Hello,
Clustering methods are new to me: I'm trying to find out which would be the best clustering method for a sample size of around 350 observations and around fifteen variables, either quantitative or binary.
I'm using SPSS: what would be the most convenient method between HCA, K-mean and a two-step cluster analysis?
Hello Juliette,
Is there some a priori number of clusters that you seek, either because of previous research, theory, or some similar consideration? If so, k-means methods are quite efficient (though, with N ~ 350 and 15 variables, efficiency won't be much of a concern).
If not, hierarchical cluster methods might be worth a look. Of course, your choice on how to handle any number of decision points can well yield a different apparent solution than would some other set of choices. Such is the challenge of clustering. As well, verifying that the resultant clusters have actual utility (vs. numerical differences alone) is where the payoff to the whole enterprise lays, in my opinion.
Among the decision points:
1. Whether to standardize scores, and if so, how;
2. How to quantify similarity/proximity;
3. What specific method is chosen;
4. What stopping criterion to use (for agglomerative clustering or divisive clustering);
5. How to affirm cluster distinctiveness;
6. How to affirm cluster utility.
Good luck with your work.
As David Morse notes, there are two main types of cluster analysis. I'd add that if interested in deciding the number of clusters to use in a kmeans procedure, the gap statistic (doi = 10.1111/1467-9868.00293) can be used. It works similarly to the scree test and like some functions for the scree test, it can help you to decide what k might be good (though of course this should be based also on the usefulness of the results for your research question). Also, on your research questions, there are lots of other procedures that could be used to classify 350 people on the baseis of 15 variables (e.g., mixture models, latent trait analysis).
For your dataset with around 350 observations and approximately fifteen variables (quantitative or binary), several clustering methods can be considered. Here are three commonly used clustering algorithms that could be suitable for your data:
The choice of the clustering method depends on various factors such as the nature of your data, the desired number of clusters, interpretability, computational efficiency, and the specific goals of your analysis. It is often recommended to try multiple clustering algorithms and compare the results to better understand the underlying structures in your data.
- Badges
- Science topic
- Similar topics
- Books