You should actually be careful if you're really seeing the third line there. That's a form of p-hacking. Don't find a statistically significant effect in your e

You should actually be careful if you're really seeing the third line there. That's a form of p-hacking. Don't find a statistically significant effect in your entire study population? Fine, just segment it into 20 subsets, and in an average study, you can now expect to randomly find at least one population subset in which the effect is statistically significant.