I am working with the cause of preference for employees in a specific sector using PLS-SEM.
My model fits well and all path coefficients are positive. But for one hypothesis or relations p value is 0.06. Can I accept that hypothesis?
Pls help me with some references
Probability level is a researcher choice, 5%, 1%, or even 10% all accepted in research. However, the lower the p value, the higher the significance. Regards.
No, you can not "accept that hypothesis". Abscence of evidence is not evidence of absence (of an effect). You are just not able to reject the hypothesis at a 5% level of significance. This means that your data contain only little (too little) information about the estimate to discern it from the hypothetical value. It is no indication that the hypothetical value would be correct.
You may have a look at the (95%-)confidence interval. It gives the range of hypothetical values that are "not too incompatoible" with your data. If this interval is fully within a range you consider "irrelevant", then you may claim that the effect is not relevant. Otherwise, your data is simply not sufficiently conclusive to say anything useful.
Hello Jochen Wilhelm. I hope all is well with you. My response to your question is it depends. If you set your p value at .05 level of significance, then, I agree with the answer you added: No, you cannot accept that hypothesis.
Even if p < ... than any level you cannot reject or accept any hypothesis. While performed or advocated in books and by teachers, the beauty of "classical" hypothesis test is that it does not attach any probability to hypothesis (https://en.wikipedia.org/wiki/Misuse_of_p-values). The hypothesis is fixed and unknown each experiment is a "random" event. We cannot look into the future and say the next experiment will succeed only backwards and cumulate all information.
The "classical" way of looking is a form of infinite regress where information is cumulated over an infinite amount of experiments. Jeff Cumming explained this in the "dance of the p-values" critisizing this (https://www.youtube.com/watch?v=5OL1RqHrZQ8&ab_channel=GeoffCumming). He adressed this as a "negative" point, but this is somehow exactly how it should behave and a "postive" point and beauty of it if you want to objectively cumulate information (Article Valid P -Values Behave Exactly as They Should: Some Misleadi...
).It would be benificial to fully report your "effect-size" and confidence intervals with the p-value and describe the "effect-size" rather than "judge" and conclude under potentially suboptimal condtions. Than you leave it to the reader what to do with the information that you provide.
In the discussion results and concepts that seem to corroborate and contradict yours can be adressed. In this way no information is lost. I would see this as an improvement of an article and the application of the frequentist framewrok. However, I experience that not a lot of people view it this way, because then it does not fit in with a dogmatic inferential framework (and journals) they get thought: find data, create a model, reject H0.
Best,
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS