/sci/ - Science & Math » Thread #12410442

88KiB, 1074x1080, rambro.jpg

View Same Google iqdb SauceNAO

How the actual fuck does null hypothesis work?

Anonymous Wed 02 Dec 05:28:23 2020 No.12410442 View Reply Original Report

Quoted By: >>12412414 >>12412464

Is H0 always "nothing happens/No relation" or agrees with the hypothesis?
For Example:
"Do Males Get Higher Scores than Females?"
Would H0 just be "Males do not get higher scores" or "Males DO get higher scores"?
This shit has always confused me. Don't even get me started on writing the shit. How the hell do I use the ? and x?

Anonymous

Anonymous Wed 02 Dec 2020 05:29:09 No.12410444 Report

Quoted By:

Everyone laugh at John for being bad at math
Tempura Fry Soldier

Anonymous

Anonymous Wed 02 Dec 2020 05:54:32 No.12410504 Report

Quoted By: >>12410509 >>12410519 >>12410526 >>12412426

The null hypothesis is that two distributions (observed through sampling, experiments, whatever) are identical. This is ultimately based in the Bayesian idea that you can find the probability an event happens conditioned on different priors, so with the null hypothesis you can find the probability the observed sample could have been sampled from the presumed-identical distribution (p-value).
So in your example "Do Males Get Higher Scores than Females," the null hypothesis is that random variables X (Male score) and Y (Female score) are identically distributed, and correspondingly that their respective means ?X and ?Y are equal. In contrast, the alternative hypothesis is that ?X > ?Y—the mean of male scores X is higher than that of female scores Y.
Hope that helps.

Anonymous

Anonymous Wed 02 Dec 2020 05:58:09 No.12410509 Report

Quoted By: >>12410519 >>12410526

>>12410504
One more thing. When you conduct a test that produces a p-value, that p-value is the probability the observed result could have occurred under the null hypothesis. That's why if your p-value is 0.05, 0.01, etc, that means the chance the null hypothesis could have generated the observed results is 5%, 1%, respectively. So then we can say: "well, since it was unlikely that the observed results were derived from two identical distributions, we have strong reason to believe that these two distributions are identical."

Anonymous

Anonymous Wed 02 Dec 2020 06:02:09 No.12410519 Report

Quoted By: >>12410526

Null is just the assumed way this data should work.

So if I have two data sets: placebo and medicine. My null is that the placebo and medicine group are basically the same.

This can be extrapolated to all sorts of different things but boils down to what is expected to occur in the absence of whatever effect you are testing for.

>>12410504
>>12410509
This is a real explanation but it's a bit more technical than I think you wanted.

Anonymous

Anonymous Wed 02 Dec 2020 06:05:02 No.12410526 Report

Quoted By:

>>12410504
>>12410509
>>12410519
okay this actually helps alot. appreciate it

Anonymous

Anonymous Wed 02 Dec 2020 20:23:38 No.12412414 Report

Quoted By:

>>12410442
You need to be 18 to post here.

Anonymous

View Same Google iqdb SauceNAO 1592497163795.png, 172KiB, 445x394

Anonymous Wed 02 Dec 2020 20:26:39 No.12412426 Report

Quoted By:

>>12410504
>This is ultimately based in the Bayesian idea that you can find the probability an event happens conditioned on different priors
it's not you mongoloid
most testing is done in a frequentist way

Anonymous

Anonymous Wed 02 Dec 2020 20:42:09 No.12412464 Report

Quoted By:

>>12410442
Null hypothesis: your idea about the world doesn't explain things (data)
Alternate hypothesis: your idea about the world does explain things (data)

In the case of gender and IQ, you think there's possibly a relationship 'mIQ > fIQ' out there in *reality* Your null hypothesis is the reality complementary to your idea, so 'mIQ = fIQ' AND 'mIQ < fIQ' and your alternate hypothesis is 'mIQ > fIQ'

You do some experiment and then see how well the pretend world not including your hypothesis of interest explains things. If it (the null hypothesis) poorly explains the data, then the alternate hypothesis is bolstered.

Capcode	All Only User Posts Only Moderator Posts Only Admin Posts Only Developer Posts
Show Posts	All Only With Images Only Without Images
Deleted Posts	All Only Deleted Posts Only Non-Deleted Posts
Ghost Posts	All Only Ghost Posts Only Non-Ghost Posts
Post Type	All Only Sticky Threads Only Opening Posts Only Reply Posts
Results	All Grouped By Threads
Order	Latest Posts First Oldest Posts First

Your latest searches