Thanks Emma. I was thinking about testing the CTR last night and today all of my questions are answers in this video! A great demonstration of presenting using Notion as well.
Hi Emma, is it true that the difference between two sample proportion follows normal distribution is not because of central limit theorem? it is just because the binomial approximation to normal distribution when np>10 & n(1-p)>10? The reason why I am asking is because central limit theorem seems only work for sample mean not sample proportion?
Thanks to Yue Cao for pointing out my mistakes at 10:50! It should be 0.5 instead of 0.05. Sorry guys, I will triple check my code before publishing it next time.
Great video as always! A question: Why do we use Z-test instead of T-test? My understanding is that T-test should be used when the standard deviation of the underlying population is unknown, which is true here since we can't run the experiment for the whole population. Could you please tell me what I am missing here? Thanks!
Emma thanks for the awesome video .. For the pooled variance or un Pooled variance I think it comes from the fact that .. if the two samples have equal or un equal variance .. so we will kind of do the F ratio first .. Then for equal variance we use a pooled variance For an un equal variance between the two samples we use unpooled variance for the standard error in the test statistic formula. 😅 hope i am not mixing things ! Thanks.
Yes I think so too! Just a follow-up question, under what scenario can we not perform an F-test? (i.e. is there a minimum size or other requirement for the F-test result to be valid?)
Hi Emma, Thanks for your sharing, and I have a question, how do you make sure ad impressions are independent Bernoulli trials, since it can be one user behind tons of ad impressions and clicks, how do you measure this impact is acceptable or not, and still keep the assumption of the data points are independent ? many thanks.
I think the most mathematical part of the procedure is the Central Limit Theorem which asserts that the sample proportion has a distribution approximately Gaussian given some statistical assumptions. The note mentions Slutsky's theorem but not Central Limit Theorem...
Many of you have asked me to share my presentation notes, and now… I have them for you! Download all the PDFs of my Notion pages at www.emmading.com/get-all-my-free-resources. Enjoy!
Great video! But some of the Python math is wrong. You applied the square root only to the last terms in two places for pooled and unpooled std. You’re just missing a set of parentheses in those two places. Otherwise, great video! Keep it up!
Thanks Emma. I was thinking about testing the CTR last night and today all of my questions are answers in this video! A great demonstration of presenting using Notion as well.
Thanks for your comment, Baizun! Glad you found my video helpful. 😊
Hi Emma, is it true that the difference between two sample proportion follows normal distribution is not because of central limit theorem? it is just because the binomial approximation to normal distribution when np>10 & n(1-p)>10? The reason why I am asking is because central limit theorem seems only work for sample mean not sample proportion?
Thanks to Yue Cao for pointing out my mistakes at 10:50! It should be 0.5 instead of 0.05. Sorry guys, I will triple check my code before publishing it next time.
Great video as always!
A question: Why do we use Z-test instead of T-test? My understanding is that T-test should be used when the standard deviation of the underlying population is unknown, which is true here since we can't run the experiment for the whole population. Could you please tell me what I am missing here? Thanks!
Emma thanks for the awesome video ..
For the pooled variance or un Pooled variance
I think it comes from the fact that .. if the two samples have equal or un equal variance .. so we will kind of do the F ratio first ..
Then for equal variance we use a pooled variance
For an un equal variance between the two samples we use unpooled variance for the standard error in the test statistic formula.
😅 hope i am not mixing things !
Thanks.
Yes I think so too! Just a follow-up question, under what scenario can we not perform an F-test? (i.e. is there a minimum size or other requirement for the F-test result to be valid?)
Hi Emma, Thanks for your sharing, and I have a question, how do you make sure ad impressions are independent Bernoulli trials, since it can be one user behind tons of ad impressions and clicks, how do you measure this impact is acceptable or not, and still keep the assumption of the data points are independent ? many thanks.
I think the most mathematical part of the procedure is the Central Limit Theorem which asserts that the sample proportion has a distribution approximately Gaussian given some statistical assumptions. The note mentions Slutsky's theorem but not Central Limit Theorem...
I liked the presentation material you put together. May I ask what tool you used to create it? Thanks!
Thanks for your comment, Junqi! I used Notion. 😊 Hope this helps!
Thank you! this is awesome!
Great vid!
10:50 the square root calculation seems wrong - should be 0.5 instead of 0.05
Thanks Yue! My mistake adding an extra 0.
Many of you have asked me to share my presentation notes, and now… I have them for you! Download all the PDFs of my Notion pages at www.emmading.com/get-all-my-free-resources. Enjoy!
Great video! But some of the Python math is wrong. You applied the square root only to the last terms in two places for pooled and unpooled std. You’re just missing a set of parentheses in those two places.
Otherwise, great video! Keep it up!