"t
et S
samples, we can assess the significance of all the interactions, and in the meanwhile
obtain further information about the main effects, by analysing all the data
together. We thus have three main effects, F, T' and t, and the interactions FT,
Tt and tF, and a residual. This analysis of variance is in Table 10.
Table 10
| SS. | d.f. M.S. |
A = 3s a a Le
Between-t ir anos [ned = 3. 12.43 PE DZ | o
TF Interaction 50140 (ny—1) (np—1) - 6 835702 (opp?| |o mo
inion 1598 (000) = 9 87 0 oo
MT Interaction | 37.89 |(n—1) (ny—1) =" 6.32) 0,2 ump” Or,
Residual ~~ | 7212 m—005—0 (=) = 15) 4010? [ener
Total | 1817.67 npnn—1 = 47 mr
To ease the interpretation of this analysis, the components of variance are
given on the right hand side of the Table: The variances or^, or? and 0°? measure
the fluctuations of shrinkage due to the three factors covered by the experiment,
namely the type of film, the temperature at which it is stored and the period of
storage. There are also the terms orr^, or? and oi? expressing the interactions of
these three factors. The practical asset of the method of analysis resides in the
fact that the significance of any of the variances and interaction terms can be
assessed, on any probability level, by comparing suitably chosen mean squares.
The component variance or” describes the temperature changes which are
common to all types of film, and all periods of storage. Similarly or^ describes
the variation of shrinkage from one type of film to the other which can be
distinguished at every storage temperature and period under consideration. The
common effect of storage time is similarly represented by the component variance
61”, It may happen that the effect of temperature, for instance, changes from film
to film; orr? represents this effect, the time factor being held constant. The same
term describes the change of shrinkage variation among films due to changing
the temperature. The order of the subscripts is therefore immaterial. This applies
to all the interaction terms. The terms oz? and o; are similarly introduced to
measure the extent to which the storage period influences the temperature effect
and is influenced by the type of film, respectively. Finally, the interaction term
opp? estimates the effect of the choice of one factor on the interaction of the
other two; and o? is the variance of the experimental error. Since we do not
know 00°, and in the absence of the original data, we have no means of assessing
the second order interaction orri”.
The routine of partitioning the total sum of the squares of the deviations
from the general mean into component sums of squares is explained in any text-
book on statistical analysis, and need not detain us here. In general each mean
square estimates a linear function of the variances and interaction terms whose
17