I'm so glad to hear that! Is there something in particular that made the content here more understandable? I ask so that I can make sure to incorporate that type of teaching in my other videos. Thanks!
@@DataDemystified I guess commenting along the way on how to interpret the results/ how all these program steps and numbers in the tables are part of the "story" was particularly helpful :)
Your videos are so easy to understand and its so amazing how many people your kindness has been helping! I have a small question and was wondering if you can share your insight about it if you have time available. A study that I am trying to replicate has categorized individuals based on whether they score above or below the mean on two variables (i.e., high high, high low, low high, low low - 4 categories). I was advised that that technique was crude and that I should instead use a cluster analysis to categorize the groups. Why would cluster analysis be a better statistical analyses than what the original authors did in categorizing the variables? Sorry to trouble you! I look forward to more of your incredibly helpful videos!
Thank you for the excellent videos on clustering. I have a dataset in which the majority of the variables are categorical. Which clustering method would be best for categorical variables? If I convert the categorical variable to have values '1', '2', '3' and so on, could i then use this 'numerised' dataset in hierarchical or k-mean clustering? I would greatly appreciate your thoughts on these questions.
Which method of hierarchical or K-means is the most appropriate for dichotomous variables with binary coding (0,1) showing the presence and absence of a variable?
Hello, I am wanting to follow this process by doing a hierarchical cluster analysis to determine the k for my k-means analysis. My question is, my variables are not all on the same scale so in the hierarchical cluster analysis I will need to convert them into z-scores or something similar so they are comparable. How does this impact the k-means cluster analysis? Do I need to do an extra step here or will my variables already be converted and able to be used again after the hierachical analysis?
During my conjoint analysis, there is no important value and utilities because spss stated "no analysis is performed because there are no valid cases" how to solve this. I did cluster analysis to get the utilities and std.error of each cluster but after performing conjoint to my one cluster, conjoint won't show results. Please help. I have no missing values, no duplication and whatsoever.
Thank you for a great explanation! I was wondering how to view demographic characteristics between each established cluster. For example, viewing percentage breakdowns of age, gender, etc. in each cluster. Thanks!
This is very helpful. Oddly enough the reason i came across this video was because i was searching on how to calculate the initial cluster centers in SPSS. I need them for my R script to perfectly replicate the K mean clusters analysis i run in spss...inputting the initial cluster centers calculated in SPSS provides the exact same results for the final cluster solution in R ...it was the first thing you said we don't need but i am pretty desperate in my search to find out how those initial cluster centers are calculated. Any help you could provide would be huge
Is it necessary to correlate first the variables before proceeding to clustering. Is it important to select only variables that are correlated, for analysis.
Hi I love your videos, really helping me analysis my dissertation data :) I was wondering if you had any videos that touched on how to identify the characteristics of your clusters? i.e. age, gender, those sort of things?
I don’t, but basically you’re just going to run either t-tests/ANOVA or cross tabs. You’d use the cluster number as the independent variable and your demographic as the dependent variable. I have a bunch of videos on those techniques in the SPSS playlist on this channel. Good luck!
Dear Jeff, I was wondering if I could ask you one more question. As I am working with z-scores and trying to compare the means (of z-scores) at the end of the cluster analysis in order to show the difference of variables within and between the clusters, I encountered very high means of z-scores ranging up to 4 or 5. Could this be an indication of outliers? Would you suggest me to remove all the outliers before the analysis or would this change the dataset too much and you would just report it as it is? Thank you!!
4-5 on a z-score is pretty high. We typically consider statistical outliers as being more than 3 standard deviations from the mean (which translates to a z-score of 3 or more). The choice to remove data, based on outliers, however, is a lot more complex. Did you pre-specifiy that you would do so? Are you doing it because your results, inclusive of the outliers don't "Look good"? The point is to make sure that your exclusion isn't going to artificially inflate Type 1 error (p-hacking). Good luck!
@@DataDemystified I did not pre-specify that I would do that. Just compared to other cluster analysis, with other data, and their results (mean z-scores), mine are very high, so I thought that I might have done a mistake and that it would be best to remove the outliers. However, I do not want to manipulate my data. Maybe it is enough to just mention the high z-scores but leave them in the data? Thank you!
@@mariabecker1803 I don't know what context you're reporting in (academic paper, school assignment, etc...) but transparency is always a good thing. At minimum, add a footnote with the explanation. Better yet is a robustness check that is explicitly exploratory: see what happens when you drop those outliers. Do the results meaningfully change? If they do, report that and speculate as to why. If they don't, report that as well with a note about how your results are robust to their removal.
@@DataDemystified Dear Jeff, it´s part of my dissertation so I really want to do a thorough job. I will definetly do a robustness check and am curious to see what will change. So thank you for your advice!
Hi Professor, very good presentation. I would like to know how can you understand your cluster in order to label them? What patterns do you find to classify your cluster?
Hello, I have a doubt and I would really appreciate if you could clarify it. So do we use Hierarchial Cluster Analysis to identify the potential clusters and then K-Means to understand how those clusters are different from each other?
We use Hierarchical Cluster analysis to identify the most likely # of clusters. We then use k-means to actually create those clusters and explore them. Hope that helps!
Hi, I was wondering how to read in cluster centers from an external file (after having done the hierarchical clustering) as SPSS always shows error messages (not correct format or one variable name is incorrect). Do you have a video for that? or any solution to my problem?
Sorry you're having trouble with that. I don't have a video on the topic and don't often import cluster centers from an external file. Is there a reason you are doing it that way rather than natively running the analysis on the data?
@@DataDemystified yes, I am using k-means clustering in order to validate the cluster centers/numbers of clusters that I have calculated with hierarchical clustering. Therefore, I want to use the cluster centers that I have (from the hierarcical clustering) as a starting point and see what changes when I do the k-means clustering. However, no matter what I do (even when I do everything according to the literature) I get error messages and SPSS has troubles reading in the cluster centres from an external file. Would you know what I could do to avoid the error messages and get my results?
@@mariabecker1803 Got it. One option is to just re-run your hierarchical clustering with the original data and then, in the same data file, run the k-means clustering. Save the cluster membership for both analyses, and then do your comparison. If that's not possible and the import isn't working, you can always do it manually. As in, sort the data by some identifier and copy and paste the column of data from your original data (where the hierarchical analysis is) into the new data file (where you plan to run k-means). I hope that helps!
@@DataDemystified Thank you! I have tried that already and it works to compare the two in the same data file. This is not the problem. However, I saw that the cluster memberships are completely different (hierarchical and kmeans), therefore I wanted to do the k-means clustering with the same cluster centers as I discovered in the hierarchical in order to see where the difference is when both have the same starting point, if that makes sense? It is just that there is no other way in order to put in the starting points (cluster centers) manually and only do it with the read in, I guess? which in my case is not working. Therefore, I do not know how to proceed.
@@mariabecker1803 My only suggestion at this point is to make sure you are using Ward's Method in your hierarchical clustering. That tends to give results closest to k-means. Good luck!
@@DataDemystified I wonder whether I should choose cluster analysis in SPSS or perform latent class analysis using Mplus to identify the underlying groups in my data, I am still a bit confused. Can you kindly provide some advice? Thank you.
@@DataDemystified Any chance you'll do one on Structural Equation Modelling? Like I know it's a bunch of regressions under the hood, but it would be nice to see a proper demo of how to use one in real life.
@@lydialim1993 Great idea, but I don't know if that'll happen any time soon. The challenge is that you need the AMOS package for SPSS, which most people don't have (including me, at the moment). That said, I'll look into how much demand there is for something like this! Thanks for the suggestion!
This is the 5th video I wath on K-Means and it FINALLY made sense. Thank you so much!
I'm so glad to hear that! Is there something in particular that made the content here more understandable? I ask so that I can make sure to incorporate that type of teaching in my other videos. Thanks!
@@DataDemystified I guess commenting along the way on how to interpret the results/ how all these program steps and numbers in the tables are part of the "story" was particularly helpful :)
@@xeniavlasenko9830 Thank you for the feedback! I will make sure to incorporate it into new tutorial videos!
Fantastic explanation! nothing to do with all those videos out there! keep going like this!
Thanks!
Can k means clustering data be represented graphically? If yes, how?
You are a hero!
Ha. Thank you!
Great advanced info, subscribed!
Your videos are so easy to understand and its so amazing how many people your kindness has been helping! I have a small question and was wondering if you can share your insight about it if you have time available. A study that I am trying to replicate has categorized individuals based on whether they score above or below the mean on two variables (i.e., high high, high low, low high, low low - 4 categories). I was advised that that technique was crude and that I should instead use a cluster analysis to categorize the groups. Why would cluster analysis be a better statistical analyses than what the original authors did in categorizing the variables? Sorry to trouble you! I look forward to more of your incredibly helpful videos!
Thank you for the excellent videos on clustering. I have a dataset in which the majority of the variables are categorical. Which clustering method would be best for categorical variables? If I convert the categorical variable to have values '1', '2', '3' and so on, could i then use this 'numerised' dataset in hierarchical or k-mean clustering? I would greatly appreciate your thoughts on these questions.
thank you for the great video, would you please explain about how to apply elbow method to find cluster number?
Which method of hierarchical or K-means is the most appropriate for dichotomous variables with binary coding (0,1) showing the presence and absence of a variable?
Hello, I am wanting to follow this process by doing a hierarchical cluster analysis to determine the k for my k-means analysis. My question is, my variables are not all on the same scale so in the hierarchical cluster analysis I will need to convert them into z-scores or something similar so they are comparable. How does this impact the k-means cluster analysis? Do I need to do an extra step here or will my variables already be converted and able to be used again after the hierachical analysis?
During my conjoint analysis, there is no important value and utilities because spss stated "no analysis is performed because there are no valid cases" how to solve this. I did cluster analysis to get the utilities and std.error of each cluster but after performing conjoint to my one cluster, conjoint won't show results. Please help. I have no missing values, no duplication and whatsoever.
Can K-means cluster analysis be performed with all categorical variables?
Hello if any video about anaaysis kmean results deeper
How do we define the names as to what is increasing and what is not
In case the Sig for some variables is bigger than .001 what should we do? Should we screen and remove them and do the cluster analysis again?
To cluster analysis for multimorbidity, is there any special format to arrange the data?
Thank you for a great explanation! I was wondering how to view demographic characteristics between each established cluster. For example, viewing percentage breakdowns of age, gender, etc. in each cluster. Thanks!
Is it the same on SPSS29?
the final conclusion of that data
how can we plot this data in spss the way R does? is there a way?
This is very helpful. Oddly enough the reason i came across this video was because i was searching on how to calculate the initial cluster centers in SPSS.
I need them for my R script to perfectly replicate the K mean clusters analysis i run in spss...inputting the initial cluster centers calculated in SPSS provides the exact same results for the final cluster solution in R
...it was the first thing you said we don't need but i am pretty desperate in my search to find out how those initial cluster centers are calculated. Any help you could provide would be huge
Is it necessary to correlate first the variables before proceeding to clustering. Is it important to select only variables that are correlated, for analysis.
Nope. Clustering does not require variables to be correlated.
How can run k-mean clustering algorithm for 40 sample with four variables to group the sample to into two?
I need to do assignments between kmean and improved kmean cluster analysis,can pls tell me how to do that
Hi I love your videos, really helping me analysis my dissertation data :) I was wondering if you had any videos that touched on how to identify the characteristics of your clusters? i.e. age, gender, those sort of things?
I don’t, but basically you’re just going to run either t-tests/ANOVA or cross tabs. You’d use the cluster number as the independent variable and your demographic as the dependent variable. I have a bunch of videos on those techniques in the SPSS playlist on this channel. Good luck!
hi can I did references for the way how you had selected the variables ? or what it is called ?
Thank you very much, this video is very easy to understand !!
How to do improved k mean cluster analysis
Dear Jeff, I was wondering if I could ask you one more question. As I am working with z-scores and trying to compare the means (of z-scores) at the end of the cluster analysis in order to show the difference of variables within and between the clusters, I encountered very high means of z-scores ranging up to 4 or 5. Could this be an indication of outliers? Would you suggest me to remove all the outliers before the analysis or would this change the dataset too much and you would just report it as it is? Thank you!!
4-5 on a z-score is pretty high. We typically consider statistical outliers as being more than 3 standard deviations from the mean (which translates to a z-score of 3 or more). The choice to remove data, based on outliers, however, is a lot more complex. Did you pre-specifiy that you would do so? Are you doing it because your results, inclusive of the outliers don't "Look good"? The point is to make sure that your exclusion isn't going to artificially inflate Type 1 error (p-hacking). Good luck!
@@DataDemystified I did not pre-specify that I would do that. Just compared to other cluster analysis, with other data, and their results (mean z-scores), mine are very high, so I thought that I might have done a mistake and that it would be best to remove the outliers. However, I do not want to manipulate my data. Maybe it is enough to just mention the high z-scores but leave them in the data? Thank you!
@@mariabecker1803 I don't know what context you're reporting in (academic paper, school assignment, etc...) but transparency is always a good thing. At minimum, add a footnote with the explanation. Better yet is a robustness check that is explicitly exploratory: see what happens when you drop those outliers. Do the results meaningfully change? If they do, report that and speculate as to why. If they don't, report that as well with a note about how your results are robust to their removal.
@@DataDemystified Dear Jeff, it´s part of my dissertation so I really want to do a thorough job. I will definetly do a robustness check and am curious to see what will change. So thank you for your advice!
Hi Professor, very good presentation. I would like to know how can you understand your cluster in order to label them? What patterns do you find to classify your cluster?
Great series of videos -- thank you! I do have one follow up question: What are the sample size requirements for Cluster Analysis? Thank you!
Hello, I have a doubt and I would really appreciate if you could clarify it. So do we use Hierarchial Cluster Analysis to identify the potential clusters and then K-Means to understand how those clusters are different from each other?
We use Hierarchical Cluster analysis to identify the most likely # of clusters. We then use k-means to actually create those clusters and explore them. Hope that helps!
@@DataDemystified yes it does thanks alot 🙌
I really thank you for this great series of videos they have been so much useful in my research
thank you so much! for this one and the hierarchical one!
Hi, I was wondering how to read in cluster centers from an external file (after having done the hierarchical clustering) as SPSS always shows error messages (not correct format or one variable name is incorrect). Do you have a video for that? or any solution to my problem?
Sorry you're having trouble with that. I don't have a video on the topic and don't often import cluster centers from an external file. Is there a reason you are doing it that way rather than natively running the analysis on the data?
@@DataDemystified yes, I am using k-means clustering in order to validate the cluster centers/numbers of clusters that I have calculated with hierarchical clustering. Therefore, I want to use the cluster centers that I have (from the hierarcical clustering) as a starting point and see what changes when I do the k-means clustering. However, no matter what I do (even when I do everything according to the literature) I get error messages and SPSS has troubles reading in the cluster centres from an external file. Would you know what I could do to avoid the error messages and get my results?
@@mariabecker1803 Got it. One option is to just re-run your hierarchical clustering with the original data and then, in the same data file, run the k-means clustering. Save the cluster membership for both analyses, and then do your comparison. If that's not possible and the import isn't working, you can always do it manually. As in, sort the data by some identifier and copy and paste the column of data from your original data (where the hierarchical analysis is) into the new data file (where you plan to run k-means). I hope that helps!
@@DataDemystified Thank you! I have tried that already and it works to compare the two in the same data file. This is not the problem. However, I saw that the cluster memberships are completely different (hierarchical and kmeans), therefore I wanted to do the k-means clustering with the same cluster centers as I discovered in the hierarchical in order to see where the difference is when both have the same starting point, if that makes sense? It is just that there is no other way in order to put in the starting points (cluster centers) manually and only do it with the read in, I guess? which in my case is not working. Therefore, I do not know how to proceed.
@@mariabecker1803 My only suggestion at this point is to make sure you are using Ward's Method in your hierarchical clustering. That tends to give results closest to k-means. Good luck!
Great! I was able to complete my entire assignment, using only this video
brilliant! thank you for uploading this video!
My pleasure!
@@DataDemystified I wonder whether I should choose cluster analysis in SPSS or perform latent class analysis using Mplus to identify the underlying groups in my data, I am still a bit confused. Can you kindly provide some advice? Thank you.
@@LXiao33 That entirely depends on your research question. Without knowing that, I really can't answer your question. Sorry!
Thanks for the demonstration
After a long search , finally something easy to understand
So clear, thank you so much!
Wonderful series! Keep it up!
Thank you! Any topics you'd specifically like to see covered?
@@DataDemystified Any chance you'll do one on Structural Equation Modelling? Like I know it's a bunch of regressions under the hood, but it would be nice to see a proper demo of how to use one in real life.
@@lydialim1993 Great idea, but I don't know if that'll happen any time soon. The challenge is that you need the AMOS package for SPSS, which most people don't have (including me, at the moment). That said, I'll look into how much demand there is for something like this! Thanks for the suggestion!
Very clear explanations. Thank you.
Thanks!
prof i need your support can you help me