practical-statistics

  • Metadata

  • Chapter 4 - Regression and Prediction

    • Confidence and Prediction Intervals

      • Confidence intervals are uncertainty intervals placed around regression coefficients and predictions
      • Bootstrap algorithm to understand this
          1. Consider each row as a single “ticket” and place n tickets in a box
          1. Draw a ticket at random, record the values, and replace it in the box
          1. Repeat step 2 n times; which is one bootstrap resample
        • 4.Fit a regression to the bootstrap sample, and record the estimated coefficients
          1. Repeat step 2 to 4, say, 1000 times
          1. Which is 1000 bootstrap values for each coefficient; find the appropriate percentiles for each one to get the confidence interval
          1. This can be re-formulated to show the prediction interval
          1. Prediction interval gives uncertainty around a single value, **confidence interval **pertains to a mean or other statistic calculated from multiple values