Similarly, variables that appear close together on the plot (such as symmetry and smoothness, or radius and perimeter) indicate clusters along the first two PCs. The negative value has no interpretation, but because all the variables are on one side means that each variable correlates with the first PC in the same direction, namely as the variable goes up, the PC1 score goes down. That means that the first principal component has a negative value for all the loadings. In the plot below from the breast cancer sample data included in Prism, we see that all the columns appear on the left hand side.
This plot is useful for identifying clusters of variables. The loadings are the correlation (or covariance) between the columns of data and the PCs. Somewhat analogous to how the PC scores plot depicts the rows of data (rotated along the PCs), the loadings plot provides information about the columns. The loadings plot simply plots the numerical values from the Loadings matrix of the specified principal components. Changing which PC is plotted on each axis using the Axis Variables section of the dialog.This allows you to customize a number of graphical features including: The Format Graph dialog can be accessed by using the button in the toolbar ( ) or by double clicking anywhere in the graphing area (except for on the axes). The underlying graphic in Prism that does this plot is the Bubble Plot, and it’s very flexible. In Prism, you can hover your cursor over points of interest to get links to that associated row or column in the data table. It’s primarily useful for clustering or deriving some other meaning based on where certain points appear in relation to the others along the two selected components. These plots offer a low dimension representation of your data. PC scores are used to plot the rows of your data along the chosen principal component axes.
Graphics are generally the most important results from PCA unless you plan to use the PC scores for further analysis.
0 kommentar(er)
