Messages
Copied over from an earlier message: As been mentioned in class a coupl' o' times, the "single sheet of paper with the candidate's own personal notes" means ONE SIDE ONLY, not one sheet of paper with notes on both sides.
" The four-hour written exam takes place on December 6 (the day after the Godt Hjort festivities, December 4-5); see separate information about where and time of day (I do think it's in Silurveien, from 15:00 to 19:00, but check other sources).
This is a No Book exam -- however, bring (1) a calculator, in case you might beed to find the square-root of 49, and (2) *one single sheet of paper with the candidate’s own personal notes*. "
Last chance with Q & A today ... og lykke til på eksamen! I might answer some questions by mail, over the few remaining days, but don't count on it. And there's Godt Hjort Monday-Tuesday, before exam on Wednesday (perhaps with themes identifiable from the Godt Hjort?).
https://www.mn.uio.no/math/english/research/groups/statistics-data-science/events/conferences/hjort70/index.html
Eksamenssettet, 4 timer skriftlig uten bok men med 1 sides personlige notater, 6. desember 2023, vil foreligge på språket engelsk -- og også de norskkompetente har full anledning til å skrive sine bevarelser på det målføret. MEN, hallo, folkens, skriv gjerne på bokmål eller nynorsk (eller latin, om den er god nok).
Nær sagt forbausende mange av dere skrev The Oblig på engelsk -- aldeles ok, men ikke glem norsken fullstendig, her på Universitas Osloensis.
1. No teaching Wednesday Nov 22nd (Nils is at INTEGREAT kickoff; how to spend 155.000.000 kroner).
2. Full teaching next week, the last week of the course: 3 hours Monday, 2 hours Wednesday. We do earlier exam questions (see previously posted list); we have Q + A (som come prepared); we go through Mist Important Things from the course.
3. I'll put up a curriculum list (essentially saying "curriculum = all those exercises and stories we've been through during the course").
4. Stian er norgesmester !!
We have had our two hours going through the basics of CDs and confidence curves, from the slim subset 1-9 of exerices from Ch 7. We have more or less one more set of things to learn in the course, related to sufficiency and best unbiased estimators, from Ch 8. From that chapter, do 2, 4, 5, 6, 7.
The rest of our time we use for earlier exam question, fore Q-and-A sessions from the full curriculum, etc. We'll also have a 3rd extra hour for a couple of the Mondays ahead.
For the coming days, we first do the Odin's Children story, both with CDs and Bayes; see PartTwo. For exam sets, this is what we've done, so far:
2022: 3
2021:
2020: 3
2019: 1, 2, 3, 4
2018: 3
2017: 1, 4
2016:
2015:
2014: 1, 3, 4
ToDo:
2018: 1, 2
2017:
2016: 1, 3 : do these first
2015: 3, 4 : and these
2014: 2
From earlier exam set questions, we've done 2019: 1, 2014: 4, 2020: 3. For the coming period, attempt to work through as many of these as you can (and we do a good subset of these on the blackboard):
2022: 3. 2021: 2. 2019: 2, 3. 2018: 3. 2017: 1, 4. 2014: 1, 2, 3.
We will have a relatively short going through of the basic ideas for CDs and confidence curves, via a slim subset of exercises from Ch7, on Mon 13th and Wed 15th November. Attempt to work through these: 1, 2, 3, 5, 6, 7, 9.
There will also be a couple of Stories to work through (more information in a few days).
We've rounded off Ch 5 on likelihood machineries, and have three chapters ahead of us, though each with relatively slim subsets. First off is the Bayesian Ch 6, where we'll be content to just the basics. We start with 6.1, 6.7, 6.10.
We also continue with earlier exam sets. We did 2019, 1. Go on with 2019: 3; 2014: 4 (Ruben); 2020: 3.
We've done Cramer-Rao. C.R. Rao was born in 1920 and died in August 2023, close to being 103. We might honour him by having a Cramer-Rao calculation for the exam in December. This is from the preface of one of his books:
" in my younger days, [my mother] woke me up every day at four in the morning and lit the oil lamp for me to study in the quiet hours of the morning when the mind is fresh. "
https:/www.facebook.com/groups/1589206911336271/posts/3554227418167534/
Thanks for your valiant efforts with The Oblig. On Mon Oct 30 I will go through some of its details, and otherwise work towards the rounding off the the likelihood chapter.
In addition to Ch 5 exercises, lifting iid likelihood theory to regression models, and some Cramer-Rao, it's time to jump into *earlier exam questions*. You find earlier exam sets on the general course website, for 2013 to 2020 (I think), and from 2021 and 2022 on this year's course website. Nils has made the sets for 2014 and 2019.
For the coming days, work with these: From 2019: 1, 3. From 2014: 4. From 2020: 3.
Nils gives a talk, torsdag 26.10 kl 17:59 i Videnskaps-Akademiet, med krig, fred, statistikk, prediksjon:
https://dnva.no/detskjer/2023/04/blir-det-fredeligere-i-verden
We've rounded off Ch 4, and now start the important Ch 5, on likelihood methods. Our programme will more or less be exercises 1, 2, 3, 4, 5, 7, 9, 12, 15, 16, then regression models with 23, 24, 25, 26. Markus does elements of 1 and 2 for Monday 23rd.
We'll soon devote time & energy to various exercises from earlier stk 4011 exam sets.
Under "Oppgaver", at the stk4011 course website (i.e. the general one, not the Autumn 2023 version), you'll find various earlier exam sets, but not those for 2021 and 2022. I've managed to find the sets, however, from colleagues, and have placed these on the Autumn 2023 website.
We'll pretty soon start working with such earlier exam set questions.
The four-hour written exam takes place on December 6 (the day after the Godt Hjort festivities, December 4-5); see separate information about where and time of day (I do think it's in Silurveien, from 15:00 to 19:00, but check other sources).
This is a No Book exam -- however, bring (1) a calculator, in case you might beed to find the square-root of 49, and (2) *one single sheet of paper with the candidate’s own personal notes*.
There should be no need for "long list of formulas". If I compose an exercise requring a certain not-so-standard formula, I might simply give it, as part of the text. The exam is not a "memory test", but is aiming at testing your understanding and ability to use what you've learned in the course.
https://www.mn.uio.no/math/english/research/groups/statistics-data-science/events/conferences/hjort70/index.html
0. Work with The Oblig!
1. We have now rounded off Ch3, and start Ch4 on 16 Oct. We have been through the Master Theorem, for testing in exponential families, see 3.31, and also the basics for multiple linear regression, see 3.33-3.35.
2. For Mon 16 Oct, Anders does a subset of 3.28. In this connection also do the following mini-version of Story II, suicide attempts data for placebo vs. Paroxetine. We have y_0 pois(m_0 \theta) and y_1 pois(m_1 \theta \gamma), and test \gamma=1 vs \gamma > 1.
2005 data: (y_0, y_1) = (1, 7), exposure times (73.3, 190.7), in terms of patient years. 2006 data: (y_0, y_1) = (1, 11), exposure times (333.0, 601.0).
3. Then go back to smallbabies-data, with y = birthweight in kg, x1 = mother's age, x2 = mother's weight before pregnancy, x3 = indicator for smoking. Do lm(y ~ x1 + x2 + x3), discuss parameter estimates, give confidence for the three regression coefficients.
...In the 9-11 October week, we round off Ch 3, and will soon start Ch 4, with the basics of large-sample approximations (CLT, Lindeberg, more delta method, applications, stories). Again, the curriculum is being defined via the exercises we go through, and for Ch 3 the subset is not so big.
The OBLIG is from Mon Oct 9 to Mon Oct 23, with reports to be delivered inside the Canvas system.
For Monday 9 Oct we have another extra hour, 14:15 to 15:00. You may send questions by mail.
We will rather soon start doing exercises from previous exam sets.
Work with these exercises:
First, for the Sachsen 8-children families, do the details for testing H0: no overdisperson against the alternative of overdisperson, using W_n = S_n^2 / (8 p0hat (1 - p0hat)) as test statistic; see Nils' FocuStat blogpost for some details.
Then, from Ch 3: 14 (Nora does some of them), 19, 25, 28.
Well, the Hjort-Stoltenberg Book-to-Be (2024) will actually have solutions to all exercises & all stories, not in the physical book, but placed in the Book-Website-to-Be (which will have all datasets, all required scripts in R and python, *and* all solutions).
We do not have the capacity and time to produce all these solutions for you, dear STK4011 crowd, autumn 2023, but I do fully understand (via Berit and Oskar) that many of you would appreciate solutions to at least *some* of the exercises.
So here's a meta-exercise, for everyone interested: (a) Go through your lists of exercises worked with, so far (up to say Oct 2), and your notes from class -- and then send me a mail, where you point to perhaps *up to 5* exercises, for which you would be happy to have written-out solutions. Write "stk4011 exercises" in the mail header, so that I find them quickly. Then I might aim for writing out such solutions, for a suitable subset of the un...
We've started Ch 3, after having rounded off Ch 2 with the basics of classic linear regression. Over the coming couple of occasions, we do as follows:
a. Rounding off STORY 4, with quantile things.
b. Then most of STORY 3, with overdispersed children; see in this connection Nils' FocuStat Blog Story.
c. Then from Ch 3: 1, 2, 3, 4, 7, 8, 10, 11.
d. Then access and read into your computers "smallbabies_data" from the website. Use y = (column 11)/1000, the baby's weight in kg; x1 = column 3, the mother's age; x2 = (column 4)*0.45349, the mother's weight (in kg) before pregnancy; x3 = column 6, an indicator for smoking or not. Use summary( lm(y ~ x1) ) etc. to run regressions on x1, on x2, on x3, and give for each a 95 confidence interval for the b in a + b x. Comment on your findings.
https://www.mn.uio.no/math/english/research/projects/focustat/the-focustat-blog!/overdispersion.html
1. We're rounding off Ch2 and will start Ch3 on 25-27 September.
2. On Mon 25 Sept we carry out the "Nils leaves the room during the break" exercise; Berit and Oskar take notes and speak with Nils a bit later on, about aspects of the course.
3. For 25-27 Sept, first do Story IV. I've placed birthweight-boys and birthweight-girls on the website, read these into your computers. As part of the exercise, make confidence plots for boys and girls, as a function of q from say 0.05 to 0.95, with (i) estimated quantiles and (ii) a 95% band. This involves estimating the variances q(1-q)/f(quantile(q)), for boys and girls. You may use "density" in R (without caring so much about which formulae are under the hood), e.g. using
fhatBval[j] <- mean( density(yyB,from=QBval[j],to=QBval[j])$y )
fhatGval[j] <- mean( density(yyG,from=QGval[j],to=QGval[j])$y )
(There might be simpler ways, but at least this works.) ...
1. The Internet Is Going Bananas Over the Real Meaning of '4011'
https://www.yahoo.com/lifestyle/internet-going-bananas-over-real-100600840.html
2. Next Monday Sept 18, Nils leaves the room during the break, and the students sum of points and views and advice (and praise). Berit and Oskar, qua student representatives, then talk to Nils later on.
3. For a few Mondays during the course, we'll use *one more hour*, 14:15 to 15:00 (in the same room), for questions and answers, some more details for exercises we've been through, etc. The first of these Mondays will be Oct 2nd.
1. There is interest in once in a while having *one extra hour of teaching* -- not to squeeze in more material or snikutvidelse av pensum, but to spend more time on various details, having questions & answers, etc. Nils will try to organise this, in that case *a few Mondays 14:15 to 15:00* (which also involves finding a room). We come back to this.
2. Next week we try to fix dates for the Oblig -- both t_1 = the day you need to hand in reports and t_0 = the day the exercises set will be made available.
3. For September 18 and 20, we proceed with Ch 2, the terrain of CLT, delta method, confidence intervals in various situations, etc. Do as much as you can for these exercises, from Ch 2: 27, 28, 29, 18, 19, 20, 34. Ruben does 27. After that we'll do a couple of the Statistical Stories, and start with Ch 3.
The congregation has appointed *Berit* (beritoo@math.uio.no) and *Oskar* (oskarhs@math.uio.no) as student representatives, tillitspersoner, for the course. Check here:
" I starten av hvert semester velges det omtrent 1+n/60 (hvor n er antall studenter på emnet) tillitsvalgte ":
https://www.mn.uio.no/math/studier/emneevaluering/index.html
We're into Ch2, later to be coupled with Ch3. Students who learn Chs 1, 2, 3 well have learned at least 51 percent of the course (!).
We have done exercises 1, 2, 3, 4, 5 (thanks, Mariem, Bård, Kristoffer), and I'll be posting R code for 2(e) on the website.
For Mon and Wed Sept 11 and 13 we do exercises 8, 9, 10, 11 (CLT with delta method and variations, supremely useful), then 12, 13, 14, 16, 18 (various applications of that theory). Peter does 14, Stian 16.
On Mon we choose 1 or 2 tillitspersoner for the course.
Again: the curriculum for the course will be essentially defined as "all exercises and all Statistical Stories we've been through".
On Wed Aug 30 (Week 2) we attempt to round off Ch 1. This will mean tending to exercises 12(e) (with Anders), 29-30 with the multinormal, 32-33-34 with the chi-squared and t, 38 with the F.
We start Week 3, 4-6 September, with *Story 19* (how strange is it, to turn 0-2 into 3-2), and then go on, perhaps using Weeks 3-4-5, with the essential parts of Chs 2 and 3. We begin with these exercises from Ch 2: 1, 2 (Mariem), 3, 4 (Bård), 5 (Kristoffer), then 8-9-10-11 with CLT and delta method theory, with applications in 12-13-14.
For our Week 1, 21-23 August, we've jumped into right subset of Ch 1 exercises, and more or less been through 1 (thanks, Oskar), 2, 3, 7, 8, 9, 14, half of 20, 21. Four Week 2, 28-30 August, we do as many as we muster of 12, rest of 14, 23, 25 (with Berit and Markus, Beta and Beta-Binomial), 15, 16, 18 the last point (computing e via uniform simulations), more from 20, 21, then 22, 25, 29, 30, 32, 33, 34.
Again, get stk4011-PartTwo into your computers (and print them out), with 21 Statistical Stories. The first of these we'll do is *Story 19* -- Aftenposten said Belgium turning 0-2 into 3-2 against Japan was "breaking a 48 year long curse" (!), and we're assessing that statement.
1. On the course website, under Exercises & course material, there is now also stk4011-PartTwo, qua pdf, with a 0.90 finished versions of a subset of 21 Statistical Stories. We will go through a subset of these again during the course.
2. Under "some R scripts" I'll place a few such, with R code to accomplish various tasks related to the exercises. Check them, run them, go through the parts, so that you understand precisely what is accomplished in which ways. You'll need to programme various things yourself, during the course, from simulations to statistical inference for given datasets.
Note from the "0.90 version" of Hjort-Stoltenberg, Chs 1-8, that we' jumping rather quickly into *exercises* (stk4011_PartOne) and *stories* (there will soon be a stk4011_PartTwo), as opposed to long sections and lectures of theory first.
For the two first weeks or so: do as many as you can of the following (for which we might be using the two first weeks):
1 (a)-(f), 2, 3 (a)-(b), 7, 8, 9, 10, 12, 14, 15, 16, 18 the last point (computing e via uniform simulations), 20, 21, 22, 25, 29, 30, 32, 33, 34, 40 ...
Also: *have a look* at what Ch 2 is about. We'll not at all do all exercises of Ch 1 before we jump to Ch 2.
I tried to make stk4011_PartOne.pdf available inside the Canvas system, under "Pensum", but it apparently too complicated for the system to give the students access to that file there. Hence I've placed that pdf directly on the course website, under "Exercises & course material".
(a) The file, which should be easily accessible & downloadable for all of you know, might be taken away later, since our publisher of the book-to-be doesn't like it. So take your copy.
(b) There will be a similar stk4011_PartTwo-pdf in a few weeks.
(c) In a little while there will be *datasets* , placed under, surprise, "Datasets", for us to work with.