great explaination for people who want to use the test but dont mind how it work, but just want to use the test. I don't use R but those details are similiar enough.
Good style as intro, but the KS test has many faces, eg. When using it for normality checks, using true vs estimated parameters, critical value calculations, test power in comparison to other tests, etc.
Hi Matthew. Very good video. I need to test only one sample to validade newcomb-benford using k-s cdf test. Have you some idea how can i do that? Thanks!
Hi, i have a little question. I used a K-S test to find out if my data have a uniform distribution, i already obtain my D and my p-value. So what if my alternative hypothesis have an exponential distribution , and i want to determine the power of the test below. Do yo know how can i do it? Thanks!
Will this test work if I am comparing non ordered categories? So if I had the proportions and cum. percentages of people attending different types of schools for example. So say I had a sample in one year and the breakdown of students in those schools, say school types A, B,C, D and another year I sampled the proportions in the school types A,B,C,D. Is the KS test a valid test to use to see if the distribution in the second year is sig. different to the previous year?
Matthew E. Clapham Thanks for replying and thank you for the video. I saw another video on here where the uploader had counts in ordered categories. Specifically he had not confident to very confident. The KS was a valid thing to use there. I just wondered whether you knew if the categories were not ordered whether the KS could be used.
+OriginalJoseyWales It doesn't seem like it would work for non-ordered data, because the test statistic is likely to differ depending on the arbitrary order you choose.
Thank you Matt! I have a question about having duplicates in the data set: how should we use the computed D value if there are many duplicates (say half of the data set)?
when I think about it, if you have duplicate values, then there would be jump in cumulative distribution function and so it would be difficult to calculate vertical distance D between two distributions. One solution I can think of is change you data by adding very small random numbers to your data. that would make sure that no value is repeated .
Excellent description. Best one I’ve seen or heard. Very useful; thank you for posting.
great explaination for people who want to use the test but dont mind how it work, but just want to use the test. I don't use R but those details are similiar enough.
Super simplified, very help. Thank you!
Concise and right to the point. I love it. Thanks.
Thanks for the awesome explanation Matt!! Loved it!
Thank you for uploading it!
It sounds like Tom Hanks is teaching us statistics.
hhh yep ...
great video, thanks. i was not sure why the data would have to be continuous
clearest video to explain KS test!
thank you for the amazing explanation. the only complaint I have is the high-frequency background noise.!
How to calculate the p value(probability of the distance) in R manually? I don't want to use the function ks.test
Hi Matthew, very clear. Rad ! Thanks
very helpful, thank you!
Very useful, thank you!
Excellent. Thank you
Hi Matt, great video!! Loved it!
Just one question, how do you then graphic the two distributions.
Thank you
In one word, lovely!
Good video. In a regression model, what will be my numeric vector for each sample? The p-value? Chi-square?
Good style as intro, but the KS test has many faces, eg. When using it for normality checks, using true vs estimated parameters, critical value calculations, test power in comparison to other tests, etc.
Hi Matthew. Very good video. I need to test only one sample to validade newcomb-benford using k-s cdf test. Have you some idea how can i do that? Thanks!
Hi, i have a little question.
I used a K-S test to find out if my data have a uniform distribution, i already obtain my D and my p-value. So what if my alternative hypothesis have an exponential distribution , and i want to determine the power of the test below. Do yo know how can i do it? Thanks!
Will this test work if I am comparing non ordered categories? So if I had the proportions and cum. percentages of people attending different types of schools for example. So say I had a sample in one year and the breakdown of students in those schools, say school types A, B,C, D and another year I sampled the proportions in the school types A,B,C,D. Is the KS test a valid test to use to see if the distribution in the second year is sig. different to the previous year?
+OriginalJoseyWales The K-S test is for continuous data - I'd use a chi-squared test if you have counts in categories.
Matthew E. Clapham Thanks for replying and thank you for the video. I saw another video on here where the uploader had counts in ordered categories. Specifically he had not confident to very confident. The KS was a valid thing to use there. I just wondered whether you knew if the categories were not ordered whether the KS could be used.
+OriginalJoseyWales It doesn't seem like it would work for non-ordered data, because the test statistic is likely to differ depending on the arbitrary order you choose.
Can someone confirm, I need to find KS score over my data, I am not sure whether it is same as KS test, please help?
Thank you Matt! I have a question about having duplicates in the data set: how should we use the computed D value if there are many duplicates (say half of the data set)?
when I think about it, if you have duplicate values, then there would be jump in cumulative distribution function and so it would be difficult to calculate vertical distance D between two distributions. One solution I can think of is change you data by adding very small random numbers to your data. that would make sure that no value is repeated .
Data are. . . .
Univariate iid data samples (vectors).