Data Levelling

I get an error when doing a PCA calculation that data cannot have zero size. What does this mean?

A: In the PCA calculation the default is to use multivariate data.  This means that there must be a valid entry for every variable used in the PCA calculation.  This error message indicates that there are no samples that contain a valid entry for all of variables used in the calculation.  To check whether the samples you wish to use are multivariate or not use the “Apply Attributes to all Rows (Selected Data Not Null)” option under the Validate menu.  You must select an attribute colour first in the Attribute Manager which will then be applied to all of the data points that contain a valid data entry for all the variables selected in the Select Variables dialog.  For more information view the “Identify Multivariate Samples” training video which will enable you see which samples are multivariate for the selected elements and which ones are not.

When levelling some data I am getting a warning message that columns could not be completed when using the Z-Score Power Transform method. Under what circumstances can this happen?

A: This can happen when using poor quality data which contains a lot of samples with the same value for a variable or with data that is heavily quantized.

When running the levelling tool I received an error that the column limit was reached and not all my columns were created. Is there a column limit in ioGAS?

A: ioGAS has a column limit of 253.

Can I have some further details on the Gauss levelling calculation used in ioGAS?

A: An overview of the mechanism used in Gauss levelling calculation can be found in the Help file under the Data Levelling section. A more detailed explanation is as follows:

  1. The data is ranked.
  2. The rank value is made into a p value using p = (Rank + 0.5)/size
  3. The inverse cumulative probability (z) is calculated using this implementation http://commons.apache.org/math/apidocs/org/apache/commons/math/distribution/NormalDistributionImpl.html
  4. These ‘z’ values are placed in a new column in ioGAS.

This is done one group at a time, and each group has no effect on any other. Small groups are ignored, so the output remains null for these groups.