Study Design (III): Formulating a hypothesis
A hypothesis assumes a relationship between two or more variables
A hypothesis is “a proposition that is to be, and can be, tested by research” (Bauman). That is to say, it is “an expected but unconfirmed relationship between two or more variables” (S & S).
Four ways to write your hypotheses (using an example)
- IF-THEN condition
If a person has worse health than the person will more likely to get a flu shot. - Mathematical statements
The probability of a person to get a flu shot = f(the degree of a person’s health) - Continuous statements
The worse a person’s health, the higher the probability of the person taking a flu shot. - Difference statements (the best way)
People with worse health are more likely to get a flu shot than those with better health.
A null hypothesis
A null hypothesis is one that it hypothesizes no relationship between variables. (Given that you expect to see relationship between variables).
Here is the tricky part. When we are doing hypothesis testing, we would put our desired outcome (there is a relationship) as the alternative hypothesis and the no-relationship outcome as the null hypothesis. This is because when we do the statistical test, if we fail to prove that the difference between two hypotheses are nonsignificant, we simply “cannot reject” the null hypothesis; it does not mean we “accept” it. In this case, if we put our desired outcome as the null hypothesis, we will never be able to “accept” our hypothesis.
Hypothesis formulation with a third variable
Typically, a hypothesis statement is like this example, “People with worse health are more likely to get a flu shot than those with better health.” It involves two variables, health and flu shot. Now recall the previous post, “Study Design (II): unit of analysis and variables”. Sometimes we are interested in the factors that intervene the relationship causally to the relationship between the two variables (intervening variables), or the factors that alter the outcome variable (modifying variables). In these scenario, we could form hypothesis to verify these variables as well.
Hypothesis testing for the intervening variable (example from the previous post)
The hypothesis statement for this relationship is “Physicians with alert receipt are more likely to do dose adjustment, and in turn, are more likely to reduce adverse drug events for patients.” Here, the event that physicians adjust dosage is the intervening variable.
Hypothesis testing for the modifying variable (example from the previous post)
The hypothesis statement for this relationship is “The relationship between alerts and adverse events varies by the care settings, such that patients from inpatient settings are more likely to have side effects than those from outpatient settings”.
How about the confounding variable?
Note that for confounding variables we do not need hypothesis testing for this since by definition, a confounding variable “contributes in a non-causal way to the relationship between the independent and dependent variables”, and it influences the dependent variable (outcome) in a causal way. In this case, to test the hypothesis for a confounding variable, it just becomes another two-variable hypothesis - the confouding variable as the new independent item and the original dependent variable.
Do scientific research with hypothesis
Recall the scientific process from Study Design (I): the principles of scientific research, the process is as follows.
Now, the steps of doing a scientific research can be expressed in this workflow.
