Science is...whatever we want it to be

SAS code of the day (simple linear regression):

proc reg;
model x = y;
plot x*y;
run;
quit;

x = dependent variable
y = independent variable

So, let's say you have a large population of data and you run a simple linear regression and get an r2 = 0.03. Pretty crappy, right? But, at least you have a scatterplot with 500+ data points, showing the independent variable cannot predict the dependent variable, or, at least, there are other factors, aside from that one variable, which need to be examined, in order to explain the dependent variable. So, instead of trying to examine other factors,or, God forbid, just say the results showed no relationship, you take a bunch of means. Leaving you with four data points. Four. Four out of 500. Then, you run a simple linear regression on the FOUR data points. Four. And your r2 is 0.98. Pretty good, right? Now you can report the highly significant predictive value of the independent variable and your research has the appearance of being legitimate. Using only FOUR effing data points. The start of any good statistical procedure.

In conclusion, science (in this case - statistics) is whatever we want it to be.