Do you want to take a class with me? Visit simplistics.net to register for a class. You can either do "live" classes, where you'll learn from me directly via zoom. Or you can register for "self-guided" courses, complete with a schedule, discussion boards, quizzes, readings, etc.
I can't believe I've never found your channel until now. Great explanations!
Месяц назад
At first, it was difficult to match your energy to concentrate on the topic. Yet, my perseverance to keep watching paid off. A really great explanation. Thank you so much.
Amazing explanation! Probably one of the simplest explanations of mixed models on RUclips. Please consider lowering the volume of the edited video to 85% or 90% as it does tend to get very loud at times
Dear Quant Psych: At around 12:30, you describe the "group" (schools) should be treated as a random intercept effect. Can you explain why we should not treat it as a fixed effect? Why not have two categorical variables, "Sex" (with two levels) and "Group" (with three levels), for a total of six combinations, to model your data?
Asking people to leave comments with their guesses is not pedagogical, it is just asking people to leave comments so that the video might get more views in the future or something. xd It's like asking people to like and subscribe. Thank you for these explanations btw, your videos are very fun to watch.
Question: Why using dummy variables to represent each condition is not sufficient? For example, what if I use a binary variable for each doctor which captures the difference, if any, between doctors and their effect on the patient they have.
Generally opting for the dummy variable approach will lead to many more parameters being added to your model (at least, when there are many groups), rather than just the group error term(s)...this is just one parameter extra for a random intercept model, for instance. It also may prevent you from modelling explanatory variables relating to the group effect, and fixed effect approaches (i.e., dummy variable approach) can't generalise beyond the groups that make up the sample.
OMG, Thanks so much for this video! Please consider the advice concerning your mic, think of all of your foreigner followers (like me) 🙏love your videos
Your video is super amazing. Take your production to the next level by using close microphone source like a lav mic clipped on. It will reduce the reverberation and increase intelligibility.
Hi there! I'm stuck on calculating needed sample size for mixed models I'm planning - is it me or is it a mess? I was hoping I'd find papers with like, explanations, tutorials, whatever - anything I can read, understand and try out. But instead I only find very theoretical papers? If you have any advice or references or so, they'd be suuper welcome.
It really sucks to calculate sample sizes for mixed models. As I recall there was a textbook written by Diggle. Maybe try googling "Diggle longitudinal power caluation." Last time I had to do it, I used the longpower package in R.
@@QuantPsych hey, thank you! I saw that most people do simulations, which takes ages (I thought my computer is ok, but for this it is not powerful enough I guess), but then I found an app/ website by Oscar Olvera Astivia. For future reference 😊
Hi there. Thanks a lot for this Video. Do you happen to know a proper way how to visualize a MLM when having 3 predictors (all with different slopes/ intercepts) into one Output? I searched almost the whole internet for this and it seems you got something like this in your video. Thanks a lot in advance for your reply
It's a very good explanation but I can only imagine you can keep on subdividing your subgroups or random effects until a clear mixed effect comes out. Are these random effects also evaluated using R^2?
If you don't have that many observations in your dataset, wouldn't you come across with some issues if you divide your data into groups? I can imagine that the worst case scenario would be ending up with only two observations per group.
Can I ask, if I follow-up a cohort longitudinally, is there any assumption about the time points participants have been assessed? For instance, can I only use assessment in time 1 and time 2 to predict an outcome variable in time 3?
For example, could be that the growth of different bacterial isolates using different growing media may have a fixed intercept since they all start at 0 growth?
I wonder if there's a difference between mixed models and the ANCOVA. I thought the ANCOVA does exactly the same: testing whether 1) there's a significant relationship between two numerical variables, and 2) whether the intercepts of these relationships differs significantly for different factors?!
A point that I would appreciate if anyone could help me understand; my understanding is that the data points within a cluster are independent with respect to other data points within the same cluster and that is why we can fit a regression model (that has independence assumption) within a single cluster. However, the data points within a cluster are not independent with respect to data points in other clusters and that is why we cannot fit a single regression model (that has the independence assumption) across clusters. Can someone please tell me if my understanding is correct? Thanks!
1:04..or maybe we measure people who share the same gender. Why can’t I see a clear reason that “gender” is not a common candidate for nesting variable (ie people usually just control for it), but classroom always is?
With gender we generally exhaust the categories we're interested (e.g., male, female, nonbinary). With classrooms we do not because we can't possibly sample all classrooms out there.
Great video. I want to ask about how linear mixed models handle missing data? Can it handle missing data on the predictor/covariate level or the response level? Or both? Or only the response level?
There's some nuance to the answer. If an entire wave of measurements is missing, that's no big deal for mixed models. (It's considered MAR or missing at random). If some are missing from a wave, but not others, it will not handle it without some additional missing data strategy (e.g., multiple imputation).
Watching this video, I felt miss a boring class. In my opinion, your classes are good, however, the exaggerations in the jokes are bad. I suggest decreasing it and your classes will be better!
My god - the presentation style is so distracting. Do we really need all the overacting and overplaying? It hurts my brain to try and pain attention to this. Please consider that different learners have different needs - this style is off putting to learners like me.
Then I say to learners like you...don't watch my videos. Find someone who fits your learning style. (That's far easier than for me to change my teaching style).
![Noob To Max With DRAGON REWORK In Blox Fruits [FULL MOVIE]](http://i.ytimg.com/vi/LnBMOinoOvA/mqdefault.jpg)
Do you want to take a class with me? Visit simplistics.net to register for a class. You can either do "live" classes, where you'll learn from me directly via zoom. Or you can register for "self-guided" courses, complete with a schedule, discussion boards, quizzes, readings, etc.
It takes a special kind of person to be so energetic for a subject like this, thanks for the explanation and unconventional format!
This video single handedly made my life sooooo much easier. Such a great explanation and so to the point. Loved it!
I can't believe I've never found your channel until now. Great explanations!
At first, it was difficult to match your energy to concentrate on the topic. Yet, my perseverance to keep watching paid off. A really great explanation. Thank you so much.
It helps to watch my videos while stoned.
9:43 This is what I have been trying to understand for a long time. Thank you.
FINALLY I understood this. Thanks a lot. Keep up the good work. Greetings from Brazil!
it's the first time to really enjoy and understand statistical modelling like that! thanks a lot!
Glad you liked it!
My thesis is based on HLM and you just saved my life.
Best explanation on youtube so far
This is such a fun and clear explanation!
This is amazing. I wish my MLM class had been this clear. Thanks for the refresher.
Best explanation ever on RUclips! And what a fun style of teaching 😎
Amazing explanation! Probably one of the simplest explanations of mixed models on RUclips. Please consider lowering the volume of the edited video to 85% or 90% as it does tend to get very loud at times
Nice fun explanation of the concept. Thanks
Dear Quant Psych: At around 12:30, you describe the "group" (schools) should be treated as a random intercept effect. Can you explain why we should not treat it as a fixed effect? Why not have two categorical variables, "Sex" (with two levels) and "Group" (with three levels), for a total of six combinations, to model your data?
You are totally amazing!!!!and fun to listen to
Awesome video! Very entertaining as well
Will you please introduce generalized estimating equations (GEE)? When to use GEE vs. Mixed Model?
_/|\_Thank you
Great video. Thanks so much! It helps that you go over examples too.
brilliant way to explain these grad level topics. Thank you
This was awesome - and quite entertaining.
Amazing explanation! I was tired of searching for a good video on this topic, glad I found this video.
Doing the lords work, thanks!
- Ph.D. Student
Thank you! I love this explanation style. It's hilarious and really sticks.
Hi Quant Psych! Do you have any tutorials for estimating sample size for multi-level models?
Best explanation, great job
Asking people to leave comments with their guesses is not pedagogical, it is just asking people to leave comments so that the video might get more views in the future or something. xd It's like asking people to like and subscribe. Thank you for these explanations btw, your videos are very fun to watch.
Exactly!
Appreciate it, thanks!
Great Explanation, Thanks
Thank you so much! This video was really helpful!
Thank you for making these lecture videos freely available without commercials. Do you have any videos on dyad analysis?
Hey! Thanks so much for these videos! Where might I find this 3 hospitals data set you used?
Thank you for your help.
Great, great explanation
Awesome video! Thank you!
Great video, thank you!
Question: Why using dummy variables to represent each condition is not sufficient? For example, what if I use a binary variable for each doctor which captures the difference, if any, between doctors and their effect on the patient they have.
Generally opting for the dummy variable approach will lead to many more parameters being added to your model (at least, when there are many groups), rather than just the group error term(s)...this is just one parameter extra for a random intercept model, for instance. It also may prevent you from modelling explanatory variables relating to the group effect, and fixed effect approaches (i.e., dummy variable approach) can't generalise beyond the groups that make up the sample.
Clear and fun! Thanks a lot!
anyway, your explanation is actually pretty good. After covering my ears I kind of understood why I should use mixed models in my analysis.
Nice video and nice explanation. I would like have seen some exemples with paired data.
OMG, Thanks so much for this video! Please consider the advice concerning your mic, think of all of your foreigner followers (like me) 🙏love your videos
your video is soooo good!
AWESOME VIDEO TYSM!!!
Hi Quant Psych, really appreciate this video. Would it be possible to get your hospital example data?
quantpsych.net/data/hospital.csv
@@QuantPsych Thank you very much for the data!
Do you have a video on Multinomial mixed effects model as I have a response with 4 levels? Thank you
This is the closest I have: ruclips.net/video/Yqf91pPzkU4/видео.html
How is a random-effects model different to a fixed-effects model with interactions (between the covariate and category)?
Your video is super amazing. Take your production to the next level by using close microphone source like a lav mic clipped on. It will reduce the reverberation and increase intelligibility.
So helpfull !! thanks so much :)
LOVE YOUR VIDEOS!!!
Damn you are good! Thank you a thousand times!!!
Hi there! I'm stuck on calculating needed sample size for mixed models I'm planning - is it me or is it a mess? I was hoping I'd find papers with like, explanations, tutorials, whatever - anything I can read, understand and try out. But instead I only find very theoretical papers? If you have any advice or references or so, they'd be suuper welcome.
It really sucks to calculate sample sizes for mixed models. As I recall there was a textbook written by Diggle. Maybe try googling "Diggle longitudinal power caluation." Last time I had to do it, I used the longpower package in R.
@@QuantPsych hey, thank you! I saw that most people do simulations, which takes ages (I thought my computer is ok, but for this it is not powerful enough I guess), but then I found an app/ website by Oscar Olvera Astivia. For future reference 😊
@@QuantPsych and I will look up the book anyway! Thanks again 🙌
Do you have a video explaining what i.i.d. means?
Hi,
Anytime i add the level 2 variables, I get a lower ICC. What is the problem or how do I interpret that?
Hi there. Thanks a lot for this Video. Do you happen to know a proper way how to visualize a MLM when having 3 predictors (all with different slopes/ intercepts) into one Output? I searched almost the whole internet for this and it seems you got something like this in your video. Thanks a lot in advance for your reply
Hi, is this what is used for Panel regressions with time series data? Thanks!
At 11:10, I guess it should be "normally you don't want to fix an intercept..."
At 15:26: "it could be a fixed intercept model..."
Yes, you're right :)
really good, can use more slides to emphasize points of discussion
It's a very good explanation but I can only imagine you can keep on subdividing your subgroups or random effects until a clear mixed effect comes out. Are these random effects also evaluated using R^2?
Yes, with some added complications. I have a video about computing R squared for HLMs
If you don't have that many observations in your dataset, wouldn't you come across with some issues if you divide your data into groups? I can imagine that the worst case scenario would be ending up with only two observations per group.
Can I ask, if I follow-up a cohort longitudinally, is there any assumption about the time points participants have been assessed? For instance, can I only use assessment in time 1 and time 2 to predict an outcome variable in time 3?
You'd use Time as a predictor (1, 2, or 3) to predict the outcome.
Why do you ask him to stop screaming ? It wakes me up when I'm at work
For example, could be that the growth of different bacterial isolates using different growing media may have a fixed intercept since they all start at 0 growth?
when you say cluster, I am assuming that you mean a variable, Ie every column. so a cluster would be the doctor column and the patient number column?
Are you related to the Mathantics videos guy?
love iiiit!
I wonder if there's a difference between mixed models and the ANCOVA. I thought the ANCOVA does exactly the same: testing whether 1) there's a significant relationship between two numerical variables, and 2) whether the intercepts of these relationships differs significantly for different factors?!
yo this video bumps
also, what is your ggplot theme its so cute
A point that I would appreciate if anyone could help me understand; my understanding is that the data points within a cluster are independent with respect to other data points within the same cluster and that is why we can fit a regression model (that has independence assumption) within a single cluster. However, the data points within a cluster are not independent with respect to data points in other clusters and that is why we cannot fit a single regression model (that has the independence assumption) across clusters. Can someone please tell me if my understanding is correct? Thanks!
You're wrong - daa within a cluster is always dependent, e. g. a group of patients which visit the very same doctor.
👍👍👍👍
1:04..or maybe we measure people who share the same gender. Why can’t I see a clear reason that “gender” is not a common candidate for nesting variable (ie people usually just control for it), but classroom always is?
With gender we generally exhaust the categories we're interested (e.g., male, female, nonbinary). With classrooms we do not because we can't possibly sample all classrooms out there.
Great video. I want to ask about how linear mixed models handle missing data? Can it handle missing data on the predictor/covariate level or the response level? Or both? Or only the response level?
There's some nuance to the answer. If an entire wave of measurements is missing, that's no big deal for mixed models. (It's considered MAR or missing at random). If some are missing from a wave, but not others, it will not handle it without some additional missing data strategy (e.g., multiple imputation).
He throws many new concepts and said dont worry? Is that a good explanation?
Seriously! This guy's an idiot!
Sir,...can you make a small video with dataset example on how REML works to find variance components ..for. eg. y= a+b+e a and b being random effect.
Hello, does anyone here knows how to analyze such model in python when you want to predict a binary variable?
Watching this video, I felt miss a boring class. In my opinion, your classes are good, however, the exaggerations in the jokes are bad. I suggest decreasing it and your classes will be better!
Why does he shout the whole time? 😮
because he is passionate 🥹
stop screaming at me bro
Calm down!
no
Stop screaming
My channel, my rules.
why are u screaming at the camera? so annoying. have some self respect and make a proper video.
… Turn your sound down if you don’t like it. These videos are great!
My god - the presentation style is so distracting. Do we really need all the overacting and overplaying? It hurts my brain to try and pain attention to this. Please consider that different learners have different needs - this style is off putting to learners like me.
Then I say to learners like you...don't watch my videos. Find someone who fits your learning style. (That's far easier than for me to change my teaching style).