Multilevel binary logistic regression example in SPSS
HTML-код
- Опубликовано: 21 мар 2018
- This video is intended to be a broad demonstration of some of the SPSS functions available for carrying out multilevel binary logistic regression using Generalized Mixed Models in SPSS. I provide a review of single level binary logistic regression, and then demonstrate how to carry out the analyses taking into account the multilevel nature of the data. You can obtain a copy of the data and follow along with the presentation by going to the following web address: drive.google.com/open?id=1irH...
NEWER VIDEO ON MULTILEVEL BINARY LOGISTIC REGRESSION (2020):
• Multilevel binary logi...
There, you can download a Powerpoint and accompanying SPSS data file!
For more instructional videos and other materials on various statistics topics, be sure to my webpages at the links below:
Introductory statistics:
sites.google.com/view/statist...
Multivariate statistics:
sites.google.com/view/statist...
this is what i'm looking for
I really really love your detailed explanations on the results! BIG thanks!
Thanks for visiting Bingxin! I'm glad this was helpful to you! best wishes.
This is very helpful. Thanks a lot.
Thank you for helping me with my first meta-analysis!
I'm glad you found it helpful, Samuel! Cheers!
Thank you very very much, you're a real lifesaver!
You are very welcome! Thanks for visiting!
Super helpful, thank you!
Thank you dear i really benefited from your presentations
Intuitive vedio! thank you.
Thank you so much ! This was very useful
Glad it was helpful. Thanks for watching!
Hello everyone, I have a newer (better quality, 2020) video on Multilevel binary logistic regression that can be found at: ruclips.net/video/roWTULimNPk/видео.html
There is an accompanying Powerpoint and SPSS data file underneath the video description. I hope you check it out!
Great lecture
Hi, thanks for the Video! It helped me alot. Do you know, if SPSS is able to offer Fit-Indizes like AIC, BIC, df, CHi2 and LL like Stata can?
Thank you very much for this!
I'm a little confused by the last few minutes when you talk about the random effect variable.
At 29:46 you say "That's the effect of adding in the treatment predictor at level two". Since the estimate became non-significant (from .000 to .117), does that mean that the model fits better or worse when the random effect is included?
Thanks again!
Very useful materials Mr Crowson!! Do you have videos on multilevel modeling of ordinal outcome variable with more than 2 categories?
Thanks, Sandile. I haven't developed a video with MLM with ordinal outcomes yet. I'll probably be working on something like that in the future. Stay tune :)
Wondering how you interpret the increasing value for model fit across the null model and the model with fixed effects - is a conclusion in this case to reject the more complex model despite what appears to be a theoretically defensible rationale/interest?
Dear Dr. Crowson! I'd like to compare the "strength" of the effect of the independent variables. What is a good value to compare? I suppose not the OR. Thank you! Adam
Hi Mike, thanks for your videos. If each patient in my study has multiple episodes of infection, and my outcome is sepsis/no sepsis, any ideas how I would present that data in this model? Would the 'patient' be equivalent to 'Group' in this case, and the episode number equivalent to 'subside.wg'? I want to do multilevel binomial logistic regression without treating each episode number as exactly the same, and accounting for random effect between these episodes. Every patient had an episode at random points during their stay. Thanks!
Hi, I am wondering how you would calculate the ICC for a multilevel analysis with a dichotomous outcome?
hi Mike,
looking at the first approach (binary logistic regression) it gives us 78,6% accuracy on the prediction and below we see the variables in the equation. How can I use these outcomes to formulate an equation for a new sample without any data of "happy.dvar" in order to find out into which "happy.dvar" they would fall?
What is the equation used by SPSS?
Hi, I had SPSS 22 and upgraded to 24. I am unable to do logistic regression . Not as straight forward as SPSS 22. Please help.
Thanks for the video! Around 24:55, you mention you won't go about allowing the slopes to randomly vary, but how would you go about doing so?
Hi Chris, I have a new video on this topic ( ruclips.net/video/roWTULimNPk/видео.html ) that also covers random slopes. Additionally, underneath the video description, you can download a Powerpoint and spss data to follow along. Hope this helps!
I followed the steps above and was only getting the model summary without error. I am using version 25.What I am I missing?
Thanks so much ....is there Bivariate multilevel logistic regression? please show us
May I ask some questions, Sir! If constant (intercept) is not significant, it is possible to interpret the remaining predictors. And if the model is not fitted to proceed (Omnibus test is insignificant), what should I do to use that model? This is because its model is critical to use in my thesis.
Hi there. The short answer to your question is 'no'. A significant intercept is not a requirement for interpretation of the predictors in the model. The test of the intercept that is printed out is nothing more than a test of whether it is different from zero, and this test is often not of substantive interest to researchers. (even though, depending on coding, the intercept is something they wish to make sense of).
FYI, you might consider visiting the following (more recent) videos I have posted on binary logistic regression (ruclips.net/video/cpWSSJHuT2s/видео.html) and multilevel binary logistic regression (ruclips.net/video/roWTULimNPk/видео.html
). Both include Powerpoints you can download that might be helpful as you work on interpreting your results.
best wishes.
hello Mike, i have watched a lot of your videos and pdf and they helped me a lot. would you be so nice to help me in solving this problem? i am running a binary logistic regression and i have identified two models:
1st model:
Nagelkerke coefficient:45,2%
prediction accuracy: 76,9%
Hosmer and Lemeshow p-value:0,004
2nd model:
Nagelkerke coefficient:39,5%
prediction accuracy: 79,7%
Hosmer and Lemeshow p-value:0,093
i don't know how to choose among them because if i look at the nagelkerke coefficient i would choose the 1st, but if i look at the hosmer and Lemeshow p-value + prediction accuracy i would opt for the second. could you help me in solving this problem?thank you
Hi Chiara, think of the pseudo-R-square, Hosmer/Lemeshow test, and classification accuracy measures as three lenses through which you judge the fit of a single model. If you are comparing two models in terms of relative fit, then there are a couple of alternatives. If you are comparing two nested models (where the set of predictors in Model 1 are a subset of those in Model 2), then you could always perform a likelihood ratio test to determine if model 2 is a significant improvement in fit relative to Model 1. What you are describing suggests to me that you might be comparing two models using the same outcome variable, but perhaps different sets of predictors between models. In that case, your models are non-nested - thus you cannot use the likelihood ratio test noted above. An option in this case might be to compare your two models in terms of a measure such as Aikaike's Information Critereria (AIC), which is a parsimony-adjusted measure of fit. In general, models with smaller AICs are preferred over those with larger AICs. [By the way, the value of the AIC within one analysis will not tell you anything. It is only used for model comparisons] You cannot obtain the AIC through the SPSS route I demonstrated in this video. However, you CAN obtain it if you run your analysis through the Generalized Linear Models option. Although I don't spend any time really discussing the AIC in this video (ruclips.net/video/P0VKXrHrFUc/видео.html), I do demonstrate (mid-way through, I believe) how to run your logistic regression through Generalized Linear Models. I also have a powerpoint that accompanies the video (see link underneath video description). Also, although unrelated to your question, I do have a newer video on logistic regression here: ruclips.net/video/cpWSSJHuT2s/видео.html I hope this is helpful. Cheers!
How does SPSS know which of the independent variables are level 1 and which are level 2? At 24:50 you drag both variables into the same place, but one of them is supposed to be at level 1 and the other at level 2. Did you make a mistake there or am I missing / misunderstanding something?
I am reasearching this because I have a binary dependent variable and nested independent variables on two different geographic levels (it's either level 1 or level 2; I don't have a single one of them for both levels), and from what I have learned so far, I need to use a multilevel binary logistic regression.
Between times 13:22-13:30, you'll see me move the level 2 group identifier over to the subjects box. So the program recognizes the level 2 identifier that way. Now, all level 1 predictor variables will have variability within groups. When you look at your level 1 predictor in your dataset, you will find the values vary within the Level 2 units. Level 2 predictor variables, in contrast, will exhibit no within-group variability (as values are constants within groups). If you look at your SPSS dataset, you will see the same value on the Level 2 predictor for each case in your dataset, but you'll see different values across groups. So everything is computed with that recognition.
Just to clarify: Are you saying that SPSS automatically notices that the values of predictor tx (in your video) are the same for each case inside a distinct group?
Because as far as I can see, you are only telling SPSS for each case in what group that case is, but not directly which predictor is on level 1 and which on level 2.
I would also be interested if you agree, that, based on my description of my research subject in the initial question, it makes sense to use this statistical method.
Thank you very much! This helps me a lot.
May I know the multilevel binary logistic regression is it same as hierarchical binary logistic?
Hi there. Multilevel binary logistic regression would not be the same as hierarchical binary logistic regression. To use kind of an analogy, think of the relationship between hierarchical multiple regression and multilevel regression. The former would involve testing a series of models with predictors added sequentially - but they are all at the same level of analysis (.e.g, person level). Multilevel modeling, however, involves inclusion of predictors and testing of random coefficients across different levels/units of analysis. Hope this helps!
@@mikecrowson2462 thank you very much.
Hi. I just would like to ask, is this regression here in the vid, the same with the nested logistic regression?
Hi there. Yes, I think you can consider it a nested logistic regression. I don't use that terminology, but I believe I've heard it referred to that way before. The term 'nested' simply connotes a multilevel data structure. Cheers!
@@mikecrowson2462 Thank you! Highly appreciated!
Hi, I am wondering how you would calculate the ICC for a multilevel analysis with a dichotomous outcome?
Hi, I am wondering how you would calculate the ICC for a multilevel analysis with a dichotomous outcome?