ge: our little
on
3) overlaid a
'"htened town
nt structures
close-range
tury, so this
2. ACCURACY
As for GIS, definition of accuracy is based upon the model
entity-attribute-value: generally three types of accuracy are
taken into account, spatial, thematic and temporal.
In the course of implementation of database, data collection
may assume different features: a homogeneity test may stress
the different degree of homogenity of collected data, for
different samples; also a correlation test may suggest possible
linkages between different attributes.
When hypotheses of normal distribution of data are supported,
normal statistic inference is used, otherwise non parametric
statistic inference has to be used.
In the last cases a great deal of significance tests alternative to
parametric ones are available; benefits are the following:
o it's not necessary to state precise hypotheses a priori
about statistic behaviour of a certain population;
o difficulties of computation for filing and processing
data are limited;
o their understanding is generally intuitive;
The following tests, at first conceived for lineage evaluation,
proved useful to increase the accuracy.
The first application of test analyzes fortified structures and
settlements datings and it's possible to notice the homogeneity
of attribute; in a second application comparing settlements and
churches datings, an opposite result is sharpened.
The distribution free test named U (Wilcoxon, Mann, Whitney)
in the version “fied ranks” is used. The Hy hypothesis is that
the means of two samples are equal, i.e. the samples are
homogeneous and originate from the same population; the
hypothesis is verified through the comparison between the
standardized normal corresponding to a certain significance
level and an observation equal to:
7 = 2 where:
3 3
mn S?-S rtj-ti
ss-p| m T m| Umi(U,U).
Luca EA
2
DU. > un + TE MER VU, + U, = mn
2 2 2 2
Sy = m En
XIII century has been esteemed because this kind of control is
significant only in a sincronic perspective.
There are 16 castles (sample A) and 12 ecclesiastic structures
(sample B) originate during this century. The value is 1 if dating
is provided exactly (a precision of an year), 2 if dating il less
precise, moreover values are ordered downstream the value and
identified with the sanple.
"t" represent the number of links in next groups in samples A
and B, while R are ranks of both samples (We remind that rank
value in case of links is computed giving a growing order to
common values, adding those values and dividing them for
number of links.
In the first case there are 16 values equal to 1 for A, and 10 for
B, no values equal to 2 for A and 2 values for B. The rank for
first 26 values equal to 1 is 13.5, while for ther values equal to 2
is 1,5. The final value for z is 1,66 and thus this value is minor
to the critic value of standardised normal Z= 1.96 for a
significance level of a = 0.05, we can say that samples are
homogeneus.
An opposite result (z=3.18) is achieved iterating the test for the
same sample B and another sample C (ecclesiastic structures)
for whom 17 values equal to 1 and 5 values equal to 2 are
counted (data refer always to XIII century).
Results could be read as following: a castle is a power control
centre of territory and such power is expressed by a foundation
act (a documentary source is a remembrance for future of such
act). À settlement foundation is often tied to a castle, while an
ecclesiastic structure, if it is a simply church, not a monastery,
has the predominant role to be a social gathering centre, and the
foundation act is not transmitted to future generations.
Now we propose, as a correlation test, the distribution free test
of Spearman, in the version “fied ranks”. The Hy hypothesis
concern the independence of two attributes; here the atrributes
are the dating and map reference accuracy. If computed value is
lower than the value issuing from table at a certain level of
significance, two attributes are not mutual connected.
where:
M =1/6(n’n)
712$ «À =r)
T, -M12V (6 - 1)
M-(XD? «T, + Ty)
WM - 2T, XM - 2T, )
Y —
The ecclesiastic structures group, built during XIII century, is
entertained: it includes, as above, 22 elements. Regarding
dating, values are the same of the previous test, while, regarding
the map reference, a value equal to 1 is assigned to structures
spotted on maps properly to map's tolerance, a value equal to 2
39
for less precise references, and at last the value 3 to unpreserved
structures that are linked to the aerial centroid of geometric map
entities.
Sample values, concerning both attributes, are organized in a
double row matrix, other two rows include rank of two values
corresponding to the same element; they are computed in the
way above described if values appear several times (tied ranks).
D is the difference in rank between samples. (the sum of D has
to be equal to 0).
Regarding dating, values 1 (appearing 17 times) have rank 9, 2
values (appearing 5 times) have rank 3; regarding map
references: values 1 (appearing 7 times) have rank 4, 2 values
(appearing 13 times) have rank 7, 3 values (appearing 2 times)
have rank 1,5.
The result r is 0.76, thus it is lower than reference value that is
the t of Student (t,=1.72 for a=0.05 and v=n-2=20). The starting
hypothesis is not refusable, so samples are not mutually
connected
We expected an opposite result (low level of territory control is
connected to a low level of searching out).