First a hypothesis is created. Like, when ice cream sales are high, the weather is hot. Therefore, my hypothesis is: selling more ice cream makes the weather hot.
So, now, let’s “test” the hypothesis. To do this, let’s determine under what conditions the hypothesis works, and where it fails.
Test #1) let’s look at all the hot days in 2016. What were the ice cream sales? Were they all high? Yes, they were. Ok. So far hypothesis looks good.
Test #2) let’s look at the difference between weekend ice cream sales, versus Monday to Friday ice cream sales in 2016, during the hot days. Hmmm. Something interesting in the data. Weekends tend to have more sales? Hypothesis starts to get a bit challenged. Seems like there is another factor influencing the behaviour.
Test #3) let’s look at ice cream sales on weekends only, and see how they correlate with temperature. Oh my. Wait a minute. Holidays during weekends that fall on cold days, like Easter, shows spike in ice cream.
Test #4) let’s look at hot food versus cold food consumption during the months of the year. Hmmm. Interesting. More soup during winter months, and more ice cream during hot months.
Test #5) let’s look at ice cream sales north versus south of the equator. Hmmmm. Interesting. The summer and winter months are opposite, north versus south of the equator. And where we see summer months, we observe higher ice cream sales.
So, I realize the testing above seems ridiculous, but this is the process of discovery…as in testing the hypothesis. The problem with bad science is when testing is rigged or customize to “prove the hypothesis”. What you really need to do is find the conditions that prove AND disprove the hypothesis.