Bootstrap coefficients / upper and lower bounds

[beallen]

Adding either the complete set of bootstrapped coefficients or the upper and lower bounds to help scenario modeling.

[psolymos]

There is covariation among the estimates, so using the full bootstrap distribution is better (try predicting based on CIs of a linear model of y=a+b*x, it will be a funnel). It also gives a handle on uncertainty in the predictions by looking at quantiles of the 100 runs.