Document

Report
General additive models
Variance and covariance
 a1 
 
 a2 
T
U     U a1
...
 
a 
 n
a2
...
an 
 
 
 
T
M     M  
...
 
 
 
n
UU
T

a
2
i

Variance
i 1
1

...
n
a
n 1
2
i

2

i 1
 a1  

 a2  
V 
...

a  
 n

1
V
n
 a
n 1
i
 
i 1
T
 a1  
2







a2  
n
VV
T

 a
...
an   
    ( n  1)Variance
2
i
i 1
Sums of
squares
Variance
1

n 1
VV
T
 
2
1
n 1
( U  M )( U  M )
 a1   A 
 b1   B



a


 2
 b2   B
A 
A 
;
B


 ...
...



a   
b  
A 
B
 n
 n
n
AB
T

 a
i







  A b i   B   ( n  1) Covariance
i 1
1
n 1
T
( A  X A ) ( B  X B )  Co var iance
T
M contains the mean
1
n 1
( A  X A ) ( B  X B )  Co var iance
T
The coefficient of correlation
r 
We deal with
samples
cov( xy ) 

X

Y 
R 
1
n 1
cov( xy )
 x
var( y ) 
 x
y
For a matrix X that
contains several variables
holds
Cov Matrix  D 
1
n 1
1
n 1
( X  Μ X )' ( X  Μ
X
)
R 
( X  Μ Y )' ( X  Μ Y )
( X  Μ X )' ( Y  Μ Y )
X
 xy
( X  Μ X )' ( Y  Μ Y )
var( x ) 
(X  Μ
y

)' ( X  Μ
X
)( Y  Μ Y )' ( Y  Μ Y )
1
1
n 1
( X  Μ )' ( X  Μ )
1
n 1
Σ X ( X  Μ )' ( X  Μ ) Σ X
1
R  Σ X DΣ X
1
1
The diagonal matrix SX contains the standard
deviations as entries.
X-M is called the central matrix.
The matrix R is a symmetric distance matrix that contains all
correlations between the variables
ΣX
 X 1

 0

0

 0

0

X 2
0
0
0
...
0
0
0 

0 
0 

 Xn 
R 
1
1
n 1
1
Σ X ( X  Μ )' ( X  Μ ) Σ X
R  Σ X DΣ X
1
1
Pre-and postmultiplication
1 /  X 1 


 ... 
X 
... 


1 /  
Xn 

R  ΣX
X
T
 1 / 
X1
...
Xn

ΣX
r  X ΣXX
T
  11

  21
 1 /  1 ... ... 1 /  n 
...


 n1
Premultiplication
n
R 11  X 1; n Σ n ; n X n ;1 
0
...
1/ 2
...
...
...
0
...


0  Σ
X

...

1 /  n 
0
 12
...
 22
...
...
...
 n2
....
 ij
n

i 1
1 /  1

 0
X 
...

 0

1/
...
  11

  21

...


 n1
j 1
i
n

 12
...
 22
...
...
...
 n2
....
 12
...
 22
...
...
...
 n2
....
 1n 

 2n 
... 

 nn 
 1n  1 /  1 


 2 n   ... 
...   ... 




 nn   1 /  n 
Postmultiplication
n
r
ij
i 1
j
  11

  21

...


 n1
 scalar
j 1
 1n 

 2n 
... 

 nn 
For diagonal matrices X
holds
R  X S X  XX Σ  ΣXX
Linear regression
European bat species and environmental correlates
N=62
ln(Area)
ln(Number
of
species)
10.26632
6.148468
11.33704
7.696213
8.519989
12.24361
10.3264
10.84344
12.40519
11.61702
8.891512
5.703782
9.068777
9.019059
10.94366
7.824046
9.132379
11.27551
10.67112
7.887209
10.71945
7.243513
12.73123
13.20664
12.78555
1.871802
11.7905
11.44094
11.54248
11.16014
12.6162
9.615805
11.07637
3.258097
0
3.218876
0.693147
2.70805
2.890372
2.995732
3.178054
2.890372
3.496508
2.197225
1.609438
3.044522
2.833213
3.526361
1.098612
2.890372
3.178054
2.639057
2.639057
2.397895
0
2.397895
3.465736
3.218876
1.609438
3.496508
3.332205
0
2.397895
3.433987
2.564949
2.772589
ln( S )  a 0  a1 ln( A )
 y1 
 x1   1
1 
 
  
 
y
 2
 x2   1
1 
Y     a 0    a1    
...
...
...
...
 
  
 
1 
y 
x  1
 
 n
 n 
x1 

x 2  a 0 
 

...  a 1 

x n 
Y  XA
Matrix approach to linear regression
X is not a square matrix, hence X-1 doesn’t exist.
X ' Y  X ' XA
 X ' X 1 X ' Y
  X ' X  X ' XA  IA  A
A  X ' X  X ' Y
1
1
The species – area relationship of European bats
3.258097
0
3.218876
0.693147
2.70805
2.890372
2.995732
3.178054
2.890372
3.496508
2.197225
1.609438
3.044522
2.833213
3.526361
1.098612
2.890372
3.178054
2.639057
2.639057
2.397895
0
2.397895
3.465736
3.218876
1.609438
3.496508
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
ln(Area)
10.26632
6.148468
11.33704
7.696213
8.519989
12.24361
10.3264
10.84344
12.40519
11.61702
8.891512
5.703782
9.068777
9.019059
10.94366
7.824046
9.132379
11.27551
10.67112
7.887209
10.71945
7.243513
12.73123
13.20664
12.78555
1.871802
11.7905
X'
1
1
1
1
1
1
10.26632 6.148468 11.33704 7.696213 8.519989 12.24361
X'X
3.5
R 
n 1
1
y = 0.2391x + 0.1468
R² = 0.4614
3
X'Y
154.2937
1647.908
2.5
2
1.5
1
-1
(X'X) (X'Y)
a0 0.146808
a1 0.239144
0.5
0
-0.5
-5
Σ X ( X  Μ )' ( X  Μ ) Σ X
0
5
10
15
20
ln (Area)
ln S  0 . 24 ln A  0 . 15
What about the part of variance
explained by our model?
1
1
1
1
1
1
1
1
10.3264 10.84344 12.40519 11.61702 8.891512 5.703782 9.068777
4
62 607.1316
607.1316 6518.161
(X'X)-1
0.183521 -0.01709
-0.01709 0.001746
ln(# species)
ln(Number
of
Constant
species)
S e
1
0 . 15
A
0 . 24
 1 . 16 A
0 . 24
1.16: Average number of species per unit area (species
density)
0.24: spatial species turnover
R 
0.769488
-2.48861
0.730267
-1.79546
0.219442
0.401763
0.507124
0.689445
0.401763
1.007899
-0.29138
-0.87917
0.555914
0.344605
1.037752
-1.39
0.401763
0.689445
0.150449
0.150449
-0.09071
-2.48861
-0.09071
0.977127
0.730267
-0.87917
1.007899
0.843596
-2.48861
-0.09071
0.945379
0.076341
1
n 1
Σ X ( X  Μ )' ( X  Μ ) Σ X
1
(X-M)'
0.473878
-3.64398
1.54459
-2.09623
-1.27246
2.451164
0.533954
1.050991
2.612741
1.824579
-0.90093
-4.08866
-0.72367
-0.77339
1.151213
-1.9684
-0.66007
1.48306
0.878671
-1.90524
0.927004
-2.54893
2.938785
3.414195
2.993105
-7.92064
1.998051
1.64849
1.750039
1.367698
2.823752
-0.17664
0.769488
0.473878
-2.48861 0.730267
-3.64398 1.54459
(X-M)'(X-M)
71.0087 136.9954
136.9954 572.8582
-1.79546 0.219442 0.401763
-2.09623 -1.27246 2.451164
(X-M)'(X-M) / (n-1)
1.164077 2.245826
2.245826 9.391119
Sx
1.078924
0
0 3.064493
4
3.5
Sx -1
0.926849
0
0 0.326318
Sx -1 (X-M)'(X-M) / (n-1)
1.078924 2.081542
0.732854 3.064493
Sx-1 (X-M)'(X-M) / (n-1) Sx-1
1 0.679245
0.679245
1
Sx-1 (X-M)'(X-M) / (n-1) Sx-1)2
1 0.461374
0.461374
1
y = 0.2391x + 0.1468
R² = 0.4614
3
ln(# species)
X-M
1
2.5
2
1.5
1
0.5
0
-0.5
-5
0
5
10
ln (Area)
15
20
How to interpret the coefficient of determination
4
3.5
y = 0.2391x + 0.1468
R² = 0.4614
3
ln(# species)

2
Y ;M

n
1
(Y

n 1
2
Total variance
2

1.5
1
2
Y ;Y ( X )

n
1
n 1

(Y i  Y ( X i ))
2
i 1
Rest (unexplained) variance
0.5

0
-0.5
2
Y ( X ); M

n
1
(Y ( X

n 1
i
)Y)
i 1
-5
0
5
10
15
20
Residual (explained) variance
ln (Area)
 Y ; M   Y ;Y ( X )   Y ( X ); M
2
n
1
2
Y)
i 1
2.5
R  1
i
Residual
variance
 1
 (Y
n 1
Total variance
n
i
 Y ( X i ))
i 1

n
1
 (Y
n 1
2
i
Y)
Statistical testing
is done by an F or
a t-test.
F 
1 R
2
)Y)
n
i
Y)
i 1
t 
2
i
i 1
 (Y
2
i 1
R
 (Y ( X
df
t 
F
R
1 R
2
df
2
2
2
2
2
ln( S )  a 0  a1 ln( A )  a 2  T  a 3 N T  0  a 4 L
The general linear model
n
Y  a 0  a1 X 1  a 2 X 2  a 3 X 3  ...  a n X n  a 0 
a
i
Xi
i 1
A model that assumes that a dependent variable Y can be expressed by a linear combination of
predictor variables X is called a linear model.
 y1

 y2
Y 
...

y
 m
 1
 
 1
  1
 
 1
 
 y1

 y2
Y 
...

y
 m
 1
 
 1
  1
 
 1
 
x 1 ,1
...
x 2 ,1
...
...
...
x m ,1
...
x 1 ,1
...
x 2 ,1
...
...
...
x m ,1
...
x1 , n   a 0 
 
x 2 , n   y1 
 XA
...   ... 
 
x m , n   y n 
X ' Y  X ' XA
 X ' X 1 X ' Y
A  X ' X  X ' Y
x1 , n   a 0    0 
   
x 2 , n   y1    1 

 XA  Ε
...   ...   ... 
   
x m , n   y n    n 
The vector E contains the error terms of each
regression. Aim is to minimize E.
  X ' X  X ' XA  IA  A
1
1
The general linear model
n
Y  a 0  a1 X 1  a 2 X 2  a 3 X 3  ...  a n X n  a 0 
a
i
Xi
i 1
If the errors of the preictor variables are Gaussian the error term e should also be
Gaussian and means and variances are additive
n
Y  a 0  a1 X 1  a 2 X 2  a 3 X 3  ...  a n X n    a 0 
a
i
Xi 
i 1
n
 (Y )  a 0   a i  ( X i )   (  )
i 1

 (Y )    a 0 

2
2
Total
variance
2

i 1

2
a i X i    ( )

Explained
variance

  a0 

2
R 
n
Unexplained
(rest)
variance

a
X
 i i   2 (Y )   2 (  )
i 1


2
2
 (Y )
 (Y )
n
ln( S )  a 0  a1 ln( A )  a 3 N T  0  a 4 L
Y
Country/Island
Albania
Andorra
Austria
Azores
Baleary Islands
Belarus
Belgium
Bosnia and Herzegovina
British islands
Bulgaria
Canary Islands
Channel Is.
Corsica
Crete
Croatia
Cyclades Is.
Cyprus
Czech Republic
Denmark
Dodecanese Is.
Estonia
Faroe Is.
Finland
France
Germany
Gibraltar
Greece
Hungary
Iceland
X
ln(Number
of
Constant
species)
3.258097
0
3.218876
0.693147
2.70805
2.890372
2.995732
3.178054
2.890372
3.496508
2.197225
1.609438
3.044522
2.833213
3.526361
1.098612
2.890372
3.178054
2.639057
2.639057
2.397895
0
2.397895
3.465736
3.218876
1.609438
3.496508
3.332205
0
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
ln(Area)
Days
below
zero
Latitude
of capitals
(decimal
degrees)
10.26632
6.148468
11.33704
7.696213
8.519989
12.24361
10.3264
10.84344
12.40519
11.61702
8.891512
5.703782
9.068777
9.019059
10.94366
7.824046
9.132379
11.27551
10.67112
7.887209
10.71945
7.243513
12.73123
13.20664
12.78555
1.871802
11.7905
11.44094
11.54248
34
60
92
1
18
144
50
114
64
102
1
12
11
1
114
1
2
119
85
2
143
35
169
50
97
0
2
100
133
41.33
42.5
48.12
37.73
39.55
53.87
50.9
43.82
51.15
42.65
27.93
49.22
41.92
35.33
45.82
37.1
35.15
50.1
55.63
36.4
59.35
62
60.32
48.73
52.38
36.1
37.9
47.43
64.13
Multiple regression
1. Model formulation
2. Estimation of model parameters
3. Estimation of statistical significance
Y  XA
A  X ' X  X ' Y
1
X'
1
1
1
1
1
1
10.26632 6.148468 11.33704 7.696213 8.519989 12.24361
144
18
1
92
60
34
53.87
39.55
37.73
48.12
42.5
41.33
1
1
10.3264 10.84344
114
50
43.82
50.9
X'X
62
607.1316
4328
2906.4
2906.4
4328
607.1316
6518.161 48545.59 29086.57
534136 228951.7
48545.59
29086.57 228951.7 141148.1
(X'X)-1
1.019166 -0.02275
-0.02275 0.002458
0.00261 -7.5E-05
-0.02053 8.3E-05
0.00261 -0.02053
-7.5E-05 8.3E-05
1.3E-05 -5.9E-05
-5.9E-05 0.000509
(X'X)-1X'
0.025783 0.163309 0.013407
0.003376 -0.00859 0.002243
-0.00017 0.000405 9.87E-05
-0.00066 -0.00195 -0.00056
a0
a1
a2
a3
(X'X)-1X'Y
2.679757
0.290121
0.002155
-0.06789
0.07203 0.060295 0.010457 -0.13031 0.170347
-0.00078 0.00013 0.001069 0.003124 -0.00097
-0.00019 -0.00014 0.000364 -0.00054 0.000676
-0.00074 -0.00076 -0.00064 0.003269 -0.00409
X'Y
154.2937
1647.908
11289.32
7137.716
(X'X)-1(X'Y)
2.679757
0.290121
0.002155
-0.06789
Multiple R and R2
The coefficient of determination
n
1
R  1
2
Residual
variance
Total variance
 1
 (Y
n 1
n
i
 Y ( X i ))
i 1
1

n
 (Y
n 1
2
i
Y)
 (Y ( X
i 1
R 
1
1
n 1
Σ X ( X  Μ )' ( X  Μ ) Σ X
y
x1
x2
 1

 r1 y

R  r2 y

 ...

 r2 y
ry 1
ry 2
...
1
r21
...
r21
...
...
...
...
...
rn 1
...
...
 1
R  
 R XY
)Y)
2
i 1
n
 (Y
2
i
i
Y)
2
i 1
1
xm
rmy 

rm 1 
... 

... 

1 
R YX 

R XX 
The correlation matrix can be devided into
four compartments.
T
R  R XY R XX
2
R 
2
1
R YX  R XY R XX
det( R XX )  det( R )
det( R XX )
 1
1
R XY
T
det( R )
det( R XX )
ln(Number
of species)
ln(Area)
3.2580965
0
3.2188758
0.6931472
2.7080502
2.8903718
2.9957323
3.1780538
2.8903718
3.4965076
2.1972246
1.6094379
3.0445224
2.8332133
3.5263605
1.0986123
2.8903718
3.1780538
2.6390573
2.6390573
2.3978953
0
2.3978953
3.4657359
3.2188758
1.6094379
3.4965076
3.3322045
0
2.3978953
10.26632
6.148468
11.33704
7.696213
8.519989
12.24361
10.3264
10.84344
12.40519
11.61702
8.891512
5.703782
9.068777
9.019059
10.94366
7.824046
9.132379
11.27551
10.67112
7.887209
10.71945
7.243513
12.73123
13.20664
12.78555
1.871802
11.7905
11.44094
11.54248
11.16014
3.4339872
2.5649494
2.7725887
1.7917595
2.6390573
2.8903718
12.6162
9.615805
11.07637
5.075174
11.08702
7.858641
Days
below
zero
34
60
92
1
18
144
50
114
64
102
1
12
11
1
114
1
2
119
85
2
143
35
169
50
97
0
2
100
133
23
18
Latitude of capitals
(decimal degrees)
41.33
42.5
48.12
37.73
39.55
53.87
50.9
43.82
51.15
42.65
27.93
49.22
41.92
35.33
45.82
37.1
35.15
50.1
55.63
36.4
59.35
62
60.32
48.73
52.38
36.1
37.9
47.43
64.13
53.43
41.8
110
124
90
130
93
52.7
56.96
47.67
54.62
49.62
X-M
X-M
X-M
X-M
(X-M)'
0.769488
-2.48861
0.730267
-1.79546
0.219442
0.401763
0.507124
0.689445
0.401763
1.007899
-0.29138
-0.87917
0.555914
0.344605
1.037752
-1.39
0.401763
0.689445
0.150449
0.150449
-0.09071
-2.48861
-0.09071
0.977127
0.730267
-0.87917
1.007899
0.843596
-2.48861
-0.09071
0.473878
-3.64398
1.54459
-2.09623
-1.27246
2.451164
0.533954
1.050991
2.612741
1.824579
-0.90093
-4.08866
-0.72367
-0.77339
1.151213
-1.9684
-0.66007
1.48306
0.878671
-1.90524
0.927004
-2.54893
2.938785
3.414195
2.993105
-7.92064
1.998051
1.64849
1.750039
1.367698
-35.8065
-9.80645
22.19355
-68.8065
-51.8065
74.19355
-19.8065
44.19355
-5.80645
32.19355
-68.8065
-57.8065
-58.8065
-68.8065
44.19355
-68.8065
-67.8065
49.19355
15.19355
-67.8065
73.19355
-34.8065
99.19355
-19.8065
27.19355
-69.8065
-67.8065
30.19355
63.19355
-46.8065
-5.54742
-4.37742
1.242581
-9.14742
-7.32742
6.992581
4.022581
-3.05742
4.272581
-4.22742
-18.9474
2.342581
-4.95742
-11.5474
-1.05742
-9.77742
-11.7274
3.222581
8.752581
-10.4774
12.47258
15.12258
13.44258
1.852581
5.502581
-10.7774
-8.97742
0.552581
17.25258
6.552581
0.769488
0.473878
-35.8065
-5.54742
(X-M)'(X-M)
71.0087
136.9954
518.6241
-95.1758
-2.48861 0.730267
-3.64398 1.54459
-9.80645 22.19355
-4.37742 1.242581
136.9954
572.8582
6163.884
625.8081
-1.79546 0.219442 0.401763 0.507124 0.689445 0.401763 1.007899
-2.09623 -1.27246 2.451164 0.533954 1.050991 2.612741 1.824579
-68.8065 -51.8065 74.19355 -19.8065 44.19355 -5.80645 32.19355
-9.14742 -7.32742 6.992581 4.022581 -3.05742 4.272581 -4.22742
518.6241 -95.1758
6163.884 625.8081
232013.7 26066.26
26066.26
4903.6
(X-M)'(X-M)/(n-1)
1.164077 2.245826
2.245826 9.391119
8.502034 101.0473
-1.56026 10.25915
2
T
1
S1
0.926849
0
0
0
0 0.326318
0
0
0
0 0.016215
0
0
0
0 0.111534
S1 D
1.078924
0.732854
0.137858
-0.17402
2.081542
3.064493
1.638448
1.144244
7.880104 -1.44613
32.97357 3.347747
61.67255 6.928784
47.66024 8.965874
R 
2
det( R XX )  det( R )
det( R XX )
S1 DS-1
0.945379 2.823752 -51.8065 -5.07742
0.076341 -0.17664 40.19355 5.822581
0.28398 1.283927 54.19355 10.08258
-0.69685 -4.71727 20.19355 0.792581
0.150449 1.294578 60.19355 7.742581
0.401763
-1.9338 23.19355 2.742581
R YX  R XY R XX
1
R XY
8.502034 -1.56026
101.0473 10.25915
3803.503 427.3157
427.3157 80.38689
S
1.078924
0
0
0
0 3.064493
0
0
0
0 61.67255
0
0
0
0 8.965874
1 0.679245 0.127773 -0.16129
0.679245
1 0.534656 0.373388
0.127773 0.534656
1 0.772795
-0.16129 0.373388 0.772795
1
Det RXX
0.286065
Det R
0.095413
T
S1 DS-1 )-1
1.408029 -0.86031 0.139099
-0.86031 3.008345 -2.00361
0.139099 -2.00361 2.496439
S1 DS-1 )-1 RXY
0.824037
0.123194
-0.56418
S1 DS-1 )-1 RXYRYX
0.666462
 1
det( R )
det( R XX )
R2
0.666462
S1 DS-1
1 0.679245 0.127773 -0.16129
0.679245
1 0.534656 0.373388
0.127773 0.534656
1 0.772795
-0.16129 0.373388 0.772795
1
R  R XY R XX
-0.29138
-0.90093
-68.8065
-18.9474
Det RXX
0.286065
Det R
0.095413
R2
0.666462
1
SE 
trace ( R )( 1  R )
2
n  k 1
R: correlation matrix
n: number of cases
k: number of independent
variables in the model
t
parameter
SE ( parameter )
D<0 is statistically not
significant and should
be eliminated from
the model.
Adjusted R2
2
R adj  1  (1  R )
2
 1 df 1
2
n 1
n  k 1
F 

2
2
df 2

R
n  k 1
2
1 R
2
k

0 . 66646 62  3  1
0 . 33354
3
 38 . 6307
A mixed model
ln S  a 0  a1 ln A  a 2 D T  0  a 3 L  a 4 L
Y
Country/Island
Albania
Andorra
Austria
Azores
Baleary Islands
Belarus
Belgium
Bosnia and Herzegovina
British islands
Bulgaria
Canary Islands
Channel Is.
Corsica
Crete
Croatia
Cyclades Is.
Cyprus
Czech Republic
Denmark
Dodecanese Is.
Estonia
Faroe Is.
Finland
France
Germany
Gibraltar
Greece
Hungary
Iceland
Ireland
Italy
Kaliningrad Region
Latvia
X
ln(Number
of
Constant
species)
3.258097
0
3.218876
0.693147
2.70805
2.890372
2.995732
3.178054
2.890372
3.496508
2.197225
1.609438
3.044522
2.833213
3.526361
1.098612
2.890372
3.178054
2.639057
2.639057
2.397895
0
2.397895
3.465736
3.218876
1.609438
3.496508
3.332205
0
2.397895
3.433987
2.564949
2.772589
2
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
ln(Area)
Days
below
zero
10.26632
6.148468
11.33704
7.696213
8.519989
12.24361
10.3264
10.84344
12.40519
11.61702
8.891512
5.703782
9.068777
9.019059
10.94366
7.824046
9.132379
11.27551
10.67112
7.887209
10.71945
7.243513
12.73123
13.20664
12.78555
1.871802
11.7905
11.44094
11.54248
11.16014
12.6162
9.615805
11.07637
34
60
92
1
18
144
50
114
64
102
1
12
11
1
114
1
2
119
85
2
143
35
169
50
97
0
2
100
133
23
18
110
124
Latitude
of capitals
Latitude2
(decimal
degrees)
41.33
42.5
48.12
37.73
39.55
53.87
50.9
43.82
51.15
42.65
27.93
49.22
41.92
35.33
45.82
37.1
35.15
50.1
55.63
36.4
59.35
62
60.32
48.73
52.38
36.1
37.9
47.43
64.13
53.43
41.8
52.7
56.96
1708.169
1806.25
2315.534
1423.553
1564.203
2901.977
2590.81
1920.192
2616.323
1819.023
780.0849
2422.608
1757.286
1248.209
2099.472
1376.41
1235.523
2510.01
3094.697
1324.96
3522.423
3844
3638.502
2374.613
2743.664
1303.21
1436.41
2249.605
4112.657
2854.765
1747.24
2777.29
3244.442
X'
1
1
1
1
1
1
10.26632 6.148468 11.33704 7.696213 8.519989 12.24361
34
60
92
1
18
144
41.33
42.5
48.12
37.73
39.55
53.87
1708.169 1806.25 2315.534 1423.553 1564.203 2901.977
1
1
10.3264 10.84344
50
114
50.9
43.82
2590.81 1920.192
X'X
62
607.1316
4328
2906.4
141148.1
607.1316
4328
2906.4 141148.1
6518.161 48545.59 29086.57 1441737
48545.59
534136 228951.7 12488619
29086.57 228951.7 141148.1 7106497
1441737 12488619 7106497 3.71E+08
(X'X)-1
6.45421 0.000497 0.001087 -0.25606 0.002409
0.000497 0.002557 -8.1E-05 -0.00092 1.03E-05
0.001087 -8.1E-05 1.34E-05 6.63E-06 -6.8E-07
-0.25606 -0.00092 6.63E-06 0.010716
-0.0001
0.002409 1.03E-05 -6.8E-07
-0.0001 1.07E-06
(X'X)-1X'
0.028519 -0.00857 -0.18332 0.227512 0.119213 -0.18587 -0.27812 -0.01106
0.003388 -0.00932 0.001402 -0.00011 0.000382 0.000229 0.002492 -0.00174
-0.00017 0.000453 0.000154 -0.00024 -0.00016 0.000419 -0.00049 0.000727
-0.00078
0.0055 0.007968 -0.00748 -0.00331 0.007864 0.009674 0.003767
1.21E-06 -7.6E-05 -8.7E-05 6.89E-05 2.61E-05 -8.7E-05 -6.6E-05
-8E-05
a0
a1
a2
a3
a4
(X'X)-1X'Y
-3.40816
0.264082
0.003862
0.195932
-0.0027
The final model
ln S   3 . 41  0 . 26 ln A  0 . 004 D T  0  0 . 196 L  0 . 0027 L
2
Negative species
density
Realistic increase of
species richness with
area
Increase of species
richness with winter
length
Increase of species
richness at higher
latitudes
A peak of species
richness at
intermediate
latitudes
ln(# species predicted)
Is this model realistic?
The model makes a series of unrealistic predictions.
Our initial assumptions are wrong despite of the high degree of variance explanation
4.5
4
3.5
3
2.5
2
1.5
1
0.5
0
-0.5
-1 0
y = 0.6966x + 0.7481
R² = 0.6973
1
2
3
4
Our problem arises in part from
the intercorrelation between the
predictor variables
(multicollinearity).
We solve the problem by a stepwise approach eliminating the
variables that are either not
significant or give unreasonable
parameter values
ln (# species observed)
The variance explanation of this final model
is higher than that of the previous one.
Multiple regression solves systems of intrinsically linear algebraic equations
2
3
2
3
Y  a 10  a 11 X 1  a 12 X 1  a 13 X 1 ...  a 21 X 2  b 22 X 2  a 23 X 2 ... a n 1 X  a n 2 X 2  a n 3 X ...
3
A  X ' X  X ' Y
1
Polynomial regression
•
•
•
•
•
The matrix X’X must not be singular. It est, the variables have to be independent. Otherwise we
speak of multicollinearity. Collinearity of r<0.7 are in most cases tolerable.
Multiple regression to be safely applied needs at least 10 times the number of cases than
variables in the model.
Statistical inference assumes that errors have a normal distribution around the mean.
The model assumes linear (or algebraic) dependencies. Check first for non-linearities.
Check the distribution of residuals Yexp-Yobs. This distribution should be random.
Check the parameters whether they have realistic values.
Multiple regression is a hypothesis
testing and not a hypothesis generating
technique!!
ln(# species predicted)
•
General additive model
4.5
4
3.5
3
2.5
2
1.5
1
0.5
0
-0.5
-1 0
y = 0.6966x + 0.7481
R² = 0.6973
1
2
ln (# species observed)
3
4
Standardized coefficients of correlation
Z-tranformed distributions have a mean
of 0 an a standard deviation of 1.
Z 
x

B  Z X ' Z X
 1 Z X ' Z Y
n
r



Z'Z  



1
 (X
i
 X )(Y i  Y )
i 1
n 1
Zx 1i Zx i1

sXsY
...
...
...
...
...
...
...
...
...
...
Zx ni Zx i1
n 1
 Zx

n
1

i 1
(X i  X ) (Y i  Y )
sX
sY

n
1
n 1
Z
X
ZY
i 1
Zx 1 n 
 r11


...

1
 ...

R

Z
'
Z


 ...
...
n 1


r

Zx ii Zx ii 
 n1
ni
B   R xx  R
....
...
...
...
...
...
...
...
r1 n 

... 
... 

rnn 
1
R 
1
n 1
1
Σ X ( X  Μ )' ( X  Μ ) Σ X
1
 R 
1
n 1
Z'Z
XY
R
XY
R
XX
B
In the case of bivariate regression Y = aX+b, Rxx = 1.
Hence B=RXY.
Hence the use of Z-transformed values results in standardized correlations coefficients, termed b-values
then b Xi  B Xi
If Y  BX
 Xi
How to interpret beta-values
Y
Beta values are generalisations of simple coefficients of correlation. However, there is an important difference. The
higher the correlation between two or more predicator variables (multicollinearity) is, the less will r depend on the
correlation between X and Y. Hence other variables might have more and more influence on r and b. For high levels of
multicollinearity it might therefore become more and more difficult to interpret beta-values in terms of correlations.
Because beta-values are standardized b-values they should allow comparisons to be make about the relative influence
of predicator variables. High levels of multicollinearity might let to misinterpretations. Beta values above one are
always a sign of too high multicollinearity
Hence high levels of multicollinearity might

reduce the exactness of beta-weight estimates

change the probabilities of making type I and type II errors

make it more difficult to interpret beta-values.
We might apply an additional parameter, the so-called coefficient of structure. The coefficient of structure ci is defined
as
ci 
riY
R
2
where riY denotes the simple correlation between predicator variable i and the dependent variable Y and R2 the
coefficient of determination of the multiple regression.
Coefficients of structure measure therefore the fraction of total variability a given predictor variable explains. Again,
the interpretation of ci is not always unequivocal at high levels of multicollinearity.
Partial correlations
2
r zy
Y
1.5
1
X
1
0.5
0.5
0
y = 1.70Z + 0.60
0
0
Y
2
Y
Z
r xy
y = 1.02Z + 0.41
1.5
X
X
r zx
2.5
0.5
1
0
0.5
Z
X  X (Y )  X ( Z )
Y  Y (X )  Y ( Z )
1
Z
The partial correlation rxy/z is the correlation of the residuals X and Y
rX Y / Z 
rX Y  rX Z rYZ
1  rX Z
2
1  rYZ
2
Semipartial correlation
r( X |Y ) Z 
rX Y  rX Z rY Z
1  rY Z
2
A semipartial correlation correlates a variable with one residual only.
Path analysis and linear structure models
Multiple regression
X1
e
X2 X3 X4
Y
e
The error term e
contain the part of
the variance in Y
that is not
explained by the
model. These errors
are called residuals
Y  a 0  a1 X 1  a 2 X 2  a 3 X 3  a 4 X 4  e
e
e
X3
Y
X2
X1
e
X4
e
Regression analysis does not study the
relationships between the predictor
variables
Path analysis defines a whole model and tries to separate correlations into
direct and indirect effects
Path analysis tries to do something that is logically impossible, to derive causal relationships from sets of
observations.
Path analysis is largely based on the computation of partial coefficients of correlation.
Y
p XY
W
p ZY
X
p XW
e
e
Z
Path coefficients
e
p ZX
e
Path analysis is a model confirmatory tool.
It should not be used to generate models or even to seek for models that fit the data set.


X  p xy Y  e


Z  p zx X  p zy Y  e 
W  p xw X  e
We start from regression functions
 p xw X  W  e  0
X  p xy Y  e  0
 p zx X  p zy Y  Z  e  0
From Z-transformed values we get
 p xw X  W  e  0
Z W  p xw Z X  e
X  p xy Y  e  0
Z X  p xy Z Y  e
 p zx X  p zy Y  Z  e  0
Z Z  p zx Z X  p zy Z Y  e

X
Z W Z Y  p xw Z X Z Y  e Z Y
pYX
Y
Z X Z W  p xy Z Y Z W  e Z W
Z Z Z W  p zx Z X Z W  p zy Z Y Z W  e Z W
p XW
Z X Z Z  p xy Z Y Z X  e Z X
W
p XZ
Z X Z Y  p xy Z Y Z Y  e Z Y
p YZ
Z Z Z Y  p zx Z X Z Y  p zy Z Y Z Y  e Z Y

Z
r W Y  p x w rX Y
rX W  p x y rY W
Path analysis is a nice tool
to generate hypotheses.
It fails at low coefficients
of correlation and circular
model structures.
rZ W  p z x rX W  p z y rY W
eZY = 0
ZYZY = 1
rX Z  p x y rY X
rX Y  p x y
rZ Y  p z x rX Y  p z y
ZXZY = rXY
Non-metric multiple regression
Target
symptom
X
1
1
0
1
0
1
0
1
1
0
1
1
1
1
0
0
1
1
0
Predicted value
Sum
A
0
0
1
0
1
0
1
0
1
1
0
0
0
0
1
1
0
0
1
B
1
1
0
1
0
1
0
1
1
1
0
0
0
1
1
1
0
0
1
Symptoms
C
1
1
0
1
0
1
0
1
1
0
1
1
0
0
0
0
1
1
0
D
0
0
0
1
0
1
0
1
1
0
1
1
1
1
0
0
1
1
1
E
1
1
0
1
1
1
0
1
1
1
1
1
1
1
0
0
1
1
1
8
11
10
11
15
Expected values
0.848615
0.848615
-0.2092
1.108631
0.106749
1.108631
-0.2092
1.108631
0.899435
0.19602
1.01936
1.01936
0.575961
0.665233
-0.11992
-0.11992
1.01936
1.01936
0.456037
X'
A
B
C
D
E
0
1
1
0
1
X'X
A
A
B
C
D
E
0
1
1
0
1
B
8
5
1
2
4
1
0
0
0
0
C
5
11
6
6
9
D
1
6
10
8
10
X'Y
1
1
7
10
10
12
0.5
0
-0.5
0
1
Observed occurrences
1
0
0
0
1
E
2
6
8
11
11
4
9
10
11
15
(X'X)-1
0.205969 -0.09304 0.098145
0.0242 -0.08228
-0.09304 0.224792 -0.05216 0.028233 -0.09599
0.098145 -0.05216 0.361387 -0.06158 -0.19064
0.0242 0.028233 -0.06158 0.368379 -0.25249
-0.08228 -0.09599 -0.19064 -0.25249 0.458457
+L$25*B23+L$26*C23+L$27*D23+L$28*E23+L$29*F23
1.5
0
1
1
1
1
(X'X)-1X'Y
-0.2092
0.089271
0.443399
0.260016
0.315945
Statistical inference
n
1
R 
2
Residual
variance
 1
 (Y
n 1
i 1
1
Total variance
n 1
Predicted value
i
 Y ( X i ))
n
 (Y
i
Y)
i 1
1.5
1
0.5
0
-0.5
0
1
2
2
Symptoms
Target
symptom
X
A
B
C
D
E
1
0
1
1
0
1
1
0
1
1
0
1
0
1
0
0
0
0
1
0
1
1
1
1
0
1
0
0
0
1
1
0
1
1
1
1
0
1
0
0
0
0
1
0
1
1
1
1
1
1
1
1
1
1
0
1
1
0
0
1
1
0
0
1
1
1
1
0
0
1
1
1
1
0
0
0
1
1
1
0
1
0
1
1
0
1
1
0
0
0
0
1
1
0
0
0
1
0
0
1
1
1
1
0
0
1
1
1
0
1
1
0
1
1
Mean
0.631579 0.421053 0.578947 0.526316 0.578947 0.789474
Predicted Predicted
Total
Explained
values
values
variance variance
Unexplain
ed
variance
0.848615
0.848615
-0.2092
1.108631
0.106749
1.108631
-0.2092
1.108631
0.899435
0.19602
1.01936
1.01936
0.575961
0.665233
-0.11992
-0.11992
1.01936
1.01936
0.456037
0.022917
0.022917
0.043763
0.011801
0.011395
0.011801
0.043763
0.011801
0.010113
0.038424
0.000375
0.000375
0.179809
0.112069
0.014382
0.014382
0.000375
0.000375
0.207969
1
SE 
n  k 1
2
0.828365
R2 (X'X)-1
0.828365 0.205969 -0.09304 0.098145
0.0242 -0.08228
2
1-R -0.09304 0.224792 -0.05216 0.028233 -0.09599
0.171635 0.098145 -0.05216 0.361387 -0.06158 -0.19064
N
df
0.0242 0.028233 -0.06158 0.368379 -0.25249
19
13 -0.08228 -0.09599 -0.19064 -0.25249 0.458457
B
1
0.047105
0.047105
0.706903
0.227579
0.275446
0.227579
0.706903
0.227579
0.071747
0.189711
0.150374
0.150374
0.003093
0.001133
0.564758
0.564758
0.150374
0.150374
0.030815
0.245614 0.249651 0.042156
Approximated R2
trace ( R )( 1  R )
0.135734
0.135734
0.398892
0.135734
0.398892
0.135734
0.398892
0.135734
0.135734
0.398892
0.135734
0.135734
0.135734
0.135734
0.398892
0.398892
0.135734
0.135734
0.398892
True R2
Observed occurrences
Rounding errors due to
different precisions cause the
residual variance to be larger
than the total variance.
1
1
0
1
0
1
0
1
1
0
1
1
1
1
0
0
1
1
0
A
B
C
D
E
-0.2092
0.089271
0.443399
0.260016
0.315945
0.035351
0.038582
0.062027
0.063227
0.078687
SE(B)
0.052147
0.054478
0.069074
0.069739
0.0778
t
-4.01163
1.638668
6.419143
3.728397
4.060988
P
0.001479
0.125246
2.27E-05
0.002529
0.001348
Logistic and other regression techniques
M a le
M a le
M a le
M a le
M a le
M a le
M a le
M a le
M a le
M a le
F e m a le
F e m a le
F e m a le
F e m a le
F e m a le
F e m a le
F e m a le
F e m a le
F e m a le
F e m a le
A
5 .9 9 8
3 .9 1 6
4 .5 1 1
5 .9 4 0
6 .5 3 2
6 .5 1 3
3 .0 5 2
3 .5 1 2
6 .6 7 6
6 .9 7 6
5 .6 4 9
5 .7 1 2
5 .1 1 2
3 .6 8 1
5 .2 3 9
5 .1 8 0
2 .1 3 3
5 .3 6 1
6 .4 6 0
6 .8 3 9
B
0 .8 3 8
0 .9 9 2
0 .9 0 4
0 .7 9 5
0 .5 7 4
1 .0 3 6
0 .5 8 4
1 .1 2 6
0 .9 9 2
0 .5 0 2
0 .9 1 3
0 .4 7 4
0 .2 7 7
0 .3 2 9
0 .9 2 2
0 .5 4 6
0 .3 0 0
0 .4 7 2
0 .3 2 1
0 .4 2 6
C
2 .2 5 3
1 .9 6 4
1 .9 3 0
1 .1 7 1
1 .3 9 0
0 .5 7 1
2 .1 7 9
1 .8 4 3
2 .2 8 8
1 .0 6 2
2 .2 3 1
2 .2 3 7
1 .0 0 9
2 .4 2 0
1 .5 9 2
2 .4 1 8
3 .0 8 7
2 .1 7 5
1 .0 0 7
3 .1 7 9
Z 
e
y
1 e

y
1
1 e
n
Y  a0 
a
i
xi
i 1
We use odds
n
a0 
n
 p 
ln 
  a0 
1

p


n
ax
i
i 1
i

p
1 p
e
a0 
 aixi
i 1
 p
e
n
a0 
Threshold
1
Z 
0 .8
0 .6
Indecisive
0 .4
region
e
 a i xi
i 1
n
1 e
a0 
 a i xi
i 1
The logistic regression model
0 .2
0
0
5
S urely m ales
10
Y
15
S urely fem ales
20
 aixi
i 1
n
1 e
1 .2
Z
y
a0 
 aixi
i 1
Z 
e
 0.19  0.2 A  6.36 B  1.77 C
1 e
 0.19  0.2 A  6.36 B  1.77 C
1
0 .6
0 .4
0 .2
S ex
F e m a le
F e m a le
F e m a le
F e m a le
F e m a le
M a le
M a le
M a le
M a le
0
M a le
Z
0 .8
Generalized non-linear regression models
Y 
b0
1  b 1e
a0 
1
Y
Y
1
 aixi
b 1= 3
b 1= 1
b 2= 4
b 2 =0 .5
0
0
0
5
x
10
0
5
10
x
A special regression model that is used in pharmacology
Y  b0 
b0
X 
1 

b
 1
b2
b0 is the maximum response at dose saturation.
b1 is the concentration that produces a half maximum response.
b2 determines the slope of the function, that means it is a
measure how fast the response increases with increasing drug
dose.

similar documents