Hypothesis testing is valid only for random variables and their different combinations. So if you say about two variables, one is independent and other is dependent from the first, this worrks for deterministic analysis such as mathematical analysis. For the random and statistical analysis you should tett about two variables each of is dependent from the other.

Hypothesis is just a random event, and sometimes, using some test, you can find the probability of other random event that is somehow connected with your hypothesis.

Maybe your researcher had a deal with several different random populations and proposed each hypothesis for each population. I think it could be.

Useful reference: Maurice Kendall, Alan Stuart. The Advanced Theory of Statistics. Volume 2 - Inference and Relationship.

Thanks, Petr. The reason for my question is that I have seen several scientists who have looked at the effect of a chemical on many different biological or behavioural tests and many parameters (dozens) within those tests given to the same animals without any biological or other type of justification for the choice of their tests and parameters. Then they run many statistical tests and present the results of the parameters that have shown a statistically significant difference as scientific conclusions of their investigation. I believe that they only generated hypotheses, and no conclusions. My interpretation of this situation is that the researchers have run an hypothesis generating study. They did not know what they were looking for, so they tried anything they could and, of course, they found some statistically significant differences between their control and experimental groups, which one would expect at least at the alpha criterion they accepted, e.g. 0.05 typically. Even if they say that they were testing dozens of hypotheses (e.g., I wanted to see if A had an effect B, and on C, and ... and on Z)., my point is that, if they tested dozens of hypotheses on the same animals, you didn't test any hypotheses. So I came to wonder about how you distinguish between a study that is truly hypothesis testing vs. hypothesis generating. I believe that many hypothesis-generating studies masquerade as hypothesis-testing. What are the criteria of a hypothesis-testing study that operationally distinguish it from a hypothesis-generating study? Thanks.

Jacques, there is exist one mathematical model, which is very useful for different theoretical outcomes: each observation in the set is the realization of separate random variables. For example, let consider that set consist of 30 observations. To study the distribution of set's mean we can assume that we have 30 separate independent random variables, each of them gave as only one observation. That is very useful theoretical approach.

Practically we can assume that all of these 30 variables have the same distribution, so we can think that we have a deal with independent observation from one population.

So it can be that genearating so much hypothesises is the result of not correct interpretation of some theoretical approaches. Generally you can generate as much hypothesises as you need. But in the lot of tests for making statistically significant decision about the hypothesis you have to have the set that consist not less than 30 observations.

So it can be that your colleague generates hypothesises without real chances to statistically valid accept or decline it.

"Generally, inductive reasoning uses the data to generate ideas (hypothesis generating), whereas deductive reasoning begins with the idea and uses the data to confirm or negate the idea (hypothesis testing)"

http://bmj-ebn.highwire.org/content/3/3/68.full

Thus there is no magic number where inductive reasoning

suddenly becomes deductive reasoning.

Thanks, Ian. Good reference!

Good Google!