0
votes

I am currently trying to understand cluster analysis (using SPSS and R). Reading up so much about it further confused me as to what clustering method to use to answer a research question.

My research question investigates whether a) certain participants can be clustered according to their change in variable A (a group that remains stable, a group that worsens, and a third group that improves over 2 assessments), and b) how these groups/clusters differ with regard to two other variables at assessment 1 (B and C). That is, do people with different patterns in B and C have a different change in A?

Question: I have standardised the data and, so far, have tried two step hierarchical and k-means clustering. However, I am unsure if this is the right method for answering my question. In the case of a fixed number of clusters, I chose 3 because I am interested in seeing clusters of people that improve/worsen/stay stable over time, and the clusters' individual pattern of B and C. Is this feasible? Am I missing something?

standardized data for variables A, B and C

For k-means clustering I used the following syntax:

QUICK CLUSTER z_A_change z_B_mean z_C_mean 
/MISSING=LISTWISE
/CRITERIA=CLUSTER(3) MXITER(10) CONVERGE(0) 
/METHOD=KMEANS(NOUPDATE) 
/SAVE CLUSTER DISTANCE 

Finally, is there any way to visualise these clusters on a 3-D plot in SPSS? I am not quite as proficient in R's ggplot2 or scatterplot3d as I would like to be.

Thank you in advance.

1
This is a question about SPSS not about R so you should probably remove the R tag. Pictures of data are not useful. It is far better to provide a reproducible example. In R that would involve pasting the output from dput() for at least a small sample of your data.dcarlson
I know, I have used dput() in the past but now I only have an example from SPSS. Not sure how to export an SPSS output in a similar way.annedroid

1 Answers

0
votes

If you use TWOSTEP CLUSTER or QUICK CLUSTER to fit a three cluster solution and save the cluster memberships as a new variable, you can create a grouped 3D scatter plot via the Chat Builder. In the menus, go to Graphs>Chart Builder. In the Gallery view, under Choose from: select Scatter/Dot. In the icons shown underneath the main canvas, the second from the right in the top row should be the grouped 3D scatter. Move that icon into the canvas. Select each of the three variables used in the clustering for the X, Y, and Z axes. Specify the cluster membership variable as the Set Color variable, then click OK.