Eleisha's Segment 23: Bootstrap Estimation of Uncertainty
To Compute:
1. Generate 100 i.i.d. random draws from the beta distribution , for example using MATLAB's betarnd or Python's random.betavariate. Use these to estimate this statistic of the underlying distribution: "value of the 75% percentile point minus value of the 25th percentile point". Now use statistical bootstrap to estimate the distribution of uncertainty of your estimate, for example as a histogram.
After generating 100 i.i.d random values, my estimated value the statistic was approximately
After bootstrapping (nboot = 100,000), the mean of my test statistic was approximately: and the standard deviation was approximately: .
Below is a histogram of the bootstrapping values:
2. Suppose instead that you can draw any number of desired samples (each 100 draws) from the distribution. How does the histogram of the desired statistic from these samples compare with the bootstrap histogram from problem 1?
When drawing from the true distribution the mean was: and the standard deviation was:
Below is a histogram of these values:
3. What is the actual value of the desired statistic for this beta distribution, computed numerically (that is, not by random sampling)? (Hint: I did this in Mathematica in three lines.) The actual value of the desired statistic for this beta distribution is approximately: 0.23295. This was calculated using the inverse cdf of the Beta Distribution with the proper quartiles.
Sample output:
Bootstrapping Values Mean of values: 0.204169593645 Standard Deviation of values: 0.0242746089495 Values when sampling from the True Distribution Mean of values: 0.231145899062 Standard Deviation of values: 0.0265314651351 Actual Value: 0.232952354264
To Think About
1. Suppose your desired statistic (for a sample of N i.i.d. data values) was "minimum of the N values". What would the bootstrap estimate of the uncertainty look like in this case? Does this violate the bootstrap theorem? Why or why not?
2. If you knew the distribution, how would you compute the actual distribution for the statistic "minimum of N sampled values", not using random sampling in your computation?
3. For N data points, can you design a statistic so perverse (and different from one suggested in the segment) that the statistical bootstrap fails, even asymptotically as N becomes large?
Back To: Eleisha Jackson