Thanks to a number of your videos, I was able to write chapter 4 of my dissertation over the course of two weekends. I get it and feel confident in my work thanks to you! I'm definitely recommending these videos to our students! Thank you.
Thank you so much!!!! I was having troubling using SPSS for my biostatistics class, and it was my first time having to manipulate real world data using this method of testing. Thank you again! Major help.
Thanks so much for these wonderful lectures. For the spreadsheet of RStats Effect Size Calculator for t Tests, somehow the version provided in the link is missing quite some components as demonstrated in the lecture, such as cohen’s d calculator. How could I access the spreadsheet presented in the lecture?
special thanks to you for your amazing teaching methods, I've been told that we don't have to check for normality of data more than 40 because the k-s test would be almost always significant, is that correct?
Actually, I think that is the Shapiro-Wilk and the value is 50 (although that is approximate). I would use K-S with samples of 50 or more, but definitely check a Q-Q plot and/or box plot for visual confirmation. You can also divide the kurtosis value by the SE of kurtosis (same pattern with skewness) and look for values greater than 1.96 as indicators of non-normality. Good luck
The example asked the question 'was there a difference between the two training techniques?.' However, you have concluded that the 'findings do to support the idea that clicker training is more effective than traditional food reward training.' Should it be concluded that 'there is no significant difference in performance by the puppies clicker and food-reward group'??? If we want to know if clicker is better than food-reward method, isn't it a directional comparison and thus would require one-tail independent sample test? Please explain my doubt. Thank you Dr. Daniel for all the wonderful material and superb teaching, I have learned more from this channel than in my class.
You can start with an "agnostic" alternative hypothesis asking about differences without indicating direction. Once you have done the test, you now know the direction (by checking the means) and it is appropriate to say so. In this case, we have two competing techniques, so we might not know which was better at the outset. If we were comparing an experimental group to control, we would have a cleared expectation of which one would change (exp) and which would not (ctrl). But if you asked if Clicker was BETTER than Food in your hypothesis, then you would use a one tail test with Food as your reference group.
So a Large Cohen D's effect size simply means that the statistically significant difference in means is indeed LARGE, correct? So what is it one could conclude other than size of effect using this Cohen D's effect size in light of the example you provided in the video?
We could use effect size for a power analysis ("how many people do I need for this new study I plan to do?") and also for comparisons between studies. They could be used in a meta-analysis and they could be interpreted directly to indicate magnitude of change in a study. In this example, the matching effect sizes between independent and paired t tests illustrates the increased power of the paired model.
The updated spreadsheet is in the google drive folder as "Effect_Size_&_tTest_MultiTool_RbD.xlsx". It has been dramatically updated from this older video.
You could use GDP as a DV but the IV has to be only two categorical groups. If you have two scale variables (level of consumption) you could do a correlation. Good luck
You can compute critical values for t tests as low as 1 degree of freedom (n = 2), but there is a lot of error. There is no minimum for a t test, but n >30 minimizes the error in the test. So bigger is better.
Thanks for the comment. I would like to share your acronym with my students. Clever. But if you think that stats in SPSS is suffering, try doing them by hand...whew. :o)
Ah...the problem of probability. Essentially, even though your effect size is large, the test is not powerful enough to "see" it. Adding more participants would give the test sufficient power to detect the large effect.
Research By Design So if for example our samples is 50 instead of 16, we can expect that the t stat will be significant and will have a large effect size of .70?
If (a) the Likert scale has 5 or more item options, and (b) it has multiple items that can be combined, then you can treat it as scale data for the purposes of a t test (see Dawes (2008) Do Data Characteristics Change According to the Number of Scale Points Used)
Thanks to a number of your videos, I was able to write chapter 4 of my dissertation over the course of two weekends. I get it and feel confident in my work thanks to you! I'm definitely recommending these videos to our students! Thank you.
Excellent! Love hearing when someone gets closer to finishing a dissertation. Glad that the video was helpful. and thanks for the referrals.
btw - you're 100% the best stats teacher in UA-cam #thankyou
Thank you so much for watching. Tell your friends and professors. :o)
Thank you so much!!!! I was having troubling using SPSS for my biostatistics class, and it was my first time having to manipulate real world data using this method of testing. Thank you again! Major help.
You're so welcome!
Awesome series of tutorials! Thank you so much! Strongly recommended!
Glad you like them!
Love your vids, well explained, detailed and easy to understand!
Much appreciated!
This an amazing tutorial. I have learnt a lot . Thanks so much
Thanks so much for these wonderful lectures. For the spreadsheet of RStats Effect Size Calculator for t Tests, somehow the version provided in the link is missing quite some components as demonstrated in the lecture, such as cohen’s d calculator. How could I access the spreadsheet presented in the lecture?
Thanks for your video and I learn from your you Tubes
What happens when Levene's test indicates unequal variance? Do we report the values along side the variance equality unassumed?
special thanks to you for your amazing teaching methods,
I've been told that we don't have to check for normality of data more than 40 because the k-s test would be almost always significant,
is that correct?
Actually, I think that is the Shapiro-Wilk and the value is 50 (although that is approximate). I would use K-S with samples of 50 or more, but definitely check a Q-Q plot and/or box plot for visual confirmation. You can also divide the kurtosis value by the SE of kurtosis (same pattern with skewness) and look for values greater than 1.96 as indicators of non-normality. Good luck
Much obliged sir
The example asked the question 'was there a difference between the two training techniques?.' However, you have concluded that the 'findings do to support the idea that clicker training is more effective than traditional food reward training.' Should it be concluded that 'there is no significant difference in performance by the puppies clicker and food-reward group'???
If we want to know if clicker is better than food-reward method, isn't it a directional comparison and thus would require one-tail independent sample test?
Please explain my doubt.
Thank you Dr. Daniel for all the wonderful material and superb teaching, I have learned more from this channel than in my class.
You can start with an "agnostic" alternative hypothesis asking about differences without indicating direction. Once you have done the test, you now know the direction (by checking the means) and it is appropriate to say so. In this case, we have two competing techniques, so we might not know which was better at the outset. If we were comparing an experimental group to control, we would have a cleared expectation of which one would change (exp) and which would not (ctrl). But if you asked if Clicker was BETTER than Food in your hypothesis, then you would use a one tail test with Food as your reference group.
Hello, I wonder where can I download the wonderful effect size spreadsheet ? I can't quite locate it on the website.
Keep going!
I cannot find Rstate in the list of google drive , what can I do ?
So a Large Cohen D's effect size simply means that the statistically significant difference in means is indeed LARGE, correct? So what is it one could conclude other than size of effect using this Cohen D's effect size in light of the example you provided in the video?
We could use effect size for a power analysis ("how many people do I need for this new study I plan to do?") and also for comparisons between studies. They could be used in a meta-analysis and they could be interpreted directly to indicate magnitude of change in a study. In this example, the matching effect sizes between independent and paired t tests illustrates the increased power of the paired model.
RStatsInstitute thank you for the explanation would love to see a video on power analysis
thank you so much
Thank you so much sir for the great workshop.
Kindly help me to download Rstatistics calculator for one sample t-test and also Cohen's d effect.
The updated spreadsheet is in the google drive folder as "Effect_Size_&_tTest_MultiTool_RbD.xlsx". It has been dramatically updated from this older video.
Can we do it with the data of GDP of 40 + countries as dependent variabe and factors of GDP like investment, consumption etc as independent variable
You could use GDP as a DV but the IV has to be only two categorical groups. If you have two scale variables (level of consumption) you could do a correlation. Good luck
@@ResearchByDesign thanks a lot but i am having problem while conducting t tests
I thought t-tests were used for sampes smaller than 30
We use t distribution when sample size less than 30 (n
You can compute critical values for t tests as low as 1 degree of freedom (n = 2), but there is a lot of error. There is no minimum for a t test, but n >30 minimizes the error in the test. So bigger is better.
SPSS:
(S)ubstantial
(P)sychology
(S)tudent
(S)uffering
Thanks for the comment. I would like to share your acronym with my students. Clever. But if you think that stats in SPSS is suffering, try doing them by hand...whew. :o)
Why do we have a Large effect size even if the T stat is not significant?
Ah...the problem of probability. Essentially, even though your effect size is large, the test is not powerful enough to "see" it. Adding more participants would give the test sufficient power to detect the large effect.
Research By Design So if for example our samples is 50 instead of 16, we can expect that the t stat will be significant and will have a large effect size of .70?
@@empaulstube6947 In theory, yes. But practically, anything can happen if you were to conduct the same experiment with a larger sample size.
perfect
what if you have 117 respondents who answers likert scalling survey
If (a) the Likert scale has 5 or more item options, and (b) it has multiple items that can be combined, then you can treat it as scale data for the purposes of a t test (see Dawes (2008) Do Data Characteristics Change According to the Number of Scale Points Used)
CovidImages need to be invested more than half19