One in Ten rule in Statistics.


From my current understanding, it means for every 1 independent variable you need to get at least 10 respondents as a sample.

In: Mathematics

If one in ten people have red hair (not true statistic) then out of 100, 10 have red hair. 1000-100, and so on.

A major part of statistics is modelling — given this data we observed, this is what we believe is the underlying shape of the data. From that model, you can then make predictions of what future data should look like. Those models have a number of parameters — for example, if your model is that the data follows a straight line, you have two parameters (the slope of the line, and its vertical offset). More sophisticated models with more parameters allow you to tune the model very closely to the data you saw, but, if the model has _too_ many parameters, you run the risk of adjusting so closely to the sample (“training”) data that you make your model _less_ accurate for future data (which is called “overfitting”).

The one in ten rule is not a hard rule, or anything. Rather, it’s just a rule of thumb for how many parameters you can have in your model before the risk of overfitting becomes too high.