Drug affects individuals but ANOVA not significant: references? Let's suppose that I measure the effect of a drug on blood pressure in several people many times. For some people, the drug increases blood pressure significantly and for others decreases it significantly. So the drug is affecting the individuals. Because the direction of the effect is different for each person, it is possible that if I perform a global statistical analysis such as ANOVA, the drug shows not significance. 
So, it is incorrect to conclude that the drug does not affect people by using the ANOVA. I would like to know references that talk about this problem for being able to write about it using more correct terminology. 
More specifically, this is what I have. I collected a dose-response curve for each subject with and without drug, fit a cumulative normal distribution to each condition and estimate the mean and confidence intervals by bootstrapping. So I collect many data for each subject. For each subject the mean with and without drug is significantly different(no overlap of confidence intervals) but, for some observers the drug increases blood pressure while for others decreases blood pressure. So if I use the means to perform an ANOVA, the effect of the drug is not significant.
So, it seems to me that report only the ANOVA without caring about the individual effects is not correct, but this is sometimes done in my field. So I was looking for some references that pointed this out. 
EDIT: I think I chose the wrong example using drugs and blood pressure. So I asked another question using an example closer to my real problem fitting a model of each subject and perform a t-test over the estimated parameters
 A: More than likely the problem you're observing is simply random variation.  From measurement to measurement some people's blood pressure will go up and some will go down. Observing this alone does not mean the drug had any effect whatsoever.
While it's possible that the drug does increase the blood pressure in some people and reduces it in others I'm willing to be that in your data the people who tended to have high blood pressure the first time they were measured had lower blood pressure the second time while those who had low blood pressure in the first session have higher in the second. That's called regression toward the mean and is still just a consequence of random variation.
A: I would test for an interaction between the drug effect and the subject in a mixed-random effect model. A statistically significant interaction will mean that the subjects differ in their response to the drug. If the differences are due to variation / regression to mean, we will not observe a statistically significant interaction.
