DISCRIMINANT ANALYSIS
A technique for analysing marketing research data when
- the criterion or dependent variable is categorical (non-metric) and
- the predictor or independent variables are interval (metric) in nature.
When the criterion variable has two categories, the technique is known as two-group
discriminant analysis, e.g., Gender – Male & Female
When three or more categories are involved, the technique is referred to as multiple
discriminant analysis, e.g., Experience: Less than 2 years, between 3-5 years and more than 5
years.
For example,
- The dependent variable may be the choice of a brand of personal computer (brand A,
B, or C) and the independent variables may be ratings of attributes of PCs on a 7-
point Likert scale.
- In terms of demographic characteristics, how do customers who exhibit store loyalty
differ from those who do not?
Discriminant Analysis Model
D =b0 + b1X1 + b2X2 +b3X3 + ... +bkXk
where
D - discriminant score
b’s - discriminant coefficient or weight
X’s -predictor or independent variable
Statistics Associated with Discriminant Analysis
Eigenvalue. For each discriminant function, the eigenvalue is the ratio of between-group to
within-group sums of squares.
- Large eigenvalues imply superior functions
- Usually, the eigen value is more than 1 implies the D function is strongly
discriminating
Wilks’lambda (ℷ). Sometimes also called the U statistic, Wilks’ lambda for each predictor is
the ratio of the within-group sum of squares to the total sum of squares.
- Its value varies between 0 and 1. Large values of (near 1) indicate that group means
do NOT seem to be different.
- Small values of (near 0) indicate that the group means seem TO BE different
1
Standardized discriminant function coefficients. These are the discriminant function
coefficients and are used as the multipliers when the variables have been standardized to a
mean of 0 and a variance of 1.
Centroid. The centroid is the mean values for the discriminant scores for a particular group.
- There are as many centroids as there are groups, because there is one for each group.
- The means for a group on all the functions are the group centroids.
Structure correlations /Structure matrix / discriminant loadings. The structure
correlations represent the simple correlations between the predictors and the discriminant
function.
Classification / confusion / prediction matrix. It contains the number of correctly classified
and misclassified cases.
- The correctly classified cases appear on the diagonal, because the predicted and actual
groups are the same.
- The off-diagonal elements represent cases that have been incorrectly classified.
- The sum of the diagonal elements divided by the total number of cases represents the
hit ratio.
- The hit ratio is the percentage of cases correctly classified by discriminant
analysis.
SPSS Commands
Go to “Analyse”
Go to “Classify”
Then choose “discriminant”
In the grouping variable input the “cluster variable” which was formed using cluster analysis
Click “defined range”, input min 1 and max 3
Click “Continue”
Inputs four anxiety factors in independents
Under Classify
In Priority probabilities, check only “all groups equal”
In Use covariance matrix check only “within groups”
Under plots check only “Combined groups”
Under display check “summary table and leave-one-out classification”
Click “Continue”
Under save click “predicted group membership” & “discriminant scores”
Click “Continue”
Click “OK”
Output interpretation
Look into the eigen values table and we can see two functions are there, which have eigen
value more than 1. This means both functions show higher discriminating power.
2
Under Wilk’s Lambda check for significance value, first is 1 through 2 and 2 both functions
have significance value. As the Wilk’s Lambda values are closer to 0, we can say that both
the group means seem to be different.
Under standardised canonical discriminant functions coefficient table, we form the equation
for D1 and D2.
D1 = (0.441 * maths_anxiety) + (0.203 * statistics_anxiety) + (0.996 * computer_anxiety) +(-
0.157 * software_anxiety)
D2 = (-0.453 * maths_anxiety) + (0.976 * statistics_anxiety) + (-0.091 * computer_anxiety) +
(0.051 * software_anxiety)
Look into structure matrix for naming the functions created (D1 and D2). Computer and
software anxiety seem to be loaded in function 1 and maths and statistics anxiety is loaded in
function 2.
- D1 can be named as technology anxiety group
- D2 can be named as aptitude anxiety group
Look at canonical discriminant function graph to observe the D1 and D2 are used to
discriminating three groups.
Look into classification results table and note the footnote ‘c’ value, which is 98.3%. So we
can conclude that we can predict the discriminant based on the clusters 1, 2 and 3 and the IVs
(maths, statistics, computers, and software anxiety).
In the variable view we can see that the discriminant variable is created (row-wise).
Name them technology anxiety and aptitude anxiety.
In columns we can also see the discriminant scores for the respective respondents. These
values can be further used in analysis, e.g. regression analysis.
Characteristic Profile: An aid to interpreting discriminant analysis results by describing
each group in terms of the group means for the predictor variables.