Multilevel binary logistic regression using Stata drop-down menus March 2020)

[@kevinheupel]

Thank you for making this difficult subject very understandable and useful. Much appreciated.

[@rapmantheoneman]

This was amazing, Mike! Thank you! I will definitley download the PowerPoint. Quick question-Do you also plan to post a svy data example?

[@VictorAgboga]

Thanks. How to we go about multilevel multinomial logistic regression?

[@javiertrujillo-y2y]

Excellent videos, Mike, congratulations. Is it possible to use multilevel binary logistic regression with a competing risk analyses? How can I do it? Thank you

[@rhencylegaspi7120]

Hi! I'm using survey data (stratified 2-stage sampling design) and I'm wondering if it's possible to add a varying intercept for the strata (cities)? The clusters/level-2 indicators are a group of households nested under a stratum. I'm getting an error of "hierarchical groups are not nested within clusters," is it theoretically wrong to add varying coefficients for variables that are above level-2 indicators?

[@seanng5018]

Hi Mike, Many thanks for taking the time to film this video. I have a question I'm hoping you can help with me with. I ran an intercept only model got an acceptable ICC. However, I am getting a very low ICC (e-30) after including the first level predictors to the model. I'm not sure how this result should be interpreted. Does it mean that I will can (or should) no longer include the slopes? Thanks.

[@tarikutesfaye447]

Thank you so much, @Mike Crowson for the tutorial! It's informative, simple, and understandable.

[@manasjyotiborah435]

Hi Mike, could you please highlight on how to incorporate time /longitudinal aspect in the above analysis.

[@lizicejasmine]

Thank you for the video! Do the LR test results have to be significant for the models? If it is not significant at model 2 or 3, does it mean we should not use the multilevel logistic regression? Thank you so much again!

[@hushjay3134]

Excuse me professor can i ask what does gpa center within classroom mean?

[@sandeepsivakumar4959]

Hi Mike, i have 2 level model, with a categorical response variable. One of my major predictor variable is also categorical. How to go about it in this set-up. Any suggestions?. Also one of I have an interaction term, which is a categorical predictor.

[@lammie10]

Thank you for many easy-to-follow tutorials! Would you please also demonstrate how you would generate plots in Stata? Where the y-axis would be the probability of outcome, and x-axis would be one of the IVs (among many IVs)?

[@andreab2114]

@Mike Crowson Thank you for this demonstration! What about multilevel multinomial? I tried to implement it using gsem but it is too complicated for me!

[@sophiyagarasia9450]

Hi Mike! This video was super helpful. Thank you. How would you approach it if there were three levels? I know you have a video on it with SPSS. Hopefully, you can do one on Stata too.

[@mikecrowson2462]

Hi there Sophiya, I'll certainly put it on my list of to-do's. :)

[@jingxu1388]

Sorry to bother you sir. My second question is that if I have 3 level one variables(x1 x2 x3) and 2 level two variables(z1 z2), but I am only interested in the influence of z2. Should I do a single z2 HLM first to see whether z2 is significant? eg: “y z2 || classes: ”?

[@mikecrowson2462]

If you are wanting to control for the other predictors in the model, then you will need to include them at some point. I suppose technically you could do one model with z2 and then add in the other variables as controls in a second model. But it does seem to be a somewhat unconventional approach to model building. I guess as long as the final model includes the controls, all would be good. Hope this helps.

[@jingxu1388]

Mike Crowson thanks a lot! And look forward to your videos. Actually I have been a big fan of your channel!

[@nguyenthilanh4513]

Dear Mike Crowson, thank you so much for your valuable video and guideline, much appreciated. I have one question: How can I calculate VIF when I run "multinomial regression?". Hope to receive your reply soon. Stay safe and healthy!

[@mikecrowson2462]

Hi Nguyen, thanks for visiting and for your question. When you say, "multinomial regression", I'm assuming you are referring to a single-level model, right? VIF pertains to the inflation of the variance of your regression parameters that occurs as a result of collinearity among your predictors. It is computed for each predictor as 1/(1-R-square), where the predictor of interest is regressed onto the remaining predictors. Because VIF is based only on the predictors, a logical way of obtaining the VIF's (that doesn't entail you having to manually regress each predictor onto the remainder and calculating the VIF) is to generate the run the regression through OLS (yes, I know this doesn't make sense given the DV, but we are not concerned with the DV, but rather the IV's) and then to type the vif command in, which will produce the VIF's and 1/VIF = tolerance. This is essentially tricking the program into giving you the VIF's. OR, you could use the 'collin' command associated with the collin package that you can download at: stats.idre.ucla.edu/stat/stata/ado/analysis/ . The package is described by Ender at www.philender.com/courses/categorical/notes2/collin.html . Either approach will give you the same results. Hope this helps!

[@nguyenthilanh4513]

Dear Mike Crowson, how happy I am. I really appreciated your kindness and guidelines sooner than I expected. Yes, I got it and deal with a well-done VIF issue by "collin" command. Stay healthy and safe from Germany!

[@lilacpeaches]

Hello Sir 🙂. I want to ask. Does multilevel binary logistics require a minimum number of sample/group at each level?

[@mikecrowson2462]

Hi Johan, I haven't spent a whole lot of time on this question in the context of multilevel binary logistic regression. However, I can point you to a few articles that might be helpful to you to inform your thinking on the topic: journals.plos.org/plosone/article?id=10.1371/journal.pone.0225427 www.rips-irsp.com/articles/10.5334/irsp.90/ www.joophox.net/publist/methodology05.pdf Cheers!

[@lilacpeaches]

@@mikecrowson2462 thank you sir

[@fatemehhadavand7505]

Hi Mike Crowson, thank you for your helpful tutorials. When can we use logistics instead of Multi-level? and You may explain more about variance-covariance structure of the random effects and it's option's,

[@jingxu1388]

Thanks again for your sharing. I just have one question, what is the difference between “melogit bieber teacher_fan gpa_cmc || classes:” and “melogit bieber gpa_cmc || classes: teacher_fan”? And when shall we use the former one? When to use the latter one? Many thanks！

[@mikecrowson2462]

melogit bieber teacher_fan gpa_cmc||classes: gpa_cmc specifies a model with random intercepts and random slopes , whereas melogit bieber teacher_fan gpa_cmc||classes: specifies a random intercept only model. I put together a video covering the syntax here (be sure to check it out!): kzbin.info/www/bejne/eIeVl4t_a62dl9k The choice between them depends in part on your assumptions regarding whether the relationship between the level 1 predictor and outcome variable varies across level 2 units. Often, this is done as a preliminary step or as part of modeling a cross-level interaction.

[@jingxu1388]

thanks for your detailed answer! Your videos are always quite helpful for quantitative analysis! Look forward to your future fantastic tutorials. I am a big fan of your channel !

[@jingxu1388]

thanks a lot for your sharing! It is really helpful ！

[@mikecrowson2462]

You are very welcome!

[@090127]

Thank you so much for sharing. It is helpful for me.

[@forextradingwithimran]

Hello Sir, I am an undergraduate student. My function is Innovation=f(Experience of Manager, Gender of Manager, Skilworkers, Technology license, Firm Age, Firm Size, Ownership). where Innovation is a dummy. The experience of the manager is quantitative. All others are binary variables. My dependent variable is Innovation. What will be my Level variable?. Plz help.

[@tawilk]

can you have more than 2 levels to nest? such as you have classes, but then school, or school district?

[@mikecrowson2462]

Yes, you certainly can. I don't have a video on that, but it is definitely possible. Cheers!

[@olly1192]

Great video! Thanks a lot!

[@albertbeaulne9114]

I can not find the power point,

[@mikecrowson2462]

Hi there. It is underneath the video description. But here it is again: drive.google.com/file/d/1Gr_XJozZS4-6iYswfmT3bAerzoK9EYTw/view . Cheers!

[@MAX-ho6wg]

Good job. could you please help me with codes or do file

[@mikecrowson2462]

Hi Max, I put together a video on the syntax/coding for the demos in this presentation. Go here to view: kzbin.info/www/bejne/eIeVl4t_a62dl9k . Best wishes.