Can't thank you enough for this. This is the sort of knowledge I didn't even realize I was missing; hey, if we pay you to review some stats textbooks for us, will you recommend one? 😅
Exactly what I need; my understanding is ad-hoc and quite poor overall — took a single uni course years ago, and ever since have just learned bits-and-pieces as-encountered (where possible: often, it hasn't been, due to the sort of large gaps in fundamentals common to the lazy autodidact–).
May not be relevant to this particular textbook (working my way through it now 👊), but one thing I've found in the past is that:
···–––······–––···
• Often — in stats as well as most other mathematical/CS topics — you can learn *how to use* the tools and techniques in a practical sense, without ever really understanding /why they work/ or /how they were derived/ in the first place.
···–––······–––···
How much importance do you place on this latter?
I get curious, so it often bothers me to not understand why I'm going through xyz process — even if able to manipulate the thing correctly and hence arrive at a useful result — but in past attempts to rectify this, I've had little luck¹....
...but I'm not sure how much I ought worry about it, really. Baby steps, perhaps...
I think a lot of autistic people get too focused on understanding basics of probability theory and mathematical statistics that they can't use the methods in practice. I don't claim to be a linear algebra wizard, but I have 10+ years of experience with applied statistics, so I have a good intuition for when something works and when it doesn't. The beauty of programming is how easy it is to run simulations to verify and build intuitions.
> Using this procedure, then, will on average yield a value of 20k * 10, or 200k USD/year, as you hired 10 people.
Shouldn't this be $20k * 10 * 0.7? I thought the $20k was a figure per-standard-deviation, but the test only improved employee quality by 0.7 deviations, right? And the real benefit will be less, because you probably weren't hiring at random before.
My understanding was that the case for hiring-by-test was (a) the quality of your hires goes up, some; and (b) it is cheaper than other hiring methods. There's an interplay between a few things:
1. The test will give you a better rating of the candidate's quality than your bespoke process would.
2. The test is cheaper, per candidate, than your bespoke process.
3. In fact, the test is so cheap that you can consider a much larger number of candidates.
Item 1 improves employee quality because you can more accurately recognize candidate quality. This will hopefully be worth money as described in most of the post.
Item 2 improves your company finances without having any effect on employee quality. This is worth money too!
Item 3 costs you money. The test is cheap, but its cost is not negative, so expanding the candidate pool is bad for your finances. However, a larger pool improves employee quality because some of the candidates that join the pool (compared to your original process) will have quality above replacement level. The top 10 candidates from a pool of 400 will be better than the top 10 candidates from a pool of 100. In the terms of the post, this lets you spend money to improve your selection ratio.
All of these are pretty significant effects, but none of them are best compared to "random hiring".
Yes, the comparison is based on random hiring. No one does random hiring. So the improvement will be the difference between how well the current method selects for job performance vs. this method, or their combination. Probably the best is testing early, then interviewing the high scoring candidates.
"As I have explained before, the same is true when it comes to polygenic embryo selection, which is essentially a hiring task for the job of being your child. 🧒🏻"
Excellent point. And also being a valuable, productive member of the human race.
Can't thank you enough for this. This is the sort of knowledge I didn't even realize I was missing; hey, if we pay you to review some stats textbooks for us, will you recommend one? 😅
I've read a bunch, but the best is this one. https://www.statlearning.com/
Thanks a ton, mi amigo!
Exactly what I need; my understanding is ad-hoc and quite poor overall — took a single uni course years ago, and ever since have just learned bits-and-pieces as-encountered (where possible: often, it hasn't been, due to the sort of large gaps in fundamentals common to the lazy autodidact–).
--------------------------------------------------
May not be relevant to this particular textbook (working my way through it now 👊), but one thing I've found in the past is that:
···–––······–––···
• Often — in stats as well as most other mathematical/CS topics — you can learn *how to use* the tools and techniques in a practical sense, without ever really understanding /why they work/ or /how they were derived/ in the first place.
···–––······–––···
How much importance do you place on this latter?
I get curious, so it often bothers me to not understand why I'm going through xyz process — even if able to manipulate the thing correctly and hence arrive at a useful result — but in past attempts to rectify this, I've had little luck¹....
...but I'm not sure how much I ought worry about it, really. Baby steps, perhaps...
.
.
--------------------------------------------------
(¹: oh God, the Wikipedia mathematical articles:
• "consider a probability P"
[okay, so far so good 🙂]
• "now, let P hyperrenormalize by taking the Hamiltonian-octonion inverse of the Lie group's Riemannian tensor-functor Q^r•x(√π^e)/∆§mx+e•{π|Ω}..."
[...😦]
—Jesus, lol! Well, perhaps to someone like Glorious Leader E.O.W. Kirkegaard they're useful... but I get *nothin'* from 'em, heh.)
I think a lot of autistic people get too focused on understanding basics of probability theory and mathematical statistics that they can't use the methods in practice. I don't claim to be a linear algebra wizard, but I have 10+ years of experience with applied statistics, so I have a good intuition for when something works and when it doesn't. The beauty of programming is how easy it is to run simulations to verify and build intuitions.
> Using this procedure, then, will on average yield a value of 20k * 10, or 200k USD/year, as you hired 10 people.
Shouldn't this be $20k * 10 * 0.7? I thought the $20k was a figure per-standard-deviation, but the test only improved employee quality by 0.7 deviations, right? And the real benefit will be less, because you probably weren't hiring at random before.
My understanding was that the case for hiring-by-test was (a) the quality of your hires goes up, some; and (b) it is cheaper than other hiring methods. There's an interplay between a few things:
1. The test will give you a better rating of the candidate's quality than your bespoke process would.
2. The test is cheaper, per candidate, than your bespoke process.
3. In fact, the test is so cheap that you can consider a much larger number of candidates.
Item 1 improves employee quality because you can more accurately recognize candidate quality. This will hopefully be worth money as described in most of the post.
Item 2 improves your company finances without having any effect on employee quality. This is worth money too!
Item 3 costs you money. The test is cheap, but its cost is not negative, so expanding the candidate pool is bad for your finances. However, a larger pool improves employee quality because some of the candidates that join the pool (compared to your original process) will have quality above replacement level. The top 10 candidates from a pool of 400 will be better than the top 10 candidates from a pool of 100. In the terms of the post, this lets you spend money to improve your selection ratio.
All of these are pretty significant effects, but none of them are best compared to "random hiring".
Yes, the comparison is based on random hiring. No one does random hiring. So the improvement will be the difference between how well the current method selects for job performance vs. this method, or their combination. Probably the best is testing early, then interviewing the high scoring candidates.
"As I have explained before, the same is true when it comes to polygenic embryo selection, which is essentially a hiring task for the job of being your child. 🧒🏻"
Excellent point. And also being a valuable, productive member of the human race.