FAQ/rankit/bcis - CBU statistics Wiki

Revision 1 as of 2013-02-14 11:17:34

Clear message
location: FAQ / rankit / bcis
  • = Bootstrap Confidence Interval for the percentage bend correlation =

It should be mentioned that in lot of cases the percentage bend correlation is so robust to outliers and changes in sampling distributions that there is little or no variation in the correlations computed for each of the bootstrap samples.

The bootstrapping, however, may be done in the EXCEL spreadsheet as follows:

  1. Ensure that you only enter complete cases in the spreadsheet in columns A and B
  2. Go to Add-ins and click on 'bootstrap' in the far left corner which opens up the bootstrap macro written by Barreto and Howland.
  3. A dialogue box should appear: prompting you to insert the original sample which are the cells in columns A and B contianing the two sets of data to be correlated
  4. Next choose a block of unoccupied cells of the same size as the ones just chosen to represent the space where the bootstrap samples are written
  5. Highlight cell AO3 which contians the percentage bend correlation and enter this in the box corresponding to the cell to be tracked
  6. You can keep the 1000 as the default number of bootstrap samples. (Wilcoxon's R programs uses 599 bootstrap samples). You can also ask for all the raw bootstrap values to be outputted.
  7. Click the 'Proceed' button
  8. The results of each computation of the selected cell (percentage bend) for each bootstrap sample are outputted. These can be used to obtain 95% confidence intervals for the correlation.