Overview

Requirements

1. Samples: 3+  
   
2. Attributes: 6+  
   

This workflow is focused on the typical test setup using only Category, or only Line Scale questions, or other scale questions not mixed with others. Please see the bottom of this page for information about PCA on CATA data.

Generate the PCA Workbook

1. In the test, click Results . If you need to exclude specific samples and/or sample sets from the analysis, click Filters .
   
   
2. Click on the Reports menu and select Create report .
   
   
3. Under the 1 Select Report Type , select PCA workbook .
   
   
4. Under the 2 Select Options , specify the options you wish to include in the report.
   
   

   
   
1. Threshold as percentage of average eigenvalues. This is a way of selecting how many factors will be presented in the workbook. A higher threshold value will only include factors with larger explained variability. By lowering the threshold, more factors may be seen. Default is set at 0.95, but can be adjusted as needed.
   

   The eigenvalue is the variance capture by each factor.
   
   
   
2. Data label text size (px). This is the font size in the report. The default is 10 pixels.
   
   
3. Appearance of loadings. Attribute loadings are the relationship (sign and degree) between the new factor and the original sensory attribute. You can specify how you want the attributes to be marked. The available options are:
   
1. Diamond
   
2. Square
   
3. Circle (default selection)
   
4. Vector
   
   
4. Display Labels. Select this option to display attribute names or attribute numbers. When the option is unchecked, no attribute information will be displayed.
   
   
5. Covariance or Correlation. PCA can either be run on the covariance or correlation matrices.
   
1. Covariance. Retains the original variance structure of the attributes. Situations where you would likely use covariance option:
   
1. Variation in the actual scores to be used.
   
2. Panel has been properly calibrated with reference standards.
   
   
2. Correlation. Standardizes the data before running the PCA computations. This amounts to re-scaling the attributes to a standard scale. Situations where you would likely use correlation option:
   
1. Variables that are measured on different scales.
   
2. Variables with unequal variance.
   
3. Using an untrained panel with little knowledge of the product.
   
   
6. 
   

   Include Only Significant Attributes. This option allows you to remove the attributes which are not statistically significant.
   You may wish to simplify the map by removing the attributes that do not have significant difference. However, you should be mindful not to reduce it to the point of no longer being meaningful.
   
   The significant difference between attributes is based on the Alpha value and , for scaled attributes, on the ANOVA type. Please refer to the points below titled Alpha and Analysis options from defaults for further details before making the decision whether or not to include only significant attributes.
   
   We recommend running the Summary Report on scaled attributes using your preferred analysis options to see whether there is a significant difference in any of them before running the PCA. That can help you determine if you want to use this checkbox, or manually deselect any of the attributes (see step 5) before generating the report.
   
   
7. Use zero for missing data. When checked, this option allows you to substitute incomplete data points with zeros. When this option is unchecked, the missing data points will not be included in the analysis.
   

   
   If you know that there are missing data points (use the Data tab to identify whether there is missing data) and you do not wish to include that panelist's data at all (their sample set), we recommend leaving this option unchecked and then filtering out the incomplete sample sets.
   
   
   To filter sample sets:
   
1. In the top right-hand corner of the Results page, click Filters .
   
2. In the top right-hand corner of the Filters window, click Sample Sets .
   
3. Select Complete and click Apply Filters .
   
   
8. Alpha. This option will allow you to set the desired alpha for determining significant difference.
   
   
9. Analysis Options from Defaults. Click Change advanced analysis options to select the ANOVA type that suits your analysis needs.
   
   
5. 
   
   Under the 3 Select Questions , select the questions and attributes that you wish to include in the report. All scaled attributes from the same question type, such as Line Scale, for example, will be combined and reported onto two sheets (PCA and Biplot).
   
   
6. The 4 Select Export Type has only one export type. Click Create my report .
   
   
7. Click the download arrow, save the report to a location on your computer or network drive and open it. 
   

Report Details - PCA/Biplots

1. PCA (means)
   
2. Biplot (means)
   

If a combination of different question types compatible with the PCA Workbook is used in the test, each question type will have its own pair of sheets. Please see the section below about PCA on CATA.

1. Loadings. Attribute loadings are the relationship (sign and degree) between the new factor and the original sensory attribute.
   
   
2. Scores. Sample scores are the value of each sample on the new factor.