Brainerd-Robinson Similarity Calculator

Compare artifact type frequencies between two archaeological assemblages to measure their coefficient of agreement on the classical 0-200 scale.

Last reviewed: May 2026

Built and maintained by Shannon Patterson · Last updated May 2026

ARTIFACT TYPE SITE A % SITE B %

TOTALS 100% 100%

Brainerd-Robinson Coefficient

200.00

Identical assemblages

Sum of Absolute Differences

0.00

Interpretation

Identical

Educational disclaimer: This calculator is provided for educational and exploratory research purposes only. Archaeological similarity coefficients depend critically on the underlying typological system, sample size, and recovery methods. Consult published methodological literature and, for publication-quality work, a qualified quantitative archaeologist before drawing chronological or cultural inferences from BR scores alone.

The Brainerd-Robinson calculator above computes the classical coefficient of agreement between two archaeological assemblages described as percentage distributions of artifact types. Enter the percentage of each type in Site A and Site B; the tool returns a value between 0 (no overlap) and 200 (identical), the sum of absolute differences that drives it, and a textual interpretation. The sections below explain exactly how that figure is derived, where it came from in 1951, why it has survived seventy-five years of methodological critique, when it breaks, and how it compares to chi-square, Bray-Curtis, and modern Bayesian alternatives. The worked example walks through a four-type pottery comparison step by step and visualizes both assemblages side by side as an inline matrix.

1. What This Calculator Does

The Brainerd-Robinson coefficient is a measure of agreement between two archaeological assemblages, each described as a percentage distribution across a shared list of artifact types. The classical definition, given simultaneously by George Brainerd (1951) and W. S. Robinson (1951) in American Antiquity, is straightforward: subtract the sum of absolute percentage differences across all types from 200. In symbols, BR = 200 − Σ|A_i − B_i|, where A_i and B_i are the percentages of type i in assemblages A and B respectively, and both percentage vectors sum to 100.

The output is bounded between 0 and 200. A score of 200 occurs only when the two assemblages have identical percentage distributions across every type — every A_i equals every B_i, every absolute difference is zero, and nothing is subtracted from 200. A score of 0 occurs when the two assemblages share no types at all: one is 100% Type X and the other is 100% Type Y, producing |100 − 0| + |0 − 100| = 200, which subtracted from 200 yields zero. Every real-world comparison sits somewhere between these extremes, and the coefficient gives you a single number summarizing how close the two assemblages are.

What the calculator does not do is tell you whether the difference between two assemblages is statistically significant in the inferential sense — BR has no canonical sampling distribution, so there is no p-value attached. It also does not correct for the granularity of your typology, the size of your samples, or the recovery bias of the excavation. Those interpretive responsibilities sit with the analyst. The calculator handles the arithmetic; the archaeology handles the meaning.

2. How to Use It

Enter the percentage of each artifact type in Assemblage A in the first numeric column and the percentage in Assemblage B in the second. The two columns should each total approximately 100% — the running totals beneath the input rows turn red if either column drifts more than 1% above or below 100, and the warning band appears explicitly when that threshold is crossed. Click + Add Artifact Type to extend the type list; click the × on the right of any row to remove that type from both assemblages. The coefficient recalculates live on every keystroke.

The result panel reports three numbers: the Brainerd-Robinson coefficient (0-200), the raw sum of absolute differences before subtraction from 200, and a textual interpretation tier. The interpretation tiers follow the conventional archaeological reading: BR ≥ 195 is treated as effectively identical; 150-195 as closely related; 100-150 as showing moderate cultural overlap; below 100 as showing low affinity. These thresholds are heuristic, not statistical, and should be reported alongside the underlying numbers in any published work — never as standalone descriptors. The share link below the results encodes the entire input table in the URL so you can bookmark, email, or cite a specific comparison without screenshotting.

If you are comparing assemblages with different native type lists (a common situation when typologies have evolved between projects), normalize the lists first: build a unified vocabulary of every type present in either assemblage, then enter 0% for any type absent from one side. This is correct procedure, but be aware that the zeros will inflate the absolute-difference sum and depress your BR score in a way that reflects taxonomic mismatch as much as cultural difference.

3. Worked Example: Two Four-Type Pottery Assemblages

Two excavations have yielded sherd counts at adjacent Late Woodland sites in the lower Mississippi Valley. After classification, the analyst reports the following percentage distributions:

Assemblage 1, Type A: 40%
Assemblage 1, Type B: 30%
Assemblage 1, Type C: 20%
Assemblage 1, Type D: 10%
Assemblage 2, Type A: 35%
Assemblage 2, Type B: 25%
Assemblage 2, Type C: 30%
Assemblage 2, Type D: 10%

Both columns sum to exactly 100, so we can proceed without rescaling. Compute the absolute difference for each type:

Type A: |40 − 35| = 5
Type B: |30 − 25| = 5
Type C: |20 − 30| = 10
Type D: |10 − 10| = 0

Sum the absolute differences: 5 + 5 + 10 + 0 = 20. Subtract from 200: BR = 200 − 20 = 180. This falls in the 150-195 band, which the calculator labels "Closely related" — substantial overlap, broadly consistent with two contemporaneous communities sharing the same ceramic tradition but with a modest divergence in the popularity of Types B and C.

Figure 1. Side-by-side type-percentage comparison for the worked example (BR = 180).

The figure shows where the divergence actually lives: Types A and B differ by 5 percentage points each, Type C by 10 points (Assemblage 2 has noticeably more of it), and Type D is identical at 10% in both. If this comparison were one cell in a larger seriation matrix of, say, twelve assemblages along a stratigraphic column, the BR of 180 would mark these two as neighbors in the chronological order even though they are not literally indistinguishable. That relative-ranking use is where BR genuinely shines — and where reporting BR alongside its component differences, as above, makes the analysis legible to reviewers.

4. Common Use Cases

The original application — pottery assemblage comparison — remains the most common. Ceramicists working on regional sequences from the American Southwest, the Mississippian Southeast, the Aegean Bronze Age, and Mesoamerican lowlands all routinely compute BR matrices across assemblages excavated from different sites or strata. The output drives both site-to-site cultural-affiliation arguments (these two villages were trading partners; these were not) and within-site chronology (this stratum sits between those two on the seriation matrix).

Lithic typology comes second. BR has been applied to flaked-stone industries where types are defined by reduction stage, retouch pattern, or formal class — for example, comparing Paleoindian assemblages by point morphology percentages, or comparing Middle Paleolithic assemblages by Bordian type counts. The mathematical move is identical to the pottery case; only the type definitions and the underlying production-system theory change. Faunal assemblages — counts or percentages of identifiable taxa per context — are the third major use case, including comparisons of subsistence economies between sites or between phases within a single site.

Seriation, the chronological ordering of assemblages on the assumption that type popularity follows a unimodal "battleship curve," is BR's signature application. The analyst computes BR for every pair of assemblages in the dataset, builds a symmetric matrix, and reorders rows and columns until the highest values cluster along the diagonal. The resulting sequence is the proposed chronological order. Site-to-site cultural-affiliation studies — where the question is not chronology but whether two communities participated in the same tradition — also rely heavily on BR, sometimes with explicit threshold rules ("assemblages with BR ≥ 150 are treated as members of the same regional ceramic tradition"). Each of these use cases has a body of methodological literature spelling out the conventional thresholds and the known failure modes; never apply BR outside that literature without documenting why.

5. Edge Cases and Limitations

Brainerd-Robinson has well-known weaknesses, every one of which becomes a publishable methodological footnote when you ignore it. The first is sensitivity to type granularity: lumping two functionally distinct subtypes into a single category, or splitting a single type into multiple subtypes, can move the BR by 20 or 30 points without any underlying change in the assemblages themselves. This is the "tyranny of typology" — the BR is only as meaningful as the type system feeding into it, and any reordering of that system propagates immediately into the results. Document your typology before computing BR, and never adjust it post hoc to chase a target score.

The second is the independence assumption. BR treats every type as a free dimension, but real ceramic, lithic, and faunal assemblages have strong type-correlations built in by production constraints, raw-material availability, and functional cohorts. A single underlying variable (say, "preferred clay source") may simultaneously raise the percentages of three correlated types and depress the percentages of two others, but BR has no way to recognize that and treats the five shifts as five independent pieces of evidence. The result is that BR can overstate the magnitude of difference when the underlying causes are correlated, an issue chi-square at least partially diagnoses via residual analysis.

The third is zero-inflation. When Type X is present in Assemblage A at 8% but completely absent from Assemblage B, the |8 − 0| = 8 difference enters the sum just like any other 8-point gap — but the qualitative implication (a type missing entirely) is arguably much stronger than the quantitative weight BR gives it. Conversely, when a unified type list contains many types absent from both assemblages, the matched zeros artificially inflate the BR by contributing nothing to the difference sum while still counting as agreement. Both directions of this problem are unresolved at the BR level and require pre-analysis decisions about which types to include.

Sample size matters more than most introductory presentations admit. The standard rule of thumb — 30 sherds or other countable units per assemblage as an absolute minimum, with 50-100 strongly preferred — comes from the observation that below 30, individual-artifact reclassifications swing the percentages enough to shift BR by 10-20 points. This "tyranny of small N" is especially severe for rare types: a single sherd reclassified from Type C to Type D can move 4 percentage points each in a 25-sherd assemblage. For very small assemblages, BR results should be reported with explicit caveats and ideally with bootstrap-derived confidence intervals; Drennan and Peterson (2004) provide the canonical Bayesian framework for that uncertainty quantification.

6. Other Similarity Metrics and Why BR Persists

BR is not the only assemblage-similarity measure, and the alternatives have been visible in the archaeological literature for as long as BR itself has. Robinson (1951) published his original derivation in the same issue of American Antiquity as Brainerd (1951), and the two papers cite each other politely as having arrived at the same coefficient by different reasoning paths. Robinson framed his version as an "index of agreement"; Brainerd framed his as a seriation tool. The Brainerd-Robinson label that stuck combines both contributions, though some older sources still distinguish the "Robinson Index" specifically from the symmetric coefficient.

Cowgill (1968), "Archaeological Applications of Factor, Cluster, and Proximity Analysis," produced the most influential methodological critique. Cowgill argued that BR's lack of a sampling distribution made it ill-suited to inferential work, that chi-square offered a genuinely statistical alternative for testing assemblage differences, and that factor and cluster analysis exposed structure in type-frequency data that BR collapsed into a single number. The critique was sharp and broadly correct, but BR survived because the alternatives required computational machinery and methodological training that were not universal in archaeology departments in the 1960s and 1970s, while BR could be calculated on a hand calculator or even by hand from a percentage table.

Doran and Hodson (1975), in Mathematics and Computers in Archaeology, formalized what they called the "Robinson Index" alongside several alternatives, including Sokal and Sneath's matching coefficients from numerical taxonomy. The Bray-Curtis similarity index, developed independently in ecology in 1957, turns out to be mathematically equivalent to a normalized BR — Bray-Curtis on percentage data with the absolute-difference formulation produces a value of (1 − BR/200), which means archaeologists and ecologists have been computing essentially the same coefficient under different names for seven decades. The ecological literature is a useful cross-reference for any archaeologist working on BR's statistical properties.

BR persists in archaeology in 2026 for three reasons. First, it is genuinely easy to compute and interpret, which matters when a dataset has dozens of assemblages and the analyst is doing exploratory seriation by hand or in a spreadsheet. Second, the archaeological literature has accumulated decades of conventional threshold values — "BR ≥ 150 implies shared tradition" — that act as informal anchors for new analyses, and switching to chi-square or Bayesian alternatives sacrifices that interpretive continuity. Third, modern best practice now reports BR alongside more rigorous methods rather than replacing it: the BR matrix is published as a transparent exploratory step, and Bayesian seriation (Drennan and Peterson 2004; Buck et al. 1996) handles the inferential work in a separate, complementary analysis.

7. Historical and Methodological Context

The mid-1951 archaeological landscape into which BR was introduced was dominated by what would now be called a chronology problem. Stratigraphic excavation could establish local sequences within a single site, but cross-dating between sites — placing Site X's lower stratum in the same period as Site Y's middle stratum hundreds of miles away — required a method of comparing artifact assemblages directly. Willard Libby's radiocarbon technique had been published in 1949, but practical, affordable C-14 dating with reasonable error bars was still a decade away from being routinely available to field archaeologists. The discipline needed a seriation tool. Brainerd's and Robinson's papers, published in the July 1951 issue of American Antiquity, gave it one.

Brainerd's contribution was the conceptual argument: assemblages of pottery types from a continuously evolving tradition should produce a similarity gradient when ordered chronologically, with adjacent assemblages most similar and distant assemblages least similar. Robinson's contribution was the explicit numerical formulation. Together they laid out the calculation, the seriation procedure, and the interpretive framework that archaeologists would use without major revision for the next twenty years. Numerous mid-twentieth-century regional chronologies — Hohokam ceramic phases in the Sonoran Desert, the Mississippian periods of the lower Southeast, the Aegean Late Bronze pottery sequences — were ordered using BR matrices computed on graph paper.

Cowgill's 1968 critique opened the second methodological era. Factor analysis, cluster analysis, and chi-square testing entered routine use by the 1970s, and BR became one tool among several rather than the dominant one. The third era began with the spread of widespread radiocarbon dating in the 1960s-70s and accelerated with the calibration revolution of the 1980s and the Bayesian chronology toolkit of the 1990s. By the time Buck, Cavanagh, and Litton published Bayesian Approach to Interpreting Archaeological Data in 1996, and Drennan and Peterson followed with their explicit Bayesian seriation framework in 2004, the inferential frontier had clearly moved past BR. Yet the tool remained in undergraduate methods textbooks, in regional ceramic monographs, and in the standard exploratory toolkit because it does one thing well — produce a transparent, easily explained similarity number — and that one thing remains useful even when it is no longer the only or the best thing available.

8. Frequently Asked Questions

There is no universal cutoff, but archaeologists conventionally interpret BR scores as: 180-200 (very strong agreement, plausibly contemporaneous or culturally identical assemblages); 150-180 (high agreement, likely closely related contexts); 100-150 (moderate agreement, partial cultural overlap); below 100 (low agreement, distinct assemblages). The thresholds are heuristic, not statistical — Cowgill (1968) showed BR values do not have a standard sampling distribution, so "significance" depends on sample size and the number of types involved. For seriation, the absolute value matters less than the relative pattern of BR scores across an ordered matrix of assemblages.

The original Robinson (1951) formulation defined the coefficient as 200 minus the sum of absolute percentage differences across all artifact types. Because two percentage vectors that sum to 100 each can differ by a maximum of 200 (one assemblage is 100% Type A, the other is 100% Type B, giving |100-0| + |0-100| = 200), the range of the raw difference is 0 to 200. Subtracting from 200 inverts it so that higher values mean more similarity. Some modern papers report a "normalized" BR ranging 0-1 by dividing by 200, but the classical 0-200 scale remains standard in North American and European archaeological literature.

Chi-square tests whether observed type frequencies differ significantly from expected frequencies under an independence model, producing a p-value tied to a known sampling distribution. Brainerd-Robinson is a descriptive similarity index without an inferential framework — it tells you how similar two assemblages are, not whether the difference is statistically significant. Cowgill (1968) argued chi-square was preferable for hypothesis testing, but BR remains popular because it is intuitive, easy to compute on small samples where chi-square assumptions fail (expected cell counts below 5), and produces a single number that ranks well in seriation matrices. Many archaeologists use both: BR for seriation and chi-square for testing specific contrasts.

Yes, but only after normalizing both assemblages to the same type list. If Assemblage A contains Types 1-5 and Assemblage B contains Types 3-7, you must build a unified type list (1-7) and enter 0% for any type missing from a given assemblage. The BR calculation then proceeds normally. The catch: assemblages with very different type lists tend to produce low BR scores even when the present types overlap heavily, because the zeros inflate the absolute-difference sum. This is a known weakness when comparing assemblages from different cultural traditions or analytic systems. Aligning your type taxonomy across all assemblages before computing BR is essential.

BR is sensitive to the granularity of your typology. If you treat "redware" as one type, two assemblages may both be 60% redware and look highly similar. If you split redware into five subtypes by rim form, paste, and surface treatment, the percentages get spread across five columns and the absolute differences can grow substantially. This is not a bug — it reflects the genuine analytic choice you have made about which distinctions matter. The practical rule: settle your typology before computing BR, document it explicitly, and never lump or split types post hoc to improve a BR score. Drennan (2010, Statistics for Archaeologists, 2nd ed.) covers this trade-off in detail.

The conventional minimum is 30 sherds (or other countable units) per assemblage, and 50-100 is preferred for stable percentages. Below 30, single-artifact decisions move the percentages enough to swing the BR by 10-20 points. Bootstrapping or jackknifing the BR — repeatedly recomputing it on resampled assemblages — gives a sense of the noise band, and Drennan and Peterson (2004) provide an explicit Bayesian framework for quantifying that uncertainty. With very small assemblages (under 20 artifacts), BR results should be treated as suggestive rather than diagnostic; reporting confidence intervals alongside the point estimate is increasingly expected in peer-reviewed work.

Seriation is the technique of ordering archaeological assemblages chronologically by patterning in their type frequencies, under the assumption that type popularity follows a unimodal "battleship curve" (rising, peaking, then declining over time). Brainerd-Robinson supports seriation by giving you a similarity score for every pair of assemblages; you then reorder rows and columns of the BR matrix until the highest values cluster along the diagonal, producing a chronological sequence. Before radiocarbon dating became affordable in the 1960s-70s, seriation was the primary tool for ordering Southwestern, Mississippian, and Mediterranean ceramic sequences. It remains useful today for fine-grained ordering within a radiocarbon plateau or in regions with poor C-14 preservation.

For inference about chronological order or assemblage relationships, yes — Bayesian methods (Drennan and Peterson 2004; Buck, Cavanagh, and Litton 1996) handle sample-size uncertainty explicitly and combine seriation with absolute dates in a single coherent framework. BR remains widely used because it is computationally trivial, transparent, and produces a single intuitive number, whereas Bayesian seriation requires specifying priors and running MCMC chains in R or Python. For exploratory work and teaching, BR is still the entry point. For publication-quality chronological inference on contested sequences, Bayesian methods with credible intervals are increasingly the standard. The two approaches are complementary rather than mutually exclusive.

Quick reference

Brainerd-Robinson Similarity Calculator — quick reference
Aspect	Detail
Purpose	Brainerd-Robinson Similarity Calculator is a free browser calculator for fast, repeatable estimates.
Runs in	Any modern browser — no install, no signup.
Privacy	Inputs are computed locally; nothing is uploaded.
Best for	Quick estimates, daily reference, study, planning.