Thank you for your continued videos. This is another good snippet. I have watched your series and they have been a huge help in getting a good grasp on the usability of factor analysis in combination with regression as well as SEM. I would love to see more additions to the already great concentration of educational videos you have.
Thank you for sharing! I was confused on how to plot residuals because I need to test my data for homoscedasticity and you've shown that SPSS can do that. :D
Thanks, great video as always. Just wondering, is this to check homoscedasticity before or after using a statistical technique? For example, if checking homoscedasticity of the variate after regression, I read that you can plot the studentized residuals against the standardized predicted residuals (both values can be saved in SPSS and then plotted on a graph). Using this plot, you would expect equal dispersion about zero, i.e., no tendency to be greater than or less than zero.
Hey James, I'm working on my dissertation and with an intent of being thorough I was wondering if (1) I should show the plots for each IV-DV combo (2) while there is no "coning" I do have my X-axis data almost in straight horizontal line increments ...any idea why that could be happening? (I am using means of IVs if that matters). Thanks!
+Val_C No need to include every plot. The columned plots are due to the nature of the scale. If you were using Likert scales, then the responses can only fit into the 5 or 7 discrete values available to respondents.
+James Gaskin That's what I figured but as you can imagine it looks so much weirder than something continuous looks like :) thanks so much for getting back to me so quickly btw. Now, I have a 2nd order related question which I'll go post on that video ;)
Thank you, James, for your wonderful work. I attended one of your 3 days boot camp at RMIT in 2017. Since then I have been following your channels. After going through some books, I am wondering is it necessary to check for homoscedasticity even after checking for normality and multicollinearity, for a PhD thesis?. If so, can we really do this homoscedasticity test on a 5-point Likert scale, whereas, authors like Tabachnick and Fidell (2019, p.73) only mentioned about continuous variables?. In my case, I have 1 IV, 1 mediator, 1 moderator, and 2 DVs. I am getting homoscedasticity for 1 DV and heteroskedasticity for the 2nd variable, when I use 1 IV and 1 DV, but when using 3 IVS together on each of the DVs separately, the results can be considered more as homoscedastic? What could be the soloution to this issue?. Many thanks.
@@Gaskination Thank you very much, James. I highly appreciate that for getting back to me very quickly. I shared your response with my supervisor too. I wish I could use your comment as a reference in the thesis. lolz. Stay blessed.
Hi Gaskin Thanks for answering my other question on the other video. My question is how do you test for linearity and homoscedasticity in SPSS when one of the variables is a latent variable? In my case it is the dependent variable. Many thanks.
Hello! Thank you for the helpful video. How would I check for homoscedasticity in my variables if I'm going to run a 2x2 RM anova? There are no 2 separate variables for me to put in the independent and dependent boxes how you did here (if that makes sense) since my design is within-subjects. Is it even necessary for RM Anova? Thank you in advance for the help!
Okay looks easy enough! :) I am confused about one thing though, if I wanted to make many comparisons/ tests with one variable then would I have to do a homoscedasticity test for every comparison? If I want to know some other things like linearity then I would only do one test per variable (i.e. looking at linearity for variables A,B,C,D) But now for homoscedasticity I would have to do tests on the relations between A-B, A-C, A-D, B-C, B-D, C-D? Also, does it matter which of these variables is entered as the dependent in their respective relationships? Because I have correlational/ observational data, so I don't feel like it should matter.
Correct, you would need to do this for all pairs of variables involved in cause and effect relationships. You do not need to do it for variables only linked by correlation.
Hi Gaskin. A very informative guide on the homoscedasticity. I would like to get your guide on testing homoscedasticity on the variables in my study. I have 2 IV, 2 MEDIATOR, 1 DV, 2 MODERATOR and 2 CONTROL VARIABLES (AGE & EDUCATION). How do I run the homoscedasticity, linearity & normality? Between which variables?
Currently, I actually recommend to skip this step, as it is not very informative and there is little one can do about it anyway... The only thing I would recommend still testing (out of the things you listed) is normality (skewness and kurtosis).
Hi Gaskin Your video really nice. just wondering how did you get variable "values" namely JOY not as indicator anymore? due to when you explain about normality test you still use every single items per variable? Thank you in advance
+sunu widianto` I'm not sure I understand the question. But I think you are asking how I calculated the JOY variable from all of its items. If so, then the answer is that I imputed factor scores in AMOS during my CFA. Here is a video: ruclips.net/video/dsOS9tQjxW8/видео.html
+James Gaskin Hi Gaskin yes you right I am asking about that but if I want to calculated the JOY variable from all of its items from SPSS how is the procedure? Is it okay it I calculate the items with mean to become JOY variable. Many thanks for your kind respond
Sir, thanx for the video. i have 2 questions (1) hair et al 2010 provides two methods for homoscedasticity first i sgraphical which you have shown here and the second is the levene test ..... kindly guide me how can i check the homoscedasticity of my complete SEM model ? and do you have any video for levene test?
Here is a video for a Levene's test: ruclips.net/video/E5VIKZU5kB8/видео.html As for testing your whole model simultaneously for homoscedasticity, this does not make much sense. The levene's test allows you to do a homogeneity of variance test, which is roughly what you're asking, but for the items, not the model.
Respected Sir, relate to prior question where I found '' No variance within groups - statistics for DV * IV cannot be computed'' I have 5 variables and measured with 5 Likert scale (strongly disagree 1 to strongly agree 5) and tested for linearity i.e. 4 IVs and 1 DV. What might be the issue. ? although respondents answered as 1 to 5 to all of these variables. Will be thanked.
That error means that for one of your groups (assuming your data has been split or filtered), at least one of the variables has no variance - meaning that the variable has all the same responses. For example, let's say you're studying clubs, and you use gender as a predictor. If one of the clubs is a female only group, then for that group, you'll have zero variance for the gender variable (i.e., all will be female). You cannot have a predictor variable that doesn't vary.
Hi Gaskin, Thank a lot for a clear explanation of Homoscedasticity. I have a question, I am working with dummy independent variable (Categorical variable). when I run the Homoscedasticity test, the graph is confusing. I am unable to interpret it. Please guide me, how to test homoscedasticity for dummy variable and how to interpret the results. Thank you.
Homoscedasticity is irrelevant to a dummy variable because the values in the dummy variable only represent the presence or absence of a trait, rather than a gradual increase of a trait.
I respond to similar requests many times each day, so please keep your email to me very concise and direct. Also, please only ask for help if you get stuck and after you've searched the youtube channel and StatWiki for answers. Thanks! james.gaskin@byu.edu
Thank you for your help, can you please provide any reference for above statement? as you mentioned Homoscedasticity's irrelevancy on dummy variable. it will help me to support my argument in my thesis. Thanks once Again.
Helpful vedio. However, I analyzed Linearity Test in SPSS-Got this answer Edit The ANOVA table shows the following text instead of values. No variance within groups - statistics for DV * IV cannot be computed. what its means?
It means that one of your variables has only the same values for one of the groups. For example, let's say you are doing ANOVA on gender, and one of your variables asked if you've ever been pregnant. For the male group, you would have all the same answer (NO), so there would be zero variance in that group.
Thank you for your continued videos. This is another good snippet. I have watched your series and they have been a huge help in getting a good grasp on the usability of factor analysis in combination with regression as well as SEM.
I would love to see more additions to the already great concentration of educational videos you have.
you actually made me smile at statistics because of your explanation... so long it has only been giving me headaches so thanks mate!
Stay blessed brother James Gaskin for your illustrative explanation that encouraged me to move a step towards completion of my Master degree.
Your tutorial helped me with my SPSS class work. Thank you for taking the time. Respect,
Thank you for sharing! I was confused on how to plot residuals because I need to test my data for homoscedasticity and you've shown that SPSS can do that. :D
omg! thank you so much for this. you are a life saver!!!
thank you, you saved my year, my life, thanks again
the "yay" at the end made me laugh as it's literally what I exclamed finding this video
So good, thank you so much for this explanation.
Good video, even like 4 years later. Well explained, and good method.
Thank you for this video!
Nice laymans guide. Cheers mate.
Thank you - this helps me a lot to understand more.
very good explanation ! it took me a while to even pronounce the word !
Thanks! Fun video!
Love u honestly thank u so much
Excellent Video
Super useful this, thanks! Now I just need to figure out how to make a 12x12 matrix of these.... Do a postgrad I said, it'll be fun I said.
Hi ! Does someone know how to do this when you have a binary independent variable ..? Can't find any way on SPSS
Thanks for the explanation, had a huge headache trying to understand this
Thanks, great video as always. Just wondering, is this to check homoscedasticity before or after using a statistical technique? For example, if checking homoscedasticity of the variate after regression, I read that you can plot the studentized residuals against the standardized predicted residuals (both values can be saved in SPSS and then plotted on a graph). Using this plot, you would expect equal dispersion about zero, i.e., no tendency to be greater than or less than zero.
SlyguyWhoeatspie I use it prior to running the causal model in order to demonstrate that we don't have heteroskedasticity.
Great. I always wanted to know how to do this with two metric variables prior to my causal model. Thanks!
Ty so much!
Hey James, I'm working on my dissertation and with an intent of being thorough I was wondering if (1) I should show the plots for each IV-DV combo (2) while there is no "coning" I do have my X-axis data almost in straight horizontal line increments ...any idea why that could be happening? (I am using means of IVs if that matters). Thanks!
+Val_C No need to include every plot. The columned plots are due to the nature of the scale. If you were using Likert scales, then the responses can only fit into the 5 or 7 discrete values available to respondents.
+James Gaskin That's what I figured but as you can imagine it looks so much weirder than something continuous looks like :) thanks so much for getting back to me so quickly btw. Now, I have a 2nd order related question which I'll go post on that video ;)
Thank you, James, for your wonderful work. I attended one of your 3 days boot camp at RMIT in 2017. Since then I have been following your channels. After going through some books, I am wondering is it necessary to check for homoscedasticity even after checking for normality and multicollinearity, for a PhD thesis?. If so, can we really do this homoscedasticity test on a 5-point Likert scale, whereas, authors like Tabachnick and Fidell (2019, p.73) only mentioned about continuous variables?. In my case, I have 1 IV, 1 mediator, 1 moderator, and 2 DVs. I am getting homoscedasticity for 1 DV and heteroskedasticity for the 2nd variable, when I use 1 IV and 1 DV, but when using 3 IVS together on each of the DVs separately, the results can be considered more as homoscedastic? What could be the soloution to this issue?. Many thanks.
I no longer assess homoscedasticity. There is little that can be done about it and its impact is minor.
@@Gaskination Thank you very much, James. I highly appreciate that for getting back to me very quickly. I shared your response with my supervisor too. I wish I could use your comment as a reference in the thesis. lolz. Stay blessed.
Hi Gaskin
Thanks for answering my other question on the other video.
My question is how do you test for linearity and homoscedasticity
in SPSS when one of the variables is a latent variable? In my case it
is the dependent variable. Many thanks.
+eric chen You would have to first create a factor score or composite variable somehow collapsing the latent variable into a single variable.
Hello! Thank you for the helpful video. How would I check for homoscedasticity in my variables if I'm going to run a 2x2 RM anova? There are no 2 separate variables for me to put in the independent and dependent boxes how you did here (if that makes sense) since my design is within-subjects. Is it even necessary for RM Anova? Thank you in advance for the help!
not necessary unless running regression-based analyses.
Okay looks easy enough! :) I am confused about one thing though, if I wanted to make many comparisons/ tests with one variable then would I have to do a homoscedasticity test for every comparison? If I want to know some other things like linearity then I would only do one test per variable (i.e. looking at linearity for variables A,B,C,D) But now for homoscedasticity I would have to do tests on the relations between A-B, A-C, A-D, B-C, B-D, C-D? Also, does it matter which of these variables is entered as the dependent in their respective relationships? Because I have correlational/ observational data, so I don't feel like it should matter.
Correct, you would need to do this for all pairs of variables involved in cause and effect relationships. You do not need to do it for variables only linked by correlation.
Hi Gaskin. A very informative guide on the homoscedasticity. I would like to get your guide on testing homoscedasticity on the variables in my study. I have 2 IV, 2 MEDIATOR, 1 DV, 2 MODERATOR and 2 CONTROL VARIABLES (AGE & EDUCATION). How do I run the homoscedasticity, linearity & normality? Between which variables?
Currently, I actually recommend to skip this step, as it is not very informative and there is little one can do about it anyway... The only thing I would recommend still testing (out of the things you listed) is normality (skewness and kurtosis).
quite easy :) thanks
Hi Gaskin
Your video really nice. just wondering how did you get variable "values" namely JOY not as indicator anymore? due to when you explain about normality test you still use every single items per variable? Thank you in advance
+sunu widianto` I'm not sure I understand the question. But I think you are asking how I calculated the JOY variable from all of its items. If so, then the answer is that I imputed factor scores in AMOS during my CFA. Here is a video: ruclips.net/video/dsOS9tQjxW8/видео.html
+James Gaskin Hi Gaskin yes you right I am asking about that but if I want to calculated the JOY variable from all of its items from SPSS how is the procedure? Is it okay it I calculate the items with mean to become JOY variable. Many thanks for your kind respond
+sunu widianto` The mean is fine, but factor scores are better.
pretty easy thank youu
Sir, thanx for the video. i have 2 questions (1) hair et al 2010 provides two methods for homoscedasticity first i sgraphical which you have shown here and the second is the levene test ..... kindly guide me how can i check the homoscedasticity of my complete SEM model ? and do you have any video for levene test?
Here is a video for a Levene's test: ruclips.net/video/E5VIKZU5kB8/видео.html
As for testing your whole model simultaneously for homoscedasticity, this does not make much sense. The levene's test allows you to do a homogeneity of variance test, which is roughly what you're asking, but for the items, not the model.
Respected Sir, relate to prior question where I found '' No variance within groups - statistics for DV * IV cannot be computed'' I have 5 variables and measured with 5 Likert scale (strongly disagree 1 to strongly agree 5) and tested for linearity i.e. 4 IVs and 1 DV. What might be the issue. ? although respondents answered as 1 to 5 to all of these variables. Will be thanked.
That error means that for one of your groups (assuming your data has been split or filtered), at least one of the variables has no variance - meaning that the variable has all the same responses. For example, let's say you're studying clubs, and you use gender as a predictor. If one of the clubs is a female only group, then for that group, you'll have zero variance for the gender variable (i.e., all will be female). You cannot have a predictor variable that doesn't vary.
Hi Gaskin,
Thank a lot for a clear explanation of Homoscedasticity.
I have a question, I am working with dummy independent variable (Categorical variable). when I run the Homoscedasticity test, the graph is confusing. I am unable to interpret it. Please guide me, how to test homoscedasticity for dummy variable and how to interpret the results.
Thank you.
Homoscedasticity is irrelevant to a dummy variable because the values in the dummy variable only represent the presence or absence of a trait, rather than a gradual increase of a trait.
Thank you very much. can i have your email address? I need your help in my analysis phase.
I respond to similar requests many times each day, so please keep your email to me very concise and direct. Also, please only ask for help if you get stuck and after you've searched the youtube channel and StatWiki for answers. Thanks! james.gaskin@byu.edu
Thank you very much for your assistance. I sent you an email regarding my issue.
Thank you for your help, can you please provide any reference for above statement? as you mentioned Homoscedasticity's irrelevancy on dummy variable. it will help me to support my argument in my thesis.
Thanks once Again.
Helpful vedio. However, I analyzed Linearity Test in SPSS-Got this answer Edit
The ANOVA table shows the following text instead of values.
No variance within groups - statistics for DV * IV cannot be computed.
what its means?
It means that one of your variables has only the same values for one of the groups. For example, let's say you are doing ANOVA on gender, and one of your variables asked if you've ever been pregnant. For the male group, you would have all the same answer (NO), so there would be zero variance in that group.
yay :)
you forgot to show how to do the plot
+ernest7 I think I must not understand your comment. The whole video is showing how to do the plot...
Thanks but yo didn't show the options in the related window how to ask spss to do the plot.
+ernest7 I think I must still not understand. In the video, I clearly show all the options I clicked on and checked.