Multilevel binary logistic regression example in SPSS

Рет қаралды 61,129

Mike Crowson

Күн бұрын

Пікірлер: 45

@tewodrosendale123 2 жыл бұрын

Thanks so much ....is there Bivariate multilevel logistic regression? please show us

@killionmokaya2572 4 жыл бұрын

I followed the steps above and was only getting the model summary without error. I am using version 25.What I am I missing?

@chriskelly1109 4 жыл бұрын

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?

@mikecrowson2462 4 жыл бұрын

Hi Chris, I have a new video on this topic ( kzbin.info/www/bejne/qKC6hYiCntKBhs0 ) that also covers random slopes. Additionally, underneath the video description, you can download a Powerpoint and spss data to follow along. Hope this helps!

@daniellacarucci8084 4 жыл бұрын

Hi, I am wondering how you would calculate the ICC for a multilevel analysis with a dichotomous outcome?

@eramresearchconsultancyser3222 2 жыл бұрын

Hi. I just would like to ask, is this regression here in the vid, the same with the nested logistic regression?

@mikecrowson2462 2 жыл бұрын

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!

@eramresearchconsultancyser3222 2 жыл бұрын

@@mikecrowson2462 Thank you! Highly appreciated!

@ironstef 4 жыл бұрын

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

@bingxinli417 4 жыл бұрын

I really really love your detailed explanations on the results! BIG thanks!

@mikecrowson2462 4 жыл бұрын

Thanks for visiting Bingxin! I'm glad this was helpful to you! best wishes.

@LaurenG191 6 жыл бұрын

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!

@SwedishRagers 3 жыл бұрын

Thank you for helping me with my first meta-analysis!

@mikecrowson2462 3 жыл бұрын

I'm glad you found it helpful, Samuel! Cheers!

@ethio-tex4209 5 жыл бұрын

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.

@petenelsonpm 6 жыл бұрын

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?

@mikecrowson2462 4 жыл бұрын

Hello everyone, I have a newer (better quality, 2020) video on Multilevel binary logistic regression that can be found at: kzbin.info/www/bejne/qKC6hYiCntKBhs0 There is an accompanying Powerpoint and SPSS data file underneath the video description. I hope you check it out!

@kabir.sokullu 4 жыл бұрын

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?

@Rooo8 6 жыл бұрын

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.

@mikecrowson2462 6 жыл бұрын

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.

@Rooo8 6 жыл бұрын

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.

@susunyein7317 4 жыл бұрын

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.

@mikecrowson2462 4 жыл бұрын

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 (kzbin.info/www/bejne/maG6hIaAfdqHaNU) and multilevel binary logistic regression (kzbin.info/www/bejne/qKC6hYiCntKBhs0 ). Both include Powerpoints you can download that might be helpful as you work on interpreting your results. best wishes.

@929bleh 4 жыл бұрын

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!

@simmingtan7241 5 жыл бұрын

May I know the multilevel binary logistic regression is it same as hierarchical binary logistic?

@mikecrowson2462 5 жыл бұрын

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!

@simmingtan7241 5 жыл бұрын

@@mikecrowson2462 thank you very much.

@sandilesimelane4813 6 жыл бұрын

Very useful materials Mr Crowson!! Do you have videos on multilevel modeling of ordinal outcome variable with more than 2 categories?

@mikecrowson2462 6 жыл бұрын

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 :)

@chiaraostuni2775 4 жыл бұрын

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

@mikecrowson2462 4 жыл бұрын

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 (kzbin.info/www/bejne/hmG5fIuofdd5i8U), 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: kzbin.info/www/bejne/maG6hIaAfdqHaNU I hope this is helpful. Cheers!