所谓方差分解的内涵是这样的。一组数据(Y)，说，我很复杂；另外一组数据说，我也很复杂(E)；第三组数据说，不好，我也复杂(S)。这个复杂，用方差来表征。而方差之间是可以相互解释的。1、2、3组数据的复杂，两两间有共通部分。2用尽全力，可以解释1中的一部分（a），3用尽全力，也可以解释1中的一部分（b）。1、2合力，也可以解释一部分（c）。而1、2合力也解释不了的部分为d。下面是这样：
Call:
rda(X = Y,
Y = E)
Partitioning of variance:
Inertia Proportion
Total
6.0000
1.0000
Constrained
5.2692
0.8782 the percentage E can
explained
Unconstrained 0.7308
0.1218
Eigenvalues, and their contribution to the
variance
Importance
of components:
RDA1
RDA2
RDA3
RDA4
RDA5
RDA6
PC1
PC2
Eigenvalue
4.92 0.1194 0.1059 0.0825 0.02988 0.00723 0.384 0.2034
Proportion
Explained
0.82 0.0199 0.0176 0.0138 0.00498 0.00120 0.064 0.0339
Cumulative
Proportion 0.82 0.8406 0.8583 0.8720 0.87699 0.87819 0.942
0.9760
PC3
PC4
PC5
PC6
Eigenvalue
0.0860 0.04641 0.0108 0.000616
Proportion
Explained
0.0143 0.00773 0.0018 0.000100
Cumulative
Proportion 0.9904 0.99810 0.9999 1.000000
Importance of components:
RDA1
RDA2
RDA3
RDA4
RDA5
RDA6
Eigenvalue
4.924 0.1194 0.1059 0.0825 0.02988 0.00723
Proportion
Explained
0.935 0.0227 0.0201 0.0157 0.00567 0.00137
Cumulative
Proportion 0.935 0.9572 0.9773 0.9930 0.99863 1.00000
Scaling 2
for species and site scores
* Species
are scaled proportional to eigenvalues
* Sites
are unscaled: weighted dispersion equal on all
dimensions
* General
scaling constant of scores: 3.0274
同样可以得到S可以解释的部分，比如说是0.7345
然后求E+S可以解释的部分，实际是求E+S不可以解释的部分
Call:
rda(X = Y,
Y = E, Z = S)
Partitioning of variance:
Inertia Proportion
Total
6.0000 1.0000
Conditioned
4.9624
0.8271
Constrained
0.7868
0.1311
Unconstrained 0.2508
0.0418这是不可解释的部分
Eigenvalues, and their contribution to the
variance
after
removing the contribution of conditiniong variables
Importance of
components:
RDA1
RDA2
RDA3
RDA4
RDA5
RDA6
PC1
PC2
PC3
Eigenvalue
0.393 0.206 0.109 0.0669 0.0127 0.000254 0.165 0.0822
0.00331
Proportion
Explained
0.378 0.198 0.105 0.0645 0.0122 0.000240 0.159 0.0792
0.00319
Cumulative
Proportion 0.378 0.577 0.681 0.7458 0.7581 0.758310 0.918 0.9968 1.00000
Accumulated constrained eigenvalues
Importance
of components:
RDA1 RDA2 RDA3
RDA4
RDA5
RDA6
Eigenvalue
0.393 0.206 0.109 0.0669 0.0127 0.000254
Proportion
Explained
0.499 0.261 0.138 0.0850 0.0162 0.000320
Cumulative
Proportion 0.499 0.760 0.898 0.9835 0.9997 1.000000
Scaling 2
for species and site scores
* Species
are scaled proportional to eigenvalues
* Sites
are unscaled: weighted dispersion equal on all
dimensions
* General
scaling constant of scores: 3.0274
Species
scores
RDA1
RDA2
RDA3
RDA4
RDA5
RDA6
氨基酸 0.05032 -0.23450
-0.03096 -0.16156 0.09800
-0.005019
糖类
0.33094 -0.06727 0.28093 -0.14253
-0.03792
0.004606
羧酸
-0.16719
0.47500
0.06466 -0.12253 0.03184
-0.002039
杂类
0.67220
0.16830 -0.08843 0.07567 0.02936
-0.001875
聚合物 0.04102
-0.01976
0.01194 -0.03390 -0.05905 -0.017668
酰胺类 0.07963 0.01456 -0.27193
-0.18408 -0.05490 0.004679
Site
scores (weighted sums of species scores)
RDA1
RDA2
RDA3
RDA4
RDA5
RDA6
1
-0.17227 -0.01607 0.32489 0.4627 0.4424 1.1243
2
0.03381 -0.06797 0.40757 0.2167 0.7958
-4.5834
3
0.12764
0.19025 -0.71172 -0.9496 -1.3782 3.7444
4
-0.31396 -0.85269 0.30681 1.9064 -0.4643
-0.3825
5
1.15855
1.55876 -0.66518 -1.1845 -0.8376 -0.5405
6
-0.26971 -0.53186 0.79232
-0.3373
1.5398
0.4933
7
-1.54211
1.21819 -0.23281 1.1187 2.9660
-2.4333
8
-0.49221 -1.51228 -1.40678 -0.3923 -0.8610 -0.3740
9
-0.13254
0.64827 -1.36074 0.8335 0.3608
-0.1458
10
-0.66728 -1.74427 1.32106 -2.4978
-1.1574
2.8262
11
0.51853
0.09139
0.33940 -1.2221 -2.3344 2.7709
12
1.49327
0.69983
1.09911
1.4157
0.1561 -1.2784
13
-0.74239
0.89492
0.65436 -1.2385 -2.3846 3.8700
14
-0.39911
0.21786 -0.08959 0.7703 -0.2570
-2.9781
15
1.39979 -0.79434 -0.77872 1.0982 3.4135
-2.1131
Site
constraints (linear combinations of constraining
variables)
RDA1
RDA2
RDA3
RDA4
RDA5
RDA6
1
-0.1168 -0.08062 0.48359 -0.08457
-1.04381
0.6948
2
-0.1709 -0.10535 0.51584
-0.70106
0.14110 -0.1560
3
0.2682
0.25749 -0.98455 0.59406 0.93431
-0.4662
4
-0.5545 -1.42253 0.48514 1.60379
-0.23390
0.2061
5
0.8365
0.74262 -0.43210 -1.45682 -0.31975 -0.3052
6
0.2400
0.72676
0.46573 -0.05336 0.39269 0.1382
7
-1.7179
0.79651
0.06406
0.07296
1.48611 -0.7389
8
-0.2515 -0.78766 -1.70120 0.18889 -0.37670
-0.4047
9
-0.2480
0.45277 -1.17910 0.05373
-0.71183
1.6386
10 -0.4708
-1.40809
1.05975 -1.41720 0.18073 0.1998
11
0.5567 -0.07437 0.18673 -0.04788
-0.39332 -0.7487
12
1.4485
0.72737
1.10766
1.22088
0.06150 -0.0895
13
-0.6971
0.93150
0.39169
0.10566
0.06296
1.2762
14
-0.4006
0.27261
0.01626
0.12896 -1.46903 -1.5917
15
1.2780 -1.02901 -0.47950 -0.20803 1.28893 0.3471
Biplot scores for
constraining variables
RDA1
RDA2
RDA3
RDA4
RDA5
RDA6
pH.H2O.
-0.302863
0.10367
0.157284 -0.042456 -0.435636 -0.30536
H2O -0.007642
-0.27559 -0.089946 -0.071777 -0.008832 -0.05420
EC
-0.031034
0.04747 -0.125103 0.007019 0.121992 0.02514
SOM
0.029698 -0.16131 -0.217162 -0.029197 -0.008583 0.18298
Nt
0.081163
0.02886
0.072383
0.004714 -0.138659 0.26608
Pt
0.277357 -0.17170 -0.006217 -0.094796 -0.116895 0.12061
Corg.Nt
-0.068781 -0.27484 -0.290561 0.091834 0.184620
-0.14944
Cmic.Corg 0.162569 0.20239 0.298060 -0.033212
-0.103654 -0.26600
然后通过简单的加减方程，计算a，b，c，d。
forward
selection是为了得到和因变量密切相关的自变量，这个选择有较强的整体性
library(packfor)
forward.sel(Y, E, alpha=0.05)
加载中…
喜欢