1
$\begingroup$

As you can read in the title I get the error message "Not enough cases to perform cluster analysis" after trying K-Means Clustering including all the variables (or columns). I will try to provide all the information that you might need to be able to help me. Ask me if need anything else.

  • The data base is from a survey with 409 participants, and the survey had 19 questions (many of them with multiple responses) getting a total of 58 columns.
  • The data base that was loaded to SPSS, was "coded" or "transformed" (I dont know the correct word, I think its Liker scale) with numbers eg: 1 if answer "unemployed" ... 6 if "self-employed" and so on with al the columns and then indicated in the variable what every number meant or tag.
  • Since there is a lot of multiple response questions, I have a lot of empty cells or "missing values". So I tried to replace these empty cells with "0" or "9" and indicate in the program that these numbers corresponded to missing values, but it didn't change anything.
  • I used the option to create the multiple response set that were necessary.
  • All the variables are nominal.
  • I tried to find any similar post that could of have helped me, but I didn't find anything.
  • I uploaded an image of the actual data base so you can see how actually looks like
  • Sorry if I made too many grammar mistakes, English is not my first language. enter image description here
$\endgroup$
3
  • $\begingroup$ K-means clustering in spss deletes cases with missing values listwise. You have many missings on some of your variables. So in the end it may occur that n is less than k. No analysis can be done. Consider removing the variables with missings or doing some imputation of missing data. $\endgroup$ – ttnphns Sep 5 '20 at 11:10
  • $\begingroup$ @ttnphns yea I was considering removing the variables with missing values, but it was important for me to use all the data. If anyone is interested, I ended up solving my problem by using the extension STATS MCSET CONVERT, which converts the multiple category sets in to a multiple dicotomy, creating dummy variables. $\endgroup$ – PabloSalgado Sep 6 '20 at 22:24
  • $\begingroup$ This isn't a remedy against missing values. $\endgroup$ – ttnphns Sep 6 '20 at 22:39
0
$\begingroup$

If missings were due to a category not being selected, then converting to multiple binaries indicating whether or not a category was chosen is probably actually a reasonable solution. The QUICK CLUSTER procedure uses LISTWISE deletion of cases with missing values by default, but you can select the PAIRWISE option, which would not delete cases with missing data, but compute Euclidean distances using only the available variables for a given case.

Filling in system missing (".") values with numbers and specifying those numbers as user missing values by itself makes no difference. If you do this and then specify MISSING=INCLUDE in QUICK CLUSTER, then the procedure will treat those values as valid numbers, so arbitrary values are not helpful. User missing values are typically useful with categorical data where for some purposes you want to treat cases with missing data as a distinct group.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.