Great video! I really like the way you explain the statistics and Jamovi, you make it really fun. I have a question about normality: My regression failed the Shapiro-Wilk test (and even though the skewness and kurtosis of my data is OK, I have some variables whose plots do not look normally distributed at all). Are there any ways to do a "non-parametric" regression with Jamovi?
Thanks! No, you cannot do a nonparametric regression. You’d have to do group tests like Wilcoxon or Spearman’s rho. However, the GLM is fairly robust for normality issues if your sample size is sufficiently large, like 50+.
@@AlexanderSwan Thank you for your answer! Do you mean "general linear model" or "generalised linear model"? The GAMjl3 jamovi module offers both options and I havent really understood where the difference is. My sample is almost 250 so I should be fine 🙂
Hi, I'm currently trying to do an assignment where we have to test the hypothesis "Men have higher starting salaries than women, even after the effects of education and work experience are controlled for" and I was wondering how you would do it in Jamovi - would education and work experience be covariates? And if so how do you then read the table to see the relationship?
Yes, you’d put all your predictors into the regression equation and then look at your coefficients. If there’s a gender difference even with those other two variables in the equation, you’ve controlled for them and still have a significant relationship with gender and salary.
Hello Mr. Alexander. Are p values in regression analysis (for example like this one in video) adjusted for multiple comparison? Or multiple comparison matters only for direct comparisons with categorical predictor?
They are not generally adjusted for multiple comparisons. Adjusting is generally reserved for post-box analyses, so if your entire model was part of your initial prediction, you don’t necessarily run the risk of a type I error
Thank you; this was helpful. I was confused as to why the R in this example is positive, because the slope is negative; the r, calculated under "Correlation matrix," IS negative. Is this difference because with regression there can be more than one predictor and not all correlations will necessarily be in the same direction (and therefore the sign of R is irrelevant)? Also, I noticed that the standard error of estimate is not provided by jamovi, despite being covered in most introductory statistics books (unless I am missing it). However, I realized that everything is provided -- the SS for the residuals and df -- and so this is easily calculated (except by jamovi 😊).
Hi, Yes multiple R is always positive, as it is an absolute value for the reason you stated. It’s also why R-squared is more important in these situations because it will always be positive. For individual relationships, it’s best to look at the predictors themselves. As for your other question, I’m pretty sure you get the SE for the unstandardized coefficients and they’re there in the video - is that what you’re referring to?
@@AlexanderSwan I am referring to the standard deviation of the predicted Y values (from using the regression formula) from the actual Y values. In other words, the sum of the squared differences between the actual data points and the predicted data points is divided by the df, and then the square root of the result is calculated. (Similar in concept to the standard deviation of the mean).
All variables that exist on the X side of the equation get put into the sections of the kinds of variables they are: continuous or categorical. The control variables you have are so-called because that’s what you’re describing them as, but the regression equation doesn’t actually care that you’re calling them that.
You are a great teacher and this example made it easy to understand. Thank you!
Thanks for the help! You explain everything so well!
Great video! I really like the way you explain the statistics and Jamovi, you make it really fun.
I have a question about normality: My regression failed the Shapiro-Wilk test (and even though the skewness and kurtosis of my data is OK, I have some variables whose plots do not look normally distributed at all).
Are there any ways to do a "non-parametric" regression with Jamovi?
Thanks! No, you cannot do a nonparametric regression. You’d have to do group tests like Wilcoxon or Spearman’s rho. However, the GLM is fairly robust for normality issues if your sample size is sufficiently large, like 50+.
@@AlexanderSwan
Thank you for your answer!
Do you mean "general linear model" or "generalised linear model"? The GAMjl3 jamovi module offers both options and I havent really understood where the difference is.
My sample is almost 250 so I should be fine 🙂
@@cristinagarciamata5081 general linear model
Hi, I'm currently trying to do an assignment where we have to test the hypothesis "Men have higher starting salaries than women, even after the effects of education and
work experience are controlled for" and I was wondering how you would do it in Jamovi - would education and work experience be covariates? And if so how do you then read the table to see the relationship?
Yes, you’d put all your predictors into the regression equation and then look at your coefficients. If there’s a gender difference even with those other two variables in the equation, you’ve controlled for them and still have a significant relationship with gender and salary.
Hypothesis 6 in this assignment is tough hey haha
I have difficulty with vision. How did you get that teal circle around your mouse pointer? thank you!
It’s an app for MacOs called Cursor Pro. You can find it on the Mac App Store.
Hello Mr. Alexander. Are p values in regression analysis (for example like this one in video) adjusted for multiple comparison? Or multiple comparison matters only for direct comparisons with categorical predictor?
They are not generally adjusted for multiple comparisons. Adjusting is generally reserved for post-box analyses, so if your entire model was part of your initial prediction, you don’t necessarily run the risk of a type I error
Thank you; this was helpful.
I was confused as to why the R in this example is positive, because the slope is negative; the r, calculated under "Correlation matrix," IS negative. Is this difference because with regression there can be more than one predictor and not all correlations will necessarily be in the same direction (and therefore the sign of R is irrelevant)?
Also, I noticed that the standard error of estimate is not provided by jamovi, despite being covered in most introductory statistics books (unless I am missing it). However, I realized that everything is provided -- the SS for the residuals and df -- and so this is easily calculated (except by jamovi 😊).
Hi, Yes multiple R is always positive, as it is an absolute value for the reason you stated. It’s also why R-squared is more important in these situations because it will always be positive. For individual relationships, it’s best to look at the predictors themselves.
As for your other question, I’m pretty sure you get the SE for the unstandardized coefficients and they’re there in the video - is that what you’re referring to?
@@AlexanderSwan I am referring to the standard deviation of the predicted Y values (from using the regression formula) from the actual Y values. In other words, the sum of the squared differences between the actual data points and the predicted data points is divided by the df, and then the square root of the result is calculated. (Similar in concept to the standard deviation of the mean).
Where can I put control variables?
All variables that exist on the X side of the equation get put into the sections of the kinds of variables they are: continuous or categorical. The control variables you have are so-called because that’s what you’re describing them as, but the regression equation doesn’t actually care that you’re calling them that.
you talked so fast...
I recommend using the play controls to slow me down if it is too fast!