Other Transformations in the Analysis of Variance

Sometimes in an analysis of variance the dependent variable is not a continuous variable but discontinuous.  

For example, in discussing blemishes in objects, the number of blemishes of course has the lower limit zero but can go to very large numbers. Technically, the distribution is probably Poissonian. Under these circumstances it is more appropriate to use lie square root of the number of blemishes as the dependent variablet ). Where the number of blemishes are in many cases less than 10, it is slightly preferable to use, where x is the number of blemishes, Vx -H 1/2 

Occasionally runs fail to produce results the sample may get lost or any similar accident may happen. In the case of a two factor experiment, of r rows and c columns, one missing value may be efficiently estimated by the formula( ) rR + cC — T 

in Table 12.1, if the value of TgWz (actually the observed value was 27) had been missing, we would have estimated it as 4 X 131 + 4 X 78 — 481 

In carrying out the analysis of variance, this value is inserted as though it had actually been observed, and the analysis then proceeds normally. However, the degrees of freedom for the Residual and for the Total are both decreased by unity. 

---