- 264
- 932 171
Mikko Rönkkö
Finland
Приєднався 30 гру 2006
This channel contains teaching material used in the doctoral programs taught at Jyväskylä University School of Business and Economics and Aalto University, Finland.
Basic course syllabus: mycourses.aalto.fi/mod/resource/view.php?id=492116
Advanced course syllabus: mycourses.aalto.fi/mod/resource/view.php?id=504734
Basic course syllabus: mycourses.aalto.fi/mod/resource/view.php?id=492116
Advanced course syllabus: mycourses.aalto.fi/mod/resource/view.php?id=504734
Simulations in Stata
This screencast explains how to do statistical simulations in Stata. The Stata file can be found at osf.io/w5z7f
Переглядів: 360
Відео
Simulations in R
Переглядів 3772 місяці тому
This screencast explains how to do statistical simulations in R. The R file can be found at osf.io/5uep6
Reading research articles with Microsoft Edge and Copilot
Переглядів 1 тис.5 місяців тому
This is a quick screencast that shows how you can use Microsoft Copilot in the Microsoft Edge browser to access GPT-4 for free and use GPT to help you understand research articles.
Metodifestivaalit 2023 Confirmatory Factor Analysis talk
Переглядів 5198 місяців тому
A live recording from Metodifestivaalit 2023 conference. I was asked to talk about confirmatory factor analysis. I give a brief conceptual introduction to the topic after which I briefly explain my diagnostics workflow with an empirical example. Link to slides: osf.io/7bh5p
AoM 2023: Everything You Wanted to Know About Moderated Regression (But Were Afraid to Ask)
Переглядів 7329 місяців тому
Get slides and materials at www.jeremydawson.com/pdw.htm This is a live recording of a professional development workshop on mediation, given by Jeremy Dawson and Mikko Rönkkö at the Academy of Management Meeting 2023 in Boston. Although the testing of interaction effects via moderated regression is commonplace in management research, many misunderstandings and gaps in knowledge persist, and kno...
GPT in higher education
Переглядів 1,3 тис.Рік тому
I was involved in creating the Jyväskylä University School of Business and Economics (JSBE) policy on the use of GPT and other language models in teaching. In the video, I talk about GPT and present four ways it could be used in higher education. Finally I tell about the JSBE policy and why we decided to have each specific guide line. The four use cases: 1) Cheating by writing entire essay assi...
Using AI to understand articles
Переглядів 2,4 тис.Рік тому
In the past, I taught students that when they read an article on a topic they are not familiar with, they should start by finding the terms and definitions. In this video, I show another approach where I use ChatGPT to explain the article to me. This can provide a shortcut to understanding articles, but is not a substitute for reading the article because the AI sometimes goes horribly wrong.
GPT yliopistossa
Переглядів 889Рік тому
Olin mukana tekemässä Jyväskylän yliopiston kauppakorkeakoulun (JSBE) politiikkaa GPT:n ja muiden kielimallien käytöstä yliopiston opetuksessa. Esittelen videossa GPT:n ja neljä tapaa miten sitä voisi käyttää yliopiston opetuksessa. Lopuksi kerron JSBE:n politiikasta ja miksi päädyimme näihin linjauksiin. Käyttötavat: 1) Huijaaminen kirjoituttamalla kokonaisia esseetehtäviä tekoälyllä. 2) Artik...
Linear model implies a covariance matrix matrix
Переглядів 414Рік тому
The video explains the application of path analysis tracing rules for calculating covariances in statistical models. It highlights the extension of these rules, originally used for correlations, to also encompass covariances. The significance of two-headed arrows in these rules is emphasized, indicating their role in quantifying correlations, covariances, and variances. The video demonstrates h...
Empirical identification checks (with Stata)
Переглядів 516Рік тому
Empirical identification checks (with Stata)
Starting values and model convergence
Переглядів 1,5 тис.Рік тому
Starting values and model convergence
Non-convergence in structural equation models and other models
Переглядів 1,2 тис.Рік тому
Non-convergence in structural equation models and other models
How to practice troubleshooting non converged models
Переглядів 1,4 тис.Рік тому
How to practice troubleshooting non converged models
Introduction to model convergence playlist
Переглядів 565Рік тому
Introduction to model convergence playlist
Troubleshooting Zotero references in a document
Переглядів 4,8 тис.Рік тому
Troubleshooting Zotero references in a document
AoM Meeting 2022 live presentation about modelling of entrepreneurial orientation.
Переглядів 1,2 тис.Рік тому
AoM Meeting 2022 live presentation about modelling of entrepreneurial orientation.
Conditional fixed-effects logistic model
Переглядів 5 тис.2 роки тому
Conditional fixed-effects logistic model
Numerical integration in ML estimation
Переглядів 2972 роки тому
Numerical integration in ML estimation
Matrix presentation of structural equation model
Переглядів 9302 роки тому
Matrix presentation of structural equation model
Estimation of nonlinear mediation models
Переглядів 9182 роки тому
Estimation of nonlinear mediation models
this is really well explained and organized. i think i am finally getting it.
Hello, Mikko. I have a question about SEM. In recent years, an increasing number of people have been using time series to build Structural Equation Models (SEMs). I wonder if it is appropriate to estimate parameters using the Maximum Likelihood method. I think there is an obstacle, since classic SEM theory requires that different observations are independent. For example, if I survey N people and ask each person for their math scores and English scores, while these scores may be correlated with each other, different individuals’ scores should be independent. This assumption, along with the assumption of normal distribution, ensures that the Maximum Likelihood estimator is consistent, asymptotically unbiased, and asymptotically efficient. It also allows the application of Z-tests and chi-square tests. However, time series data do not meet this assumption. For instance, consider two variables: GDP(t) and Investment(t). We can consider (GDP(2020), Investment(2020)) and (GDP(2021), Investment(2021)) as two observations (or two participants’ scores). These two observations are obviously correlated. Therefore, in such a case, if Maximum Likelihood estimation is adopted, will the estimators still possess these desirable properties? If not, is there any other method that could be used? Thank you very much.
Subject: Request for Data Set and Guidance on Random Effects Model for Longitudinal Data Dear Mikko Rakko, I hope this message finds you well. My name is [Your Name], and I am currently a PhD student in a nursing program. I am reaching out to kindly request if you have a data set available that I could use to check assumptions on the random effects model for longitudinal data. I have been following your UA-cam videos and they have been incredibly helpful in enhancing my understanding of various statistical concepts, including the assumptions regarding Random Effects Models (REMs) and others. Your guidance has been invaluable to me as I delve deeper into my research. I would greatly appreciate any assistance or advice you could provide in this regard. Your expertise in this field would be immensely beneficial as I continue my studies and research in nursing. Thank you for considering my request. I look forward to hearing from you soon. Warm regards, Ritbano Ahmed
Subscribed. Very good
can I find somewhere examples of random coefficient models where the variable of the random coefficient is not continuous but categorical? ideally written with STATA or SPSS?
Hi. What do you think about stationarity when dealing with GMM. Does it matter if you use system gmm even if data is not stationary?
thanks bro
You are welcome
Smooth! Understandable, solid.
You are welcome!
fking cool, man
You are welcome!
I have a question as follows and the answer is c). Can you please help to explain why? The Spanish Army used to have a 2-year forced-conscription military service at 21 years of age. Because each year there were more conscripts that were needed, recruits could apply for military service exemption. Not surprisingly, there used to be more applications for exemptions than exemptions available, and the army used to make a lottery to randomly choose which applicants were awarded the exemption (and which applicants were forced to serve). A comparison of the earnings of exempted applicants and unsuccessful applicants some years after the service would allow us to evaluate: a) The average earnings effect of serving in the army. b) The average earnings effect of serving in the army for those who served c) The average earnings effect of serving in the army for those who did not serve
This is a really good question. You need to think about what the two groups are. You have everyone who did not serve, but only a subset of those that did. Therefore you cannot generalize to the full population or those that did serve. Here is a GPT4 generater explanation: The correct answer is c) "The average earnings effect of serving in the army for those who did not serve." This might initially seem counterintuitive, but it's based on a fundamental concept in econometrics known as the "Local Average Treatment Effect" (LATE). In the scenario described, there is a random assignment of military service exemptions through a lottery. This randomization ensures that, on average, the groups of exempted (treatment group) and non-exempted (control group) applicants are similar in all respects except for the treatment-here, serving in the military. This similarity is crucial because it mimics the conditions of a randomized controlled trial, allowing us to infer causality from the comparison between the two groups. Now, let's break down why each answer choice is what it is: a) "The average earnings effect of serving in the army" suggests we are looking at the impact on all individuals, regardless of their inclination or circumstances that led them to apply for the exemption. This is not what the comparison would reveal since the analysis only includes individuals who applied for the exemption and were subjected to the lottery system. b) "The average earnings effect of serving in the army for those who served" might seem like a reasonable answer, but it's not the focus of this comparison. This is because the analysis isn't solely focused on those who served; it also includes those who were exempted. The aim is to understand the impact of not serving (being exempted) versus serving. c) "The average earnings effect of serving in the army for those who did not serve" is correct because the analysis effectively compares individuals who wanted to be exempted (and thus, by extension, did not want to serve). Those who win the lottery (and are exempted) serve as the treatment group, and those who lose (and thus serve) are the control group. The comparison then reveals the effect of not serving on the group that applied for exemptions but had to serve due to losing the lottery. In essence, this setup allows us to estimate the impact of military service on those who would have preferred not to serve but were compelled to do so because they did not win the exemption lottery. It's a subtle but important distinction in understanding the causal effects in this scenario.
Sir, thank you so much for this video. You are amazing. I'd like to ask a trivial question, if you (or anyone who sees this) could answer me: when differentiating, why doesn't beta_0 get canceled as well? I mean, when we subtract y_t-1 on both sides, wouldn't it imply in getting a negative beta_0 canceling the positive beta_0 on the right hand side? Thank you so much!
First differencing does eliminate beta_0, but there is an error in my slides. My video on first differencing has the correct equation ua-cam.com/video/hQWSh_j3Oy0/v-deo.html
Hi, why is R studio producing different results while using the same call.
I do not understand the question. What call are you referring to?
Hi! Why would we not use a t test in this case?
You could use t for testing a single constraint. But in practice we use F because it can handle both single constraint and multiple constraints. A single parameter F distribution is simply a square of the corresponding t distribution.
@@mronkko thanks!! Then why in some cases we should use a wald test instead of a t test like in this video?
@@junbeombahk3668 I am not using a t test in the video. The test is a single parameter Wald test. See this for an explanation of the tests ua-cam.com/video/AbhwpFX2Xdw/v-deo.html
With “characteristic” you mean “the effects of treatment”? Thanks in advance!
No, I meant that we measure the characteristic of each individual after the treatment. We cannot measure the effects of treatment. We can only measure the characteristics of interest and then use these measured characteristics to calculate an estimate for the average treatment effect.
I think a lot of applied researchers see/hear the phrase "alpha underestimates reliability" and think that alpha is a good default because it is conservative in being careful to not overestimate reliability. However, getting a nearly acceptable .70 from a covariance matrix describing two orthogonal factors is scary (2:46). The total score of this scale could be a meaningless average between two unrelated constructs. As someone who does applied research, it was important for me to read the McNeish 2017 article and do my own simulation to see that alpha can actually overestimate reliability as well.
Alpha does not overestimate reliability in the example that you mention, but you are correct that alpha can be misleading. The problem with alpha is that it does not quantify whether the variables measure one thing or multiple things. If you have a score generated from scales measuring two orthogonal factors, the scale score is almost certainly meaningless. Nevertheless, we can estimate the reliability of that scale score (think e.g. test-retest reliability). Alpha underestimates this reliability. The problem is that an acceptable alpha does not guarantee that you have a useful scale score. At minimum, you should also do a factor analysis as well.
I have been trying to find workflow videos on regression analysis for a while now, this is the first (and only one) that I found. It helped me immensely, thank you.
You are welcome. It is surprising that very few people teach how to actually use the analyses in empirical research practice.
@@mronkko that's true. Most videos cover only interpretation of results or are focused on let's say one part of the analysis but no one covers the whole process in a single video, with a single dataset. Just an idea - You could maybe consider doing a workflow series focusing on how to do analysis with different combinations of explanatory/response variables? Let's say one categorical explanatory variable, 1 exp and 1 quantitative, 2 categorical exp var, and so on. And the same logic with explanatory - quantitative vs qualitative. I'm not sure if you've done it already, but it'd be so so helpful! Thanks again, keep up the good work. I wish you good luck!
So you have a general factor G that you confirm with a bifactor CFA. How would you calculate a score for G? For example, in your bifactor model at 14:39, how would you estimate somebodies Verbal Intelligence score? I ask because I am interested in estimating these general scores for each person in my dataset, and then adding those estimates as columns in my dataframe. Curious whether you'd take the mean of the total 7 items *or* calculate separate means for Verbal Comprehension and Working memory, add them, and divide by two. Or maybe something else? Great content as always! It's the only content on youtube that really feels like a high quality grad-stats class. I've suggested to other students to look for the guy with the blue hoodie if they are looking for stats enlightenment :)
You can use factor scores or take a mean of all items to get a score for G. However, make sure you understand the challenges in using scores to represent factors: I have a video on factor scores that explains this issue. Whether you take the mean of all items or the mean of two subsets and then a meanwill not make a of these means meaningful difference. This is because taking a mean of all item vs the mean of subscale means approaches are just two different ways of calculating a weighted average. In practice, indicator weights rarely make a difference as long as the items are at least moderately correlated. Thanks for the compliments! These videos are actually a part of my doctoral level course and most of the content is something that I have also presented in class..
So clear!! Thank you so much for your effort!
You are welcome!
Amazing video, thanks from Norway :)
You are welcome!
Hi im korean student studying data science in Spain in spanish. I'm literally suffering due to language barrier. And your videos here help me a lot. Thank you so much❤❤❤
Thanks. I am glad that the videos are helpful.
Like the others here I appreciate the clarity. Thanks for making this video.
You are welcome!
Thanks for making your helpful videos freely available
You are welcome
Good Job 👍👍👍
Thanks!
Excellent content! I came across your post linking This video in a lavaan Google group. However, I’m still a little confused. It sounds like what you are saying undermines the use of SEM, no? Presumably, factor scores are passed into the structural part of an SEM. Yet, you are saying that summing up the factor indicators is sufficient for representing a construct for a given observation. Any clarification would be helpful! Thanks
SEM does not use factor scores but estimates the relationships between latent variables without estimating latent variable values. " factor scores are passed into the structural part of an SEM" is thus not what SEM does.
Excellent tutorial
You are welcome!
baadiya tha
You are welcome, I suppose.
Thank you Professor for wonderful video! It would be even better if you prepare video from your paper of simulation! It will be productive for us!
Here you go: osf.io/mjzyw I actually recorded these screencasts specifically for that paper. I will link it to the video description after it is published. (Or if it is published)
Thanks for the video especially the parallel processing command. I have two questions. 1: isn't better to simulate the standard errors separately instead of calculating them from the simulated beta estimates? They might not be always identical. 2: How to store whether an SEM model converged or not when simulating? I am currently (probably) using a sub-optimal way to do this. I am currently simulating the RMSEA and Chi-square to see if they are missing and also checking whether standard errors are missing. Is there any way to directly store the convergence error in the simulations' files as a separate variable?
1) I do not understand the first question. SEs are calculated based on the simulated datasets. 2) You can get convergence status from e(converged). For example. run webuse census13 sem ( <- mrgrate dvcrate medage), standardized ereturn list
@@mronkkoThanks. I meant that instead of getting the standard deviation of estimates by the command summarize, we directly simulate the standard error for beta from each sample and store them. That is instead of (simulate _b) and using the summarize command to get the standard deviation of bs, we directly simulate the standard errors for each sample (simulate _b _se). Do they differ?
@@Rezayyyyyyyyy It depends on what you want to study with the Monte Carlo simulation. If you want to study the precision of estimate, you look at the SD of the estimates and ignored the SEs. If you want to study whether the SEs are unbiased, you store the SEs and compare the average SE against the SD of the estimates.
Thank you very much Professor. I have been looking for a tutorial like this for some time. It is very helpful
You are welcome
Any chance there is a simulations in SAS coming? Thanks for all of the content you make.
Thank you Mikko!
Thank you for such a clear explanation.
You are welcome!
Thanks a lot for this awesome explanation! It helped me a lot with my thesis and saved me a lot with my help!
You are welcome.
Thank you so much for your explanation, sir. This video was very helpful!
You are welcome
I like your videos soooo much. thank you!
You are welcome!
Great job, but what about missingness that exist in a single column and also it's more than 50%? Is deep models like GAN would be useful for imputation?( In time-series prediction). Many thanks🙏
I assume GAN refers to some kind of neural network. Imputation works regardless of the amount of missing data, under these three conditions: 1) You are doing multiple imputation and not single imputation so that you can quantify the uncertainty introduced by the imputation process. 2) The imputation model contains all features of your data that are relevant for the analysis. 3) The missingness does not depend on the missing value itself. (i.e. data are MAR or MCAR) I do not really see what neural nets would add over throughfully developed imputation model but they are likely to increase sample size requirements.
@@mronkko "Hi again Mikko, I'm tackling a unique challenge with my dataset and believe your insights could greatly help. Could you share any contact info for more brief discussion? Thanks!"
@@mohamadmatinhavaei9859 I take consulting orders through instats.org/expert/mikko--rönkkö-829.
amazing channel, amazing content. simply superior
Thanks!
A good example is geography. The amount of rainfall (observed) causes the agricultural value (latent/unobserved) of a given municipality.
Rainfall affects agricultural value, but it does not measure agricultural value. You could use rainfall as an instrumental variable. If aggricultural value is the dependent variable, then I do not see how rainfall would be useful for its measurement.
Great explanation, i'm forever gratefull Mr. Rönkkö !!
You are welcome!
THANK YOU SIR
Most welcome
Bookmarking video, thanks
You are welcome!
You saved me for my Quantitive Theory Methods exam. Thanks!
You are welcome!
Many thanks for this helpful video; I have long asked myself the value of centering for linear regression.
You are welcome.
Thank you, great explanation!
You are welcome!
Thanks a lot sir
Most welcome
Thank you so much for this content. I am finishing my doctoral thesis work and this has been so helpful for me to better understand the differences in the estimation of the mediation models under a non-linear framework. I was wondering if you had any thoughts on mediation analysis when both the mediator and the outcome are modeled as time-to-event outcomes. If we face similar issues as in the non-linear setting you discussed or entirely new issues when trying to interpret the relative NDE and NIE. Again thanks for the content.
Sorry for a delayed response. In your case, you need to clearly define what the causal effect of interest is conceptually. For example, if you are assuming that X causes M with some delay and M cause Y with some delay, you could be intersted in the total time for Y to occur and how it depends on the value of X. I would start by predicting the time to Y with different levels of X and M and then comparing. But this really depends on your research question.
Hi, would not be better not to include the sample size to calculate degrees of freedom but having K-1 instead? This might imply that we can compare two models when outliers are dropped in one but the number of parameters is the same (comparing non-nested models using F test). So, the question is can we compare two models with only different numbers of observations (k being the same) using the F test?
No. F tests is designed to be used with nested models. This means that they 1) use the same observations and 2) the same variables, but differ in parameters.
Kiitos hyvästä videosta! Mikä toi text editori oli mitä käytit tossa?
BBEdit. Mutta toi tekstieditorin kautta käyttäminen on turha välivaihe koska nykyään nämä työkalut ymmärtää suoraan PDF:ää.
very useful! keep up the good work, can we get a video about predictive modeling?
What do you mean by predictive modeling? Like neural networks?
Thank you for your reply! So I'm working on a project where I would like to predict the outcome of a certain variable based on specific patient characteristics, i have been using linear and logistic regression to check for predictive power of the patient characterisitcs i am interested in. What i know about this subject is from googling and reading about it independently, i would appreciate if you could walk me through the details. Neural networking is another technique i would like to learn about, but if I had to choose, i would like to master the classical way using regressions.
@@alikh.3225 I do not teach or do predictive analytics myself. ( I have reviewed a few prediction papers, though). For these reasons, it is unlikely that I would be posting anything about that on the channel. But I can recommend a couple of books on prediction Hastie, T., Tibshirani, R., & Friedman, J. H. (2013). The elements of statistical learning: Data mining, inference, and prediction (2nd ed. 2009. Corr. 10th printing 2013 edition). Springer. statweb.stanford.edu/~tibs/ElemStatLearn/printings/ESLII_print10.pdf James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An Introduction to Statistical Learning (Vol. 103). Springer New York. doi.org/10.1007/978-1-4614-7138-7
I see, thank you for the resources! Keep up the good work 👍🏼