p-hacking is a serious methodological issue. It is also referred to as data fishing, data butchery, or data dredging. It is the misuse of data analysis to detect patterns in data that can be statistically meaningful. This is done by conducting one or more tests and only publishing those that come back with higher-significance results.

We have seen in the previous section, Hypothesis testing, that we rely on the P-value to draw a conclusion. In simple words, this means we compute the P-value, which is the probability of the results. If the P-value is small, the result is declared to be statistically significant. This means if you create a hypothesis and test it with some criteria and report a P-value less than 0.05, the readers are likely to believe that you have found a real correlation or effect. However, this could be totally false in real life. There could be no effect or correlation at all. So, whatever is reported is a false positive. This is seen a lot in the field of publications. Many journals will only publish studies that can report at least one statistically significant effect. As a result of this, the researchers try to wrangle the dataset, or experiment to get a significantly lower P-value. This is called p-hacking.

Having covered the concept of data dredging, it's now time we start learning how to build models. We will start with one of the most common and basic models—regression.