Your video is very clear! Thank you! can random effect be continuous data? I was thinking using GPS (Xvalue and Yvalue, 2 separate columns) as a random effect for the response variable 'sick pigs' and 'healthy pigs' under some environmental explanatory variables. but I dont know if it makes sense
This video is quite old but maybe you see it. Can I have multiple cluster variables? my dataset is over multiple years and in different areas. So as I understand it I would cluster it by the year but also to the areas?
Mixed models are a great topic. One suggestion I have is it not oversimplify the scope of statistical dependence. I am alluding to the following phrase near the beginning of the video: "... and when we don't meet the assumption of independence, everything gets screwed up, so we need to take a mixed models approach." There are many, infinitely many, ways to violate the assumption of statistical independence. Mixed effects models cover many of the practical cases, and are therefore worth teaching to people that will go on to do statistics, however they are not sufficiently general to handle the broader scope of modelling statistical dependence. An example that is very similar to the case of mixed effects is that of mixture distributions. Your samples may not be independent, but perhaps they are conditionally independent on their membership with a cluster/grouping. Only problem is, let's say, you don't actually have labels for the clusters. Taking a mixture distribution approach, which I'll admit sounds very similar to the term "mixed effect", is more appropriate to the problem. The point of my suggestion is just to offer students some caution that there are topics beyond mixed effects due to the fact that they are not 'always' the appropriate approach even if they are worth knowing.
That was easy to understand, why do people say it's difficult to understand these??. Why is fixed effects modeling called what it is if it's about learning a slope and intercept for each person?
Hello there! A corr of negative value indicates that individuals starting (age 14) with high alcohol increase (larger random intercept) tend to increase their alcohol consumption in a slower rate as they grow older, and vice versa. Visually this linear (or non-linear relationship in this case) can be represented in a scatter plot of random intercepts vs random slopes.
Where it might be useful for interpretation: Do children that start speaking earlier than their peers, then develop more slowly (in educational terms) than those peers who don't? Do people who inherit more than x amount of money have faster income growth compared to others ?
To be fair, you're a generalist too if you don't have a PhD in Statistics. You're right there with the woman with a PhD in neurology. You have expertise in quantitative psychology and it's no different really than her having a PhD in neuroscience. You even admit "math stat" isn't your thing. Math stats is Statistics. The applied statistics you know wouldn't exists without mathematical statistics and statistical theory.
I think that's a pretty narrow definition of a specialist. If math stat is the criteria, then having a PhD in stats would be a generalist and only mathematicians would be able to claim expertise. I think you misunderstand quantitative psychology. It's even more specialized than a graduate degree in statistics. A quant psych degree is a very specific branch of statistics that tends to specialize in measurement (though not always). And there are quantitative psychologists that could run circles around most PhD stats graduates when it comes to math stats. (My advisor was one of them). My weakness in math stat is unique to me, not my profession. (And that statement should be qualified that it's a Dunning-Kruger statement--compared to my colleagues, I'm weak, but certainly more qualified than a regular psych PhD graduate). I hope that clarifies things.
Q: Why do you have AGE_14 as both a random AND a fixed effect in the lmer() formula (around 2:40)? This seems counterintuitive to me. You're accounting for the within group variance of AGE_14 as a random effect, but also want to measure that as a fixed effect?
First you need to consider what AGE_14 is in terms of being a fixed effect predictor in the model. Here it represents the predicted increase in drinks per week as age increases (above age 14). However, you may not expect that increase to be the same across all individuals, therefore we allow each individual to have their own slope (or change in drinks per week as age increases). If you do not have AGE_14 specified as a random slope, you are in effect assuming that all individuals increase their alcohol use at the same rate with each increase in age.
@@tonycarlsen1627 Ahh. Not used to seeing something that is not "1" before the vertical bar and it threw me off there. I should have stopped to think that one through more thoroughly. Thanks.
Hey Dustin, did you create the FlexPlots library in R? That’s amazing! I’m sold. I had bought some R courses on Udemy, but after watching your videos, I’ve decided to take your course instead. The "Intro to Simplistics" is your foundational course, right? I just wrapped up a predictive analytics course, and now I’m just one course and a Capstone project away from finishing my Master’s. By the way, keep making these videos! You’re going to absolutely blow up. People are so confused about statistical knowledge, and I believe a ton of folks will be motivated to learn about statistics, especially after the debacle in the 2016 election. Keep them coming!
I highly recommend buying this guy’s ‘Introduction to Simplistics’! As a psychology honours student, I’ve learned more from this Professor in 3 hours than I did in three years of my Psychology bachelor’s. Thank you, Professor!
You cannot assign a probability to a parameter since it is supposed to be constant in frequentist statistics, while samples are random. Flipped if you think with a Bayesian brain 😂😂
Most challenging class in stats was math. stats in my case. It took me almost 2 yrs to really understand the math. stats. Today I love it to read about math stats contents and proofs. If you really like stats you have to spend time and resources to do research and studies, which the most people do not do.
There there. You might be relieved to know that Dunning Krueger effect also doesn’t exist. Just a regression to the mean effect mistaken for something real. The world is full on bad stat-analysis.
Definitely NOT gatekeeping. Her saying that she is an expert because she teaches intro stats is like a highschool AP calculus teacher claiming they are an expert in mathematics. Its INSANE. Don't worry! The students who are eager to learn will have your back!
60 credit hours in STATs and I am still humble to know there’s too much crap to remember. Also, too much mathematical derivations and the mathematical techniques to remember.
The worst teachers I had were experts in their fields. I'm not saying that's you, I'm saying that using expertise as a criterion for "intro X" curriculum is not the correct argument to make in this case. Also the dunning-kruger effect doesn't actually exist, at least not in the way you show it.
I'm a medical doctor. Dunning-Kreuger effect is rampant in medical interations. Really your mother's friend's cousin is a nurse's aid and she says I should treated your pyelonephritis with cranberry juice? Sure, I'll hold the ICU bed ready for you. Seriously, though, it's a matter of trust - people will trust their mother's friend's cousin because their connection, however thin, is thicker than it is to the stranger in the white coat. Similarly, they will trust their passing familiarity with stats moreso than the funny guy in the gray polo.
Wasn’t it that the statistical proof on the Dunning Kruger Effect was not good? I mean, I think it can be a helpful concept to describe your own ignorance and maybe a learning curve you’re on. When it became an argument on social media I always thought ppl don’t realize how much of a bomerrang insult that is. By the way - thanks for the videos. I find them to be really helpful. I used to work in media business and the most useful think I learned is, that you can create wonderful things if you hire a team of experts and everyone is allowed to put their expertise in the project. I know that there are a few sciences having a discussion about reproducibility etc. But I think your rant hints on a bigger problem. You need to work on your acceptance of the expertise of others. Especially when you are in an environment where some fields within a group may have only one person to represent it. I would argue even outside filmmaking nowadays most things are so complex that they are influenced or done by many people to be completed. Yet, we still have this ideology of the single genius saving the day. In media we celebrate the show runner, when you have a “writers room” coming up with ideas and tons of people to realize them. In science we celebrate the genius paper from the first person who gets his name on it. Ignoring the others that follow, kind of ignoring the work that is quoted in the paper and most definitely not talking about all the failed papers and trials that helped the successful ones to avoid pitfalls.
I do think that's part of the problem--only one person gets credit when so many deserve it. And I think many people realize that, so it ends up being a zero-sum game to get that credit.
During engineering school, I mentioned I took 'statics' (introductory branch of mechanics/physics) and soooo many people pipe up. I've also taken statistics....... 😂
This very gatekeepy to the point of vanity. Especially since you said this a matter of an intro stats curriculum, and you're railing against the person with years teaching this curriculum.
Having years of experience doing something stupid doesn't make you an expert. Walter Jackson Freeman II did thousands of lobotomies, but that doesn't mean I'd trust him with an ice pick to my cranium. Of course I'm vain. I'm a RU-vidr.
I am a data analyst and bioinformatician in microbial ecology, I use R every day but I am not a statistician. I am teaching stats for bio this semester for the first time and the course was structure the same way as a similar level course I took 30 years ago… And, by the way, I love your content and I hope you keep doing more videos. It helps me as many others!
Hello professor, my name is Teslim. How can I contact you directly?. I have filled a contact form in your website waiting for your reply.. Thank you so much
I'll share what I think, knowing I could be wrong and also knowing that, even if I'm right, the probability that you'd take my opinion to heart is close to zero. I think you've done more than your part in the fight against bad statistics education. I think it might be time for a career change. I think academia is a completely toxic work environment. I think you're overworked and underpaid and that your expertise and services are outrageously underappreciated. I think you're basically in a toxic abusive relationship with your employer and your colleagues. And I think you're probably smart enough to figure out how to do and teach statistics once you've freed yourself from that hell hole. There are many ways to do that which would remunerate you fairly and more importantly wouldn't be detrimental to your mental health. And I think you have family and friends and students who are negatively affected by your mental health struggles too. Taking care of yourself isn't a selfish thing to do.
Quant Psych: Huh. Isn't it crazy how people think they know more about statistics than they really do, thanks to the Dunning-Kruger Effect? Comment Section, having a full-blown Imposter Syndrome episode: Oh my god, he's onto me. I shall never speak about statistics again.
Hey! Quick question, if it's bad practise to dichotomize continuous variables (which totally makes sense), then how does one estimate say for example, the effect of weather shocks on agricultural produce? This is something I have been working on, where I take within entity deviations from the long run average and then dichotimize it based on an officially documented cutoff. But if there is no cutoff, how do I define what is a shock in my case?
So ok, just came from the last video… been already 4 years into my stat phd… and about halfway I had same realization. Everything something like a linear model… now your approach kind of confirms it, at least to me. I’m curious about your book! Will take a look at your GLM videos!
So funny you mention derivatives wrt eigenvalue. Just had to learn some of it for the real part of a complex eigenvalue this summer for application in my own PhD (in stats) research. Well… I FULLY agree, after quite some years studying stats and digging myself more and more into research, intro stat curriculum makes no sense. Everything feels taken out of a magic box without the fundamentals. This is my first interaction with your content, I’ll check your stats teaching directions, now that I’m also curious about the topic.
I heard one paper make the argument that the way we teach statistics is like teaching surgeons to perform operations without them knowing biology. It's ridiculous!
One suggestion to make sure is in your lessons on random effects could be to clarify the difference between a random effect and an interaction, although it may be too much to go into a full explanation I feel like a disclaimer/warning could at least be valuable. I feel like you explained random effects very well, but an inexperienced person coming into this may follow the theoretical explanation and then when you show the plots think, "OH! I know how to do that, you run 'lm(y~x1*x2), vs lm(y~x1+x2)!' and then be in for some pain later. I remember learning stats and having that misconception for a brief period. Thank you for the content you produce, it is valuable and appreciated, your content is excellent for learning and great for back-to-the-basics review.
I feel your pain! Sad that this world is so fucked up. Everything just seems to be done as a quick fix for the immediate problems. I'd love to learn statistics from you! Sadly for now I can't really afford it because it costs a bit less than I make a month. But I have a dream to save this money in the future and take your course. Somehow the field of statistics really fascinates me.
@@QuantPsych I think it's the course simplistics (Live) in november, right? the program looks awesome :) oh wow.. I didn't see that there's a self-guided option for 95 euro. That's really nice! thanks for such affordable prices! Would be fun to be in the live class with you though) maybe some day in the future.
I'm definitely not an expert in statistics. I'm just an applied data scientist. The thing that blew my mind during my master's degree, was when a statistician (meteorologist) showed as multiple papers as a case study on econometrics. It took two weeks as there were a lot of papers. He wanted to showcase HOW MANY GRAVE MISTAKES were made by all sorts of prestigious -nonetheless- people, particularly on regression analysis. At first the whole class was skeptical. After the first lecture it was really interesting for everyone. I didn't have an idea by then how bad this thing is. It goes without saying that all those papers were reviewed and were published by the best of the best journals. This grounded me hard and it was one of the most humiliating experiences I had. And I'm thankful for that because I thought I was really good at these stuff.
I immediately doubt the expertise of anyone that looks puzzled if I mention "sigma algebras" or "pullback measure". I mean, can you really call yourself a statistician is you don't know what a Radon measure is? ;)
Your latest video brought me here. Now, I get it what is the problem, you are absolutely correct, we are all taught Statistics following the old (and still the same) curriculum under graduation. Things need to be changed! But how ? :(
