2 ms·

Ask HN: Any books that focus on understanding hypothesis testing?

including probability distributions, confidence intervals, p-values(hacking (https://xkcd.com/882/) ), sampling and methods of conducting experiments? would like a book that explains intuitively and directly instead of having me extract them from examples.

mindcrime 4y ago
Sounds like you want a book that would typically be titled "Design of Experiments" or "Experiment Design" or something similar. There are tons of them on the market[1][2][3][4], probably any mainstream one used for a university course on the subject would be adequate. Here's a freely available book / lecture notes doc that covers some of the ground you're interested in. http://www.stat.rice.edu/~dobelman/courses/texts/Notes.411_616.Tab6.pdf http://www.stat.rice.edu/~dobelman/courses/texts/Notes.411_6... [1]: https://www.amazon.com/Design-Analysis-Experiments-Springer-Statistics/dp/3319522485/ https://www.amazon.com/Design-Analysis-Experiments-Springer-... [2]: https://www.amazon.com/Design-Experiments-Mike-Peralta/dp/1494378582/ https://www.amazon.com/Design-Experiments-Mike-Peralta/dp/14... [3]: https://www.amazon.com/Design-Experiments-Approach-Bradley-Jones/dp/1119746019/r https://www.amazon.com/Design-Experiments-Approach-Bradley-J... [4]: https://www.amazon.com/Design-Analysis-Experiments-Douglas-Montgomery/dp/0470128666/ https://www.amazon.com/Design-Analysis-Experiments-Douglas-M...
- BOOSTERHIDROGEN 4y ago
  Let say I'm using statistics program like SPSS. I have 20 different datasets that explain 20 different variables. The Scientific Method would be to propose a hypothesis, determine that the hypothesis would be supported is A was related to G, and then tell SPSS to run a crosstabulation of A and G. Alas, the software can just as easily run a cross tabulation of all 380 possible combinations of variables. There is even a special command to do this. With p-Hacking I look through these 380 tables to find the one with the best correlation (let's say that is C and R). Then I try and guess up a theory that C being correlated to R supports. I can't logically validate a hypothesis but only reject the opposite hypothesis.

mindcrime 4y ago

Sounds like you want a book that would typically be titled "Design of Experiments" or "Experiment Design" or something similar. There are tons of them on the market[1][2][3][4], probably any mainstream one used for a university course on the subject would be adequate. Here's a freely available book / lecture notes doc that covers some of the ground you're interested in. http://www.stat.rice.edu/~dobelman/courses/texts/Notes.411_616.Tab6.pdf http://www.stat.rice.edu/~dobelman/courses/texts/Notes.411_6... [1]: https://www.amazon.com/Design-Analysis-Experiments-Springer-Statistics/dp/3319522485/ https://www.amazon.com/Design-Analysis-Experiments-Springer-... [2]: https://www.amazon.com/Design-Experiments-Mike-Peralta/dp/1494378582/ https://www.amazon.com/Design-Experiments-Mike-Peralta/dp/14... [3]: https://www.amazon.com/Design-Experiments-Approach-Bradley-Jones/dp/1119746019/r https://www.amazon.com/Design-Experiments-Approach-Bradley-J... [4]: https://www.amazon.com/Design-Analysis-Experiments-Douglas-Montgomery/dp/0470128666/ https://www.amazon.com/Design-Analysis-Experiments-Douglas-M...

BOOSTERHIDROGEN 4y ago
Let say I'm using statistics program like SPSS. I have 20 different datasets that explain 20 different variables. The Scientific Method would be to propose a hypothesis, determine that the hypothesis would be supported is A was related to G, and then tell SPSS to run a crosstabulation of A and G. Alas, the software can just as easily run a cross tabulation of all 380 possible combinations of variables. There is even a special command to do this. With p-Hacking I look through these 380 tables to find the one with the best correlation (let's say that is C and R). Then I try and guess up a theory that C being correlated to R supports. I can't logically validate a hypothesis but only reject the opposite hypothesis.