Session Ready
Exercise

Confidence intervals on log-transformed models

The previous exercise highlighted that the model output for a log-transformed response is in terms of the logarithm of the response variable.

The effect size for a log-transformed value is in terms of change of logarithm per unit of the explanatory variable. It's generally easier to interpret this as a percentage change per unit of the explanatory variable, which also involves an exponential transformation: 100 * (exp(__effect_size__) - 1).

Instructions
100 XP
  • Build the model of log-transformed price as shown in the editor.
  • Create an ensemble of 100 bootstrap replications of the model.
  • Calculate the effect size ~ Age on the bootstrap replications.
  • Transform the numerical value of the slope to a percentage change per unit of the explanatory variable.
  • Find the 95% confidence interval on the percentage change per year of age. You can use the usual method: mean plus-or-minus twice the standard deviation.