Skip to content

Loading Bar Chart VIP Exclusive

📊 What is this?

The Loading Plot shows the distribution of variables on components using scatter points, which is good for seeing structural relationships.

The Loading Bar Chart uses a more direct expression: it lays out the loading values of all variables on one specified component as bars.

💡 One-line distinction: the loading scatter plot answers "what is the relationship between these variables"; the Loading Bar Chart answers "on this one component, which is heavy and which is light".


🧐 How to read?

Bar direction and length

FeatureMeaning
Bar pointing up (positive value)The variable is positively correlated with the score on this component
Bar pointing down (negative value)The variable is negatively correlated with the score on this component
The longer the barThe greater the variable's contribution to this component
Bar close to 0The variable hardly plays any role on this component

Use it in contrast with the score plot

This is the key idea for understanding the loading bar chart:

The score plot looks at samples, the loading plot looks at variables. A sample with a very high score on PC1 means that it also has extreme values on those variables whose "absolute loading on PC1 is large".


🛠️ How to use?

Configuration items

Configuration itemDescription
Number of componentsChoose the principal component / latent variable to view (from 1 to the optimal number of components)

Click Draw to generate. You can switch between components and view them one by one.

Typical uses

PurposeMethod
Interpret the meaning of a principal componentFind the variables with the largest loadings on that component → together they define the business meaning of the component
Variable screeningLook across multiple components; if a variable is close to 0 on all components → consider removing it
Find collinear variablesTwo variables with highly similar loadings on all components → redundant information
Validate business intuitionCheck whether the dominant variables match process common sense; if not, go back and check the data

The platform provides three loading-related views, each with a different purpose:

ChartPresentationBest for
Loading PlotScatter (p1 vs p2)Seeing the structural relationships between variables
X+Y Loading PlotScatter (X vs Y)Seeing the direction of association between X and Y
Loading Bar ChartBars (single component)Seeing which variable matters most on a given component

💡 Workflow suggestion: first use the scatter loading plot to see the overall structure and lock onto the directions of interest; then use the Loading Bar Chart to dig into each component, quantifying "exactly which few variables are dominating".


⚠️ Notes

  • Only supervised models are supported (PLS / PLS-DA / OPLS / OPLS-DA)
  • The comparability of loading values depends on the preprocessing method: whether the data is standardized and whether the scaling method is consistent both affect the magnitude of the loadings; be careful when comparing across models
  • The special nature of OPLS models: their loadings reflect the predictive component after orthogonal correction, and are not directly comparable with PLS loadings
  • The choice of component count affects the conclusions——it is recommended to view within the optimal component range; components too far back are mostly noise

Let data speak, make decisions simpler.