Jacqui Mulville wrote:
> This is a query about how much of an assemblage you need to
> get valid results.
The simple answer is that depends on what questions you wish to answer and
how much error you will tolerate in your estimated numerical values.
The basic reference has to be:
Orton C (2000) Sampling in Archaeology. Cambridge Manuals in Archaeology.
Cambridge University Press, Cambridge.
The next question is whether you want to know about the excavated
assemblage or to make wider deductions about the populations from which it
is derived. I.e are you ignoring taphonomy, particularly biostratinomy?
The best statistical treatment of taphonomic factors I have seen are the
recent papers by Alan Rogers:
Rogers AR (2000) Analysis of bone counts by maximum likelihood. Journal of
Archaeological Science 27 (2) 111-125
Rogers AR (2000) On equifinality in faunal analysis. American Antiquity 65
(4) 709-723
Rogers AR (2000) On the value of soft bones in faunal analysis. Journal of
Archaeological Science 27 (7) 635-639
After that when you are asking questions about the assemblage, it is a
case of simple sampling theory. If you want to know the male:female ratio
and you believe it to be 1:100 or 2:100 then you will need a sample of
300+, if you want to distinguish 1:100 from 2:100 you will need much
larger samples. If you want to know the proportions of major species to
the nearest 10%, then a sample of 100 is more than sufficient. However
the samples in both cases must be truly random samples from the
assemblage.
> What would you do with an assemblage of 17,000
> ident fragments, from a restricted early medieval time
> period, derived from pits......and a similar site has been
> reported on from close by? Can you justify full analysis,
> if not what would you do?
If you think the two sites are very similar then analysing small samples
(up to a few hundred) will tell you that is not much different, if you
want to know what those small differences are then you may have to look at
every fragment.
> As for small assemblages, to me it always depends on what
> they are - so I will analyse tiny amounts if I think they
> add some exciting/useful info, but would not consider them
> valid sample sizes for statistical analysis.
And in some periods/areas an accumulation of small samples is all we have
in order to get large enough samples for statistical analysis.
> To start the debate here are two examples.....
> Umberto, Mark and I for Castle Mall, Norwich, used only
> -Over 100 NISP per period for relative percentages of
> animals, and the MNI over 30 for rel percent from MNI.
<snip>
> I also know that Hambleton (1999) has suggested that the
> optimum sample size for a comparative study of species
> proportions in British Iron Age faunal assemblages is a
> total NISP of greater than 300.
Again it depends what you want to do. The Norwich method is no good if
you want to be sure of species percentages to the nearest 1%, but I doubt
that you wanted that. Hambleton's sample size will allow more precise
estimates of proportions, and is also chosen to allow a statistical
confidence at better than 95% that a species present at 1% of NISP has not
been missed.
best wishes
Andrew
=========================================================================
Dr. Andrew Millard [log in to unmask]
Department of Archaeology, University of Durham, Tel: +44 191 374 4757
South Road, Durham. DH1 3LE. United Kingdom. Fax: +44 191 374 3619
http://www.dur.ac.uk/a.r.millard/
=========================================================================
