On the other hand, predictive modelling uses methods + statistics to peek into the future. While generally a function of historical data, the methods applied to the data are designed to make the best possible prediction based on
We may never perfectly model a football match due to the infinite number of variables and complex interactions between them. What we can do is reduce these variables by making assumptions. Assumptions are like a recipe: too many
assumptions and our model can become too complex to implement. A powerful assumption made time and time in academic papers that study football is that the number of goals in a match follows a Poisson distribution. I would even
dare to say that researchers concluded this by looking at the descriptive statistics that summarized a dataset containing the number of goals scored!
So we made 1 assumption...
Great! We have reduced football goals to a statistical problem we have studied extensively, know a great deal about and behaves well, the Poisson distribution. Now let's assume that football goals depend on the teams playing which
is not too crazy to assume.
Now we have 2 assumptions...
Even better, our second assumption means that goals scored is a function of the teams playing. We can now use powerful methods to obtain the best estimate of goals scored in a match as a function of two teams playing. In fact,
both teams might have never played before, but we can still plug them into our function and obtain a result. What is more, this result will behave according to the Poisson distribution (due to our 1st assumption) such that we
can extract a probability and make statistical inferences about it (requires an entirely new article to expand on this concept).