Suppose I have a $2 \times 2$ contingency table, and I know for a fact that I can determine the proportion in every cell in this table with one parameter. How many degrees of freedom would I have for the $\chi^2$ goodness-of-fit test?
@Glen_b thanks :) still need to take some time to familiarize myself with the mod tools but it's probably nothing pathbreaking
@gung congrats on your sixth stewardship! First one to get it in all categories, very impressive and a big signal for the huge commitment you put into this site. Thanks a lot for this!
@Glen_b I agree with Senn's quotation, but I think reactions to it might (ought) to be quite different when we replace "desires" by either "capabilities" or "needs." The trick of engineering a good course lies in not confusing these three things with one another. (This is the subject of many of Plato's dialogs, in which Socrates tries hard to demonstrate that a rational person's desires will be the same as what is good for him.)
Hi, I have a doubt and I am unsure whether it is big enough to ask in the questions. The questions we generally assume the failure of any object in real life, say distance moved by a car before failing to be an exponential random variable. Another important feature of this random variable is that it is memoryless, I don't think in real world failure functions are memoryless, so i cant understand the reason why failure function is chosen to be exponential
@Glen_b I think the summary statement: "Stephen Senn argues that viewing university students as 'customers' can be dangerous" is definitely accurate. I had a chemistry teacher who remarked years ago that education is one of the few places where people would pay more to get less. When a student is more concerned about a GPA and a diploma than knowledge they act like Senn's hypothetical student asking "Is this on the test."
This is a shame, and defeats the stated goal of paying for an education, but as long as grades are a proxy for "knowledge accumulated" then students will continue to optimize for grades instead of knowledge. In a way this makes it easier for those who wish to distinguish themselves because it is easier to stand above the crowd if you put in the effort.
@Glen_b "the pressures" <-- When I was in college I heard some teachers lament that they get complaints from underachievers that their course was "too hard", and as a result administration asked the teachers to tone it down. I think that is more a problem with spineless management/admin than anything else, but what do I know?
By and large my time in higher education has demonstrated the death of critical thinking. People are happy to be spoon fed a perspective without question. This is seems to be in direct contrast with the statement:
> Statistics is a critical subject that requires critical thinking.
If people aren't willing to think critically then they will certainly have a problem with teachers who actually require it.
@whuber Presumably the driving test was devised based on the needs of the student in the real world, even though the student might irrationally desire to change the test so they can legally drive. Even though a rational person desires only what is good for them humanity is proof that people are irrational.
I once taught a mathematics course (in the US) for future elementary school teachers. This was far and away one of the most discouraging experiences of my academic life.
I ended up submitting failing grades for a majority of the class, which the course manager was not pleased about. In the end, I walked away and told the course manager they were free to do what they wanted with my gradebook, but I would have nothing to do with passing students who had taken an antagonistic attitude to the material.
@Varun Your premise about what's generally assumed for failure times is not an accurate reflection of reality. Exponential failure time is sometimes used for certain kinds of very simple devices, and it's very commonly used in introducing students to a number of problems but that doesn't mean it's the most commonly used failure time distribution. There are a variety of distributions used in practice.
@Varun Consider the survreg function for fitting parametric survival-time models in R. It defaults to Weibull but includes a number of other survival time distributions, including log-logistic and lognormal.
@Varun also google "frailty models" and "accelerated failure time" ... the AFT models often assume a log-logistic survival.
...
@Varun But you really should have asked on the main site. What's the point in working to answer a question for one person here when you can answer it for many there? .... .... .... ....
@Erik At the university I sometimes teach at, they get a very large proportion of their Masters students from overseas (whom I am teaching a group of now); the income is crucial to the university. It's hardly surprising that the administration doesn't want to see them fail in the short term, even though in the long term it makes the degree valueless.
Of course the consequences of failure are immediate, while the consequences of devaluation are delayed. Admin people frequently move on to other universities or other jobs altogether long before the consequences of their choices can impact their own career.
@Glen_b I think you hit the nail on the head there. The short-term income worry causes them to mortgage their reputation. Unfortunately a school trades largely on their status, at least among the upper echelons.
@Glen_b lol that sounds suspiciously like politicians and their public policies
I work part time as a researcher in industry and pick up teaching (mostly post-grad) as it arises.
I don't work in a stats department. I used to, before and during my PhD (different ones before and during). Since then however, I've worked in a different area (more related to my work), still teaching stats or mostly statistically related material.
@Glen_b very interesting. Sorry for the ignorance but what does it mean "researcher in industry?" Does a business give you a bunch of data and say "find something useful in this mess?" or is it more structured like "we are having a problem with XYZ what are the probable causes and tests to ensure it was corrected?"
@Erik I work in industry. It's more the second than the first (we are having an issue with XYZ...), though sometimes it's myself coming up with the problems. The "here is a bunch of data, find something useful" seems to me, mythical.
I have had the "find something useful" once or twice. It's usually not a productive way to operate (usually you can find some things that are interesting, but if its not something you can make any operational changes from, what's the point?)
Usually the main thing I find out is that the people with the data hold some surprising delusions about what it contains -- even the people that supposedly are most familiar with it.
Anything from "This data is super clean; it's been carefully checked"... even after you prove that there's all kinds of problems with the data, there's denial ("But that's not possible!") to "I need you to help me write a report on how the number of new households is decreasing, to explain to my supplier why I am ordering less. Here's the data" ... (which very clearly show the number of new households was increasing). Denial and anger over that one.
Ive had "I have a big data problem, and need advice about what software I should get to deal with huge data. It won't fit into R."
Seriously, I don't think anyone can have ever actually looked at the data. They really had no idea what was actually there. They insisted they did, but they clearly didn't.
I run into a lot of that sort of thing with businesses. All too busy pontificating in meetings about stuff they've never actually dealt with.
several times I've seen worse versions of things I did pop up 10-15 years later in a paper someone wrote. Can't say anything...
It certainly has its frustrations. At least when you do research in academia, you get to talk about it.
I've had some tricky times with refereeing papers sometimes. I know something in a paper I am reviewing is wrong (because I've proved it in my own work). Can I cite anything? no.
I once was reviewing two different papers at the same time. Both made the same error. In both cases I asked them to fix the error. In both cases they dropped the entire section of the paper. Which of course meant that nobody else knew it was wrong --- so the first time I wasn't a referee on such a paper, it was published.
Reports for consulting I keep. But most of my work isn't consulting, it's mostly research. That's just the occasional but of work on the side.
@Glen_b I don't know if this is possible with the types of papers you've refereed, but when you can't prove them wrong due to a NDA can you suggest an edge case they haven't accounted for?
It wasn't like anyone was trying to do an end-run around me. Just that 15 years after I showed something couldn't be done, about 4 different groups of people wrote papers that "did it" in the space of two years.
I was able to convince two of those groups they were wrong. But I'm not that big a name (since almost all my good work is unpublished) so there's only so many papers I get to referee.
