WORKING METHODS FOR COMPUTERS 475

TABLE VII

SATURATION COEFFICIENTS

A.
By Simple Summation

B. By Weighted Summation.

i. First Approximation.
ii. Minimum Rank.
Fin
al Approximation.

ist Factor.
and Factor.
3rd Factor,
ist and Factor. Factor.
3rd Factor.
ist Factor.
2nd 3rd Factor. Factor.

Comp.
•8l8
-197
-042
•822 •
199
•°73
•8325
•1594 ^567

Read.
"5^9
•490
— •163
'597 '
503
—159
•6225
'4577 —1952

Spell
-512
•355
-071
•506
342
•086
•5238
•3185 -0610

Hand.
'473
—405
— •140
-472 —
401
—-129
•4486
— 4336 — 1006

Draw.
*592
-•488
— 139
'597 —
498
— -144
•5690
— 5387 — 1126

Writ.
•549
—139
•262
•546 -
144
-272
•539°
—1553 -2795

(Unweighted)

(Weighted)

Totals

•oio
—-067
001
— -ooi

----- OOO2 '0008

is based (principles that are also those adopted by Spearman and Thurstone)
the proposal would force us to assume that with each specific factor the
weighting for every test except its own is exactly zero, and therefore exactly
equal. This would be intelligible if we could assume that each of the pro-
cesses tested—writing, for example—called into play a large and important
specific * ability * or * brain-centre/ peculiar to itself and sharply localized.
Actually, what is specific to each set of test-measurements, taken in isolation,
with only a single application of each test, is not some unique ability or highly
specialized neural process, but certain minor accidental conditions accom-
panying its administration or involved in its construction. Yet even these
conditions can hardly be considered as remaining strictly constant for all the
other tests. If they do not influence them positively, they are likely (in some
small measure) to influence them negatively and in varying degrees. Thus,
to be consistent, all general-factor methods of analysis (including Thurstone's
centroid method) should treat the specific factor, not as a large primary
ability in a * simple structure ' with a positive saturation for one test and zero
saturations for the remainder, but as a small bipolar factor. With Hotelling's
method of principal components (which also assumes the variances to be unity)
there is at least one such bipolar factor which is strictly specific (in the sense
of having a positive saturation for one test only); but this is mainly an
artefact, since its large negative saturations are required merely to neutralize
the enlarged positive saturations which appear in the preceding factors owing
to the enlargement of their variances.
For special purposes yet other working methods may be tried. For