Customer Relationship Management: A Database Approach

1
Customer Relationship ManagementA Database
Approach
MARK 7397 Spring 2007
Class 11 Discriminate Analysis
James D. Hess C.T. Bauer Professor of Marketing
Science 375H Melcher Hall jhess_at_uh.edu 713
743-4175
2
Discriminant Analysis of Toaster Customers
GS location of Goldstar buyer
Size of
Family
BD location of Black Decker buyer
10
BD
BD
9
GS
8
BD
7
BD
6
BD
GS
GS
5
4
GS
GS
3
2
1
Taste for
0
10
20
30
40
50
60
70
80
90
100
Speed
3
Toaster Customers Means and Confidence Circles
Size of
Family
Location of average Black Decker buyer
10
BD
BD
9
Location of average Goldstar buyer
GS
8
BD
7
BD
6
BD
GS
GS
5
4
GS
GS
3
2
1
Taste for
0
10
20
30
40
50
60
70
80
90
100
Speed
4
Toaster Customers Means and Confidence Ellipses
Size of
Family
10
BD
BD
9
GS
8
BD
7
BD
6
BD
GS
GS
5
4
GS
GS
3
2
1
Taste for
0
10
20
30
40
50
60
70
80
90
100
Speed
5
Discriminant Analysis of Toaster Customers
Size of
Family
10
BD
BD
9
0-2.0-.013TS.456SF or SF4.2.03TS
GS
8
BD
7
BD
6
BD
GS
GS
5
4
GS
GS
3
2
1
Taste for
0
10
20
30
40
50
60
70
80
90
100
Speed
6
(No Transcript)
7
(No Transcript)
8
(No Transcript)
9
(No Transcript)
10
Assignment of Observation to Population
Suppose we have computed discriminant
coefficients b and have observed a value X for an
observation. Suppose that a priori we think the
two populations have probabilities qBD and qGS
and the costs of misclassification are C(BDGS)
and C(GSBD). Let the group centroids be XBD and
XGS then the observation should be allocated to
GS if
Note if prior probabilities and costs are
symmetric then the last term above is ln(1)0.
However, suppose that qGS2/3 and misclassifying
a GS as BD is 5 times more costly than
misclassifying a BD as GS. Then the threshold
for allocating to GS is easy since ln(10)2.3.