Hypothesis Testing (Statistics)

Photo by Crissy Jarvis

Old but Gold

From the perspective of modern statistics and data science, which is computationally more efficient and relies on statistical software and programming to conduct most types of hypothesis testing, it may make little sense in going over the classic hypothesis testing procedure.

But, in this article let’s try to get a feel for the most useful and critically meaningful parts of the entire hypothesis testing procedure that is so common in standard introductory statistics textbooks.

So, About those Steps

Step 1: State the null hypothesis (H0) and alternative hypothesis (Ha)

The way I like to think about this as “know what you’re looking for”. Put another way if you think a mean value is higher for one group than another group, or the group means are not the same, then be sure you “know” what you’re expecting (alternative hypothesis) the null hypothesis then is just the idea that what you think is not true (no difference).

Step 2: Compute the test statistics

This step now might as well become “select the correct hypothesis testing procedure”. Now a-days most software gives us this when we type the code or click the buttons. Of course, it is a good idea to know how a few simple test statistics are calculated and be comfortable looking up the math equations when you need to.

The main point about this step now is to be sure you know and can nail-down the appropriate statistical test for the hypotheses testing you want to do.

Step 3: Determine the p-value

This step should become “know which tail of the p-value to interpret”. When using software or a programming language be careful when just interpreting any p-value, you want to interpret the p-value that is right for your hypothesis (either one-tail or two-tail depending on your hypothesis).

And it goes without saying that you should know how to interpret the p-value and know what it means (we talk about p-values in a whole separate article and video).

Step 4: Decide between your hypotheses

Luckily this step is still very much alive (that is until we have the computers make this decision for us in the future, which may be closer than we think!).

The key point for this step is basically having enough critical judgement to make a decision between the null or alternative hypothesis; we also need to be comfortable in deciding when to throw in the towel and say we need more data when our p-value is ambiguous.

And finally Step 5: State the real-world conclusion

Luckily machines haven’t experienced the real world as much as humans have. The crux of this step is basically stating the accepted hypothesis (again we accept either the null or alternative hypothesis) in practical terms with respect to real people, places or things.

A Modern Take on the Hypothesis Testing Procedure

So, if I could give a nice modern rehashing of the classic hypothesis testing procedure then it would be:

1. State your null and alternative hypotheses
2. Select the correct statistical procedure for your hypotheses
3. Interpret the correct statistical results p-value / confidence intervals for your hypotheses
4. Decide between your hypotheses
5. State the real-world conclusion

Hope you guys like my modern twist to this old (and rather drab but extremely important) topic.

For more content on the types of hypothesis testing procedure check out more of our content on our channel. Until then. Happy mining!

For more articles and content check out my YouTube Channel if you like what I have to offer then please show your support and connect with me on Facebook and follow me on Twitter