Worksheet effect
Limiting ourselves to data of one informant (because of the
fieldworker effect), we notice that our measurements
are effected by worksheets.
subset: fieldworker: Lowman
worksheets: all
826 informants
grouping: fieldworker + worksheet + community: 426 locations
(two locations merged because of identical coordinates)
maps: clustering, Ward's method
Lexical differences
The existence of a worksheet effect on measuring differences of lexical choice
is to be expected. Different questions prompt the informant to different alternatives.
data: lexical items
method: Gewichteter Identitätswert
With only three clusters, the Preliminary South Atlantic worksheet stands out
as a single cluster, with only very few exceptions:
With a division into seven clusters, there is a strong suggestion of a
worksheet effect between Middle Atlantic and South Atlantic:
Phonetic differences
Strangely enough, there seems to be a worksheet effect in the phonetic data.
With all lexical variants removed and all data transcribed be a single
fieldworker, this effect is unexpected. Perhaps our method is too susceptible
for secondary effects, such as word frequencies, although we try to
compensate for that effect by normalisation.
data: raw phonetic strings, no lexical variants
method: Levenshtein
With only four clusters, hints appear of an effect of all three major worksheets used
by Lowman:
With seven clusters, a worksheet effect between Middle Atlantic and South
Atlantic is evident:
With eleven clusters, a nearly complete worksheet effect for the Preliminary
South Atlantic appears: