Is it acceptable to have combination of both Directional and Non Directional Hypotheses in a study? This happens because there are conflicting outcome for certain hypotheses, so no direction (positive or negative) can be stated.
Thank you
Suba Shini What is the meaning of a directional hypothesis? Please clarify.
There is actually no indication for a "directed hypothesis" in research. You would use a directed hypothesis only in NP testing, where you have a directed loss function (you don't have a loss function, do you? - you probably don't even know what this is, do you?).
In research, the test should (very usually) help you to judge whether or not your data is sufficient to interpret the direction of the estimate (of the interesting effect). For this reason, the tested hypothesis is usually "the effect is zero", and then you determine how likely your data is under this assumption. If your estimate is positive and your data is already very unlikely under the hypothesis"the effect is zero", then it is even more unlikely under any hypothesis about negative effects. So in this case your data would be sufficient to expect a positive effect, because your data gives a positive estimate and would be way too unlikely under any non-positive hypothesis about the effect.
This is why the "undirected test" is the correct test to decide on the sufficiency of the data (to interpret a direction), and why this is the only test that makes sense in research questions.
Hi Dr Joseph Tham . The directional hypothesis indicates positive/ negative relationship between the variables, based on past studies etc. Conversely, the non-directional hypotheses is used when there is no prior studies or studies with both positive and negative outcome. Hence, no positive/ negative relationship is mentioned in the hypothesis statement.
Thank you Dr Jochen Wilhelm ..This is certainly a new piece of information!
Suba Shini are you doing NP (Non parametric) Estimation, with a loss function?
Note: when I wrote "NP testing", "NP" is for "Neyman-Pearson", not for "non-parametric"!
see e.g. here: Article Fisher, Neyman-Pearson or NHST? A Tutorial for Teaching Data Testing