Discriminant Analysis

Green and Salkind (2017) described discriminant analysis as a tool that can classify people into groups or distinguish groups from one another, based on a variety of measures. For instance, job applicants could be grouped based on a psychological profile test, school GPA, years of experience, and other factors. In this post I will show a hypothetical example of a discriminant analysis as it pertains to quantitative research. The study referenced below was assigned with variables that I could derive from a quantitative consultation with any company or organization. The data points can change but the analysis gives clear actionable data upon which to make informed decisions. So, without further delay, let’s dig into it.

Research Question

Will the publications, grant funding, teaching rating, and committees served be enough to predict the type of professor the applicant will be?

Hypotheses

H0: Publications, grant funding, teaching rating, and committees served will not be enough to predict the type of professor the applicant will be.

H1: Publications, grant funding, teaching rating, and committees served will be enough to predict the type of professor the applicant will be.

Results

In this subheading, I will present descriptive statistics, discuss testing of the assumptions, present inferential statistic results, and conclude with a concise summary. Descriptive Statistics

A total of 25 professors participated in the study. Table 1 depicts descriptive statistics for the study variables for each group. Figure 1 depicts a bar graph of research scientist type and Figure 2 depicts the teaching mogul type.

Table 1

Means and Standard Deviations for Quantitative Study Variables

Type of professor Variable M SD
Research Scientist Number of publications in last 2 years 2.9200 .81240
  Grant funding (in $10000) over the last 5 years 4.3692 2.47581
  Mean teaching rating for last 3 semesters 2.8782 .60132
  Number of committees served on during last 5 years 2.6400 2.34307
Teaching Mogul Number of publications in last 2 years 3.4800 .82260
  Grant funding (in $10000) over the last 5 years 4.5471 3.17875
  Mean teaching rating for last 3 semesters 3.5122 .42025
  Number of committees served on during last 5 years 3.4800 2.23830
Total Number of publications in last 2 years 3.200 .85714
  Grant funding (in $10000) over the last 5 years 4.4581 2.82125
  Mean teaching rating for last 3 semesters 3.1952 .60511
  Number of committees served on during last 5 years 3.0600 2.30713
       

Inferential Results

One discriminant function was found to be statistically significant (Wilks’s Λ = .609, (4) = 22.849, p = .000 for discriminant function 1). The first discriminant function explains 64.3% of variance and the second discriminant function explains the rest of variance. Canonical correlation is .626 for both discriminant functions 1, indicating that 63% of variances were explained by the relationship between predictors and group membership by DF1 and DF2, respectively. Discriminant function 1 has the largest relationship with grant funding, followed by number of committees, number of publications, and teaching rating. Pairwise comparison showed only one pair. DF1 scores indicates that differences on DF1 were found in the following: Research Scientist (M = -.786) vs. Teaching Mogul (M = .786).

References

Green, S. B., & Salkind, N. J. (2017). Using SPSS for Windows and Macintosh: Analyzing and understanding data (8th ed.). Pearson..

5

Appendix – Discriminant Analysis SPSS Output

 

Analysis Case Processing Summary
Unweighted Cases N Percent
Valid 50 100.0
Excluded Missing or out-of-range group codes 0 .0
At least one missing discriminating variable 0 .0
Both missing or out-of-range group codes and at least one missing discriminating variable 0 .0
Total 0 .0
Total 50 100.0
Group Statistics
Type of professor Mean Std. Deviation Valid N (listwise)
Unweighted Weighted
Research Scientist Number of publications in last 2 years 2.9200 .81240 25 25.000
Grant funding (in $10000) over the last 5 years 4.3692 2.47581 25 25.000
Mean teaching rating for last 3 semesters 2.8782 .60132 25 25.000
Number of committees served on during last 5 years 2.6400 2.34307 25 25.000
Teaching Mogul Number of publications in last 2 years 3.4800 .82260 25 25.000
Grant funding (in $10000) over the last 5 years 4.5471 3.17875 25 25.000
Mean teaching rating for last 3 semesters 3.5122 .42025 25 25.000
Number of committees served on during last 5 years 3.4800 2.23830 25 25.000
Total Number of publications in last 2 years 3.2000 .85714 50 50.000
Grant funding (in $10000) over the last 5 years 4.4581 2.82125 50 50.000
Mean teaching rating for last 3 semesters 3.1952 .60511 50 50.000
Number of committees served on during last 5 years 3.0600 2.30713 50 50.000
Tests of Equality of Group Means
  Wilks’ Lambda F df1 df2 Sig.
Number of publications in last 2 years .891 5.865 1 48 .019
Grant funding (in $10000) over the last 5 years .999 .049 1 48 .826
Mean teaching rating for last 3 semesters .720 18.675 1 48 .000
Number of committees served on during last 5 years .966 1.680 1 48 .201
Pooled Within-Groups Matricesa
  Number of publications in last 2 years Grant funding (in $10000) over the last 5 years Mean teaching rating for last 3 semesters Number of committees served on during last 5 years
Covariance Number of publications in last 2 years .668 .067 -.077 .219
Grant funding (in $10000) over the last 5 years .067 8.117 -.043 3.907
Mean teaching rating for last 3 semesters -.077 -.043 .269 -.057
Number of committees served on during last 5 years .219 3.907 -.057 5.250
Correlation Number of publications in last 2 years 1.000 .029 -.181 .117
Grant funding (in $10000) over the last 5 years .029 1.000 -.029 .599
Mean teaching rating for last 3 semesters -.181 -.029 1.000 -.048
Number of committees served on during last 5 years .117 .599 -.048 1.000
a. The covariance matrix has 48 degrees of freedom.
Covariance Matricesa
Type of professor Number of publications in last 2 years Grant funding (in $10000) over the last 5 years Mean teaching rating for last 3 semesters Number of committees served on during last 5 years
Research Scientist Number of publications in last 2 years .660 .426 -.004 .137
Grant funding (in $10000) over the last 5 years .426 6.130 -.155 3.185
Mean teaching rating for last 3 semesters -.004 -.155 .362 .108
Number of committees served on during last 5 years .137 3.185 .108 5.490
Teaching Mogul Number of publications in last 2 years .677 -.292 -.149 .302
Grant funding (in $10000) over the last 5 years -.292 10.104 .068 4.630
Mean teaching rating for last 3 semesters -.149 .068 .177 -.222
Number of committees served on during last 5 years .302 4.630 -.222 5.010
Total Number of publications in last 2 years .735 .091 .016 .335
Grant funding (in $10000) over the last 5 years .091 7.959 -.014 3.866
Mean teaching rating for last 3 semesters .016 -.014 .366 .080
Number of committees served on during last 5 years .335 3.866 .080 5.323
a. The total covariance matrix has 49 degrees of freedom.
Log Determinants
Type of professor Rank Log Determinant
Research Scientist 4 1.638
Teaching Mogul 4 .859
Pooled within-groups 4 1.542
The ranks and natural logarithms of determinants printed are those of the group covariance matrices.
Test Results
Box’s M 14.104
F Approx. 1.283
df1 10
df2 11015.139
Sig. .234
Tests null hypothesis of equal population covariance matrices.
Eigenvalues
Function Eigenvalue % of Variance Cumulative % Canonical Correlation
1 .643a 100.0 100.0 .626
a. First 1 canonical discriminant functions were used in the analysis.
Wilks’ Lambda
Test of Function(s) Wilks’ Lambda Chi-square df Sig.
1 .609 22.849 4 .000
Standardized Canonical Discriminant Function Coefficients
  Function
1
Number of publications in last 2 years .567
Grant funding (in $10000) over the last 5 years -.118
Mean teaching rating for last 3 semesters .890
Number of committees served on during last 5 years .280
Structure Matrix
  Function
1
Mean teaching rating for last 3 semesters .778
Number of publications in last 2 years .436
Number of committees served on during last 5 years .233
Grant funding (in $10000) over the last 5 years .040
Pooled within-groups correlations between discriminating variables and standardized canonical discriminant functions  Variables ordered by absolute size of correlation within function.
Canonical Discriminant Function Coefficients
  Function
1
Number of publications in last 2 years .694
Grant funding (in $10000) over the last 5 years -.041
Mean teaching rating for last 3 semesters 1.716
Number of committees served on during last 5 years .122
(Constant) -7.891
Unstandardized coefficients
Functions at Group Centroids
Type of professor Function
1
Research Scientist -.786
Teaching Mogul .786
Unstandardized canonical discriminant functions evaluated at group means
Classification Processing Summary
Processed 50
Excluded Missing or out-of-range group codes 0
At least one missing discriminating variable 0
Used in Output 50
Prior Probabilities for Groups
Type of professor Prior Cases Used in Analysis
Unweighted Weighted
Research Scientist .500 25 25.000
Teaching Mogul .500 25 25.000
Total 1.000 50 50.000
Classification Function Coefficients
  Type of professor
Research Scientist Teaching Mogul
Number of publications in last 2 years 5.743 6.834
Grant funding (in $10000) over the last 5 years .570 .505
Mean teaching rating for last 3 semesters 12.417 15.113
Number of committees served on during last 5 years -.027 .166
(Constant) -28.157 -40.560
Fisher’s linear discriminant functions
Classification Resultsa,c
    Type of professor Predicted Group Membership Total
    Research Scientist Teaching Mogul
Original Count Research Scientist 20 5 25
Teaching Mogul 3 22 25
% Research Scientist 80.0 20.0 100.0
Teaching Mogul 12.0 88.0 100.0
Cross-validatedb Count Research Scientist 18 7 25
Teaching Mogul 4 21 25
% Research Scientist 72.0 28.0 100.0
Teaching Mogul 16.0 84.0 100.0
a. 84.0% of original grouped cases correctly classified.
b. Cross validation is done only for those cases in the analysis. In cross validation, each case is classified by the functions derived from all cases other than that case.
c. 78.0% of cross-validated grouped cases correctly classified.

Let’s look at your data together. Book Time Here

Tags

No responses yet

Leave a Reply

Your email address will not be published. Required fields are marked *