Timeline for Can the scaling values in a linear discriminant analysis (LDA) be used to plot explanatory variables on the linear discriminants?
Current License: CC BY-SA 4.0
25 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Sep 3, 2020 at 23:39 | history | edited | ttnphns | CC BY-SA 4.0 | added 549 characters in body |
| Sep 3, 2020 at 23:12 | history | edited | ttnphns | CC BY-SA 4.0 | added 235 characters in body |
| Apr 13, 2017 at 12:44 | history | edited | CommunityBot | replaced http://stats.stackexchange.com/ with https://stats.stackexchange.com/ | |
| Aug 18, 2014 at 22:41 | comment | added | ttnphns | Loading squared is the share of the variable's variance that is accounted for by the component. | |
| Aug 18, 2014 at 22:28 | comment | added | TLJ | @ttnphns thanks so much, I am seeing it now. You are the PCA master! So just to clarify, the coefficients in the loadings should be used for interpretation because they account for the differences in variance among the different variables? So if each variable had equal variance, than the loadings would match the eigenvectors (in an ordinal sense), correct? | |
| Aug 18, 2014 at 20:36 | history | edited | ttnphns | CC BY-SA 3.0 | added 84 characters in body |
| Aug 18, 2014 at 20:33 | comment | added | ttnphns | @TLJ, "these and those" = variables and components. You said you computed raw component scores. Standardize each component to variance=1. Compute covariances between the variables and the components. That would be the loadings. "Standardized" or "rescaled" loading is the loading divided by the st. deviation of the respective variable. | |
| Aug 18, 2014 at 19:28 | comment | added | TLJ | @ttnphns thanks for the response. Would you mind clarifying the last sentence of your comment-- I am not sure what you mean by "these and those". Also, in your above answer, what was your method of rescaling the loadings? | |
| Aug 18, 2014 at 17:46 | comment | added | ttnphns | @TLJ, should be: between variables and standardized components. I've inserted the word. See please here: Loadings are the coefficients to predict... as well as here: [Footnote: The components' values...]. Loadings are coefficients to compute variables from standardized and orthogonal components, by virtue of what loadings are the covariances between these and those. | |
| Aug 18, 2014 at 17:38 | history | edited | ttnphns | CC BY-SA 3.0 | added 140 characters in body |
| Aug 18, 2014 at 16:22 | comment | added | TLJ | @ttnphns your PCA answers have been very helpful. You say about the loadings: "eigenvectors normalized to respective eigenvalues; loadings are the covariances between variables and components." However, when I calculate raw PC scores, the correlations between variables and the raw PC scores are different from the loadings-- even in an ordinal sense. How may I interpret this? Thanks | |
| Apr 1, 2014 at 9:06 | history | edited | ttnphns | CC BY-SA 3.0 | added 121 characters in body |
| Mar 21, 2014 at 7:01 | comment | added | ttnphns | @Vitomir, please go down the link in my answer. | |
| Mar 21, 2014 at 2:18 | comment | added | Vitomir Kovanovic | Can you describe how you calculated standardized LDA loadings? | |
| Jan 24, 2014 at 16:56 | comment | added | ttnphns | @Etienne I added details you asked for to the bottom of this answer stats.stackexchange.com/a/48859/3277. Thank you for your generosity. | |
| Jan 24, 2014 at 15:13 | comment | added | ttnphns | You can plot unstandardized coefficients, standardized ones, or correlations. A plot itself is not a necessary tool for interpretation, it's just a convenient visualization. Correlations - I'll add to my other answer and inform you. | |
| Jan 24, 2014 at 15:10 | history | bounty awarded | Etienne Low-Décarie | ||
| Jan 24, 2014 at 15:10 | comment | added | Etienne Low-Décarie | How does one calculate "Pooled within-groups correlations between variables and discriminants"? I see it mentioned a few places online but can find a reference for its calculation. | |
| Jan 24, 2014 at 14:57 | comment | added | ttnphns | I think yes, why not? You see, biplot is simply an overlay of two scatterplots in the same axes. For example, PCA biplot is loading plot + scores plot; axes being the components. See here as example. You are in right to do similarly with LDA. | |
| Jan 24, 2014 at 14:46 | comment | added | Etienne Low-Décarie | So I can plot, after arbitrary scaling, either "Standardized discriminant coefficients" or "Pooled within-groups correlations between variables and discriminants" on the same axis as "Discriminant scores" to interpret the results in two different ways? In my question I had plotted "Unstandardized discriminant coefficients" on the same axis as the "Discriminant scores". | |
| Jan 24, 2014 at 4:41 | history | edited | ttnphns | CC BY-SA 3.0 | added 512 characters in body |
| S Jan 24, 2014 at 1:09 | history | suggested | Etienne Low-Décarie | CC BY-SA 3.0 | Thank you! I am working through your explanation linking to R code. |
| Jan 24, 2014 at 1:04 | review | Suggested edits | |||
| S Jan 24, 2014 at 1:09 | |||||
| Jan 23, 2014 at 14:29 | history | edited | ttnphns | CC BY-SA 3.0 | deleted 1 characters in body |
| Jan 23, 2014 at 13:56 | history | answered | ttnphns | CC BY-SA 3.0 |