Tuesday, April 14, 2015

Statistical Analysis of Hively and Horn's Archaeoastronomical Claims at the Newark Octagon 2004

Statistical Analysis of Hively and Horn's
Archaeoastronomical Claims at the Newark Octagon

Christopher S. Turner 

Poster presented at the 2004 joint meeting of the Southeastern Archaeological Conference and the Midwestern Archaeological Conference


October 21-23, 2004 St. Louis


Abstract

In 1982, Earlham College professors Ray Hively and Robert Horn published an analysis of calendrical sightlines at a Hopewell geometrical earthwork. The octagonal enclosure in Newark, Ohio, was shown to define specific lunar rise and set points along the horizon surrounding this notable site. This poster presents statistical analyses of these claims, using both chi-square and resampling tests. Using these methods, the Hively and Horn results are found to be statistically significant. Such results lend credence to their other, "ad hoc" sightlines (as the authors called them), and to the overall plausibility of other substantiated Hopewell archaeoastronomical claims.

Introduction

            Research published in 1982 by Ray Hively and Robert Horn detailed the presence of astronomical sightlines at the octagon component of the Newark earthworks (33Li10)(Figure 1). The Newark Octagon, believed to date to the late Middle Woodland (c AD 250), is one of what was once a score of such sites scattered throughout southern Ohio.  These geometric earthworks, with dimensions exceeding 300 meters, are endemic to the heartland of the Ohio Hopewell.
                    Standard to the methodology of archaeoastronomical studies are statistical analyses of proposed alignments. Here I present such an analysis of the Hively and Horn Newark Octagon sightlines.

Null Hypothesis

            I will restrict my analysis to sightlines specific to the octagon embankments and the eight vertices formed where they converge. Between these eight points, Hively and Horn claimed to find seven of the eight lunar extrema azimuths marked. As with all such assertions in archaeoastronomy, the obvious question is “What are the chances that these sightlines are the result of random association?”   How likely is it that the alignments formed by these eight points would randomly indicate seven out of eight specific sightlines?
            The null hypothesis is, then, this:

                        Events marked by the octagon number no greater than what would be expected from random occurrence.

      As is conventional, if a statistical probability under 5% likelihood can be demonstrated for the observed outcome, the null hypothesis is considered false.

Framing the Problem
      What are at issue here are visual sightlines. The inter-vertex lines at the octagon define these. Specific directions are expressed in degrees of azimuth around the horizon. Thus, statistically, the 360-degree horizon effectively becomes the sample universe.
      The number of total possible sightlines between the eight vertices of the earthwork is simple to calculate: it is n (n-1), where n is the number of vertices. Thus, there are
8 x 7 = 56 possible sightlines defined between the eight gateways of the octagon (Figure 2).
      To calculate probability, one must define a sampling interval. Here it is defined in degrees of azimuth. It would be meaningless to define the sampling interval as smaller than the known accuracy of the survey data upon which the archaeoastronomical analysis is based. A value of one or one-half degree is commonly used. It is problematic to insist on greater accuracy due to questionable archaeological context, such as erosion of earthen embankments, or displacement of megalithic or architectural features, for instance.
      Lastly, as noted earlier, the celestial events marked by the octagon, according to Hively and Horn, are the lunar extrema, of which there are eight. It is these eight independent azimuths along the horizon that are considered as “hits” in the probability analysis.


The Hively and Horn Data and the Statistical Sampling Interval
      As mentioned already, Hively and Horn posited that seven of the eight lunar extrema are indexed by sightlines defined by the octagon vertices. However, as shown in their data table, three of the seven sightlines involve azimuth errors greater than one-half degree, putting them outside of the range of the one-degree sampling interval used in this statistical analysis. This leaves only four valid sightlines. Hence, when using the one-degree sampling interval, the statistical likelihood of only four “hits” will be tested.
      Looking at the data in another way, I expanded the sampling interval to twice the largest error value found amongst the proposed sightlines. The largest error in the Hively and Horn data table is 1.3°, so I set the sampling interval to 2.6°. In this way I am able to test the likelihood of the occurrence of all seven of their proposed lines.
      Given these conditions, I calculated the expected frequencies of lunar extrema sightlines that would occur randomly. With a sampling interval of one degree and 56 randomly generated sightlines, we should expect 1.24 lunar extrema sightlines. With the sampling interval set to 2.6°, we should expect 3.24 lunar extrema sightlines. These expected frequencies were compared with the Hively and Horn results using chi-square (Guassian) statistics, and with resampling (Bayesian) statistics.
     
Chi-Square Analysis
      The chi-square results are shown in Figures 3 and 4. Because we are examining whether there are more sightlines than would be expected randomly, the null hypothesis is said to be directional. Also, because the test is of a simple binary choice (hit or miss of the lunar extrema azimuth), it is said to have one degree of freedom.
      Both chi-square results negate the null hypothesis. With the one-degree sampling interval, the probability of finding 4 lunar extrema sightlines is 1.3 %. When the sample interval is 2.6°, the probability of finding 7 such sightlines is 3.1%. 
     
Resampling Results
      The chi-square test is an approximation, and it is said to overstate significance, especially when the number of degrees of freedom are low, or when the value of the “expected” factor is low, as are both the case in the above examples. For this reason I have also included statistical resampling results. Resampling is done with computer software (Simon 1999). The program repeatedly generates (100,000 times in these examples) the probability model being tested, and counts the results. Here, 56 sightlines are repeatedly and randomly simulated, and the program counts the number of times the target intervals (in this case, the eight lunar extrema) are “hit”. The mean of this value should (and does) converge on the “expected” frequency.
      Additionally, the software program counts the number of times that the frequency of lunar sightlines meets or exceeds that found by Hively and Horn. This value is expressed as a percentage as with probability. Again, because we are examining what percentage “meets or exceeds” the critical value, the hypothesis is directional (or “one-tailed”).
      As anticipated, the resampling probabilities indicate somewhat less statistical significance than the chi-square results, but they are still under the 5% critical value. With the sampling interval at one-degree, the probability of finding 4 lunar sightlines is 3.7% (Figure 5). With the 2.6° sampling interval, the probability of 7 sightlines is 4.3%  (Figure 6).

Independence of Sightlines
            To meet theoretical constraints, samples being tested with chi-square must be mutually independent. That is to say, in this case, the random sightlines must be independent of each other. The actual sightlines at the Newark earthwork, however, are obviously interrelated because they are defined by a regular polygon. The key word here, though, is regular. It is arguable that a given symmetry of sightlines defined by an octagon could favor the likelihood of multiple lunar extrema azimuths being selected, more so than especially a randomly generated set of such lines. It is important to note, then, that the octagon is not strictly symmetrical. Hively and Horn demonstrated how given angles formed by the embankments were distorted from regular symmetry in the direction of the lunar extrema azimuths.
      The twin earthwork to the Newark Octagon, the High Bank Octagon, is even more distorted from a regular polygon. The High Bank earthwork complex, in Chillicothe Ohio, was also analyzed by Hively and Horn (1984). It was also found to index the lunar extrema and the four solstice azimuths.
      The Hopeton Works, also in Chillicothe, exhibit minimal geometric regularity, the polygon anomalously somewhere between an octagon and a square. Yet it too indexes the lunar and solar extrema, as well as the equinox and cross-quarter dates (Turner 1982, 1983, 2004).
      Geometric regularity was important to the Hopewell, but was secondary in importance to the accuracy of the calendrical sightlines. Surely the Newark Octagon was the apex of Hopewell geometric earthwork construction, the octagon being the most complex figure they crafted. It is not however, and never was intended to be, an equilateral octagon. As noted by Hively and Horn, the angles at the vertices alternate, so we are left with a flattened, or bilateral octagon. It seems evident that the builders chose this particular bilateral octagon, out of the many such possible, to best match the azimuths of the lunar extrema. The intent was not to adhere to a perfect equilateral octagon.
      Given the variation in geometric symmetry between Hopeton, High Bank, and the Newark Octagon, but remembering that each one indexes the corpus of lunar extrema, it is arguable that the overarching intent of such earthwork construction was to mark the calendrical sightlines. Any symbolism incorporated into the architecture by virtue of the geometric design should be interpreted as being secondary to calendrical sightlines themselves. Hence the non-independent nature of the sightlines, rather than being a theoretical obstruction, is exactly that for which we are testing: that the interdependence of the sightlines is a result of the calendrical functioning of the earthwork. The criticism that the sightlines are non-independent by virtue of the octagonal construction is answered by evidencing the primary influence of the calendrical intent, and the intentional distortions from idealized geometric regularity as found in these earthworks.

An Earthwork Phylogeny
      The above three cited earthworks, Newark Octagon, High Bank, and Hopeton, share morphology: they fall under the rubric of what Martin Byers calls the “C-R” groups, i.e. circle-rectilinear (Byers 1998:139). The majority of Hopewell geometric sites have a three-component, or tripartite arrangement. The “C-R” groups by contrast consist of a circle, or more properly an ovoid, conjoined to a polygon.
     The two sites, Newark Octagon and High Bank, are further distinguished by having each an octagonal polygon, and because their near-perfect circular components share nearly the exact same diameter. Hopeton, their awkward cousin, fails to trace an octagon or boast precise circularity, but it joins the two in a group: C-R earthworks that mark lunar sightlines.
      This apparent interrelation has been long noticed, and was commented on by Maclean (1879:83) “It would appear that the Hopeton and High Bank Works were either modeled after that at Newark, or else the last was a combination of the other two”.
      In this triad, we see a clinal variation of both geometrical and calendrical precision. Also, though carbon dates are not yet conclusive, I would assert that presently they suggest, or are consistent with,  a chronology of: Hopeton, 1st century AD; High Bank second century; and Newark Octagon third century AD. This temporal ordering may be evidenced in the degree of geometrical complexity found at Newark Octagon relative to High Bank, and Hopeton. Similarly, there appears to be increased specialization in the selection of specific types of celestial sightlines marked. Hopeton, which is the least geometrically regular of the three, and perhaps the oldest, notably indexes the largest number and greatest variety of celestial alignments. Also, Hopeton has many multiple or redundant sightlines. 
      High Bank, clearly intermediate in geometric symmetry and complexity between the other two, also marks an intermediate number of celestial events. And this is not some spurious parsing of the types of calendrical sightlines. High Bank indexes all of the lunar extrema events, and the four solstice azimuths, and nothing else. It trends toward an octagonal outline (and accompanying angular constraints) at the expense of omitting Hopeton’s equinox and cross-quarter sightlines.
      Finally we have Newark Octagon, the most morphologically symmetrical of the three, but what’s more, yielding the most restricted set of celestial phenomena: its sightlines mark only the lunar extrema. This statement can be restated as such: the Newark Octagon has been designed, especially in comparison to High Bank and Hopeton, to not mark any solar phenomena. Arguably and demonstrably, calendrical accuracy of sites such as Hopeton or High Bank was sufficient. No practical reason exists compelling the avoidance or non-inclusion of solar sightlines at the Newark Octagon. We could conjecture that a taboo as such, or perhaps a refinement of celestial mana, was at play, restricting the embodiment of solar angles and the symbolism they carried, magnifying or concentrating things lunar.
      Hence it appears that the Hopewell intentionally included the lunar lines and excluded the solar at the Newark Octagon. The statistical question thus arises: what is the probability that this octagon could index some of the lunar lines but still miss all of the solar ones in question?

A Final Statistic
      Remembering the earlier discussion, we have 4 lunar extrema sightlines marked with the one-degree sampling interval, or we can include all seven Hively and Horn lines by using the larger sampling interval (2.6°).
      There are ten rising and setting solar events that are not marked by the Newark Octagon that are marked at other Hopewell sites: these are the solstices, equinoxes, and cross-quarter dates.

      Using the one-degree sampling interval, the likelihood of 56 random sightlines both:
§  marking 4 out of 8 lunar extrema azimuths, and
§  not-marking any of 10 solar azimuths
      is P = 0.74%.

      When using the 2.6° sampling interval, i.e. when including all seven of the Hively and Horn sightlines, the likelihood is one in 1492 (Figure 6).

 (Author’s note: In the original text of this paper, which I distributed to accompany the conference poster presentation in 2004, the last probability figure of 1 in 1492 was stated as for a directional hypothesis. Since then, specifically when I was editing my 2011 article in Time and Mind,  I reasoned that because the above combined probability result is for a combination of “hits” and “misses”, that the  analysis would then  be  non-directional, or “two-tailed”. Therefore, in the Time and Mind article (page 308), I cited the combined resultant likelihood as per a non-directional hypothesis as 1/746, half of the directional result of 1/1492  Now, I have been reminded that in order to posit non-directionality in the statistical results, that the two hypothesized events, though independently tested directionally and using both “tails”, (the right tail in the lunar case and the left tail in the solar case), must be mutually exclusive in order to warrant the “non-directional” interpretation.  This is definitely not the case here. The avoidance of solar sightlines and the “hitting” of lunar sightlines are not mutually exclusive events which are being tested simultaneously by the same null hypothesis. The mutual exclusivity presented in each null hypothesis is, in the first case: do the inter-vertex octagon sightlines “hit” azimuths that index the 8 lunar extrema more than the expected number of times , or do they not?  The second null hypothesis, concerning the solar sightlines, asks: do the  sightlines “miss” the 10 azimuths that index solar events less often than what would be expected, or do they not. Therefore I would further adjudge my conclusion here and suggest that my original combined probability figure of 1 in 1492 is sound. Note that in each example above the terms “more than” or “less often than” denote these hypotheses as bring directional.
 

Summary

            Hively and Horn revealed that the Newark Octagon, a Middle Woodland Ohio Hopewell geometric enclosure, marks seven of eight lunar extrema azimuths, as defined by inter-vertex sightlines. Using chi-square and resampling tests, I analyzed these calendrical alignments for statistical significance. Using two different sampling intervals, all results negated the proposed null hypothesis.
            The tests are not invalidated by the non-independence of the sample (sightlines). The very interdependence that the sightlines encode is due to their intentionally being aimed at specific calendrical azimuths, not to the overarching geometric outline of the enclosures. This calendric intent is exactly what the archaeoastronomical results suggest in the first place. Geometric symmetry was not the primary goal in constructing these earthworks. The degree of geometric regularity present in the earthworks varies greatly from site-to-site. Specific embankments are skewed away from geometric symmetry and toward calendrically significant azimuths.
            Within the triad of enclosures, Hopeton-High Bank-Newark Octagon, there are clinal variations in geometric and calendrical complexity that may also reflect a relative chronology between the three sites. Hopeton embodies the greatest number of types of calendrical sightlines, High Bank less, and Newark Octagon the least. The latter earthwork marks only lunar events and appears to be constructed to specifically avoid solar azimuths.
            I tested this last point statistically, and found extremely low probabilities that this particular combination of circumstances occurs randomly.
            Statistical analyses are fundamental to argument veracity in archaeoastronomy. However, the methods are often questioned as suspect. The above paper is not intended as the final word on Hopewell archaeoastronomy, but it can serve as a jumping off point for further discussion, or more importantly as an adjunct to the range of other forms of argument available to anthropologists.








Figures























Figure 6


References Cited

Byers, A. Martin
1998    Is the Newark Circle-Octagon the Ohio Hopewell “Rosetta Stone”? A Question of Archaeological Interpretation. In Ancient Earthen Enclosure of the Eastern Woodlands,  edited by Robert C. Mainfort and Lynne  P. Sullivan, pp. 135-153. University Press of Florida, Gainesville.

Hively, Ray  and Robert  Horn
1982    Geometry and Astronomy in Prehistoric Ohio. Archaeoastronomy  
supplement to the Journal for the History of Astronomy 13(4):S1-S20.
1984  Hopewellian Geometry and  Astronomy at High Bank. Archaeoastronomy
  supplement to the Journal for the History of Astronomy 15(7):S85-S100.

MacLean, J.P.
1879      The Mound Builders. Robert Clarke and Co., Cincinnati.

Simon, Julian L.
            1999   Resampling Stats: the New Statistics. Version 5.0.2. Resampling Stats, Inc.

Turner, Christopher S.
1982    Hopewell Archaeoastronomy. Archaeoastronomy 5(3):9.
Center for Archaeoastronomy, College Park, Maryland.
1983    An Astronomical Interpretation of the Hopeton Earthworks.  Manuscript on file at the Hopewell Culture National Historical Park, Chillicothe, Ohio; and at the Ohio Historical Center, Columbus.

            2004  Middle Woodland Archaeoastronomy in Ohio. Paper presented at the 7th Oxford Conference on Archaeoastronomy, Flagstaff, AZ.

           

      

No comments:

Post a Comment