The quality of recording and instruction is on point. It’s been a long time since I’ve seen a straight to the point walkthrough of setting up an R analysis example without rambling and confusing tangents.
Hello Jon! Appreciate it. In honesty, the quality of recording isn't what I would like it to be, but I feel it's fine enough for what this is. I am glad to hear your second point as my goal was to streamline knowledge as quickly as possible. Soon I'll add in timestamps so you can skip to whereever you like (e.g., a lot of the later videos have the same first couple minutes of creating data, which can be skipped once you get the hang of it).
Thank you for this! Please bear with my long message. I have a dataset that doesn't follow the logic of experimental design. Trying to use operational data to predict customer perception obtained through a survey. Say video streaming for example. Looking at different variables related to streaming quality and using it to predict customers' survey response. So, customers watch many times and each instance of video streaming has different operational data, which makes the analysis a repeated one. I was told each row of video streaming for a particular will have the same perception value obtained through survey for the customer. Will linear mixed model work here? I am guessing operational variables will be random effect and customer ID fixed effect? Thanks!
Hello Williams! Thank you for the nice video. Could you please make a video about performing a post-hoc test (for example with LSD or Turkey test) after the ANOVA? Based on your example, may we perform a post-hoc test like this?: emmeans::emmeans(score, ~ Factor1:Factor2) %>% multcomp::cld(Letters=letters) Thank you. Qui.
Sorry for the delay in response. This thread breaks down the documentation from lme4 fairly well: www.researchgate.net/post/What_does_singular_fit_mean_in_Mixed_Models
Hello, thanks for watching! Unfortunately, I don't have a proper recording setup right now, but not to worry! This tutorial (tutorial 2 by Bodo Winter) covers what you're looking for: bodo-winter.net/tutorial/bw_LME_tutorial2.pdf You may also want to read tutorial 1 on linear regression: bodowinter.com/tutorial/bw_LME_tutorial1.pdf
Does the random factor mean it is a repeated measure or can this also be used to identify random terms in the model. Such as blocks in an experimental design?
Thank you very much for the video, so useful! How can I then run some sort of Post Hoc test to figure out which groups are different? Thanks in advance
Generally, we don't do post-hocs with linear mixed effect models because it's predicting a pattern of change across all of the levels rather than saying something in there is different from something else. We instead visualize the data and the make claims based on that (but again claims saying it changes across levels rather than this level is larger than that level). ANOVAs and such are where post-hocs come in and which ones you do depends on your design. One example is the Tukey Analysis, see here for an idea of how to do it in R: rpubs.com/aaronsc32/post-hoc-analysis-tukey
@@ChadCWilliams Thank you very much! I'm finally starting to understand what lmer does thanks to your explanation. New question: how do you fit your results in a table? My lmer results are confusing because they compare groups using one group as a baseline. My anova results, like yours above, are clearer to present, but somehow I can't fit a table with them. I tried with sjPlot and gtsummary, and neither have worked. Sorry if it's a stupid question, i'm new to R and to mixed effect models, so i'm struggling to make sense of everything.
@@MF-un3wi Hi, first is to look up contrast coding of your groups to stop one of your groups from becoming the baseline group (it will instead compare all groups to the grand average). I don't actually create tables in R generally, so I've little experience with it but it does seem like sjPlot is the way to go - what happens when you try and use this function? Might be easier to figure out sjPlot rather than using a less useful function.
@@ChadCWilliams No problem. In your example then, is there a possibility in make two variables fixed effect (time and type of company) and one random effects (company ID).
Hi Chad, many thanks for the great video. I hope you have time to reply to my concerns :) May I know what it mean when you said "we have a significant factor 1, a significant factor 2 and a significant interaction"? Does it mean: for Factor 1, the mean value of the "Scores" are different between the two levels of this Factor (Factor 1)? same for Factor 2, it also means the mean value of "Scores" are different between the two levels of this Factor (Factor 2), am I correct? For the interaction between Factor 1 and Factor 2, does it mean: among the 4 conditions, there are difference in mean value of "Scores"? Also, from the result, can we interpret anything about the random effect (because I see the 1|Subject in the command but I did not see how you interpret its result, if any. Again, thanks very much, Chad!!
Hello! Yes, you're right about factor 1 and factor 2 main effects. As per the interaction, it's a bit more complicated. Essentially, for factor 1 the effects of the two levels change across factor 2. A good example is height across gender (male and female) and age. There is a main effect in that males are more often taller than females. However, when they are young, the difference is very small if there is any, but the difference in height between males and females grows across age (I.e. as we get older). Make sense? As per random effects. Random effects are essentially 'controlling' for things. So, they are not assessed directly but instead helps us see a clearer picture of the fixed effects.
@@ChadCWilliams Thanks Chad for the prompt response! Really appreciated it! Your example makes sense to me. As regarding the random effect, I understand that they are "controlling" for things, but how can it help us see a clearer picture of the fixed effects? I meant where is it in the result of anova(reg_results) that helps us to see its "controlling" actions.
@@trinh123456 Hello Tien - you are misunderstanding what random effects are, as there is no 'results' for them. Instead, in this video it allows for participants y-intercept to be different and thus only focuses on the change from the y-intercept (i.e., the slope). In this video, I discuss this briefly: ua-cam.com/video/sg_Q3Cz9WRs/v-deo.html. Also, this tutorial (specifically, part 2) article is a good one to understand linear mixed effect models: www.researchgate.net/publication/256097211_Linear_models_and_linear_mixed_effects_models_in_R_with_linguistic_applications
The quality of recording and instruction is on point. It’s been a long time since I’ve seen a straight to the point walkthrough of setting up an R analysis example without rambling and confusing tangents.
Hello Jon! Appreciate it. In honesty, the quality of recording isn't what I would like it to be, but I feel it's fine enough for what this is.
I am glad to hear your second point as my goal was to streamline knowledge as quickly as possible. Soon I'll add in timestamps so you can skip to whereever you like (e.g., a lot of the later videos have the same first couple minutes of creating data, which can be skipped once you get the hang of it).
Hey, excellent video! short, sweet, and I feel as if I have the tools to begin exploring deeper. Great work!
Thanks so much! Glad it helped.
What a great video, thank you so much!
Thank you for the kind words! Glad it helped!
Awesome!
cool!
Clear explanation of the code, thanks! I wish you spent more time on explaining the output.
Ah, good feedback - thanks! I'll add it to the list of to-dos!
Thank you for this! Please bear with my long message. I have a dataset that doesn't follow the logic of experimental design. Trying to use operational data to predict customer perception obtained through a survey. Say video streaming for example. Looking at different variables related to streaming quality and using it to predict customers' survey response. So, customers watch many times and each instance of video streaming has different operational data, which makes the analysis a repeated one. I was told each row of video streaming for a particular will have the same perception value obtained through survey for the customer. Will linear mixed model work here? I am guessing operational variables will be random effect and customer ID fixed effect? Thanks!
Hello Williams!
Thank you for the nice video. Could you please make a video about performing a post-hoc test (for example with LSD or Turkey test) after the ANOVA? Based on your example, may we perform a post-hoc test like this?: emmeans::emmeans(score, ~ Factor1:Factor2) %>% multcomp::cld(Letters=letters)
Thank you.
Qui.
Is there another solution for non-parametric data using lmer but smth else than an anova at the last command?
Hello, I am not entirely sure what you mean. What result are you looking for exactly?
What does the Singular fit warning mean? Should we be worried about it if I get the same thing?
Sorry for the delay in response. This thread breaks down the documentation from lme4 fairly well: www.researchgate.net/post/What_does_singular_fit_mean_in_Mixed_Models
Great video! Could you make a video about interpreting the results?
Hello, thanks for watching! Unfortunately, I don't have a proper recording setup right now, but not to worry!
This tutorial (tutorial 2 by Bodo Winter) covers what you're looking for: bodo-winter.net/tutorial/bw_LME_tutorial2.pdf
You may also want to read tutorial 1 on linear regression: bodowinter.com/tutorial/bw_LME_tutorial1.pdf
@@ChadCWilliams Cool, thanks a lot :)
Does the random factor mean it is a repeated measure or can this also be used to identify random terms in the model. Such as blocks in an experimental design?
No, you are correct - it can be used to identify random terms.
Thank you very much for the video, so useful! How can I then run some sort of Post Hoc test to figure out which groups are different? Thanks in advance
Generally, we don't do post-hocs with linear mixed effect models because it's predicting a pattern of change across all of the levels rather than saying something in there is different from something else. We instead visualize the data and the make claims based on that (but again claims saying it changes across levels rather than this level is larger than that level). ANOVAs and such are where post-hocs come in and which ones you do depends on your design. One example is the Tukey Analysis, see here for an idea of how to do it in R: rpubs.com/aaronsc32/post-hoc-analysis-tukey
@@ChadCWilliams Thank you very much! I'm finally starting to understand what lmer does thanks to your explanation. New question: how do you fit your results in a table? My lmer results are confusing because they compare groups using one group as a baseline. My anova results, like yours above, are clearer to present, but somehow I can't fit a table with them. I tried with sjPlot and gtsummary, and neither have worked. Sorry if it's a stupid question, i'm new to R and to mixed effect models, so i'm struggling to make sense of everything.
@@MF-un3wi Hi, first is to look up contrast coding of your groups to stop one of your groups from becoming the baseline group (it will instead compare all groups to the grand average). I don't actually create tables in R generally, so I've little experience with it but it does seem like sjPlot is the way to go - what happens when you try and use this function? Might be easier to figure out sjPlot rather than using a less useful function.
Can we do this with the plm() function? What would be in the: index = c("","") part? and what would the model = "" and effect = "" be?
Hello, thanks for the question. Unfortunately, I've never used this function before, so am unable to comment on your query.
@@ChadCWilliams No problem. In your example then, is there a possibility in make two variables fixed effect (time and type of company) and one random effects (company ID).
@@Lucas747G I believe this is what you're looking for: ua-cam.com/video/sg_Q3Cz9WRs/v-deo.html
@@ChadCWilliams thank you!
Hi Chad, many thanks for the great video. I hope you have time to reply to my concerns :)
May I know what it mean when you said "we have a significant factor 1, a significant factor 2 and a significant interaction"? Does it mean: for Factor 1, the mean value of the "Scores" are different between the two levels of this Factor (Factor 1)? same for Factor 2, it also means the mean value of "Scores" are different between the two levels of this Factor (Factor 2), am I correct?
For the interaction between Factor 1 and Factor 2, does it mean: among the 4 conditions, there are difference in mean value of "Scores"?
Also, from the result, can we interpret anything about the random effect (because I see the 1|Subject in the command but I did not see how you interpret its result, if any.
Again, thanks very much, Chad!!
Hello! Yes, you're right about factor 1 and factor 2 main effects. As per the interaction, it's a bit more complicated. Essentially, for factor 1 the effects of the two levels change across factor 2. A good example is height across gender (male and female) and age. There is a main effect in that males are more often taller than females. However, when they are young, the difference is very small if there is any, but the difference in height between males and females grows across age (I.e. as we get older). Make sense?
As per random effects. Random effects are essentially 'controlling' for things. So, they are not assessed directly but instead helps us see a clearer picture of the fixed effects.
@@ChadCWilliams
Thanks Chad for the prompt response! Really appreciated it!
Your example makes sense to me.
As regarding the random effect, I understand that they are "controlling" for things, but how can it help us see a clearer picture of the fixed effects? I meant where is it in the result of anova(reg_results) that helps us to see its "controlling" actions.
@@trinh123456 Hello Tien - you are misunderstanding what random effects are, as there is no 'results' for them. Instead, in this video it allows for participants y-intercept to be different and thus only focuses on the change from the y-intercept (i.e., the slope). In this video, I discuss this briefly: ua-cam.com/video/sg_Q3Cz9WRs/v-deo.html. Also, this tutorial (specifically, part 2) article is a good one to understand linear mixed effect models: www.researchgate.net/publication/256097211_Linear_models_and_linear_mixed_effects_models_in_R_with_linguistic_applications
@@ChadCWilliams This was a great explanation of how to think about what a significant interaction term means from a conceptual standpoint. Thank you!
@@AffectiveApe Not a problem! Happy to help.
Can you consult me using R personally? I will pay.