topic Re: Eigenvalue in Statistical Procedures

Eigenvalue

zahidhasandipu — Fri, 04 Sep 2020 05:08:44 GMT

Assume you have 26 variables and have performed a principal component analysis. Your first 3 eigenvalues are 14, 7.8, and 1.015. Is that good? Is that bad? Does the answer depend on other issues in your research?

Re: Eigenvalue

IanWakeling — Fri, 04 Sep 2020 09:48:35 GMT

It's difficult to answer your question without seeing more detailed output. I assume you are running a correlation PCA (i.e. PROC PRINCOMP without the COV option), in which case the eigenvalues will sum to 26. So roughly 84% of the variance is summarised by 2 dimensions and it is possible to give an excellent graphical summary of the data with just one plot of PC 1 vs PC 2. That might be good or bad. Were you expecting a high degree of correlation among the 26 variables? Or did you believe you had measured 26 different things? The number of observations makes a difference too, the results would be more remarkable if you had 80 observations rather than 8.

Re: Eigenvalue

PaigeMiller — Fri, 04 Sep 2020 10:39:49 GMT

@zahidhasandipu wrote:

Assume you have 26 variables and have performed a principal component analysis. Your first 3 eigenvalues are 14, 7.8, and 1.015. Is that good? Is that bad? Does the answer depend on other issues in your research?

"Good" is not a mathematical or statistical term. "Bad" is not a mathematical or statistical term. There's no way for us to answer this.

Now that you have these eigenvalues, what are you going to do with them?