What p-value should we use for research?

What p-value should we use for research?

Should we not use it all? Should it vary by field?

idk, I just put error bars of 1sigma on my data

>What p-value should we use for research?
Depends on the experiment.

Literally no one thinks the p value means the probability of the hypothesis, being true when you have some observation. But it's a good way to measure if it's true when you see that an observation is very unlikely to occur if your premise is the hypothesis is true.

Why not 2sigma? It would correspond to 95% of results as opposed to the 68.3% of 1sigma.

Just to be clear, everyone should be using 0.01 for their P-values.

We should not use it at all, if not then how will psychologists ever get published?

Because my error bars would be larger than the total variation of my data, kinda hard to convince anyone with that.

I'm just a simple spectrocopist, no PLANCK collaboration

There are different accepted p values for different sciences. Physics uses like .01. Biology uses .05. Idk what chemistry uses. Psychology uses .09 iirc.

Do like particle physicists and use around [math]\pm 250\sigma[/math], just to make sure

Chemist here, specifically inorganic with characterization focused on single crystal x-ray diffraction. We just solve the structure to an R-squared

>We just solve the structure to an R-squared

it doesn't really matter. Statistics should only be used for preliminary studies. If your field uses it for actual hypothesis verification, it's a shit field.

0.5. As long as more replications are right than wrong, everything's fine.

make 100 observations to determine P?

>R-Squared ranges from 0 to 1
>We just solve the structure to an R-squared a = 0.5
A fifty percent false positive rate seems absurdly high, and I would never trust a scientific finding that said they had a 50-50 shot of being wrong, but you do you.

what ive meant, calibrate equipment
drunk idiot here

that's not what that means though

Then maybe you should redesign your experiment.

Psychology uses .05 as the standard, .01 is also be used occasionally.

.05 is almost always fine, you just need to adjust your p-values upward when looking at more than one variable to prevent data dredging

This guy gets it. Pharma is heading this way.

Nah, the high precision experiments ate someone else's problem.
Btw, I do disclose the error bar size in my publications, it's just that a graph with error bars greater than the total variation really isn't nice to look at, that's the only reason.
I could push them to be lower by simply doing more trials, but I also want to graduate in a finite amount of time, so large errors will have to do.

No, not that extreme :)
Higgs Boson was measured with 9 sigma.

1/20 is the standard in my department

And yet they were wrong about one decay channel, weren't they?