Wrote this 30-page essay that aims to explore an approach to statistics for the layman - from simple average to stochastic gradient descent.
Open source, free culture code - happy sharing.
(Great project. I lecture stats to biochemists and getting them interested is half the battle; convincing them why it's worth learning what the computer does the other half)
Interpreting the expression «Worth the price» from Pinegulf is not immediate.
Surely, as rvbissell wrote nearby, there are contributions around of negative value - though maybe rarely in technical publications of this kind.
If you want to donate to Carlo Occhiena, information does not seem immediately available, but his personal website - carloocchiena.com - contains his E-Mail address at the bottom of the JS generated typing, linked at «mi puoi scrivere una mail».
I appreciate this greatly... I am working with people that don't understand even the basics (such as a survey with 49 responses means that the margin of error is over 10% and thus comparing 1 month's results of say '68 per cent' with next month's '75' is meaningless); hope that this will help me better explain what they are missing.
> I am working with people that don't understand even the basics (such as a survey with 49 responses means that the margin of error is over 10%
I don't understand this, if the population if just 49 people then the margin of error is zero. So intuitively the bigger the population the bigger the bound for the margin of error.
Sorry for not being clear, the population is about 900-1200 (customers who have opened tickets in the calendar month) and the survey count (responses) is (can be) 49.
My favorite is when a survey result is presented with “as many digits as my calculator showed”, often allowing me to derive how many responses were likely received.
In 10.4 A/B testing is just a list with several points and there is not warning about having a deep understanding. For example, the point of selecting a sample is not easy, if you take a sample of something on 1 july of 2020, you have to consider if the weather, the day of week, people on vacation or anyone of thousands of factor is going to make your sample not adequate to generalize the result to other circumstances. Using statistics correctly requires neutralizing many sources of errors. It is not easy to get a good representative sample.
22 comments
[ 3.3 ms ] story [ 52.5 ms ] thread(Great project. I lecture stats to biochemists and getting them interested is half the battle; convincing them why it's worth learning what the computer does the other half)
Most people never use GitHub.
As a reader I clicked on the link to read about statistics, not licensing.
If you can’t click around a bit to get access to a free resource then I think you are better served by Amazon.
How quickly people feel they are owed something which they didn’t know it existed 10min ago is mind boggling to me.
To me, a GitHub repository seems a bit at cross purposes with that. Hence my comment.
The author doesn’t owe me anything.
And I owe them nothing as well.
...
Surely, as rvbissell wrote nearby, there are contributions around of negative value - though maybe rarely in technical publications of this kind.
If you want to donate to Carlo Occhiena, information does not seem immediately available, but his personal website - carloocchiena.com - contains his E-Mail address at the bottom of the JS generated typing, linked at «mi puoi scrivere una mail».
National exam pass rate of 67% does not imply that every class of 20 will have 14 or 15 passes!
I really appreciate the author providing access to the LaTeX source as well.
I don't understand this, if the population if just 49 people then the margin of error is zero. So intuitively the bigger the population the bigger the bound for the margin of error.