Forums  > Books & Papers  > Statistics book suggestions?  
     
Page 1 of 1
Display using:  

electrical


Total Posts: 47
Joined: Jan 2006
 
Posted: 2007-01-29 23:16
Writing to request suggestions a good introduction to statistics, ideally for an electrical engineer with a strong background in stochastic processes and signal processing. I am also looking for help with understanding what areas (in statistics) should I try to read about to make the best use of my time. I am quite clueless about the terminology, so searching hasn't helped (see end of post though).

Reason for asking and details about my background follow.

Reason: In two interviews that I had recently, I have been asked "what background do I have in statistics." My answer has been - "none," and it hasn't been received well by my interviewers.

Background: PhD candidate in EE. I suspect though that there are bits and peices of "(statistics) background" in my electrical engineering background. I have done courses like signal detection and estimation (was an introduction to hypothesis testing, estimation theory) and grad and undergrad signal processing. In fact I have done some research on hypothesis testing (mostly was large deviations applied to some problem).

I looked up the "applied statistics" course in my university, and they are using Miller and Wichern. Some more searching on Amazon led to Ruppert. It seems to contain a lot of elementary stuff (pages after pages probability distributions?) but still may be what I am looking for because it is applied stuff that I want (I suppose).

Thanks for reading, and for your suggestions.

jungle
Chief Rhythm Officer
CSD LLC
Total Posts: 3163
Joined: Jul 2004
 
Posted: 2007-01-29 23:29

the best stats book i have read is casella and berger, which quantie recommended to me some time ago.  you should never need another stats textbook after that. 

it's not very applied, though.  for applications, you may wish to read some books on the uses of stats in the markets, e.g. stat tools for finance and insurance, applied quant financemarket models (horrible), quant methods in finance (excellent, but you may find a little simple).  does this help? 


electrical


Total Posts: 47
Joined: Jan 2006
 
Posted: 2007-01-29 23:40
Jungle,

Yes, it helps, thanks! Going through the contents of these books on Amazon is immediately useful. I have so much to learn...

I had come across the post in which quantie recommended C&B - I had looked the book up and decided that it seems to require more time than what I have right now. But given how little I know from the contents of "applied quant finance" "stat tools for finance and insurance," I don't know where exactly to start...

jungle
Chief Rhythm Officer
CSD LLC
Total Posts: 3163
Joined: Jul 2004
 
Posted: 2007-01-30 01:42

at the risk of stating the obvious, those books are fairly broad in scope, so i would focus on learning techniques applicable to the areas you're interviewing for. 


nuno


Total Posts: 277
Joined: Jul 2006
 
Posted: 2007-01-30 01:45
you might want to make sure you are literate in some non-orthodox statistics as well by perusing hastie's book

urnash


Total Posts: 561
Joined: Sep 2006
 
Posted: 2007-01-30 09:00
A very nice book to start (but very elementary) is Statistics, a guide to the use of statistical methods in the physical sciences by Barlow. It was written by a physicist, for physicists. IMHO it has just the right amount of mathematics, not too much, nor too little.

Constants aren't.

Veegan


Total Posts: 684
Joined: Jun 2004
 
Posted: 2007-01-30 09:26

Slightly off-topic but this is a great statistics handbook to have to hand.

V.


"The Stranger within my gates, He may be evil or good, But I cannot tell what powers control-- What reasons sway his mood; Nor when the Gods of his far-off land Shall repossess his blood." ~ Kipling

Dimatrix


Total Posts: 539
Joined: May 2006
 
Posted: 2007-01-30 10:40
If you don't like to spend money, you could try one of the free e-books available here: E-books Most of them are all related to finance. Hope this helps.

Ctrl - L.

Graeme


Total Posts: 1629
Joined: Jun 2004
 
Posted: 2007-02-03 19:34

Given what you say about the stuff you have done which is related to stats how do you come up with "none" in your interviews? You might try to say something like

"there are bits and pieces of statistics background in my electrical engineering studies. I have done courses like signal detection and estimation (was an introduction to hypothesis testing, estimation theory) and grad and undergrad signal processing. In fact I have done some research on hypothesis testing (mostly was large deviations applied to some problem)."

If they then ask, "yes, but do you know the parameterisation of the Pearson IV distribution?" THEN you say no. Sure as hell sounds like you are selling yourself short.


Graeme West

AVt


Total Posts: 833
Joined: Jun 2004
 
Posted: 2007-02-03 21:57
I second Graeme, you are a 'humble idiot' and would guess from your
background you cover lots of what practical is needed and it would be
'easy' for you to learn the desired rest.

But to second your attitude: I never had applied math, numerics, stats
or even coding, so my my answer on background is 'none' as well :-)

Just dare, fight for it!


BTW: these are nice suggestions for books

electrical


Total Posts: 47
Joined: Jan 2006
 
Posted: 2007-02-03 23:03
Everyone,

Thanks for the sugestions and the links. A follow up on my original post:

Graeme, AVt, you are probably right. The book that I learnt what I know about "statistics" from is Poor, and the course was roughly like this (pdf) one. I had forgotten a lot of what I had learnt. However, I should have been more explicit about what I meant by "my answer (none) not having gone well" - it was followed by the questions "how would you identify outliers in a data sample" and "how would you deal with missing data in a data set." I have never handled big data sets, so my responses was "if I know something about the law behind the sample set, I could set up thresholds within which I would accept data - we could set up a simple test for validity" and "I guess I could interpolate between points," respectively. The interviewer wanted me to be more specific, which I did not know enough to be. So it was (is) a real deficiency in my learning, not short selling.

Jungle, I ultimately could not check out the books that you mentioned. I did find a decent set of course notes, which was probably quicker (and dirtier) reading. Veegan, that "handbook" is excellent. For people with background similar to mine, parts of section IV in Dave MacKay's book are also worth reading, though more for fun than anything useful.

Finally, I also got my hands on one of the books I mentioned in my original post (Ruppert). It is probably worth getting on a loan from a library (which is what I could do) and glancing over if you are as clueless as me - gave me a quick idea of all the lingo that I was unsure about - but not worth buying.

zee4


Total Posts: 59
Joined: May 2010
 
Posted: 2016-06-08 13:53
Since 2007 several books have been published. Any recommendations?

Бухарский

Rashomon


Total Posts: 171
Joined: Mar 2011
 
Posted: 2016-07-06 22:56
Not that these are all since 2007, but my statistics reco list would be very different to what was shared here in 2007.

• If you only read one statistics book, make it Angrist & Pischke.
• Check R-bloggers at least a few times to get an idea of what people are doing and what gets people excited.
• Faraway's applied linear regression book is down-to-earth, as is Sanford Weisberg's applied statistics book.
• I guess pick up one of the machine-learning books, or just watch videos of Esarey and Tibshirani, to get a flavor of the ML stuff. MIT OCW has good videos on separating hyperplanes and you can find some discussion of the kernel trick somewhere. Caveat hype though.
• I actually started with Shalizi and Jeffrey type stuff which was probably grasping at the wrong end.
• John Cochrane has some good time series notes. They may not be the best.
• You want some exploratory data analysis stuff. Not sure what a good reference is. Skim Tufte and look up some of Tukey's ideas, somehow. This is way more important than knowing asymptotic results on specific distributions (or on nonparametrics) cold. And by the way, make sure you can use SQL or dplyr/pandas before spending too much time on theory.
• Why not pick up All of Statistics? After too much stats you wonder if you've covered everything.



For what that's worth.

"My hands are small, I know, but they're not yours, they are my own. And they're, not yours, they are my own." ~ Jewel

Polter


Total Posts: 130
Joined: Jun 2008
 
Posted: 2016-07-07 01:22
Not a book, but for someone with programming background but without any statistical background, this may be a nice intro (in particular, familiarity with the bias-variance trade-off, overfitting, and cross-validation--in the last part--can't hurt in any case):
https://speakerdeck.com/jakevdp/statistics-for-hackers
https://www.youtube.com/watch?v=Iq9DzN6mvYA

zee4


Total Posts: 59
Joined: May 2010
 
Posted: 2016-07-07 05:33
Thank you guys. Appreciate it.

Бухарский

Rashomon


Total Posts: 171
Joined: Mar 2011
 
Posted: 2016-07-07 19:50
Thanks Polter. I forgot to say anything about cross-validation and it's the most important thing. Plus randomly drawing subsamples of your index / key rows is something any programmer can do but maybe hasn't realized they should.

Along that line, programmers should please throw away insignificant trailing digits. Graphics people might make the handle or the trailing digits larger depending on the target audience.

"My hands are small, I know, but they're not yours, they are my own. And they're, not yours, they are my own." ~ Jewel

jslade


Total Posts: 1096
Joined: Feb 2007
 
Posted: 2016-07-22 03:13
I don't have too much to say here that is constructive. I looked at and dislike most of these books for different reasons. There must be something out there for "folk statistics." There's lots of things out there that works when it isn't supposed to, and doesn't work when it is supposed to. Also, looking at tests like Kendall's tau (or Kolmogorov-Smirnov, or Mann-Whitney) from the information theory point of view: surely others have noticed such things come perilously close to being Shannon measures. There's lots of stuff I've ended up learning about which is tremendously useful which never really ends up being treated anywhere.

One example thing I've never seen treated well is stuff like T-tests on lift. You can read about it in database marketing books, but it's never done well. It's one of the single most useful tricks I've ever learned, and it was taught to me by a biostat guy, word of mouth.

Oh yeah, and while learning the bejeepers out of linear regression is awesome, survival models can also be helpful. Whenever I read a book on them (inevitably written for medical swine who use it to cook the books for drug companies or some such thing), I feel my eyes glazing over. Yet, you can use such things to model order book hits; and if you're clever, you can use them as starting points for better models for order book hits.

Classical statistics and statistical models are, like, really good. But the treatments are all either rubbish that makes me dumber, or theoretical claptrap that doesn't help me to solve real world problems. Signal processing books are vastly better, and solve almost the same kinds of problems, but somehow humble old stats can't muster up a decent author. At least none I have seen.

Oh yeah, a book which I found charming which approaches what I'd consider a good book on statistics; "A Guide to Econometrics" by Peter Kennedy. At least it shows you some of the piles of dogshit not to step in.

"Learning, n. The kind of ignorance distinguishing the studious."

TonyC
Nuclear Energy Trader

Total Posts: 1245
Joined: May 2004
 
Posted: 2016-07-24 05:33
>One example thing I've never seen treated
> well is stuff like T-tests on lift.

ok, i'll bite ... whats a "lift" and why would i want to take a t-stat of it?

flaneur/boulevardier/remittance man/energy trader

jslade


Total Posts: 1096
Joined: Feb 2007
 
Posted: 2016-07-26 09:09
http://scientificmarketer.com/2007/09/uplift-modelling-faq.html

Basically, measuring small potentially extremely multivariate effects on a noisy baseline. Used in medical research as well, but mostly in marketing.


"Learning, n. The kind of ignorance distinguishing the studious."

phy


Total Posts: 16
Joined: Nov 2013
 
Posted: 2017-07-14 20:50
The classics, won't find better books to learn mathematical statistics. Used in the best PhD statistics programs:

Theory of Point Estimation (Springer Texts in Statistics)
by Erich L. Lehmann and George Casella

Testing Statistical Hypotheses (Springer Texts in Statistics)
by Erich L. Lehmann and Joseph P. Romano

Elements of Large-Sample Theory (Springer Texts in Statistics)
by E.L. Lehmann
Previous Thread :: Next Thread 
Page 1 of 1