You are for real the first person that was able to tell me that mixed models, mixed effects models, hierarchical models etc are basically the same - i always asked myself that but could never find an answer to that question, thank you so much!
I love the craziness, as someone who enjoys stats it's great to hear different pitches to keep you focused when trying to understand things a little clearer
I use this type of model to predict time series with an additional variable "sequence" to account for time variable. So there may be an I,j,k. Not sure if I am calling it the correct name but I assume it is called multivariate multisample time series modeling. It's all linear regression at the end of the day
Question: how come we don’t usually treat individual differences like gender as a nesting factor? These variables usually just get controlled for in a single level model, but all of the sudden if we change “gender” to “school” we should be using MLM, cluster robust standard errors, etc. Any insight?
Loved it. Actually watched it in sweatpants on the couch on a rainy Friday evening while feasting on an obscenely large pizza. Guess I'm a true scientist now. Say, would you be so kind to share the slides you showed in the video? I'd like to add some notes, so it's easier for me to go through it add up some things you said but didn't write down. Kind regards!
The fixed effect is the regression line for the average intercept and average slope of all the mixed effects? Each subject’s regression line is a mixed effect? Thanks!
ish. You can roughly interpret the fixed effect as the average slope/intercept random effects. Each subject's regression line is a RANDOM effect, not a mixed effect.
Is it possible to apply Multilevel models when data set consists of 3 IVs of categorical nature and 1 or more continuous DVs ?.......... My data set consists of 3 categorical IVs (1st IV consists of 2 categories, Second IV also consists of 2 Categories and 3rd IV consists of 3 categories).
I actually have a question now that I have slept on this concept. What happens when the groups of dependent data share the same predictor (x) values? Even though they may share different responses (and thus slopes and y-int), it seems to me that treating the data in a disaggregate treatment vs. a HLM treatment (averaging the individual fits) would possibly produce the same thing? An aggregated treatment, however, I would expect to still be incorrect ('answering a different question'). A second question, is if instead of aggregating (averaging) the values within groups to a single value, if one were to average the data between groups. For example, the left most point of Donald averaged with the left most point of Hillary and also that of Obama. Sample size reduction would obviously occur, but would this not produce the intended 'average' behaviour for the intended question (per person)? Then a fit could be performed between these three points, and even a fit considering weights for the spread of these three average points?
First question: Not quite. I used to think that until I simulated it and found that they weren't the same. The reason why is because, when estimating each individual slope, the model "borrows" information from all the other clusters to estimate that slope. So, if you did a disaggregated analysis on that cluster, the actual slope for that cluster from HLM will be somewhere between that disaggregated line and the "grand line" that is computed from all clusters. As for your second question: that's kind of an ad hoc way of doing things that may work sometimes and may not other times. It's better to just do HLM :)
Hi in economics is this called fixed effect model? That is difference out time invariant charactics? Or in other words add more dummies in individuals? As in the aggregated analysis slide?
This was a great video, probably one of the greatest at explaining HLMs in my opinion! keep it up, also love the format and how you keep it funny and engaging
@@QuantPsych I have come up with a couple questions and I have not been able to find satisfying answers to this on the internet: 1. How do we choose when to model a variable as a random or fixed effect variable (for example teacher in the student class example). 2. How is stratified analysis different to mixed effect modelling?
@@lucha6262 1. If a characteristic varies within cluster, it can be modeled as a random effect, otherwise it cannot. For example, if you have students within classrooms, but you have a *class* characteristic (e.g., pedagogical philosophy of the instructor), that can only be modeled as a fixed effect. Once you decide if it CAN be modeled as a random effect, whether to actually model it as a random effect can be determined theoretically or empirically. You can always do a model comparison to see if it matters whether you model it as a fixed or random effect. 2. By stratified analysis, do you mean fitting a different slope for each cluster? If so, conceptually they are quite similar. The difference is that mixed effect models tend to bias within-cluster regression lines to be closer to the fixed effect line. So, if you actually plotted a regression line versus a mixed model random line, the two will be different (with the mixed model line being a weighted average of the random line and the fixed line).
Unfortunately the humor did not work for me at all. It just created random targets, that made what I was actually here to learn hard to follow. Some middleground between dry theory and absolute bonkers comedy might work better. +1 for effort though
All of a sudden in the internet there decides to be a guy who can explain stuff in plain english. Thanks!!
I LOVE YOUR CHANNEL
You are for real the first person that was able to tell me that mixed models, mixed effects models, hierarchical models etc are basically the same - i always asked myself that but could never find an answer to that question, thank you so much!
They are not the same
You have the best videos ever. I love and appreciate your enthusiasm and wicked sense of humour
I love the craziness, as someone who enjoys stats it's great to hear different pitches to keep you focused when trying to understand things a little clearer
I found this to be an exceptionally efficient intro to the HLM model. Thank you!
Thank you very much! This is the content that I have been looking for a long time for my analysis
This is an old video. This one might be a bit more contemporary :) ruclips.net/video/c_tYZxQLoDA/видео.html
This is the greatest. Humour + learning. Thank you 😊
I use this type of model to predict time series with an additional variable "sequence" to account for time variable. So there may be an I,j,k. Not sure if I am calling it the correct name but I assume it is called multivariate multisample time series modeling. It's all linear regression at the end of the day
That may be what some people call it. It's still considered a multilevel model--you just have three levels instead of 2 :)
nice way of explanation. I was learning and getting refresh same time.
Fantastic video! You explain it so unbelievable well.
This was phenomenal. Thanks so much!
You can see the updated video (with updated R code) here: ruclips.net/video/c_tYZxQLoDA/видео.html&ab_channel=QuantPsych
a crazy but clear explanation, thanks!!
you are really good. Understood and enjoyed it alot. Thanks
Thank you for explaining this!
That was so good and helpful, thank you!
Question: how come we don’t usually treat individual differences like gender as a nesting factor? These variables usually just get controlled for in a single level model, but all of the sudden if we change “gender” to “school” we should be using MLM, cluster robust standard errors, etc. Any insight?
Wow this was a great video, everything makes so much more sense
I really enjoyed it !
Thank you for the video! it was really clear (even the mathy part)
Thanks!
Hey! this video is super helpful! Could you also recommend any textbooks or text based resources for these? Thanks!
Eventually I'll add a mixed model chapter to my textbook....eventually.
Loved it. Actually watched it in sweatpants on the couch on a rainy Friday evening while feasting on an obscenely large pizza. Guess I'm a true scientist now.
Say, would you be so kind to share the slides you showed in the video? I'd like to add some notes, so it's easier for me to go through it add up some things you said but didn't write down.
Kind regards!
The fixed effect is the regression line for the average intercept and average slope of all the mixed effects? Each subject’s regression line is a mixed effect? Thanks!
ish. You can roughly interpret the fixed effect as the average slope/intercept random effects. Each subject's regression line is a RANDOM effect, not a mixed effect.
I don't like the style but appreciate the effort and the content. Thumb up for trying to teach serious stuff differently.
Great explanation! Thank you
Awesome man! You remind me a bit of Andrew Hales (he's a youtuber who also used to be on Adderall). Again, great vid. Thanks for sharing.
I call him the Brendan Fraser of Statistics
Is it possible to apply Multilevel models when data set consists of 3 IVs of categorical nature and 1 or more continuous DVs ?..........
My data set consists of 3 categorical IVs (1st IV consists of 2 categories, Second IV also consists of 2 Categories and 3rd IV consists of 3 categories).
you make me laugh sooooo loud, it's impossible whenever i learn my stats!!!!!
omg thank you so much, my dude.
I actually have a question now that I have slept on this concept. What happens when the groups of dependent data share the same predictor (x) values? Even though they may share different responses (and thus slopes and y-int), it seems to me that treating the data in a disaggregate treatment vs. a HLM treatment (averaging the individual fits) would possibly produce the same thing? An aggregated treatment, however, I would expect to still be incorrect ('answering a different question'). A second question, is if instead of aggregating (averaging) the values within groups to a single value, if one were to average the data between groups. For example, the left most point of Donald averaged with the left most point of Hillary and also that of Obama. Sample size reduction would obviously occur, but would this not produce the intended 'average' behaviour for the intended question (per person)? Then a fit could be performed between these three points, and even a fit considering weights for the spread of these three average points?
First question: Not quite. I used to think that until I simulated it and found that they weren't the same. The reason why is because, when estimating each individual slope, the model "borrows" information from all the other clusters to estimate that slope. So, if you did a disaggregated analysis on that cluster, the actual slope for that cluster from HLM will be somewhere between that disaggregated line and the "grand line" that is computed from all clusters.
As for your second question: that's kind of an ad hoc way of doing things that may work sometimes and may not other times. It's better to just do HLM :)
Hi there, i really like your videos but i really struggeling with which method to apply in my rerport.
Can i contact you to talk about this?
Thank you!!!
is there part 2 of this? thanks, good Explanation.
There's two:
ruclips.net/video/UJsoZu1ylNY/видео.html
ruclips.net/video/lf6lxc9yi8I/видео.html
Thanks for stopping by!
Hi in economics is this called fixed effect model? That is difference out time invariant charactics? Or in other words add more dummies in individuals? As in the aggregated analysis slide?
I'm not sure. I'm not familiar with econ terminology
thank you
I really appreciate that you're trying to make this video entertaining, but its unbearable.
Have you tried watching in reverse? I hear that makes it more bearable.
I loved that you kept it interesting
Maybe he's doing it to keep it interesting for himself, not you!
@@adilsarbay3181 Exactly :)
although you are nuts, this is the first video covering why we actually need this stuff. Thanks!
ummmm....thanks?
This was a great video, probably one of the greatest at explaining HLMs in my opinion! keep it up, also love the format and how you keep it funny and engaging
Glad you enjoyed it!
@@QuantPsych I have come up with a couple questions and I have not been able to find satisfying answers to this on the internet: 1. How do we choose when to model a variable as a random or fixed effect variable (for example teacher in the student class example). 2. How is stratified analysis different to mixed effect modelling?
@@lucha6262 1. If a characteristic varies within cluster, it can be modeled as a random effect, otherwise it cannot. For example, if you have students within classrooms, but you have a *class* characteristic (e.g., pedagogical philosophy of the instructor), that can only be modeled as a fixed effect. Once you decide if it CAN be modeled as a random effect, whether to actually model it as a random effect can be determined theoretically or empirically. You can always do a model comparison to see if it matters whether you model it as a fixed or random effect.
2. By stratified analysis, do you mean fitting a different slope for each cluster? If so, conceptually they are quite similar. The difference is that mixed effect models tend to bias within-cluster regression lines to be closer to the fixed effect line. So, if you actually plotted a regression line versus a mixed model random line, the two will be different (with the mixed model line being a weighted average of the random line and the fixed line).
Great content but the jokes made it harder for me to follow the stream of information.
Unfortunately the humor did not work for me at all. It just created random targets, that made what I was actually here to learn hard to follow. Some middleground between dry theory and absolute bonkers comedy might work better. +1 for effort though
AGREED
had to stop watching bc of annoying interjections
not funny