1 / 20

Bootstrap spatobotp ttaoospbr

Bootstrap spatobotp ttaoospbr. Hesterberger & Moore, chapter 16. Bootstrap. Why is that important? Bootstrap can be used to test almost any index that you think is interesting No underlying assumptions Intuitive. Bootstrap. How does it build on what we did before?

zuriel
Télécharger la présentation

Bootstrap spatobotp ttaoospbr

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. Bootstrapspatobotpttaoospbr Hesterberger & Moore, chapter 16

  2. Bootstrap • Why is that important? • Bootstrap can be used to test almost any index that you think is interesting • No underlying assumptions • Intuitive

  3. Bootstrap • How does it build on what we did before? • For each analysis, we saw that they had conditions of applications and were often at a loss about how to deal with what to do if the conditions were not met. • We saw one test statistic that we didn’t know how to test: sobel mediation test (c-c’)

  4. Bootstrap • How is it conceptually different from other measures? • It is non parametric and so does not refer to a hypothetical underlying distribution of the test statistic or of the data

  5. Today’s subject • Principles of the bootstrap: how to do it? • How to estimate bias, standard error and confidence interval of parameters with bootstrap?

  6. The process of frequentist statistics If one of those is unusual or unknown, bootstrap or permutation is useful

  7. Plan for the day • Present basic concepts of bootstrap • bias, • Standard deviation over bootstrap samples

  8. Historically • 1979: seminal paper of Efron. But there were some predecessors • Popularized in the 80’s due to availability of computer. • Strong mathematical background (even before 1979)

  9. In practice • Almost always based on simulations • It is often used to: • Estimate standard errors, • Estimate bias • Construct confidence intervals

  10. Advantages • Minimal assumption: the sample is a good representation of the unknown population • Works for almost any test statistic you can think of

  11. Bootstrap algorithm • Draw a sample x* with replacement from your sample x. Both samples have the same size n. • Compute TS* for this bootstrap sample • Repeat steps 1 and 2, B times. • We obtain TS* = (TS*1, TS*2,…,TS*B) • TS* is a sample from the unknown distribution of TS.

  12. Bootstrap Standard Errors • The standard deviation of TS* over the Bootstrap samples is an estimation of the standard error for a single sample

  13. Bootstrap estimate of bias • Bias(TS)=mean(TS*B)-TS • Because it is an estimate, it will not be exactly zero. Thus, the distance from zero must be estimated relative to the bootstrap standard deviation.

  14. Types of bootstrap • Parametric: we assume that the TS follows a specific distribution. Bootstrap is used to obtain the estimates of the parameters • Non-parametric: we do not know the distribution of the TS and obtain the empirical distribution

  15. Bootstrap estimate of covariance • Let TS1 be a test statistic and TS2 be another test statistic (e.g., TS1 is mean and TS2 is variance): • Because you estimate each TS in each replication, you can then obtain the covariance(TS1,TS2)

  16. Confidence interval: parametric • Estimate mean and standard error of the supposed distribution by using bootstrap • Use these estimates to compute a parametric (because you assume a distribution) confidence interval • Example: assume normality of TS, bootstrap TS to obtain mean and standard deviation over bootstrap (remember this is similar to standard error in a single sample)

  17. Confidence interval: percentile • Obtain a few thousands replications • Compute the TS • Order the TS from the smallest to the largest • The 95% percentile confidence interval lower (resp. higher value) is the value of TS such that 2.5% of the replications have a lower (resp. higher) TS than this value. • This interval is not symmetric

  18. Bias Corrected CI (Bca) • The TS may be biased. • The percentile confidence interval then is centered on the central tendency bootstrap value and not on the real sample estimate.

  19. Which CI to choose? • Percentile CI are easily applicable and intuitive. • In order to estimate consistently extreme percentiles, we need a very large number of replications • Bca is equal to percentile CI if TS is not biased.

  20. Take away message • 3main reasons to use bootstrap. Each research question generates one or several TS. • TS can be new and thus not implemented in any software. • H0 can be complex (not mean1=mean2) • TS can follow an unknown distribution • Procedure of bootstrap: • Bias, SE, confidence intervals

More Related