Spss For Discriniment Analysis

SPSS Syntax for creating a scoring model using all segmenting variables:

1. First, select �analyze’, then �discriminant’ from the top menu.

2. Then select your grouping variable as the variable you use to name your clusters. In this example, ours is called �clus4’

a. You will have to manually define the range of this variable. The range is equal to the total # of clusters. In this example, the range is 1 to 4

3. Then select as your �independents’ all of the segmenting / basis variables. In this example we have 68 in total

4. Then select �enter independents together’ underneath the field that shows which independent variables you’ve selected

5. Then, under the �statistics’ sub-menu, select �Fisher’s’ under the �Function Coefficients’ sub-menu

a. Note: Fisher’s coefficients would provide you with the eventual coefficients that would be input into the Excel scoring model tool to help type respondents

i. The �COEFF’ after the �/STATISTICS’ command is the syntax for producing the �Fisher’s Coefficients’

6. Under the �classify’ sub-menu, you need to select the following

a. �Compute from group sizes’ under �prior probabilities’

b. �Within groups’ under �Use Covariance Matrix’, or �separate groups’ if your Box M results are significant. By selecting �separate groups’ if the Box M results are significant, you may be able to get a better cross-validation score in the end

c. �Summary Table’ and �Leave-one-out classification’ under �Display’

d. �Territorial Map’ under �Plot’

7. Then select �paste’ and you should see a syntax window come up with the following code:

DISCRIMINANT

/GROUPS=clus4(1 4)

/VARIABLES=g4a g4b g4c g4e g4f g4g g4h g4i g4j g4k g4l g4m g4n g4o g4r g4s

g4t g4v g4w g4bb g4cc i1a i1b i1c i1d i1f i1h i1i i1j i1u i1w i1x i1y i1z

i1aa i1bb i1cc j2b j2c j2d j2e j2f j2g j2h j2i j2j j2l j2m j2n j2o j2p j2q

l1a l1b l1c l1e l1f l1h l1j m1b m1c m1d m1e m1f m1g m1j m1l m1q

/ANALYSIS ALL

/PRIORS SIZE

/STATISTICS=COEFF TABLE CROSSVALID

/PLOT=MAP

/CLASSIFY=NONMISSING POOLED .

8. Then select �run’ in the syntax menu bar and you will see the results plot out. Here’s what you need to look for in your results

a. Look under �summary of canonical discriminant functions’ at the table called: Eigenvalues. See how well the two discriminant functions explain the variance of the market. If it’s over 80-85%, then we’re probably OK to proceed

b. Then look at the chi-square column under the �Wilks’ Lambda’ table. If the proportion of the chi-square for 1 through 3 and 2 through 3 are by and large greater than 3 alone, then we’re probably OK to proceed

c. Look at the �Territorial Map’ plot and see how the clusters map out relative to one another on the two discriminant functions. If it looks like there is good separation among all clusters, then we’re probably OK to proceed

9. If a, b, and c from #8 seem indicative that a two discriminant function solution is valid to show a perceptual map, then we’ll want to add one command in the syntax to force SPSS to collapse down the solution into 2 functions only. Copy and paste the �/FUNCTION=2’ command into the appropriate spot after the �ANALYSIS’ command. See below. Now we will be able to run a new DA and be able to see how well the map coordinates will look (these will come from the structure matrix for attitudes and the group centroids for clusters). We will then export these to the Excel DA Pmap solver to finalize the segment positioning map

10. Finally, we’ll look at the bottom of the output for two tables

a. The table �classification function coefficients’ will be the Fisher coefficients that get dumped into the Excel scoring model template to be multiplied by future respondents’ answers to the segmenting attitudes

b. The table вЂ?classification results’ shows the cross-validated accuracy results which tell us how accurate our scoring model is. In general, we’d like to see our вЂ?cross-validated accuracy’ to be п‚Ñ- 80%. This is equal to saying that we have an 80% chance of accurately assigning an MD to the correct segment based on their responses

DISCRIMINANT

/GROUPS=clus4(1 4)

/VARIABLES=g4a g4b g4c g4e g4f g4g g4h g4i g4j g4k g4l g4m g4n g4o g4r g4s

g4t g4v g4w g4bb g4cc i1a i1b i1c i1d i1f i1h i1i i1j i1u i1w i1x i1y i1z

i1aa i1bb i1cc j2b j2c j2d j2e j2f j2g j2h j2i j2j j2l j2m j2n j2o j2p j2q

l1a l1b l1c l1e l1f l1h l1j m1b m1c m1d m1e m1f m1g m1j m1l m1q

/ANALYSIS ALL

/FUNCTION=2

/PRIORS SIZE

/STATISTICS=COEFF TABLE CROSSVALID

/PLOT=MAP

/CLASSIFY=NONMISSING POOLED .

SPSS Syntax for creating a scoring model using five variables only:

1. Copy the syntax command below into a new syntax window

2. Change the �clus4’

...