These patients have over the last 18 months been referred to a clinic where they received lifestyle classes in order to stop them using the emergency services with such regularity. The time they received instruction was between 1 month and 8 months, average 5 months, depending on how well they responded.

We have summed their ED attendances in 6 month blocks covering four years prior to referral and summed the attendances 6 months after discharge. Prior to referral the number of attendances per 6 month period was, 95, 90, 80, 107, 123, 125, 139 and 180. In the six months after discharge the number of attendances was 112. Note the periods do not necessarily correspond to the same calendar periods as both referrals and discharges were staggered over the 18 months.

The business question is, has the clinic succeeded in changing behaviour of these patients? Unfortunately due to national IG laws and patient confidentiality, until a person is referred we cannot track their activity even though the information is available and consent to use their data is only given on referral. I was thinking of pairing each patient with a similar patient who had not been referred and carry out the same analysis for them. ]]>

(1) True/False: Clustering is an unsupervised learning problem.

(2) True/False: Principal Components Analysis can be used to create a low dimensional projection of the data for use with clustering.

3) True/False: Common factors estimated using maximum likelihood estimation with a PROMAX rotation are orthogonal.

4) True/False: Common factors estimated using Iterated Principal Factor Analysis with a VARIMAX rotation are orthogonal.

5) True/False: In cluster analysis the choice of similarity measure will affect the cluster assignments.

6) True/False: When computing principal components the data should be standardized, i.e. the data should be centered and scaled to a (0,1) distribution.

7) True/False: Cluster analysis can only be performed on continuous variables.

8) True/False: Hierarchical clustering requires that the number of clusters be specified in advance.

9) True/False: Factor Analysis and Principal Components Analysis have the same objective of modeling the correlation structure in multivariate data.

10) True/False: Since cluster analysis is an unsupervised learning method, two different cluster partitions cannot be compared.

My answers:

1) True

2) false

3) true

4) false

5) true

6) true

7) false

8) false

9) false

10) true ]]>