Project: Modeling mpg

Ray Jones
24/05/2015

Why ?

Did you feel good about your Regression Models project ?

Remember, all you had to do was model mpg in mtcars
But… you felt you could have done better ?
You saw other results and thought “If only I'd tried …?”
You asked yourself “Where did it all go wrong ?”
Where it went wrong was - you didn't have this shiny app !
But now you do !

What ?

A simple lm() problem
Come up with your best model, say, mpg ~ I(1/hp):

data(mtcars)
par(mar=c(4,4,2,2))
plot(1/mtcars$hp, mtcars$mpg)
abline(lm(mpg ~ I(1/hp), data=mtcars))

plot of chunk unnamed-chunk-1

… and possible combinations with up to 9 other variables !

How ?

Run the Project shiny app
Control your model from the left side panel
Slider bars adjust the predictor exponents in your model
To exclude a predictor, set its exponent to zero
When ready, hit [Fit!]
Marvel at your results !

Results from your model are shown in the main panel
Model fitted: your model
Adjusted R-squared: your model's predicted variance vs sample (1.0 is perfection)
Diagnostic plots: the 4 diagnostic plots
Summary fit: the summary text describing your model results

Here's what you have to beat ...

… if you feel up to a challenge:

For a 1 predictor model try: mpg ~ I( disp^-1.0 )
For a 2 predictor model try: mpg ~ I( wt^-0.5 ) + I( hp^-0.5 )
Then see if you can beat them …
…. good luck !