いくつか(p個)の変量の値を情報の損失をできるだけ少なくして、
少数変量(m個、m<p)の総合的指標(主成分)で代表させる方法として
主成分分析(Principal Component Analysis, PCA)と
因子分析(Factor Analysis, FA)がある。
いくつかのテストの成績を総合した総合的成績、
いろいろな症状を総合した総合的な重症度、
種々の財務指標に基づく企業の評価
等を求めたいといった場合に用いられる。
p変量(p次元)の観測値をm個(m次元)の主成分に縮約させるという意味で、
次元を減少させる(reduce)方法と言うこともでき、
多変量データを要約する有力な方法である。
両者は似た目的に使われるが、元になっている考え方は異なるので
利用する場面では注意が必要である。違いにも焦点を当てながら説明する。
(制約条件 : の下で)
なお、 。
ただし、 の分散を 、共分散を とする。
[参考:立体の測定: イメージとして]
Sunday, October 31, 2021 05:00:50 PM 30 Plot of shintyou*taijyuu. Legend: A = 1 obs, B = 2 obs, etc. shintyou | | 190 + | | A | A A A | B A A A 180 + A A C C AAAB A | B A BA B A A | ABABA B C BAA AA | A B BA BAAB AA A A | AA AEA A DA 170 + A D CDCED BA A A A A | ADADAABB D | C BAABAAAAA | A C AA A B | A AA B A A A 160 + BCDA AA | AAAAB | A AABAA A | AA A | A A 150 + A | A | B | | 140 + | ---+------------+------------+------------+------------+-- 20 40 60 80 100 taijyuu Sunday, October 31, 2021 05:00:50 PM 31 The PRINCOMP Procedure Observations 199 Variables 2 Simple Statistics shintyou taijyuu Mean 168.8497487 60.44773869 StD 7.9248036 10.50548113 Covariance Matrix shintyou taijyuu shintyou 62.8025126 56.7763504 taijyuu 56.7763504 110.3651337 Total Variance 173.16764631 Eigenvalues of the Covariance Matrix Eigenvalue Difference Proportion Cumulative 1 148.139527 123.111408 0.8555 0.8555 2 25.028119 0.1445 1.0000 Sunday, October 31, 2021 05:00:50 PM 32 The PRINCOMP Procedure Eigenvectors Prin1 Prin2 shintyou 0.553923 0.832568 taijyuu 0.832568 -.553923 Sunday, October 31, 2021 05:00:51 PM 33 s h t k c i a j o a t n i k i d r s P P t j y t u r u r r O s y y o a k y u i i b e o u u k a e w n n s x u u i u i r a 1 2 1 F 145.5 42.0 76 自宅生 0 3700 -28.2930 -9.2216 2 F 146.7 41.0 85 自宅生 10000 Vodafone 6000 -28.4608 -7.6686 3 F 148.0 43.0 80 自宅生 50000 DoCoMo 4000 -26.0756 -7.6941 4 F 150.0 46.0 86 40000 . -22.4700 -7.6908 5 F 151.7 41.5 80 自宅生 35000 . -25.2749 -3.7827 6 F 152.0 35.0 77 自宅生 60000 DoCoMo 2000 -30.5204 0.0675 7 F 153.0 46.5 87 下宿生 10000 . -20.3920 -5.4700 8 F 153.0 55.0 78 自宅生 30000 . -13.3152 -10.1784 9 F 154.4 44.0 75 自宅生 9000 au 2000 -21.6979 -2.9196 10 F 155.0 48.0 83 下宿生 180000 . -18.0353 -4.6358 11 F 156.0 42.0 85 自宅生 0 DoCoMo 15000 -22.4768 -0.4797 12 F 156.0 46.0 82 自宅生 10000 Vodafone 7000 -19.1465 -2.6953 13 F 156.0 48.0 70 自宅生 30000 . -17.4814 -3.8032 14 F 156.0 49.0 85 自宅生 25000 . -16.6488 -4.3571 15 F 156.0 50.0 82 自宅生 40000 Vodafone 10000 -15.8162 -4.9110 Sunday, October 31, 2021 05:00:51 PM 34 Plot of Prin2*Prin1. Legend: A = 1 obs, B = 2 obs, etc. 20 + | | | | | | | | | | A| A | | A A 10 + | C | A A|AAAA AA | AA C A Prin2 | A A A A | B B A | B AA BA | B B AA | AA B CABB B| A B AAAA AA | A B A AB A CBBB A AA A AA 0 +---------A------A----AA---A--BA--A-A-DA---A--B-A----------------------- | A AACCA A A AB ABABBACA A A A | A A B B A A| B AA BBB A A A | A BAA A A A B A A A A | A A A | A A | A A A A A AA A A A | A | A -10 + A | A A | A | A | | A | | A | | A | | | | -20 + | --+----------------+----------------+----------------+----------------+- -40 -20 0 20 40 Prin1
(制約条件 : の下で)
合成変量の分散を最大化する軸を決定する。
身長、体重、胸囲での総合指標を求めてみよう :
Sunday, October 31, 2021 03:26:02 PM 59
The PRINCOMP Procedure
Observations 199
Variables 3
Simple Statistics
shintyou taijyuu kyoui
Mean 168.8497487 60.44773869 86.34221106
StD 7.9248036 10.50548113 8.88626378
Covariance Matrix
shintyou taijyuu kyoui
shintyou 62.8025126 56.7763504 20.7993036
taijyuu 56.7763504 110.3651337 47.7707020
kyoui 20.7993036 47.7707020 78.9656840
Total Variance 252.13333029
Eigenvalues of the Covariance Matrix
Eigenvalue Difference Proportion Cumulative
1 175.540573 121.609468 0.6962 0.6962
2 53.931105 31.269454 0.2139 0.9101
3 22.661651 0.0899 1.0000
Sunday, October 31, 2021 03:26:02 PM 60
The PRINCOMP Procedure
Eigenvectors
Prin1 Prin2 Prin3
shintyou 0.464556 -.478107 0.745387
taijyuu 0.749904 -.235289 -.618290
kyoui 0.470991 0.846199 0.249230
Sunday, October 31, 2021 03:26:03 PM 61
s
h t k c
i a j o a t
n i k i d r s P P P
t j y t u r u r r r
O s y y o a k y u i i i
b e o u u k a e w n n n
s x u u i u i r a 1 2 3
1 F 145.5 42.0 76 自宅生 0 3700 -29.5524 6.7527 -8.5761
2 F 146.7 41.0 85 自宅生 10000 Vodafone 6000 -25.5059 14.0300 -4.8203
3 F 148.0 43.0 80 自宅生 50000 DoCoMo 4000 -25.7571 8.7069 -6.3340
4 F 150.0 46.0 86 40000 . -19.7523 12.1220 -5.2027
5 F 151.7 41.5 80 自宅生 35000 . -25.1631 7.2908 -2.6487
6 F 152.0 35.0 77 自宅生 60000 DoCoMo 2000 -31.3111 6.1382 0.8462
7 F 153.0 46.5 87 下宿生 10000 . -17.5127 11.4163 -3.0265
8 F 153.0 55.0 78 自宅生 30000 . -15.3775 1.8005 -10.5250
9 F 154.4 44.0 75 自宅生 9000 au 2000 -24.3890 1.1807 -3.4280
10 F 155.0 48.0 83 下宿生 180000 . -17.3427 6.7223 -3.4601
11 F 156.0 42.0 85 自宅生 0 DoCoMo 15000 -20.4356 9.3483 1.4935
12 F 156.0 46.0 82 自宅生 10000 Vodafone 7000 -18.8490 5.8686 -1.7273
13 F 156.0 48.0 70 自宅生 30000 . -23.0011 -4.7564 -5.9547
14 F 156.0 49.0 85 自宅生 25000 . -15.1863 7.7013 -2.8345
15 F 156.0 50.0 82 自宅生 40000 Vodafone 10000 -15.8494 4.9274 -4.2005
Sunday, October 31, 2021 03:26:03 PM 62
Plot of Prin2*Prin1. Legend: A = 1 obs, B = 2 obs, etc.
20 + |
| | A
| A | A A A
| A A | A
| B A A A A AB | AA A A
| AA AABCD A BAB| AAAB A A
| AABAAABB DB CAC E A AAB AA A A
0 +-----------A----A--DA--BBBBC+BBD-ADCAA-AA-ABAA-------------------------
| A A ABA AC E D DCBAA AA
| A A A BBAA BB AAA AA A
Prin2 | A AA BA A A
| AA A| A A A A
| A |
| A |
-20 + |
| A |
| |
| |
| |
| |
| |
-40 + |
| A
| |
| A |
| |
| |
| |
-60 + |
-+-------------+-------------+-------------+-------------+-------------+
-40 -20 0 20 40 60
Prin1
Sunday, October 31, 2021 03:26:03 PM 63
Plot of Prin3*Prin2. Legend: A = 1 obs, B = 2 obs, etc.
Prin3 | |
| |
20 + |
| |
| |
| |
| A|
10 + A A A B |
| A A BB |
| ABB B B A|AB A
| A A AC CBBCA ABA
| BBACCDHEBD AA A A
0 +---------------------------------A---A--AAA-B---AAADGAAADC-B----A-A----
| A B BBCCCADBBADB A
| A B A AEABBA BC AA A B
| AA AAA A A A A
| A A A
-10 + | A A A
| | A
| A A |
| A
| |
-20 + |
| A |
| |
| |
| |
-30 + |
| |
-+---------+---------+---------+---------+---------+---------+---------+
-50 -40 -30 -20 -10 0 10 20
Prin2
Sunday, October 31, 2021 03:26:03 PM 64
Plot of Prin3*Prin1. Legend: A = 1 obs, B = 2 obs, etc.
Prin3 | |
| |
20 + |
| |
| |
| |
| | A
10 + A A| B A
| A AAA B
| A A DB AAA AAA
| A AA B AABA AAA A BAA BA
| A A AAAA DBBACBABD BBAA AB A A
0 +------A--------BA-ACBBAAB-BA+AAD-AA-C-CA--A----------------------------
| A ABAABBA AB D|ACACAAAA A A A
| AA AABA B A B |ABAA ABAB A A A
| A A A A A |A A A A
| A | A A
-10 + A A | A
| | A
| A | A
| | A
| |
-20 + |
| A
| |
| |
| |
-30 + |
| |
-+-------------+-------------+-------------+-------------+-------------+
-40 -20 0 20 40 60
Prin1
Sunday, October 31, 2021 03:26:46 PM 77
The PRINCOMP Procedure
Observations 199
Variables 3
Simple Statistics
shintyou taijyuu kyoui
Mean 168.8497487 60.44773869 86.34221106
StD 7.9248036 10.50548113 8.88626378
Correlation Matrix
shintyou taijyuu kyoui
shintyou 1.0000 0.6820 0.2954
taijyuu 0.6820 1.0000 0.5117
kyoui 0.2954 0.5117 1.0000
Eigenvalues of the Correlation Matrix
Eigenvalue Difference Proportion Cumulative
1 2.00917168 1.28938921 0.6697 0.6697
2 0.71978247 0.44873662 0.2399 0.9097
3 0.27104585 0.0903 1.0000
Sunday, October 31, 2021 03:26:46 PM 78
The PRINCOMP Procedure
Eigenvectors
Prin1 Prin2 Prin3
shintyou 0.581024 -.562895 0.587844
taijyuu 0.644596 -.122703 -.754613
kyoui 0.496898 0.817370 0.291546
【参考】分散共分散行列と相関行列を使ったときの違いを見てみたければ、 shintyou のみを mm 単位で測定したと仮定して、 10倍したものをデータとして両者の出力を比較してみよ。
明確に決まっているわけではないが、以下のような基準が一般的に
用いられている。また、結果の解釈の都合上、多少増減させることもある。
[ノウハウ] 軸の特性を把握するにはラインマーカーが重宝する。 一つの軸内で大きい(もしくは小さい)固有ベクトルにマークすると、 何がグルーピングされているかが理解し易くなる。 ただ、今週の例示では3軸までしか出現しないので、ご利益は感じにくいとも思う。 多変量(例えば以下の食品の嗜好性データ(PCA/FAとも))になればその有効性が理解できるであろう。
100種類の食品(ごはん、お茶漬け、…、リンゴ、パイ缶)に対する 性・年齢で分割した10グループの嗜好度調査のデータをMoodleに掲載しておく。 グループ1から5は男性、グループ6から10は女性であり、その中の番号の小さい方から順に 15才以下、16~20才以下、21~30才以下、31~40才以下、41才以上の10群を構成している。 また、測定尺度は「1: おらく食べる気がしない」、「2: もし強制されれば食べる」、…、 「8: いつも食べたい」、「9: もっとも好きな食品に入る」までの9段階であり、 各グループごとに尺度の平均値を取ったものが測定値として格納されている。
Sunday, October 31, 2021 03:28:00 PM 124 Obs X01 X02 X03 X04 X05 X06 X07 X08 X09 X10 1 7.69 7.31 7.47 7.76 7.87 7.51 7.24 7.70 7.91 7.95 2 6.59 5.56 6.21 6.04 5.81 6.64 6.11 6.53 6.44 6.64 3 4.55 4.18 4.36 4.25 4.53 4.60 3.66 4.04 3.68 4.43 4 6.78 6.11 6.30 5.98 5.56 6.37 6.29 5.43 5.32 5.28 5 6.47 6.24 6.02 5.42 5.88 6.00 5.60 4.60 5.40 5.95 6 6.96 6.81 6.91 6.48 6.23 7.09 7.27 7.13 6.86 7.36 7 6.57 5.70 5.89 5.16 5.30 6.07 5.56 4.50 4.92 5.33 8 7.32 6.95 6.02 4.98 4.88 6.82 6.40 5.53 5.61 5.33 9 6.51 6.15 5.51 4.68 4.16 5.17 4.81 4.70 4.86 3.82 10 6.86 6.05 5.85 6.14 6.75 6.71 5.39 5.42 6.03 6.59 Sunday, October 31, 2021 03:28:00 PM 125 The PRINCOMP Procedure Observations 100 Variables 10 Simple Statistics X01 X02 X03 X04 X05 Mean 6.038100000 5.784800000 5.947100000 5.669500000 5.640600000 StD 1.239147389 1.034139939 0.825972699 0.915395124 0.884228614 Simple Statistics X06 X07 X08 X09 X10 Mean 5.781300000 5.563900000 5.379400000 5.517400000 5.542100000 StD 1.294327683 1.182607883 1.121124814 1.016264322 1.130856737 Correlation Matrix X01 X02 X03 X04 X05 X01 M(-15) 1.0000 0.8708 0.5158 0.3701 0.1723 X02 M(16-20) 0.8708 1.0000 0.7588 0.6043 0.4021 X03 M(21-30) 0.5158 0.7588 1.0000 0.8524 0.7262 X04 M(31-40) 0.3701 0.6043 0.8524 1.0000 0.8742 X05 M(41-) 0.1723 0.4021 0.7262 0.8742 1.0000 X06 F(-15) 0.9384 0.8207 0.5164 0.3580 0.2077 X07 F(16-20) 0.8107 0.8381 0.6584 0.4875 0.3543 X08 F(21-30) 0.6161 0.7095 0.6990 0.6199 0.5235 X09 F(31-40) 0.5003 0.6470 0.7013 0.7207 0.7101 X10 F(41-) 0.3298 0.4569 0.5584 0.6321 0.7479 Sunday, October 31, 2021 03:28:00 PM 126 The PRINCOMP Procedure Correlation Matrix X06 X07 X08 X09 X10 X01 0.9384 0.8107 0.6161 0.5003 0.3298 X02 0.8207 0.8381 0.7095 0.6470 0.4569 X03 0.5164 0.6584 0.6990 0.7013 0.5584 X04 0.3580 0.4875 0.6199 0.7207 0.6321 X05 0.2077 0.3543 0.5235 0.7101 0.7479 X06 1.0000 0.8888 0.7465 0.6215 0.4932 X07 0.8888 1.0000 0.8949 0.7679 0.6415 X08 0.7465 0.8949 1.0000 0.8528 0.7741 X09 0.6215 0.7679 0.8528 1.0000 0.9112 X10 0.4932 0.6415 0.7741 0.9112 1.0000 Eigenvalues of the Correlation Matrix Eigenvalue Difference Proportion Cumulative 1 6.82795512 5.06608201 0.6828 0.6828 2 1.76187311 1.00742187 0.1762 0.8590 3 0.75445124 0.49207487 0.0754 0.9344 4 0.26237637 0.14082435 0.0262 0.9607 5 0.12155202 0.02358655 0.0122 0.9728 6 0.09796547 0.02586580 0.0098 0.9826 7 0.07209967 0.02801926 0.0072 0.9898 8 0.04408041 0.00832792 0.0044 0.9942 9 0.03575249 0.01385842 0.0036 0.9978 10 0.02189408 0.0022 1.0000 Sunday, October 31, 2021 03:28:00 PM 127 The PRINCOMP Procedure Eigenvectors Prin1 Prin2 Prin3 Prin4 Prin5 X01 M(-15) 0.286033 -.446335 0.193512 0.428019 0.162365 X02 M(16-20) 0.331337 -.239842 0.336063 0.022488 -.559594 X03 M(21-30) 0.323345 0.166337 0.442291 -.436029 -.168594 X04 M(31-40) 0.299329 0.358627 0.375366 0.063449 0.367912 X05 M(41-) 0.260727 0.507209 0.127419 0.375425 0.146879 X06 F(-15) 0.308635 -.407882 -.083695 0.267375 0.286866 X07 F(16-20) 0.344271 -.252697 -.171400 -.295655 -.025050 X08 F(21-30) 0.347877 -.032314 -.289087 -.507508 0.452377 X09 F(31-40) 0.345636 0.164368 -.322236 0.040012 -.388944 X10 F(41-) 0.303334 0.267273 -.522559 0.251270 -.190507 Eigenvectors Prin6 Prin7 Prin8 Prin9 Prin10 X01 -.016413 -.062138 -.038493 -.141617 0.668052 X02 -.212367 0.479465 0.283325 -.013739 -.225064 X03 0.476929 -.416354 0.136150 0.085922 0.163960 X04 -.562491 -.066245 -.114301 0.403713 -.068344 X05 0.385123 0.325310 -.167534 -.437833 -.148648 X06 0.209878 -.335058 0.176137 0.090538 -.618107 X07 0.137469 0.236104 -.762654 0.204382 -.046351 X08 -.128390 0.256135 0.382983 -.303270 0.106863 X09 -.387189 -.488821 -.161974 -.425188 -.030381 X10 0.181955 0.100632 0.270185 0.543107 0.229904 Sunday, October 31, 2021 03:28:00 PM 128 Obs X01 X02 X03 X04 X05 X06 X07 X08 X09 X10 Prin1 Prin2 1 7.69 7.31 7.47 7.76 7.87 7.51 7.24 7.70 7.91 7.95 5.88693 1.44204 2 6.59 5.56 6.21 6.04 5.81 6.64 6.11 6.53 6.44 6.64 1.65842 0.13686 3 4.55 4.18 4.36 4.25 4.53 4.60 3.66 4.04 3.68 4.43 -4.44537 -0.34692 4 6.78 6.11 6.30 5.98 5.56 6.37 6.29 5.43 5.32 5.28 0.72138 -0.63217 5 6.47 6.24 6.02 5.42 5.88 6.00 5.60 4.60 5.40 5.95 0.15339 -0.18363 6 6.96 6.81 6.91 6.48 6.23 7.09 7.27 7.13 6.86 7.36 3.65322 0.09908 7 6.57 5.70 5.89 5.16 5.30 6.07 5.56 4.50 4.92 5.33 -0.65902 -0.78995 8 7.32 6.95 6.02 4.98 4.88 6.82 6.40 5.53 5.61 5.33 0.76044 -1.96919 9 6.51 6.15 5.51 4.68 4.16 5.17 4.81 4.70 4.86 3.82 -1.96687 -1.71968 10 6.86 6.05 5.85 6.14 6.75 6.71 5.39 5.42 6.03 6.59 1.35649 0.51749 11 7.04 6.03 6.53 6.02 6.68 6.78 5.91 6.26 5.76 5.95 1.76317 0.15476 12 6.59 6.30 6.29 5.94 6.10 5.93 5.52 5.35 5.45 5.85 0.72383 0.14551 13 5.93 4.76 5.09 5.51 5.79 5.49 4.97 4.69 5.30 5.61 -1.20892 0.34668 14 7.00 6.31 6.82 6.26 5.26 6.69 6.27 5.94 5.78 5.26 1.42272 -0.75706 15 6.63 5.47 5.54 4.88 4.70 5.89 4.64 4.43 4.00 3.98 -2.13183 -1.49511 Obs Prin3 Prin4 Prin5 Prin6 Prin7 Prin8 Prin9 Prin10 1 -0.07682 0.40809 0.25348 -0.08404 0.00629 0.07518 -0.29432 0.07804 2 -0.90098 -0.05516 0.42231 -0.06712 -0.42537 0.15924 0.03126 0.21633 3 -0.55169 0.23782 0.37097 0.08301 0.22924 0.36927 0.13546 -0.13545 4 0.55357 -0.08449 0.21583 0.10836 -0.04069 -0.30130 0.23414 0.03747 5 0.21693 0.67993 -0.55569 0.43975 0.12407 -0.03449 0.20098 0.03292 6 -0.63902 -0.35301 -0.06665 0.08296 0.01967 0.09429 0.27099 0.10991 7 0.26301 0.42015 -0.16052 0.48267 -0.13359 -0.13508 0.28383 0.14238 8 0.06701 -0.04720 -0.60733 0.05720 0.07830 0.10980 -0.04606 0.07099 9 0.66740 -0.33431 -0.54215 -0.50605 -0.09164 0.28167 -0.34880 0.33721 10 -0.17674 1.31452 0.21057 0.17171 0.08563 0.18324 -0.13665 -0.09062 11 0.23456 0.33063 0.60092 0.58481 0.05801 0.16376 -0.43135 0.07070 12 0.49777 0.35538 -0.09548 0.21636 0.19362 0.18252 0.04441 0.14858 13 -0.53251 0.85212 0.41764 -0.06351 -0.06122 -0.26835 -0.02047 0.09838 14 0.71704 -0.53272 0.19744 -0.12540 -0.54510 0.00683 0.21138 0.12406 15 0.88751 0.23480 0.36191 0.24293 -0.09797 0.25997 0.00113 0.14547 Sunday, October 31, 2021 03:28:00 PM 129 Plot of Prin2*Prin1. Legend: A = 1 obs, B = 2 obs, etc. 4 + | | | | | | | | | | A A| | |A A 2 + A | A A A | A A A A | |A A A A Prin2 | A A A B| A A | A B AA A| A A | A A A A AA A A B | A A A A A | A A A A 0 +----------------------------------A----A+--A----A---A-----AA-------A--- | A A |A A A | A A A | A BA A | A A | A B AA | A | A A | A | A A A | A A A | A A -2 + A A A| A | A | A A | A | | | A | | | | | A | -4 + | -+---------+---------+---------+---------+---------+---------+---------+ -8 -6 -4 -2 0 2 4 6 Prin1 Sunday, October 31, 2021 03:28:00 PM 130 Plot of Prin3*Prin2. Legend: A = 1 obs, B = 2 obs, etc. Prin3 | | | | 3 + | | | A | | | | A | A | A 2 + | A | | A | | A | | AA | A | 1 + | AA | AA AAA | | A A A | A | A A |A A | A A A A AA BBA AA A A AA A A 0 +-------------------AA-A-A-A-------------+-------------A---------------- | A A AAA | A AA A | A A A A A A A AA | A A B |B A A A A B A | B A | A A A -1 + A AAA B A A | | | | A A A | | AA | | -2 + | | | -+---------+---------+---------+---------+---------+---------+---------+ -4 -3 -2 -1 0 1 2 3 Prin2 Sunday, October 31, 2021 03:28:00 PM 131 Plot of Prin1*Prin3. Legend: A = 1 obs, B = 2 obs, etc. Prin1 | | | | 7.5 + | | | | | | A A| 5.0 + A A | | A A | A A | A | A | A A 2.5 + A A A | AAA A | B A A A A A A A | A A AB | AB B | A A A A AA AA AA A 0.0 +---------A----A-A--AAA---A-A+--BA-------------A------------------------ | A A | A A | A A | A A | A A | A A A A A -2.5 + A A A A C A | B A | AA | A A | BA | | A | -5.0 + | A | A A | A | | | | A -7.5 + | | | -+-------------+-------------+-------------+-------------+-------------+ -2 -1 0 1 2 3 Prin3-->
上述の主成分分析の場合は、
データの散らばり方(分散)を捉えてデータ特性を把握する手法であった。
一方、因子分析は、変数間に(潜在的な)構造を持ち込んで関係を探る手法である
(少し理解しにくいかもしれないが)。
この手法は心理学の分野で広く利用されている。
例えば、i番目の学生の3科目の試験成績 が得られていたとする(i=1,2,…,n)。 fiをi番目の学生の特徴を表す要素、ajをj番目の科目の特徴を表す要素とし、 それらが以下のような積の形で表現できたとしたら、学生個人と科目の特性が分離できることになる。
回転の不定性: となる回転行列Tを用いると、
となり、AとFの対は無限組存在する。一意には決まらない。 何らかの基準を条件に確定させる。
Monday, November 1, 2021 09:36:28 PM 5 Obs X01 X02 X03 X04 X05 X06 X07 X08 X09 X10 1 7.69 7.31 7.47 7.76 7.87 7.51 7.24 7.70 7.91 7.95 2 6.59 5.56 6.21 6.04 5.81 6.64 6.11 6.53 6.44 6.64 3 4.55 4.18 4.36 4.25 4.53 4.60 3.66 4.04 3.68 4.43 4 6.78 6.11 6.30 5.98 5.56 6.37 6.29 5.43 5.32 5.28 5 6.47 6.24 6.02 5.42 5.88 6.00 5.60 4.60 5.40 5.95 6 6.96 6.81 6.91 6.48 6.23 7.09 7.27 7.13 6.86 7.36 7 6.57 5.70 5.89 5.16 5.30 6.07 5.56 4.50 4.92 5.33 8 7.32 6.95 6.02 4.98 4.88 6.82 6.40 5.53 5.61 5.33 9 6.51 6.15 5.51 4.68 4.16 5.17 4.81 4.70 4.86 3.82 10 6.86 6.05 5.85 6.14 6.75 6.71 5.39 5.42 6.03 6.59 Monday, November 1, 2021 09:36:28 PM 6 The FACTOR Procedure Input Data Type Raw Data Number of Records Read 100 Number of Records Used 100 N for Significance Tests 100 Monday, November 1, 2021 09:36:28 PM 7 The FACTOR Procedure Initial Factor Method: Principal Components Prior Communality Estimates: ONE Eigenvalues of the Correlation Matrix: Total = 10 Average = 1 Eigenvalue Difference Proportion Cumulative 1 6.82795512 5.06608201 0.6828 0.6828 2 1.76187311 1.00742187 0.1762 0.8590 3 0.75445124 0.49207487 0.0754 0.9344 4 0.26237637 0.14082435 0.0262 0.9607 5 0.12155202 0.02358655 0.0122 0.9728 6 0.09796547 0.02586580 0.0098 0.9826 7 0.07209967 0.02801926 0.0072 0.9898 8 0.04408041 0.00832792 0.0044 0.9942 9 0.03575249 0.01385842 0.0036 0.9978 10 0.02189408 0.0022 1.0000 2 factors will be retained by the MINEIGEN criterion. Monday, November 1, 2021 09:36:28 PM 8 The FACTOR Procedure Initial Factor Method: Principal Components Factor Pattern Factor1 Factor2 X01 M(-15) 0.74741 -0.59244 X02 M(16-20) 0.86579 -0.31836 X03 M(21-30) 0.84491 0.22079 X04 M(31-40) 0.78216 0.47602 X05 M(41-) 0.68129 0.67325 X06 F(-15) 0.80647 -0.54140 X07 F(16-20) 0.89959 -0.33542 X08 F(21-30) 0.90901 -0.04289 X09 F(31-40) 0.90316 0.21817 X10 F(41-) 0.79262 0.35477 Variance Explained by Each Factor Factor1 Factor2 6.8279551 1.7618731 Final Communality Estimates: Total = 8.589828 X01 X02 X03 X04 X05 0.90961791 0.85094991 0.76262367 0.83837129 0.91741340 X06 X07 X08 X09 X10 0.94352040 0.92177476 0.82814690 0.86329813 0.75411185
Monday, November 1, 2021 09:37:03 PM 10 The FACTOR Procedure Input Data Type Raw Data Number of Records Read 100 Number of Records Used 100 N for Significance Tests 100 Monday, November 1, 2021 09:37:03 PM 11 The FACTOR Procedure Initial Factor Method: Principal Components Prior Communality Estimates: ONE Eigenvalues of the Correlation Matrix: Total = 10 Average = 1 Eigenvalue Difference Proportion Cumulative 1 6.82795512 5.06608201 0.6828 0.6828 2 1.76187311 1.00742187 0.1762 0.8590 3 0.75445124 0.49207487 0.0754 0.9344 4 0.26237637 0.14082435 0.0262 0.9607 5 0.12155202 0.02358655 0.0122 0.9728 6 0.09796547 0.02586580 0.0098 0.9826 7 0.07209967 0.02801926 0.0072 0.9898 8 0.04408041 0.00832792 0.0044 0.9942 9 0.03575249 0.01385842 0.0036 0.9978 10 0.02189408 0.0022 1.0000 3 factors will be retained by the NFACTOR criterion. Monday, November 1, 2021 09:37:03 PM 12 The FACTOR Procedure Initial Factor Method: Principal Components Factor Pattern Factor1 Factor2 Factor3 X01 M(-15) 0.74741 -0.59244 0.16808 X02 M(16-20) 0.86579 -0.31836 0.29190 X03 M(21-30) 0.84491 0.22079 0.38417 X04 M(31-40) 0.78216 0.47602 0.32604 X05 M(41-) 0.68129 0.67325 0.11067 X06 F(-15) 0.80647 -0.54140 -0.07270 X07 F(16-20) 0.89959 -0.33542 -0.14888 X08 F(21-30) 0.90901 -0.04289 -0.25110 X09 F(31-40) 0.90316 0.21817 -0.27989 X10 F(41-) 0.79262 0.35477 -0.45389 Variance Explained by Each Factor Factor1 Factor2 Factor3 6.8279551 1.7618731 0.7544512 Final Communality Estimates: Total = 9.344279 X01 X02 X03 X04 X05 0.93786990 0.93615660 0.91021020 0.94467297 0.92966229 X06 X07 X08 X09 X10 0.94880526 0.94393897 0.89119742 0.94163724 0.96012863 Monday, November 1, 2021 09:37:03 PM 13 The FACTOR Procedure Initial Factor Method: Principal Components Scoring Coefficients Estimated by Regression Squared Multiple Correlations of the Variables with Each Factor Factor1 Factor2 Factor3 1.0000000 1.0000000 1.0000000 Standardized Scoring Coefficients Factor1 Factor2 Factor3 X01 M(-15) 0.10946 -0.33626 0.22279 X02 M(16-20) 0.12680 -0.18069 0.38691 X03 M(21-30) 0.12374 0.12531 0.50920 X04 M(31-40) 0.11455 0.27018 0.43215 X05 M(41-) 0.09978 0.38212 0.14670 X06 F(-15) 0.11811 -0.30729 -0.09636 X07 F(16-20) 0.13175 -0.19038 -0.19733 X08 F(21-30) 0.13313 -0.02434 -0.33282 X09 F(31-40) 0.13227 0.12383 -0.37099 X10 F(41-) 0.11609 0.20136 -0.60162 Monday, November 1, 2021 09:37:03 PM 14 Plot of Factor1*Factor2. Legend: A = 1 obs, B = 2 obs, etc. Factor1 | | | | 3 + | | | | | | | A 2 + |A | A A | A | A|A A | A A | A A 1 + A A A A A A | A A C A |AB A | A AA A AA | A A A | A A A BA A AA A A A 0 +-------------------A------------------A-A-----B--A-AAA---A-----A--A-- | A A A | A | A A| A A | A A A A A A A | A B A -1 + A A | A AB A A | A | A A | | AA A | A | -2 + A A | A A | | | | A | | -3 + | | | --+------------+------------+------------+------------+------------+-- -3 -2 -1 0 1 2 Factor2 Monday, November 1, 2021 09:37:03 PM 15 Plot of Factor2*Factor3. Legend: A = 1 obs, B = 2 obs, etc. Factor2 | | | | 2 + | A | A A | A | A | A | A | AA A | A | A A 1 + A A AA A| A | A A B A | A | A AA A A A| A A | A A A | A A AA A | AA AAA A | B A A 0 +----------A----B-A-----+-AB------------------------------------------ | B | BA | A A A| A A A A | A AA| A AAA | A | A -1 + A A A AA | A A A A | A AAA | A A A| A | A | -2 + AA | | | | | | | | | A -3 + | | | --+----------+----------+----------+----------+----------+----------+- -2 -1 0 1 2 3 4 Factor3
Monday, November 1, 2021 09:38:49 PM 51 The FACTOR Procedure Input Data Type Raw Data Number of Records Read 100 Number of Records Used 100 N for Significance Tests 100 Monday, November 1, 2021 09:38:49 PM 52 The FACTOR Procedure Initial Factor Method: Principal Components Prior Communality Estimates: ONE Eigenvalues of the Correlation Matrix: Total = 10 Average = 1 Eigenvalue Difference Proportion Cumulative 1 6.82795512 5.06608201 0.6828 0.6828 2 1.76187311 1.00742187 0.1762 0.8590 3 0.75445124 0.49207487 0.0754 0.9344 4 0.26237637 0.14082435 0.0262 0.9607 5 0.12155202 0.02358655 0.0122 0.9728 6 0.09796547 0.02586580 0.0098 0.9826 7 0.07209967 0.02801926 0.0072 0.9898 8 0.04408041 0.00832792 0.0044 0.9942 9 0.03575249 0.01385842 0.0036 0.9978 10 0.02189408 0.0022 1.0000 3 factors will be retained by the NFACTOR criterion. Monday, November 1, 2021 09:38:49 PM 53 The FACTOR Procedure Initial Factor Method: Principal Components Factor Pattern Factor1 Factor2 Factor3 X01 M(-15) 0.74741 -0.59244 0.16808 X02 M(16-20) 0.86579 -0.31836 0.29190 X03 M(21-30) 0.84491 0.22079 0.38417 X04 M(31-40) 0.78216 0.47602 0.32604 X05 M(41-) 0.68129 0.67325 0.11067 X06 F(-15) 0.80647 -0.54140 -0.07270 X07 F(16-20) 0.89959 -0.33542 -0.14888 X08 F(21-30) 0.90901 -0.04289 -0.25110 X09 F(31-40) 0.90316 0.21817 -0.27989 X10 F(41-) 0.79262 0.35477 -0.45389 Variance Explained by Each Factor Factor1 Factor2 Factor3 6.8279551 1.7618731 0.7544512 Final Communality Estimates: Total = 9.344279 X01 X02 X03 X04 X05 0.93786990 0.93615660 0.91021020 0.94467297 0.92966229 X06 X07 X08 X09 X10 0.94880526 0.94393897 0.89119742 0.94163724 0.96012863 Monday, November 1, 2021 09:38:49 PM 54 The FACTOR Procedure Rotation Method: Varimax Orthogonal Transformation Matrix 1 2 3 1 0.65777 0.53529 0.52990 2 -0.73396 0.61357 0.29126 3 0.16922 0.58051 -0.79647 Rotated Factor Pattern Factor1 Factor2 Factor3 X01 M(-15) 0.95490 0.13415 0.08963 X02 M(16-20) 0.85255 0.43757 0.13357 X03 M(21-30) 0.45872 0.81076 0.20605 X04 M(31-40) 0.22027 0.90003 0.29343 X05 M(41-) -0.02727 0.84202 0.46896 X06 F(-15) 0.91555 0.05731 0.32756 X07 F(16-20) 0.81272 0.18932 0.49758 X08 F(21-30) 0.58692 0.31451 0.66919 X09 F(31-40) 0.38658 0.45484 0.76506 X10 F(41-) 0.18417 0.37847 0.88485 Variance Explained by Each Factor Factor1 Factor2 Factor3 3.9249494 2.8740019 2.5453282 Monday, November 1, 2021 09:38:49 PM 55 The FACTOR Procedure Rotation Method: Varimax Final Communality Estimates: Total = 9.344279 X01 X02 X03 X04 X05 0.93786990 0.93615660 0.91021020 0.94467297 0.92966229 X06 X07 X08 X09 X10 0.94880526 0.94393897 0.89119742 0.94163724 0.96012863 Monday, November 1, 2021 09:38:49 PM 56 The FACTOR Procedure Rotation Method: Varimax Scoring Coefficients Estimated by Regression Squared Multiple Correlations of the Variables with Each Factor Factor1 Factor2 Factor3 1.0000000 1.0000000 1.0000000 Standardized Scoring Coefficients Factor1 Factor2 Factor3 X01 M(-15) 0.35650 -0.01839 -0.21738 X02 M(16-20) 0.28150 0.18161 -0.29360 X03 M(21-30) 0.07559 0.43873 -0.30350 X04 M(31-40) -0.04982 0.47796 -0.20481 X05 M(41-) -0.19000 0.37303 0.04733 X06 F(-15) 0.28692 -0.18126 0.04983 X07 F(16-20) 0.19300 -0.16084 0.17154 X08 F(21-30) 0.04912 -0.13688 0.32854 X09 F(31-40) -0.06666 -0.06858 0.40164 X10 F(41-) -0.17324 -0.16356 0.59933 Monday, November 1, 2021 09:38:49 PM 57 Obs X01 X02 X03 X04 X05 X06 X07 X08 1 7.69 7.31 7.47 7.76 7.87 7.51 7.24 7.70 2 6.59 5.56 6.21 6.04 5.81 6.64 6.11 6.53 3 4.55 4.18 4.36 4.25 4.53 4.60 3.66 4.04 4 6.78 6.11 6.30 5.98 5.56 6.37 6.29 5.43 5 6.47 6.24 6.02 5.42 5.88 6.00 5.60 4.60 6 6.96 6.81 6.91 6.48 6.23 7.09 7.27 7.13 7 6.57 5.70 5.89 5.16 5.30 6.07 5.56 4.50 8 7.32 6.95 6.02 4.98 4.88 6.82 6.40 5.53 9 6.51 6.15 5.51 4.68 4.16 5.17 4.81 4.70 10 6.86 6.05 5.85 6.14 6.75 6.71 5.39 5.42 11 7.04 6.03 6.53 6.02 6.68 6.78 5.91 6.26 12 6.59 6.30 6.29 5.94 6.10 5.93 5.52 5.35 13 5.93 4.76 5.09 5.51 5.79 5.49 4.97 4.69 14 7.00 6.31 6.82 6.26 5.26 6.69 6.27 5.94 15 6.63 5.47 5.54 4.88 4.70 5.89 4.64 4.43 16 6.56 6.57 5.74 4.76 4.39 6.56 6.29 5.61 Obs X09 X10 Factor1 Factor2 Factor3 1 7.91 7.95 0.66956 1.82121 1.58069 2 6.44 6.64 0.16626 -0.19916 1.19252 3 3.68 4.43 -1.03468 -1.43973 -0.47173 4 5.32 5.28 0.63900 0.22553 -0.50004 5 5.40 5.95 0.18242 0.09152 -0.20811 6 6.86 7.36 0.74034 0.36710 1.34854 7 4.92 5.33 0.32215 -0.32438 -0.54816 8 5.61 5.33 1.29334 -0.70969 -0.33933 9 4.86 3.82 0.58581 -0.75180 -1.38820 10 6.03 6.59 0.02089 0.39898 0.55070 11 5.76 5.95 0.40396 0.58950 0.17643 12 5.45 5.85 0.19873 0.54822 -0.27773 13 5.30 5.61 -0.59976 -0.44330 0.31921 14 5.78 5.26 0.91645 0.42072 -0.53512 15 4.00 3.98 0.46299 -0.53468 -1.57421 16 5.22 4.72 1.10983 -1.07931 -0.45219 ≪中略≫ Monday, November 1, 2021 09:38:49 PM 64 Plot of Factor1*Factor2. Legend: A = 1 obs, B = 2 obs, etc. 2 + | A | A | | | | A A A | | A A | A A | A |A A | A A | A A 1 + | A | A B | B A | A A AA A A A A Factor1 | A A A A A A | A A | A A A A | A | A A | A |A A A 0 +--------------------------------A-+-A-A------------------------------ | A A A | AA | A AA | A A A | A A| A A | AA | | | | A A|A BA A -1 + A AA A | A | A A | A | A A A | A A | AA A | A A | | A | | | A A | -2 + A| A --+----------+----------+----------+----------+----------+----------+- -3 -2 -1 0 1 2 3 Factor2 Monday, November 1, 2021 09:38:49 PM 65 Plot of Factor2*Factor3. Legend: A = 1 obs, B = 2 obs, etc. Factor2 | | | | 3 + | | A | | A | | | 2 + A | | | A A | A A | A A | A A | A A 1 + | AAA A A A A | | A A A | A A A A A AA |AAA A | B A AA A| A AA A A A 0 +-------------------------------------A-----A-+--AA---AAAA------------ | AA |AB A AB A | A A AA| AAA A A A | A A A A | A A -1 + A A| A A | A A A A | A A A | A A | A AA | A A A | -2 + A | | | | | | | -3 + | | | --+----------+----------+----------+----------+----------+----------+- -4 -3 -2 -1 0 1 2 Factor3 Monday, November 1, 2021 09:38:49 PM 66 Plot of Factor3*Factor1. Legend: A = 1 obs, B = 2 obs, etc. Factor3 | | | | 2 + | | A | A | A A |A AA | A | A AA 1 + A A A A | AA | A AAA AA CB | A A | A A A AA A A | A A A A A A A | AB AA B A AA 0 +---------------------A------------------+--A---------------B--------- | A | AA A AA A A | A A A A| A A A A A A | A A | A AAA A -1 + A A A| AA | A A | | A A |A A A A | | -2 + | | A | | | A | A | -3 + | | | | A | | | -4 + | | | --+------------+------------+------------+------------+------------+-- -3 -2 -1 0 1 2 Factor1
[例] 関西学院大学: 中央値補正法とは...
.、
科目別平均点の中央値補正前後比較
.
[例] 関西大学: 各学部での合否判定の方法