I am currently working on my dissertation, after days of data analysis confusion, I've found your video and instantly understands everything. thank you so much!
ok im at minute 1.22 and so far this looks like the most promising help i found so far about how to analyze my ordinal cateogrical variable for my master thesis in spss.
Thank you so much, Sir. It really enlightened me about how to conduct OLR and MLR. This knowledge will be so helpful in my planned research which will take place soon.
Dear Mike, I have 9 independent variables - motivational factors ( convenience, product variety, purchase surrounding, information depth, and website design) and demographic factors (age, gender, income, education level). Each of the motivational factors have 3 items based on a likert scale from 1 to 5 (5 =strongly agree; 1 = strongly disagree). My dependent variable is ordinal based on likert scale also from 1 to 5 (5 always; 1 never) - frequency of online shopping (how often do you shop online). Firstly, I wanted to find out the most meaningful factors. First, I did factor analysis and frequency distribution. Afterwards, I want to ordinal regression where shopping frequency is the DV and the IV are the 5 motivational factors. The items for these factors have been computed. So, from the 15 items from strongly agree to strongly disagree, i have now 5 variables with their mean. I computed every 3 item with their corresponding factor. I want to find out with ordinal regression which factor are related to online shopping (either positively and negatively as well as significantly or insignificantly). I did this. Now, I want to find out if the demographic factors have a relationship with shopping frequency. How do I do this? And afterwads, I want to find out also if the demographic factors have a relationship with the motivational factors. How do I do this? Please respond asap!
This helped me more than anything else, so thank you for that! Just curious, whether it matters to put ordinal/nominal variables in as factors or covariates- when should they be treated as factors?
Great explanation. Please answer the following questions. Should both route one and route two analysis be performed when reporting the results of data analysis or you just select one analysis say route one and model your equation and report its results? How do you specify the model in an equation? Where do you get the values to model the equation is it from route one results or route two results? Whichever route is selected, which values do you use to model the equation?
Hi Samson, the models in either route are equivalent. It's just that the presentation format differs - and it looks like there are things you can get in one route that you may not be able to get via the other route. A lot of times folks want information that is uniquely presented by way of one route when they are writing things up and that's why I illustrated both. For example, you'll notice that you can get McFadden's pseudo R-squared (which is commonly reported) only via one route. The same goes for odds ratios. So, really the information you desire to include in your write up is up to you. Generally, you want to report on global model fit (which generally will include the likelihood ratio chi-square test and pseudo-R-square, and/or add in Pearson's or deviance chi-square) and the regression coefficients and tests, and odds ratios. At any rate, I literally just finished up with a new presentation on ordinal logistic regression and hope you check it out at: ruclips.net/video/CdOHB3U5YHk/видео.html Cheers!
Thank you very much! This video is so useful, especially route 2 you explained. Just a question, If Goodness-of-Fit are not significant but the test of parallel lines is significant. Is it OK to use ordinal logistic regression or Multinominal logistic regression should be used?
Thanks for your explanation. So for the dependent variable box, it's only possible to put one dependent variable? In my case i have 2 dependent variables, so should i analyze each of them seperately with the studied dependent variables. Thanks.
My Analysis is to Test the effect of One Independent Variable on an Ordinal Dependent Variable... So what would I use to measure/interprete "effect" I'm not after the model
How can we do ordinal regression where the dependent variable and independent variables have large number of items whose observation is collected in likert scale? can we use the transformed mean for analysis of ordinal regression?
Thankf for your video!! I keep getting the warning at the start of the output saying there are X % dependent variable levels by observed combinations of predictor variable values) with zero frequencies.. does this not matter? I saw you had it come up too
Thank you for the explanation. I will like to know if OLR is a correct way to analyze ordinal variables when we want to follow this vairable over the time. (dependent variable: ordinal; independente variable: time) by treatment groups.
Thanks for the explanation. Please, i would like to know what should i do when the Pearson and deviance results are significant (under 0.05) in the Goodness-of -fit table? Is it an indicator that the model do not fit the data, and i should not do regression analysis? Also, If the test of parallel lines is significant (under 0.05), can we still use the results of the regression? Thanks, your reply will help a lot!
thank you for nice presentation and i have one question about response codding of the dependent variables, should start with zero? should it be always in an ascending order? thank you.
How come you put the 'pass' variable as a covariate even though it is a categorical variable coded as 0 and 1 and should have been put in the 'factors' box?
Hi Niklas, thanks for visiting. No - unfortunately I do not have any papers that you could cite directly. However, I suppose you could cite the slide presentation (download underneath the video description). best wishes!
Hello Mike, you have used (1=low interest, 2=medium interest, 3=high interest) for the threshold level of interest. but parameter estimates, I mean in final calculation it is showing only (1=low interest, 2=medium interest) two of them. can you please clarify on that.
Dear Dr. thank you very much for your great presentation about ordinal logistic regression. But I do have to questions that needs more clarification . 1. You didn't show us how to test the assumptions of ordinal logistic regression and what does they mean 2. Interpretations of the final result (relationship of the independent and independent variables)
Hi there. Did you download the Powerpoint under the video? Much of this is presented there (drive.google.com/file/d/19TriKKQ8tzwrhGoYr_LxmpfpAjSTW3Ov/view) . Best wishes!
Thank you for amazing work. I have a question regarding Goodness of Fit table. When Pearson and Deviance test disagree, which one we should report? How to actually report that table? Thank you so much in advance.
How would the standardised coefficient be presented in a paper for data analysed by ordinal regression? Let's say that it's a requirement for the paper to present standardised coefficient. I tried running generalised linear model and ordinal regression but can't seem to find standardised coefficient. I found the value on linear regression tho. Looking for any pointers! :)
can u explain the interpretation of the multinomial logistic regression shown at the end of the video? because the dependent variable was coded as 2.00 and 3.00. how would we interpret the data?
Ho Jo. Thanks for visiting. Unfortunately, I don't really do anything with SAS. But I have videos for Stata, R, and jamovi in addition to SPSS. best wishes!
it depends how you want the results to be interpreted. if you interpret in terms of Beta coefficient than use beta values and report the sig values otherwise if you want to report in terms of OR you can report Exp (B) values. Route II is better bacause it gives you both the options
Thanks so much for this! I'm working on dissertation data analyses and this walk through was helpful. One question that I had remaining is where/how do we enter variables that we would like to control for? Is there a way to enter variables in blocks?
Hi Shelly, technically anytime you add any set of predictors in your model the predictors are controlling for the presence of others. So mathematically any control variables you enter will still be partialled out of the relationship between any focal predictors and the DV. Regarding your question about blocks: Predictors are generally entered in blocks so that one can test for the increment in fit/predictive power as a result of adding predictors in different steps. In the context of linear regression, we may enter control variables early on and then add in our substantive predictors to test the increment in R-square as a result of adding in our focal predictors (the idea being that a significant increment in explained variation means that the focal variables are adding explanatory power above and beyond the control variables). That is the general logic beyond the strategy of adding variables hierarchically in this fashion. If you are performing binary logistic regression in SPSS, there is a mechanism to do the same thing as described above using 'Blocks'. Unfortunately SPSS does not give you the option for adding variables into blocks when performing ordinal logistic regression. However, if you want to use a hierarchical strategy (as described above) it is not terribly complicated. Just perform a series of logistic regressions, where you are adding in predictors across a set of models (like you would in hierarchical multiple regression). In effect, each regression would be treated as a Block. You can then use a chi-square difference test to test the difference in fit with each addition of predictors across the models. As an example, let's consider a scenario where you are running an ordinal logistic regression with three predictors (x1, x2, x3), where x1 is your control variable. Step 1: Perform the LR with x1 included as the sole predictor. Obtain the chi-square value and degrees of freedom for this model (printed out in the Model Fitting Information Table - Final model). The df=1 in this model since there's only 1 predictor. Step 2: Re-run the analysis after adding in X2 and X3 and obtain the same output. (The predictors in the model are x1, x2, x3 & the df will be 3 since there are three predictors. Step 3: Subtract the model 2 chi-square from the Model 1 chi-square to obtain a chi-square difference value. Subtract the df from model 1 from model 2 (in this example the df diff = 3-1 = 2. Step 4: Go into any chi-square table assuming alpha = .05 and df to obtain the chi-square critical value. If the chi-square difference from Step 3 exceeds the tabled chi-square value (from this step), then you have a significant improvement in fit. That's all there is to it. Hope this helps!
Hi Mukhtar, if you have a categorical (nominal or ordinal) independent variable with more than three categories, then you must treat it as a factor and move it to that box. The program will dummy code the factor for you. If you want to do the dummy coding yourself, then you can enter any dummy variables under the Covariates box. Because "pass" and "gender identification" both only have two levels, you can include them as Covariate or Factor. I prefer to enter them as Covariates because I like having more control over which group gets treated as a reference category. For more information on dummy coding: ruclips.net/video/XGlbGaOsV9U/видео.html
@@mikecrowson2462 Hey Mike, thank you so much for your fast reply; now everything works fine - I was able to download the powerpoint and dataset. THANK YOU!
Hi, thank you for the explanation! I have some doubts about it! What's more important? Model significance or independent variables significance? Thank you!
Hi Irene. Thanks for your question. When you say model significance, I'm assuming you are referring to the chi-square test? This test is fundamentally a test where the null hypothesis is that 'all regression slopes are 0' and the alternative hypothesis is 'not all regression slopes are equal'. If you reject the null hypothesis at the model level, then you generally proceed to interpreting and testing the individual predictors in the model. Hope this helps! Cheers!
@@mikecrowson2462 thank you so much! I explained it wrong, sorry. What's more important? PseudoR higher or parameter estimates more significant? So sorry, these things are new for me and I'm trying yo analyse my Medicine thesis results :)
@@ireneblanco3191 Well, you have to look at the whole picture really. The pseudo R-square is really just an analogy to the R-square we are accustomed to with OLS regression to help you provide an overall picture of model fit. But that's one lens for describing the overall fit. That's why you also want to report on the likelihood ratio chi-square under Model Fitting Information and perhaps the Pearson chi-square and Deviance chi-square under Goodness of fit tests. Also, keep in mind that not all fit measures necessarily agree. Regarding pseudo-r-squares, I prefer McFadden's which can be interpreted as the proportion of improvement in fit of your model relative to a no predictors model (see Pituch & Stevens, 2016). Once you've decided the overall model exhibits reasonable fit to the data, then you examine each of the regression coefficients to identify which predictors are likely contributing to prediction of category membership on the dependent variable. Cheers!
If the proportional odds assumption is violated, then an alternate approach is to use multinomial logistic regression. (I briefly refer to that in this powerpoint; slide 2: drive.google.com/file/d/1nKkBGrM90yRbo7ErvWe8ZxT_qiQAdgMo/view). To see it in action you can watch the accompanying video: ruclips.net/video/6S_878RheL8/видео.html
I wish all professors were eager to share their knowledge so willingly. Thank you so much for your work, this has been extremely useful!
I am currently working on my dissertation, after days of data analysis confusion, I've found your video and instantly understands everything. thank you so much!
Hi Ed, thanks for your comment! I'm glad this was helpful to you! best wishes
ok im at minute 1.22 and so far this looks like the most promising help i found so far about how to analyze my ordinal cateogrical variable for my master thesis in spss.
Best video I have found on this! Thank you so much!
Thank you Sir,
The way you explain and interpret is more much appreciable and admirable. Thank you for helping us in understanding in a deep way.
Hi Syed, thank you for your feedback! I'm glad you found the video helpful. best wishes!
I could say I learn most advanced stats knowledge from your channel. Thanks a million
I am working on my dissertation and have been struggling a bit, but this video is SOO helpful! THANK YOU.
Well conveyed, clearly communicated content. Profuse thanks for placing your material in the public domain. Definitely helpful!!
Hi Andy. You are very welcome! Thanks for visiting!
this is definitely the best instruction in this topic! Thanks so much!
Thank you very much for visiting and for your kind comment, Nguyen. Best wishes to you!
Thank you very much for your time and efforts, this greatly helped me to dissolve my doubts and misunderstandings on this topic!
thank you very much!! Congratulation from Brazil!!!
You are very welcome! Best wishes!
Mike, thank you so much for the clear explanation and for sharing as well the presentation with us!
Thank you so much, Sir. It really enlightened me about how to conduct OLR and MLR. This knowledge will be so helpful in my planned research which will take place soon.
I'm glad you found it helpful! Best wishes!
This is pure gold! Thank you very much!
Thank you very much, I am working in my dissertation and your video has been very very useful. Thank you for providing the ppt as well
Thanks a lot for teaching it in a simple way. Great job!
Thank you for sharing your knowledge! Easy to understand! Love it!
Dear Mike,
I have 9 independent variables - motivational factors ( convenience, product variety, purchase surrounding, information depth, and website design) and demographic factors (age, gender, income, education level). Each of the motivational factors have 3 items based on a likert scale from 1 to 5 (5 =strongly agree; 1 = strongly disagree).
My dependent variable is ordinal based on likert scale also from 1 to 5 (5 always; 1 never) - frequency of online shopping (how often do you shop online).
Firstly, I wanted to find out the most meaningful factors. First, I did factor analysis and frequency distribution.
Afterwards, I want to ordinal regression where shopping frequency is the DV and the IV are the 5 motivational factors. The items for these factors have been computed. So, from the 15 items from strongly agree to strongly disagree, i have now 5 variables with their mean. I computed every 3 item with their corresponding factor.
I want to find out with ordinal regression which factor are related to online shopping (either positively and negatively as well as significantly or insignificantly). I did this.
Now, I want to find out if the demographic factors have a relationship with shopping frequency. How do I do this?
And afterwads, I want to find out also if the demographic factors have a relationship with the motivational factors. How do I do this?
Please respond asap!
I appreciated your expression style. Salute your presentation.
Great video on OLR, thank you very much!
Hi everyone, please be check out my newest video (August 2021) on ordinal logistic regression here: ruclips.net/video/CdOHB3U5YHk/видео.html
Excellent presentation and thank you for the additional materials.
THANK YOU!!!! So helpful
Thank you Mike, You been had help me so much.
You're very welcome!
This helped me more than anything else, so thank you for that! Just curious, whether it matters to put ordinal/nominal variables in as factors or covariates- when should they be treated as factors?
Thank you for the video. very clear and useful for my study
I found this very useful. Thank you.
oh is it allowed to put any ordinal variable in covariate box in ordinal logistic regression?
Great explanation. Please answer the following questions. Should both route one and route two analysis be performed when reporting the results of data analysis or you just select one analysis say route one and model your equation and report its results? How do you specify the model in an equation? Where do you get the values to model the equation is it from route one results or route two results? Whichever route is selected, which values do you use to model the equation?
Hi Samson, the models in either route are equivalent. It's just that the presentation format differs - and it looks like there are things you can get in one route that you may not be able to get via the other route. A lot of times folks want information that is uniquely presented by way of one route when they are writing things up and that's why I illustrated both. For example, you'll notice that you can get McFadden's pseudo R-squared (which is commonly reported) only via one route. The same goes for odds ratios. So, really the information you desire to include in your write up is up to you. Generally, you want to report on global model fit (which generally will include the likelihood ratio chi-square test and pseudo-R-square, and/or add in Pearson's or deviance chi-square) and the regression coefficients and tests, and odds ratios. At any rate, I literally just finished up with a new presentation on ordinal logistic regression and hope you check it out at: ruclips.net/video/CdOHB3U5YHk/видео.html
Cheers!
Thank you. Gracias. 🙋♂
Thank you for the excellent presentation!
How do you set up the equation to model your results? What should be written in the constant value?
Thank you very much! This video is so useful, especially route 2 you explained. Just a question, If Goodness-of-Fit are not significant but the test of parallel lines is significant. Is it OK to use ordinal logistic regression or Multinominal logistic regression should be used?
Thanks for your explanation. So for the dependent variable box, it's only possible to put one dependent variable? In my case i have 2 dependent variables, so should i analyze each of them seperately with the studied dependent variables. Thanks.
THANKS A MILLION
Great video
Interesante vídeo, muchas gracias por su aporte !!!
My Analysis is to Test the effect of One Independent Variable on an Ordinal Dependent Variable... So what would I use to measure/interprete "effect"
I'm not after the model
very informative
How can we do ordinal regression where the dependent variable and independent variables have large number of items whose observation is collected in likert scale? can we use the transformed mean for analysis of ordinal regression?
Thankf for your video!!
I keep getting the warning at the start of the output saying there are X % dependent variable levels by observed combinations of predictor variable values) with zero frequencies.. does this not matter? I saw you had it come up too
Thank you for the explanation. I will like to know if OLR is a correct way to analyze ordinal variables when we want to follow this vairable over the time. (dependent variable: ordinal; independente variable: time) by treatment groups.
This was very helpful. I do have a question. Is it okay to analyze ordinal values which was encoded in a descending order?
Thanks for the explanation. Please, i would like to know what should i do when the Pearson and deviance results are significant (under 0.05) in the Goodness-of -fit table? Is it an indicator that the model do not fit the data, and i should not do regression analysis? Also, If the test of parallel lines is significant (under 0.05), can we still use the results of the regression? Thanks, your reply will help a lot!
Thank you very much, sir!! 😃😃😃
thank you for nice presentation and i have one question about response codding of the dependent variables, should start with zero? should it be always in an ascending order? thank you.
How come you put the 'pass' variable as a covariate even though it is a categorical variable coded as 0 and 1 and should have been put in the 'factors' box?
Is there a way to account for covariates in an ordinal logistic regression? I am trying to run a hierarchical ordinal logistic regression.
Thank you, Mike! Very clear explanation. Do you maybe have a book/paper or anything of that sort that I could cite and mention in my thesis?
Hi Niklas, thanks for visiting. No - unfortunately I do not have any papers that you could cite directly. However, I suppose you could cite the slide presentation (download underneath the video description). best wishes!
Hello Mike, you have used (1=low interest, 2=medium interest, 3=high interest) for the threshold level of interest. but parameter estimates, I mean in final calculation it is showing only (1=low interest, 2=medium interest) two of them. can you please clarify on that.
In my analysis the test of parallel lines had as a result Sig. ,000 (statistically significant), how should I proceed with my analysis in this case?
Thanks!
Dear Dr. thank you very much for your great presentation about ordinal logistic regression. But I do have to questions that needs more clarification .
1. You didn't show us how to test the assumptions of ordinal logistic regression and what does they mean
2. Interpretations of the final result (relationship of the independent and independent variables)
Hi there. Did you download the Powerpoint under the video? Much of this is presented there (drive.google.com/file/d/19TriKKQ8tzwrhGoYr_LxmpfpAjSTW3Ov/view) . Best wishes!
Thank you very much.
Thanks Mike!
You're welcome, Yulin!
Thank you for amazing work. I have a question regarding Goodness of Fit table. When Pearson and Deviance test disagree, which one we should report? How to actually report that table? Thank you so much in advance.
How would the standardised coefficient be presented in a paper for data analysed by ordinal regression? Let's say that it's a requirement for the paper to present standardised coefficient. I tried running generalised linear model and ordinal regression but can't seem to find standardised coefficient. I found the value on linear regression tho. Looking for any pointers! :)
can you use this for when the dependent variable is measured with a within-participants design?
Is an ordered logistic regression the same as an ordinal logistic regression? I'm quite confused with the terminology!
can u explain the interpretation of the multinomial logistic regression shown at the end of the video? because the dependent variable was coded as 2.00 and 3.00. how would we interpret the data?
so many thanks!
very helpful
Love your videos, will you do any SAS videos?
Ho Jo. Thanks for visiting. Unfortunately, I don't really do anything with SAS. But I have videos for Stata, R, and jamovi in addition to SPSS. best wishes!
@@mikecrowson2462 I am picking up SAS and I still watch your videos they help me understand the output even in SAS. Thank you much
very good, thanks
on your data samples. can I put pass and gender as factors? considering they are nominal variable
Hi. I have a question. I got a warning that XXX of my dependent variable cells have zero frequencies. How do I resolve that? Thanks!
can we find relation between independent variable( gender having two categories, with follow-up compliance- that is multi and ordered) ?
how can I understand which one is my reference group while doing ordinal regression?
Hi ! You explained very well.I have some confusion regardind route 1 and route 2 which values i take for under parameter estimates.Thanks
it depends how you want the results to be interpreted. if you interpret in terms of Beta coefficient than use beta values and report the sig values otherwise if you want to report in terms of OR you can report Exp (B) values. Route II is better bacause it gives you both the options
there anyone know the equation of Odds Ratio from Ordinal Logistic Regression
Thank you!!
Sir ,what if model fitt p>0.05 what is next?
Thanks so much for this! I'm working on dissertation data analyses and this walk through was helpful. One question that I had remaining is where/how do we enter variables that we would like to control for? Is there a way to enter variables in blocks?
Hi Shelly, technically anytime you add any set of predictors in your model the predictors are controlling for the presence of others. So mathematically any control variables you enter will still be partialled out of the relationship between any focal predictors and the DV.
Regarding your question about blocks: Predictors are generally entered in blocks so that one can test for the increment in fit/predictive power as a result of adding predictors in different steps. In the context of linear regression, we may enter control variables early on and then add in our substantive predictors to test the increment in R-square as a result of adding in our focal predictors (the idea being that a significant increment in explained variation means that the focal variables are adding explanatory power above and beyond the control variables). That is the general logic beyond the strategy of adding variables hierarchically in this fashion.
If you are performing binary logistic regression in SPSS, there is a mechanism to do the same thing as described above using 'Blocks'. Unfortunately SPSS does not give you the option for adding variables into blocks when performing ordinal logistic regression.
However, if you want to use a hierarchical strategy (as described above) it is not terribly complicated. Just perform a series of logistic regressions, where you are adding in predictors across a set of models (like you would in hierarchical multiple regression). In effect, each regression would be treated as a Block. You can then use a chi-square difference test to test the difference in fit with each addition of predictors across the models.
As an example, let's consider a scenario where you are running an ordinal logistic regression with three predictors (x1, x2, x3), where x1 is your control variable. Step 1: Perform the LR with x1 included as the sole predictor. Obtain the chi-square value and degrees of freedom for this model (printed out in the Model Fitting Information Table - Final model). The df=1 in this model since there's only 1 predictor. Step 2: Re-run the analysis after adding in X2 and X3 and obtain the same output. (The predictors in the model are x1, x2, x3 & the df will be 3 since there are three predictors.
Step 3: Subtract the model 2 chi-square from the Model 1 chi-square to obtain a chi-square difference value. Subtract the df from model 1 from model 2 (in this example the df diff = 3-1 = 2. Step 4: Go into any chi-square table assuming alpha = .05 and df to obtain the chi-square critical value. If the chi-square difference from Step 3 exceeds the tabled chi-square value (from this step), then you have a significant improvement in fit. That's all there is to it. Hope this helps!
@@mikecrowson2462 Thanks for your quick response!
@@mikecrowson2462 Thanks for this breakdown!
very nice, one thing i want to ask, how do we classify to move the independent variables into the factors list instead of the co-variates
Hi Mukhtar, if you have a categorical (nominal or ordinal) independent variable with more than three categories, then you must treat it as a factor and move it to that box. The program will dummy code the factor for you. If you want to do the dummy coding yourself, then you can enter any dummy variables under the Covariates box. Because "pass" and "gender identification" both only have two levels, you can include them as Covariate or Factor. I prefer to enter them as Covariates because I like having more control over which group gets treated as a reference category. For more information on dummy coding: ruclips.net/video/XGlbGaOsV9U/видео.html
Hello Mike,
unfortunately it is not possible to download the powerpoint slides :/. Is there a way to get the powerpoint anyways?
Hi Manuel, I just fixed the links. Let me know if there are any more problems. Best wishes!
@@mikecrowson2462 Hey Mike, thank you so much for your fast reply; now everything works fine - I was able to download the powerpoint and dataset. THANK YOU!
@@manuelleitner1996 You are very welcome! Cheers!
why the threshold just 2?
can i use OLR without removing outliers?
Hi. can you do jamovi on this? thanks
Hi, thank you for the explanation! I have some doubts about it! What's more important? Model significance or independent variables significance? Thank you!
Hi Irene. Thanks for your question. When you say model significance, I'm assuming you are referring to the chi-square test? This test is fundamentally a test where the null hypothesis is that 'all regression slopes are 0' and the alternative hypothesis is 'not all regression slopes are equal'. If you reject the null hypothesis at the model level, then you generally proceed to interpreting and testing the individual predictors in the model. Hope this helps! Cheers!
@@mikecrowson2462 thank you so much! I explained it wrong, sorry. What's more important? PseudoR higher or parameter estimates more significant? So sorry, these things are new for me and I'm trying yo analyse my Medicine thesis results :)
@@ireneblanco3191 Well, you have to look at the whole picture really. The pseudo R-square is really just an analogy to the R-square we are accustomed to with OLS regression to help you provide an overall picture of model fit. But that's one lens for describing the overall fit. That's why you also want to report on the likelihood ratio chi-square under Model Fitting Information and perhaps the Pearson chi-square and Deviance chi-square under Goodness of fit tests. Also, keep in mind that not all fit measures necessarily agree. Regarding pseudo-r-squares, I prefer McFadden's which can be interpreted as the proportion of improvement in fit of your model relative to a no predictors model (see Pituch & Stevens, 2016). Once you've decided the overall model exhibits reasonable fit to the data, then you examine each of the regression coefficients to identify which predictors are likely contributing to prediction of category membership on the dependent variable. Cheers!
@@mikecrowson2462 thank you very much, so kind of you! ♥️ Cheers!
What about the results for "test of parallel lines?" You asked for that result but never discussed it. What if it is violated?
If the proportional odds assumption is violated, then an alternate approach is to use multinomial logistic regression. (I briefly refer to that in this powerpoint; slide 2: drive.google.com/file/d/1nKkBGrM90yRbo7ErvWe8ZxT_qiQAdgMo/view). To see it in action you can watch the accompanying video: ruclips.net/video/6S_878RheL8/видео.html
@@mikecrowson2462 Very good. Thanks for the help and the links.
It is good explanation. yet the appropriate design is not lear
Thank you very much, I am working in my dissertation and your video has been very very useful. Thank you for providing the ppt as well