n of the wave-length is found from the
_spectrum_ of the stars.

There are blue (B), white (W), yellow (Y) and red (R) stars, and
intermediate colours. The exact method is to define the colour through
the mean wave-length (and not conversely) or the _effective_ wave-length
as it is most usually called, or from the _colour-index_. We shall
revert later to this question. There are, however, a great many direct
eye-estimates of the colour of the stars.

_Colour corresponding to a given spectrum._

_Sp._ _Colour_ _Number_
B3 YW- 161
A0 YW- 788
A5 YW 115
F5 YW, WY- 295
G5 WY 216
K5 WY+, Y- 552
M Y, Y+ 95
-----------------------------
Sum ... 2222

_Spectrum corresponding to a given colour._

_Colour_ _Sp._ _Number_
W, W+ A0 281
YW- A0 356
YW A5 482
YW+, YW- F3 211
WY G4 264
WY+, Y- K1 289
Y, Y+ K4 254
RY-, RY K5 85
--------------------------------
Sum ... 2222

The signs + and - indicate intermediate shades of colour.

The preceding table drawn up by Dr. MALMQUIST from the colour
observations of MUeLLER and KEMPF in Potsdam, shows the connection
between the colours of the stars and their spectra.

The Potsdam observations contain all stars north of the celestial
equator having an apparent magnitude brighter than 7m.5.

We find from these tables that there is a well-pronounced _regression_
in the correlation between the spectra and the colours of the stars.
Taking together all white stars we find the corresponding mean spectral
type to be A0, but to A0 corresponds, upon an average, the colour
yellow-white. The yellow stars belong in the mean to the K-type, but the
K-stars have upon an average a shade of white in the yellow colour. The
coefficient of correlation (_r_) is not easy to compute in this case,
because one of the attributes, the colour, is not strictly graduated
(_i.e._ it is not expressed in numbers defining the colour).[5] Using
the coefficient of contingency of PEARSON, it is, however, possible to
find a fairly reliable value of the coefficient of correlation, and
MALMQ