The non-interaction model looked like this:


Current SCC = 33159.253(Injection Binary) - 15598.810(Non-injected binary)

-473.830(Age) - 0.423(Average production) +

		 1.146(previous SCC) + 6722.368(body score)

To illustrate the significance of this model, it is helpful to use an imaginary

situation.  If two animals had the same age, average production, previous SCC,

and body score, the only difference in their predicted scores would come from the
 
coefficients of the two binary variables.  If the animal had been treated with

BST, there would be a ³1² as the value for the ³Injection² predictor and a ³0² as
 
the value for the ³non-injected² predictor.  If the other animal had not been

injected, it would have a value of ³0² as the ³Injection² predictor and a value

of ³1² as the ³non-injection² predictor.  Therefore, there would be around a

 48,748.063 somatic cell difference between the predicted SCC of the injected and

non-injected animals.  This value was obtained from model one (see appendix), and

the t-test applied to the injection binary gave a value of .012, which means the

hypothesis that there is no difference in the scores is rejected.  This means

that there is a significant difference between the predicted scores of the

experimental and control groups while controlling for other measurable factors. 

	This was the original model, and was used to gauge the predicted difference

in somatic cell count between the two groups.  There were other questions that

had to be solved as well, however.

	The first additional question that arose following the first regression model

was: are the co-variables necessary in analyzing the observed data?  This

question was answered with the following model:

	Current SCC = 243904.118(Constant) + 89751.152(Non-injected)

The t-test value of the binary variable was .266, and this value accepts the


hypothesis  that there is no significant difference between the means of the two

 groups.  Therefore, it can be concluded that the co-variables are necessary for
 
accurate prediction because there was a significant difference between the two

groups with the co-variables in the model, but when they were removed there was

 not a significant difference.  

 
	The next question to be looked at concerned the variables used as

 co-variables in the experiment.  Were they all necessary to ensure accurate

 prediction?  Based upon the t-tests applied to all the variables in the original

 model, it was determined that the pre-test SCC was the only significant factor

in determining an animal¹s response to BST.  It was significant at the .05 level,

 with a value of .000.  (see appendix)  In all models, the only variable that was

 found to be significant in predicting the response variable was the pre-test

value.


	The final question that came up in analyzing the data was a question of

 interaction.  This question was solved with a command file that executed an

F-test on the observed data.  The F-test failed to reject the hypothesis that

there was no interaction.  The probability that there is interaction was .390. 

This is not significant, and the interaction model (model 3, see appendix) is

rejected.