Null Hypothesis - A short Conversation

Student 1:

our correlation problem the professor said that the null hypothesis was that they are not correlated .. so if we chose the null hypothesis as ‘Related’ then the same data set will prove that they are not related. What decides what is to be chosen as a null hypothesis ?

Student 2:

As far I read and understood Null hypothesis should be an evenly or normally applicable cases which obeys guassian bell shaped population distribution curve, z and t distribution and highly probable. With tests like F-test, chi square test, ANOVA based on type of data and significance factor, we always like to disprove null hypothesis to state that we discovered something which is not NORMAL, to claim EUREKA moment. If we fail we will say we failed to reject null hypothesis rather saying NULL hypothesis got proved or won. Considering the professor answer, he says that the data is already "not related" in general sense and it is under the bell shaped distribution population. Anyway we will fail to reject NULL hypothesis, if we take "not relating" as NULL hypothesis. Taking the data as "related" as our NULL hypothesis does not make any sense if that is not the highly probable happening and falling under bell shaped curve or distribution so we could disprove and say we discovered something against it.

Student 3:

Null Hypothesis by definition states that there is no relation between the compared fields. So you can't choose the null hypothesis as "related". For example the field "Employee ID" will most likely have no relation to the field "COVID Positive". So if we were to compare the chi-square of the 2, you'll see it fall below the required value, which would mean that null hypothesis is valid and thus you can ignore the field