Weighing up degrees of freedom and variance?

Just checking my thinking - if we had two models for 12 experimental cases.
Model 1 has 10 parameters, and accounts for 95% of variance
Model 2 has 2 parameters, and accounts for 90% of the variance

- If we replicated the study and predicted scores from each model, which model would be more accurate?

I think that model 2 will be more accurate as a predictor, Even though there is less variance accounted for, the model is simpler and therefore less constrained by losses of degrees of freedom and has less chance of a type 1 error due to a smaller amount of parameters.

Would you agree?


Ambassador to the humans
It's impossible to say for sure. But yeah if you have 10 parameters for 12 observations then you're severely over fitting that model.