Comments

I am trying out StatistiXL and I think it's great. Just a couple of points so far, having to do with PCA:

1. The selection of ranges (Input Range and Output Range) would be made easier if one could use the standard keyboard sequences (Shift, End, Arrow-Down, End, Arrow-Right to select a whole block).

2. The Casewise PCA scores do not equal the results of the calculation shown in the Help file. They are de-meaned; maybe that should be made clear.

More generally, have you thought about adding some time series capabilities to the product?

Comments

  • The keyboard shortcuts can be used to select ranges in most versions of Excel. Unfortunately there is a bug in Excel 2000 that stops this from working with custom forms (the type that statistiXL uses). Microsoft have released a fix for this and you can find more info on it in our Knowledge Base (See the article titled Problems using shortcut keys to select cell ranges)

    I will check out the PCA calculations that you mention and update the Help File accordingly ... thanks for the pointer.

    We have thought about time series stuff and realise that it would be a valuable addition to statistiXL. We currently have a number of other additions that we are working on, however, and don't have any ETA for time series functions just yet.

    Cheers

    Alan Roberts
    statistiXL
  • I have been unable to determine the inconsistency you describe between the PCA output and that shown in the help file. I get the same Case Scores for my output as are presented in the Help File. Can you provide any further detail?

    Thanks

    Alan Roberts
    statistiXL
  • I will email you a test file.
  • I just emailed the excel file to support@statistixl.com
  • re: query about PCA scores.

    It is standard for PCA analyses to have the scores output in standardised form (to a mean of 0), for the following reason.

    For PCA based on the correlation matrix, the raw data columns are standardised for the mean (to 0) and standard deviation (to 1) prior to analysis, to account for differences in scale of the different variables. It therefore makes sense to output the standardised PCA scores (to column means of 0) because the data have already been standardised.

    For PCA based on the covariance matrix, where the raw data are NOT required to be standardised for the mean and standard deviation, it might make less sense to output the standardised PCA scores (to a mean of 0) because the data do not have to be standardised for analysis. However, it seems that for mathematical reasons (convenience of matrix calculation?) that the raw scores are first standardised to a mean of 0 before the PCA analysis, and the PCA scores are output standardised to a mean of 0 . Note that when using the covariance matrix, the raw scores are NOT standardised to a standard deviation of 1. So, you get different results from statistiXL if you use the raw data or standardised data for the covariance approach (but you get exactly the same results for the correlation approach because the raw data have been standardised for both mean and SD).

    In statistiXL, we also provide the option of standardising the PCA scores to a standard deviation of 1, as an option in the dialog box. Having the PCA scores standardised to a mean of 0, and potentially to a SD of 1, can be useful when further analysing the PCA output (e.g. in a MANOVA).
Sign In or Register to comment.