Pertemuan 21. Analisis Struktur Peubah Ganda (I): Analisis Komponen Utama. Matakuliah : I0214 / Statistika Multivariat Tahun : 2005 Versi : V1 / R1. Learning Outcomes. Pada akhir pertemuan ini, diharapkan mahasiswa akan mampu :
Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.
Analisis Struktur Peubah Ganda (I):Analisis Komponen Utama
Matakuliah : I0214 / Statistika Multivariat
Tahun : 2005
Versi : V1 / R1
Pada akhir pertemuan ini, diharapkan mahasiswa akan mampu :
No less than 50 observations, better to have 100. Rule of thumb is to have at least 20 observations per variable.
Theory assumes continuous variables. If all variables are binary use correspondence analysis.
Original or Standardized Variables
Principle component analysis can be performed on standardized variables (i.e. assessing the correlation matrix), or unstandardized values (I.e. using the covariance matrix). Standardized scores aid comparisons among different variables, especially when those variables have quite different variances. The difference in variances can be very important in the definition of components.
Number or Variables
PCA can be performed on any number of variables. With large numbers of variables there is a much higher chance that some of the components will have zero or very small eigenvalues indicating exact or near collinearity. Variables that do not weigh highly in the more significant components may be dropped and components recomputed. Removed variables can be analyzed in their own separate analysis.
Number of Components to Plot
How many components to plot will depend on the relative values of the
eigenvalues and the analysts criteria as to how much of the total variation
must be explained. Typical criteria are described below:
Plot combinations of components having eigenvalues >1. Use fewer factors if the number of variables is less than 50 and more if the number of variables greater than 50
Consider components important until the fraction of explained variance exceeds some pre-specified level, say 95% in the natural sciences or 60% in the social sciences, or when the last component added adds less than 5%
Examination of the Scree Plot to identify the number of components where the curve first begins to straighten out