Arun, a hypothesis test does not require a p-value and it is not about interpreting a p-value at all. A hypothesis test is about balancing error-rates in decisions between two substantial alternatives. At least in my field of reaserch I have not yet encountered a single occasion where a hypothesis test was correctly used, i.e. to devise a rational decision between two substantial alternatives.

Given you mean a significance test, which is based on interpreting a p-value, p-values are just a measure for a signal-to-noise ratio. This "ratio" is simply not interpretable in any sensible way as long as we do not consider the context and the actual signal and the actual noise. But if we do consider context, signal, and noise - then calculating some signal-to-noise ratio (like a p-value) does not add any relevant information or insight.*

Better do good science instead of adhereing to a some rituals (especially not some flawed procedure like NHST; NHST is not compatible with good science!). And don't confuse significance tests (ala Fisher) with hypothesis tests (ala Neyman/Pearson).

* Nevertheless, p-values can be of some help when signal or noise can not be interpreted in any meaningful way. But one should be aware that this is a compromise solution, where the researcher actually declares that he/she has not enough understanding about the studied effects to be able to particularily interpret the signal and the noise. And further, p-values loose their meaning/interpretability when they are taken "out of context". Very often, the context is obscured in papers that report p-values. For instance, experiments were repeated until the desired effect was observed and only this is reported; or data were removed because they were "outliers", and this is not reported; or several kinds of analyses were tried until one shows some interesting result, and only this is reported; or a story is made up on findings of the "significant results" of a larger series of experiments; or competeing models and hypotheses have not been adressed experimentally; or the "test" or analysis is selected based on the actual data, etc etc ...  P-values as such are not bad. The research practice behind them is bad, and sometimes also their interpretation. If research practice and interpretation are put right it turns out that p-values are not so very interesting. However, this will make the life harder for many authors (and also for reviewers and editors!).

Hello Arun,

Find the paper you cited :

“The problem is not that people use P-values poorly, it is that the vast majority of data analysis is not performed by people properly trained to perform data analysis” (Leek, 2014).

I think that everything is there...

As Jochen says, "people" are mainly looking for stars whatever the tool, the data or the initial idea...

Finally, I didn't see ASA banned p-values, they revealed a misunderstanding or misuse... Only one journal banned him.

What I understood is that p-value can still be used for H0 rejection but NOT to determine if H0 is true or whether to evaluate how results are important.

Regards

I agree completely with Dr Cyril and Dr Martin.

They did not say "no" but they stated what we already knew. That one needs to use all available resources (e.g., graphs, design, openness in reporting etc) or what I call exhaustiveness instead of reliance on only p-value. That is, good scientific reasoning !! Did we not know that p-value does not measure the size of an effect? or that p-value did not measure the probability of a hypothesis being true. The major issue is that reliance on a single test or technique is not a final answer. That is "the march of science." No one thing is correct all the time. There is always uncertainty which goes back to the definition of statistics.

Good that many people will not 'discard' pvalue.

pvalue is a measure of 'strength' of rejecting (or against) Null hypothesis, it is a very useful information. Please note that a researcher has stated the null, alternative hypotheses, propose the appropriate stats methodology to test the hypothesis and then derive the pvalue. We will never known the TRUTH, but stated the 'beliefs' as hypothesis and there are many chances of being wrong. 

pvalue will indicate if the sample data will 'support' one of the stated hypotheses. It may be generalized as a 'fact' only after many data samples confirm the same finding.  

Put differently, it is the smallest level of alpha (significance level) at that which, the null hypothesis can be rejected. It is often referred to as the observed level of significance. There is another, which I will call "actual alpha." This is the alpha set by the researcher before collecting or performing any data analysis. Note that I have indicated that we have 2 alphas or level of significance.

We reject the null hypothesis whenever the "actual level of significance" is greater than the "observed level of significance." For example, in a  doctoral or thesis dissertation you stated, your actual alpha  for the study is  0.05, however, when you did the study,  you obtained or observed a p-value of 0.03, then the null must be rejected. This suggests that  you have contradicted the null. {One can look at this from a trial court situation. The null is that no crime occurred (indeed the judge wasn't there to witness the crime) and the alternative, says a crime occurred. It is the job of the police etc to provide all necessary evidence to contradict the null. WE CAN NEVER PROVE THE NULL. Not guilty does not imply innocent. It simply says there is not enough evidence}.    

{One can look at this from a trial court situation. The null is that no crime occurred (indeed the judge wasn't there to witness the crime) and the alternative, says a crime occurred. It is the job of the police etc to provide all necessary evidence to contradict the null. WE CAN NEVER PROVE THE NULL. Not guilty does not imply innocent. It simply says there is not enough evidence}.    

The  above holds true as long as the researcher can show that the critical assumptions of the model concerning say, homoskedasticity especially in cross-sectional data or serial -correlation (time series)  or both homo. and serial) in panel data, are   statistically insignificant. Remember that it is the job of the researcher to convince the committee. Simply, rejecting the null without well-reasoned statistical arguments, statistical practice or complete transparency defeats the goal of good scientific reasoning, Remember the null hypothesis assumes that the researcher is candidly trying to know the answer to a question, consequently exhaustiveness is hugely releveant.

.