555	1,549
12	550	612	663	665	716	779	2,347
13	765	796	838	856	917	979	2,934
14	1,106	1,129	1,154	1,160	1,169	1,237	2,088
15	833	857	872	898	955	991	2,116
16	831	842	861	884	930	985	1,768
17	231	306	358	338	376	403	2,341
18	313	349	372	367	387	437	2,552
HU 10 without Wyoming	557	631	693	679	718	765	2,001
Wyoming, HU 10 portion alone	1,069	1,118	1,178	1,188	1,265	1,315	1,421

Note that HU 10 was run with and without the portion in Wyoming.

Fig. 3.

Box and whisker diagrams showing the estimated peak ages for each HU, arranged in chronological order, by years CE. Note that two estimates are presented for HU 10, with and without (“No WY”) the portion in Wyoming.

Potential Biases

The distribution of ¹⁴C dates over time can be biased by taphonomic loss, poor dates (e.g., due to contamination), unequal rates of production (affecting the ratio of dates/people), researcher bias, sampling bias due to archaeological visibility (e.g., small ephemeral camp vs. large pithouse village) or taphonomic bias (e.g., deep burial of older sites; 27, 28), and calibration effects. Some of these can be solved by a judicious database cleaning protocol, large sample sizes, smoothing of the SPD, and rcarbon’s thinning and binning procedures. Given the time range of interest, the last 2,000 y, systematic deep burial or erosion is less concerning. This leaves two other potential biases: 1) the replacement of ¹⁴C by other dating methods (dendrochronology and archaeomagnetic dating, primarily in the Southwest, ceramic seriation in the Southwest and Southeast, and time-delimited historic artifacts for post-1500 CE sites) and 2) the possible tendency for archaeologists to avoid running samples from shallow, and presumably young, sediments. Combined, these factors could exacerbate the tendency for SPDs to terminate in a decline.

Dendrochronological and archaeomagnetic dates do replace ¹⁴C dates in late precontact Southwest sites and yet all three show a similar late precontact decline (16, 18). Architectural and ceramic seriation proxies for population also show a similar late precontact decline (16). A bias against dating shallowly buried samples seems unlikely as an explanation, since it requires explaining why archaeologists in some parts of the country date shallow samples while those in other areas do not. Nonetheless, we tested and found a) no bias against dates from the top 50 cms of sites and b) no bias toward ages <2,000 cal BP at shallow depths (SI Appendix, Part IV and Fig. S3http://www.pnas.org/lookup/doi/10.1073/pnas.2419454122#supplementary-materials and Dataset S1). An independent eastern US database of sites divided into rough chronological phases based primarily on artifact types (e.g., ceramics) also found a strong post-1000 CE decline in site frequency (SI Appendix, Part IV). We also tested how strong bias must be to produce the observed decline at 1150 CE. Using 10,000 hypothetical dates representing exponential growth until present, we removed dates in 200-y increments, regenerating the SPD. This exercise showed that for the decline at ~1150 CE to be the result of sampling bias or edge effects there must be a complete bias against dating all materials <800 to 1,000 uncal years BP (i.e., <~1032 to 1243 CE; see SI Appendix Part IV and Fig. S4http://www.pnas.org/lookup/doi/10.1073/pnas.2419454122#supplementary-materials). This is not true of our sample, and we conclude the decline observed in the SPD is not a product of sample bias or SPD edge effects.

Results

Examining the data by HU, we find three major periods of declines based on peak date (Table 1 and Figs. 2 and 3). In all but one case, New England (HU 01), peaks are followed immediately by a decline. There are undoubtedly demographic processes and events masked by aggregating data by HUs. Some HUs cover large regions and could be biased by incorporating a region more densely dated than others, or different ecologies, and/or separate Indigenous populations that responded differently to conditions. The Northern Plains (HU 10), for example, includes much of the state of Wyoming, which is more densely dated/documented than other states contributing to HU 10. We therefore analyzed HU 10 (without Wyoming) and Wyoming separately. The peak of Wyoming by itself falls much closer to that of the Upper Colorado drainage (HU 14); without Wyoming, the Northern Plains has a significantly later peak (Table 1).

The early set includes three hydrologic units whose peaks fall between ~770 and 830 CE but that overlap sufficiently to be simultaneous:

• •Wyoming (w/o portion in HU 14), 772 CE
• •Upper Colorado drainage (HU 14; including Four Corners area/San Juan drainage, eastern Fremont), 796 CE
• •Extreme Northern Plains (HU 9), 827 CE

HU 9, which extends into Canada (the data for which are not included here), has only a small portion in the continental United States and its sample (n = 233) falls below our sample size cutoff for reliability. The timing of the Upper Colorado decline might be influenced by the portion lying within Wyoming. It also contains the Chaco/San Juan River area of northwestern New Mexico, where archaeologists have long known the decline begins later, ~1100 CE, and thus its peak might be a misleading average of different demographics (see also comment below).

Starting ~250 to 300 y later, the peak estimates record that much of the interior US declines:

• •Arizona (HU 15), 1078 CE
• •Great Basin (HU 16), 1089 CE
• •Upper Mississippi (HU 07), 1107 CE
• •Rio Grande/SW Texas (HU 13), 1112 CE
• •Cumberland (HU 06), 1167 CE
• •Southeast US (HU 03), 1169 CE
• •Lower Mississippi (HU 08), 1177 CE
• •Northern Plains (HU 10), 857 CE; w/o WY, 1257 CE
• •Southern Plains (HU 12), 1287 CE
• •Ohio River (HU 05), 1296 CE)

As expected, there is variation within some hydrologic units. Arizona’s (HU 15) Hohokam region, for example, began to decline ~1350 CE (17), after the peak our analysis established. The Upper Mississippi (HU 07) decline estimate is likely driven strongly by the abandonment of Cahokia and the American Bottom. The Rio Grande (HU 13), on the other hand, was home to large PIV/V pueblos at its northern end, and to many foraging populations at its southern extremity, but removing the Texas dates from HU 13 only shifts the peak from 1098 to 1112 CE. Cobb et al. (20) place the population decline in the Middle Cumberland at ~1300 CE, slightly later than our analysis, with most depopulation happening 1450 to 1480 CE.

Several declines have peak dates after, or ranges significantly overlapping with, European contact:

• •New England (HU 01), 1410 CE
• •Central Plains (HU 11), 1483 CE
• •Mid-Atlantic (HU 02), 1495 CE
• •Great Lakes (HU 04), 1500 CE
• •California (HU 18), 1578 CE
• •Northwest (HU 17), 1592 CE

The declines of this set most likely reflect the effects of introduced disease, warfare, forced labor, and export slavery of European colonization. Although the New England peak is just prior to European contact, the peak is a plateau (Fig. 2), and the decline starts after 1500 CE. The California result is strengthened by analysis of a large burial population (29), whose characteristics do not reflect contagious disease until ~1770 CE. Our California peak is close to this estimate, lying between when the Spanish first arrived in Alta California (1542 CE), and the first permanent European settlement there (1769 CE). The earliest direct contact between Europeans and Indigenous peoples of the Northwest Coast and middle Columbia River was slightly later (1774 CE and 1805 CE, respectively; 30), but Rorabaugh’s (31) analysis of ¹⁴C data suggests coastal and interior populations of Washington declined after ~300 cal BP (~1650 CE). Our postcontact results also broadly agree with Jones’s (30) findings based on ethnohistoric data.

Two potentially confounding factors require consideration: shell dates and “old wood.” We excluded shell dates from samples (SI Appendix, Parts II and III) as the marine reservoir correction varies, sometimes significantly, along coastlines (32), and in many cases, the correction factor is uncertain. In some cases, there is not a large difference in marine shell and other dates (e.g., ref. 31). But if there were a late precontact shift to a coastal adaptation, the abundance of shell in middens could result in an archaeological bias toward shell dates, and removing such dates could thus systematically bias an SPD away from younger-aged sites. Including marine-reservoir-corrected shell dates for California does not noticeably affect SPD peak placement, although uncertainty is high due to local reservoir corrections (SI Appendix, Part VI and Dataset S3).

By necessity, our analysis uses many legacy dates that are only identified as “charcoal” in the database. Some of these dates could be affected by the old wood issue. In humid regions, such as the eastern US, wood rots soon after death, rendering it less useful as fuel. However, in arid regions, wood can remain viable firewood long after its death and can provide misleadingly old ages. This is why researchers today aim to date short-lived remains, e.g., carbonized seeds, twigs, pine needles, bark, etc. The old wood effect is difficult to control and still maintain a viable sample size in most HUs. However, the Colorado Plateau (HU 14) offered an opportunity to evaluate the possible scale of the old wood effect since it was occupied at one time by maize horticulturalists. Separating wood charcoal from Zea mays dates (SI Appendix, Part VII) reveals a difference in peaks of ~400 y, with charcoal dates peaking ~750 CE and maize dates at ~1150 CE (see also ref. 33). The Colorado Plateau’s decline, therefore, might be closer to ~1150 CE. On the other hand, dendrochronological and ¹⁴C dates (including short-lived and charcoal samples) both show a decline at ~1275 CE in the Upland Southwest (16), and there is little difference in short-lived and wood charcoal samples in the Southwest’s Mimbres or Jornada regions (15, 18). Small numbers of identified short-lived samples in the database prevent analysis in other areas of the western US, but this issue requires future consideration for in-depth regional analyses of ¹⁴C dates.

Discussion

The ¹⁴C profiles suggest that many Indigenous populations of the continental United States underwent significant declines in the past ~1,200 y before European arrival. Some of the individual HU declines reflect emigration to other HUs (e.g., as HU 07 declines, the neighboring HU 05 increases; and when HU 05 declines, the neighboring HUs 04, 02, and 01 increase). However, many emigration events must have been associated with mortality prior to 1500 CE to produce the continental pattern seen in Fig. 1.

We cannot use an SPD to directly estimate population numbers since date frequencies align more closely with energy consumption than population (34); put more simply, the ratio of date production to population is not constant, especially between different subsistence regimes (foraging, horticulture, agriculture). Consequently, we cannot offer a maximum continental US population for 1150 or 1500 CE. However, population is an important factor in energy consumption, and the continental SPD, taken at face value, suggests a relative population loss between 1150 and 1500 CE of ~30%. However, large settlements typical of some regions ~1150 CE and later (e.g., PIII/IV period pueblos, Mississippian settlements, and Central Plains villages), are most likely underdated by ¹⁴C relative to their population size compared to earlier small settlements of nomadic foragers or horticulturalists. If so, then the peak at ~1150 CE could be greater, making 30% a minimal figure. The reconstruction of declines is best done at the level of individual HUs, or smaller regional units, where sample biases could be better controlled and Bayesian modeling efforts could tighten the ages of specific sites by increasing the date density of large sites.

The SPDs (SI Appendix, Part II) suggest a general pattern to population decline/emigration, one that began in the central Rocky Mountains. Then, about 250 y later, populations in the North American interior east of the Rocky Mountains declined/emigrated in rapid succession over a period of ~200 to 250 y, from ~1080 to 1300 CE. Parts of the continent that may have accepted some migrant populations—New England, the Mid-Atlantic, the Great Lakes, California, the Northwest, and the Central Plains—did not decline until after European contact.

These declines/emigrations entailed both agricultural and foraging populations. The cause(s) are unknown but may have entailed a mix of three factors. First, endogenous diseases of the Indigenous population could have emerged to greater effect in large settlements such as those of Late Woodland/Mississippian communities in the east (35), the PIV pueblos of the southwest (14), and Plains villages. These diseases could have included “treponemiasis...tuberculosis...tularemia, Giardia, rabies, amebic dysentery, hepatitis, herpes, pertussis, and poliomyelitis” (36). None leave unambiguous traces on bone and thus are difficult to detect. Although there was a very short-lived Scandinavian settlement in eastern Canada, L’Anse aux Meadows, ~1020 CE (37), it was probably not the disease source as the earliest declines are not in the northeast.

Emerging diseases, however, are expected to have had less effect on foraging populations that lived in small, nomadic settlements; and yet, their populations also declined and emigrated (e.g., ref. 38). Thus, a second factor could be climate change. Drought frequency increased due to ENSO (El Nino Southern Oscillation) activity, which in the past 10,000 y was most frequent ~50 BCE to 1150 CE (39); this overlapped with the Medieval Warm Period (MWP, ~950 to 1250 CE; 40). MWP events included the late 13th century droughts in the northern southwest that preceded Ancestral Pueblo migration to the northern Rio Grande, Hopi, Zuni, and Acoma areas (16); the mid-to-late 12th century droughts in the American Bottom that resulted in the depopulation of Cahokia (12) and movement of people up the Ohio River, an area of greater effective moisture at the time (41); severe droughts ~1075 to 1250 CE on the northern Plains (13, 42); and a 13th century drought in the Great Basin (43). The Savannah River valley’s population centers declined after severe droughts, 1360 to 1377 CE, and the valley was depopulated by ~1450 CE as people emigrated to the coast (9, 10). Fremont horticultural villages in the eastern Great Basin disappear by 1300 CE and perhaps ~1150 CE in the adjacent Colorado Plateau (33). It was ~950 CE that Athabascan-speaking foragers migrated from southwest Alaska (44), through the Rocky Mountains into the American southwest and southern Plains, becoming the Navajo (Dené) and Apache. In sum, Europeans might have arrived soon after a great reshuffling of North America’s Indigenous populations.

Third, and perhaps because of that reshuffling, the Indigenous population witnessed increased warfare, especially as large horticultural, aggregated populations coped with the periodic stress produced by drought, disease, and increased social competition. Warfare can produce observable declines in ¹⁴C SPDs because of increased mortality and fear-driven regional abandonment (45). Conflict increased in the east after 950 CE, evidenced in part by stockaded population centers ~1200 to 1300 CE (35, 46, 47). The political collapse of Mississippian chiefdoms by ~1350 CE dispersed the population into smaller settlements that continued to suffer chronic warfare (46). In the 16th century, Europeans encountered an arc of endogenous warfare from the Haudenosaunee (Iroquois) region in New York state/Ontario, south through the Appalachians, then to the west across the northern southeast (46). Periodic violence also erupted in the northern southwest in the 11th through the early 13th centuries (e.g., ref. 48), as well as on the Plains, where violence, apparent by ~400 CE, intensified ~1300 to 1450 CE (49) due to droughts and encroachment by emigrant communities from the east.

The continental Indigenous population may have only begun to recover from the ~1150 CE decline before Europeans arrived in numbers after 1500 CE. The SPDs of many HUs remain significantly below the continental SPD except in the Great Basin (HU 16), where, after a decline at ~1090 CE, population recovered ~1600 CE; the population of the Northwest and California continued to grow after 1150 CE (30, 31) perhaps a result of migration from the Great Basin, and from the north and east (e.g., the Na-Dene and Algic speakers of California’s northern coast). Late precontact/postcontact emigration is also recorded in some Plains oral histories, e.g., of the Apsáalooke (Crow; 50), and the Sahnish (Arikara), Chaticks-si-Chaticks (Pawnee), and Kirikir?i (Wichita), whose movements are also recorded in ¹⁴C data (13).

Conclusion

North America’s Indigenous precontact population decline was by no means unique in world history, for all populations rise and decline. Population loss can result from declines in local carrying capacities due to climate change and increases can result from technological and social innovations that are responsive to those conditions (51–53). At other times, population declines are due to contagious disease, especially where populations are aggregated. The Black Plague in Europe and adjacent portions of Asia, 1346 to 1353 CE, eventually killed 75 to 200 million people, and required >200 y for recovery; this was not a unique occurrence (e.g., the Justinian plague, 541 to 549 CE). Likewise, warfare can result in significant population declines. Therefore, demonstrating a pre-Columbian decline in the Indigenous population of North America in no way means that the Indigenous population was “a dying race” as European colonialist ideology conveniently assumed without knowing anything of the continent’s history (54). Instead, it means the Indigenous people of the western hemisphere were a human population like any other and underwent periods of growth and decline, as well as of local instances of emigration and immigration.

Unknown to them, Europeans were simply lucky to set foot in the western hemisphere when they did. Had they arrived a few hundred years earlier, they would have encountered a considerably larger Indigenous population, which in places was organized into large, cohesive polities that could have mounted a more severe response to the European incursion. Had that happened, the history of North America might have been very different.

Materials and Methods

We used a database of >60,000 archaeological radiocarbon dates from CARD compiled under the supervision of RLK, MEM, and ER (8). We statistically compared the ¹⁴C records among HUs using the “permtest” function in the rcarbon package in R (21) with ages calibrated using IntCal20. First, however, we used the “binsense” function to test whether binning radiocarbon samples (0 to 300 y) from the same site made a difference in the shape of the SPD; we found no evidence to warrant binning (SI Appendix). We then examined three different samples (AMS only, AMS and normalized, AMS, normalized and assumed normalized) of grouped geographically into HUs based on county using R covering the last 2,500 y cal BP (SE ≤ 100) to check for sample size. Permutation tests used 200-y running means with 1,000 Monte Carlo simulations identified statistically significant peaks and troughs in each HU by comparison with a simulated null model derived from the permutation of all HUs. This approach provides a control for sampling biases in the radiocarbon records of different HUs. We found no significant differences among the three samples.

Using only the last 2,000 y of dates with SE ≤ 100 for each HU (Table 1), we identified peaks for each HU using a unique method programmed in R (SI Appendix, Section V). The method identifies the single greatest year (i.e., mode) for each of 500 running mean windows between 1 and 500 y. Peaks are the median values for these 500 modes.

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (CSV)

Dataset S02 (CSV)

Dataset S03 (CSV)

Dataset S04 (CSV)

Dataset S05 (CSV)

Dataset S06 (CSV)

Dataset S07 (CSV)

Data, Materials, and Software Availability

The radiocarbon data used in this analysis are archived in CARD, the Canadian Archaeological Radiocarbon Database: www.canadianarchaeology.ca and in supporting information. The data in CARD are freely available to vetted researchers. The relevant R codes for analyses are in SI Appendix. Previously published data were used for this work (8).