Cluster Analysis

Question 1

What does cluster analysis involve?

Answer 1

converting a number of observations in to fewer groups of observations

Question 2

What are the 3 pairing methods for creating clusters?

Answer 2

Euclidean = shortest distance between points
Manhattan = sum of absolute difference between two scores
Pearson

Question 3

What is the main objective for pairing methods?

Answer 3

Maximise similarity within clusters whilst also maximising dissimilarity between clusters

Question 4

What are the 4 linkage procedures?

Answer 4

Single - combining clusters based on most similar points
Complete - combined based on most different members of two clusters
Centroid - combining based on the central value within the clusters
Average - based on average linkage distance between all cases

Question 5

What are the two structures for clustering?

Answer 5

Hierarchical = combining clusters sequentially from most to least similar
Non-hierarchical - informed by prior knowledge we combine them in a specific way

Question 6

What are dendrogams?

Answer 6

Visual ways of clustering they show the distances/processes that went in to clustering

Question 7

When looking to denote clusters using dendrograms what are we using?

Answer 7

Looking for the longest lines as that outlines where a big jump in the data was made to combine those clusters and therefore suggests that they were not very similar

Question 8

What can assist a dendrogram?

Answer 8

Agglomeration schedule

Question 9

What is an agglomeration schedule?

Answer 9

It contains the coefficients of the clusters

Question 10

What are 5 key considerations for when conducting a cluster analysis?

Answer 10

1. logic/subjectivity of a cluster - realistic?
2. Have missing or suitable data?
3. Awareness of outliers and skewness impact?
4. Clusters produced may not be all clusters that exist in real world
5. Clusters in the real world may not have been identified by spss clustering process