And of course, those with legitimate complaints with the Prof simply drop the class a few days or weeks in, and never get a chance to fill out an evaluation...
In deteriming whether course evaluations are corrlelated with teacher quality, they compared them to grades received in a follow on course, and found that they had a negative correlation. So far so good. But if grades received in follow on courses are the gold standard for teacher effectiveness, then why not use that, instead of the other proxies the article goes on to recommend (peer evaluations primarily)? Certainly it won't be possible in every case as not every class has a follow on, but at the very least they should subject their new proposed methods to the same analysis.
> if grades received in follow on courses are the gold standard for teacher effectiveness, then why not use that
Because grades received in later courses isn't actually a meaningful metric. There was a theory that you could measure teacher quality this way using something called Value Added Metrics (VAM), but it's now been completely debunked. C.f.
Your links do not remotely come close to debunking the concept of VAM.
Your first (accurately) says "measurements are noisy" and the inaccurately says "and other things matter for learning too". Unless I missed something, it doesn't provide a reason to expect bias in the other things.
The second link says that some VAM models assume student assignment is random when it is not, and also that measured educational outputs don't matter much for outcomes anyway. The logical conclusion of this is simply that VAM needs modification, and maybe we should reduce education. Claiming this debunks the idea of VAM is like saying "hah, this paper proves OLS doesn't always work well, therefore regression is debunked".
The third is really long and I couldn't figure out it's point. The fourth one says VAM is noisy.
If you believe VAM is unable to measure the effects of teachers, are you willing to follow through with the logical conclusions? Namely, the effect teachers have on students is so small it can't be measured and we should therefore focus on other things (e.g. reducing cost) instead?
> Unless I missed something, it doesn't provide a reason to expect bias in the other things.
"VAM’s instability can result from differences in the characteristics of students assigned to particular teachers in a particular year, from small samples of students (made even less representative in schools serving disadvantaged students by high rates of student mobility), from other influences on student learning both inside and outside school, and from tests that are poorly lined up with the curriculum teachers are expected to cover, or that do not measure the full range of achievement of students in the class."
> If you believe VAM is unable to measure the effects of teachers, are you willing to follow through with the logical conclusions? Namely, the effect teachers have on students is so small it can't be measured ...
We already knew that the contributions of teachers to student achievement was relatively small long before VAM was invented. The only people who seem to think that teachers have a very large impact are the pro-VAM folks (e.g. Bill Gates and Malcolm Gladwell), who all cite the same flawed study.
That said, the reasons why VAM can't measure teacher performance isn't because the contributions of teachers are small per se, but rather because there are all sorts of problems with all of the proposed measurement systems.
In order to create a VAM system that actually worked, assuming such a thing is possible, you'd have to make a number of changes to the way that the school system is designed... e.g. assigning students to teachers at random, designing standardized tests to measure teacher quality, implementing standardized tests in all subject areas, etc. But implementing all of those changes may well be worse for student learning than firing all the bad teachers or whatever. Not to mention that there is currently a major crisis in teacher retention, with half of all teachers either getting fired or quitting in their first five years. (After spending two years and god knows how much money getting their masters degrees to go into the profession.)
(The main exception is the bit about "full range of achievement". It's quite true that teachers of upper class Asian students will likely have lower VAM scores than they deserve since the test caps out at 100%.)
We already knew that the contributions of teachers to student achievement was relatively small long before VAM was invented.
You dodged the question. Do you think policymakers and politicians should stop trying to improve teachers, and instead simply try to reduce costs? I.e., pay teachers less, increase their turnover (so that we don't have to pay expensive but ineffective senior ones), etc?
What were you originally referencing in terms of bias?
> Do you think policymakers and politicians should stop trying to improve teachers, and instead simply try to reduce costs? I.e., pay teachers less, increase their turnover (so that we don't have to pay expensive but ineffective senior ones), etc?
I mean the answer to any of those types of questions really depend on what kind of school system you want to create as a whole, you could go either way on any of them and have either a good or a bad school system depending on the overall design choices. But my point was that I don't think that VAM has much to say about any of these questions, because whether or not it's theoretically possible to measure teacher performance doesn't say much about whether or not good and bad teachers actually exist and whether or not they have an impact on students learning/lives. My point was that the reason we can't measure teacher performance isn't because the difference between good and bad teachers is small, even though it is relatively small in the grand scheme of things.
Seeing as the entire use case of VAM is as a tool to decide which teachers to give bonuses to and which ones to fire, variance is pretty much the only thing that matters.
E.g. if the scores have no stability so who gets fired is basically just a coin toss (as these links suggest) then that's really important, but if these tests systematically overestimate or underestimate teacher performance then that doesn't really matter.
If scores have high variance but are unbiased then you simply need more samples. I.e., wait 2 years and take a teachers aggregate VAM over ~12-16 classes.
"As expected, the level of uncertainty is higher when only one year of test results are used [...] as against three years of data [...]. But in both cases, the average range of value-added estimates is very wide. For example, for all teachers of math, and using all years of available data, which provides the most precise measures possible, the average confidence interval width is about 34 points (i.e., from the 46th to 80th percentile). When looking at only one year of math results, the average width increases to 61 percentile points. That is to say, the average teacher had a range of value-added estimates that might extend from, for example, the 30th to the 91st percentile. The average level of uncertainty is higher still in ELA [English Language Arts]. For all teachers and years, the average confidence interval width is 44 points. With one year of data, this rises to 66 points."
"The Mathematica models, which apply to teachers in the upper elementary grades, are based on two standard approaches to value-added modeling, with the key elements of each calibrated with data on typical test score gains, class sizes, and the number of teachers in a typical school or district. Specifically, the authors find that if the goal is to distinguish relatively high or relatively low performing teachers from those with average performance within a district, the error rate is about 26% when three years of data are used for each teacher. This means that in a typical performance measurement system, more than one in four teachers who are in fact teachers of average quality would be misclassified as either outstanding or poor teachers, and more than one in four teachers who should be singled out for special treatment would be misclassified as teachers of average quality. If only one year of data is available, the error rate increases to 36%. To reduce it to 12% would require 10 years of data for each teacher."
So even with three years of data, the scores are fairly unstable, and this isn't even taking into account the fact that even if they are stable they aren't necessarily correct -- there are many more sources of error that would still be stable over time.
Furthermore, short term student gains correlate poorly with longterm student gains, so if you wanted to measure longterm student gains (presumably the most important metric) then you'd have to wait an addition couple of years after each grade before testing. This means that, in addition to the 3+ years of data you'd need to have even a semi-stable variance, it would take at least 5 years total to have even decent data:
Economist Jesse Rothstein found that students' learning decays over time, and that the teachers who had the biggest added value gains at the end of the first year didn't necessarily have the biggest gains at the end of years two or three when the same students were retested. According to the Rothstein paper, "Three aspects of the results are of note. First, there is much more variation in fourth grade teachers' effects on fourth grade scores than in those same teachers' effects on fifth grade scores. [...] Second, the average persistence of fourth grade teachers' effects one year later is only around 0.3, again consistent with recent evidence. Third, the data are not even approximately consistent with the notion that this persistence rate is uniform across teachers: The correlation between teachers' first-year effects and their two year cumulative effects is much less than one, ranging between .33 and .51 depending on the model and subject. Three-year cumulative effects show a similar pattern, correlated around .4 with the immediate effect. Even if we assume that the VAM-based estimates can be treated as causal, a teacher's first-year effect is a poor proxy for his or her longer-run impact. [...] Only about a third of teachers in the top quintile of the distribution of two year cumulative effects are also in the top quintile of the o...
It sounds like a) we don't have a good measure of what we hope to accomplish by having teachers and b) most of their measurable effect vanishes quickly.
Sounds to me like "teachers don't matter, lets try to cut costs". Do you disagree with this?
It wasn't just the measure, it was that the experiment was randomized. Peer evaluations can be done for courses without follow ons and are more useful without randomization.
I agree that he should be more critical of his own recommendations. It's not clear that faculty would be free of the "gender, and ethnicity, and attractiveness" biases studies suggest students have -- I was hoping he'd have a clever way to blind evaluations.
The methodology used in the study from Italy[1] mentioned in the article was quite interesting, because it had actual random assignment of comparable groups of students to different instructors. The students were then followed up over their further study in the same university. That's a good kind of data set to have for examining issues like this.
The alternative means of evaluating teachers mentioned in the article are also quite reasonable. Both peer review and content analysis of instructor-prepared materials can identify better teachers with different sources of bias in the evaluation, allowing a triangulation with student ratings.
Here on Hacker News, from another participant, I learned about a more rigorous method of student ratings of teachers[2] that ought to be applied (with appropriate adaptations) to higher education teaching. It appears to work will in K-12 teaching.
When I was at Columbia, we used an interesting website called CULPA ("Columbia Undergraduate Listing of Professor Ability"; http://culpa.info/). The site is interesting is that unlike many other sites, students can only give written evaluations and there are no numerical scores. Reviews are generally thoughtful and, in some cases, were particularly useful in choosing a professor for a course.
It's funny how most of the good universities in the US started out as being explicitly Christian, but over the last couple centuries the Christianity has been pretty much wholesale replaced with numerology -- student grades, teacher evaluations, impact factors, statistical significance, etc. The entire system is built upon the worship of numbers, and every phenomenon is assigned some sort of numerical value (to of course be aggregated and statistically 'analyzed') even if it's plainly non-quantitative in nature.
You've got to wonder whether this is some sort of temporary societal blind spot, or whether this ideology is so toxic and so subtle and complex to refute that it will just be around forever as the endpoint of scientific history.
Humans are prone to generalisation, reducing complex objects or states to several numeric parameters is extremely useful. Full of errors and generalization errors, but nevertheless usefull.
Humanity will just have to build more complex models to quantify complex, dynamic objects, such as courses. There is nothing special in human interaction data that inherently prevents its quantification. We just need more data and more efficient algorithms.
Were evaluations mandatory? Did they have any required length? I've done plenty of evaluations in the past which only asked for written responses but was generally disappointed to hear from classmates that they either scribbled gibberish to make it look like they were filling it out or, if the teacher was good, just writing things like "Thanks" or "Great class" in each area.
Yes, I used something like this as a student as well, and found it much more useful than a numerical rating. In cases where there were enough thoughtful reviews, it let me figure out (with admittedly some error) that certain professors tended to teach in a style that either was or wasn't attractive to me. Which is not the same as "good" or "bad"; the kinds of courses I find most valuable were not the ones everyone would find most valuable.
Sometimes even negative comments would give me information that I should take a course. I remember one detailed negative review that complained that the lectures were generally too low on covering information needed to successfully complete the assignments, mostly referring to aspects of C++ (it was a "data structures" course using C++). Instead, the review complained, the lectures were too high-level and theoretical and not helpful in writing code. Which is exactly how I like my lectures: I want lectures to give me a map of the landscape to help orient me towards what's important, how things connect, and where I should look for more information. And I really hate the kind of class that spends all the class time belaboring details of what I should type into a C++ source file, which I can look up myself. If it had been a numerical rating, this very useful piece of positive information (for me) would've been treated negatively!
I don't think it has anything to do with Spanish though, since it's really of Latin origin like many Spanish words. Culpa and words like culpable are derived from Latin (like the Latin phrase, mea culpa [my fault]).
At the end of the semester, students who enjoyed the course should just tip their professors. Adding a real free-market incentive would increase class size and teaching quality, right? Let's let the invisible hand of the market decide; it is certainly more motivating than course evaluations.
I have taught at two higher-ed institutions. In one, each student was given a 20-question evaluation. Of those twenty questions, two were of any importance to the instructor: overall instructor and overall course. The other eighteen were frequently ignored by administration and had absolutely zero influence on tenure, promotion, raises or retention. Those two examined scores were averaged over all course over the entire year, but not weighted - you'd have a six-person course count as much toward your final average as a sixty-person course. The scores were from 1 (bad) to 5 (good). You needed an overall average of 3 on each scale to avoid being fired. Higher scores yielded no tangible (or intangible) rewards. Every faculty member taught enough grad-level courses (PhD in particular tend to yield all 5s) that nobody ever really worried about their evaluations from an enforcement perspective. Those of us who cared about actually improving our practice had better ways to get honest reflections from their students. Personally, I tended to get low-ish evaluations from students expecting an easy A ("Professor __ expects his students to work really hard and learn" "He thinks he's the King of Science" etc etc), and high evaluations from students who wanted to learn the stuff the course was about.
Of course, I also got better over time and so did my course evaluations. I also got better at gaming the system. Add too much reading to the syllabus, and remove an assignment or two so you can be "responsive to student needs", for example.
At the other institution, course evaluations were ignored entirely. They played no role whatsoever in anything. On the other hand, I had a better time teaching those students, and according to my evaluations that had no impact on my career, I was doing a better job of it to boot.
Of course, anecdotes vs data, etc etc. Still, I think tips are the way to go. We already know mean course evaluations can be predicted by the grades students expect to get [1], so let's just flip that around a little.
At the end of the semester, students who enjoyed the course should just tip their professors. Adding a real free-market incentive would increase class size and teaching quality, right? Let's let the invisible hand of the market decide
First of all, broke college students certainly aren't going to 'tip' their professors. I guess perhaps you're thinking part of tuition will be used for this?
Still, what students 'enjoy' is not necessarily what is most beneficial for them in a college course. I think you're setting up the incentives wrong.
Still, what students 'enjoy' is not necessarily what is most beneficial for them in a college course. I think you're setting up the incentives wrong.
Exactly. The majority of them just want the highest grades with the least effort, and the professors who will quite naturally want to get tipped more, will optimise toward that. It'll create a situation where teaching next-to-nothing and giving everyone perfect scores gets the highest rewards; i.e. not much of an education.
That may be true for some, but not for me. Not only was I paying most of the tuition bill, I selected classes based on what I though was knowledge I needed for the career I wanted. An "easy A" didn't figure into it.
I wasn't interested in paying lots of money for nothing. I also wasn't interested in attending a university known for easy grading.
Quoting article: "There's an intriguing exception to the pattern: Classes full of highly skilled students do give highly skilled teachers high marks. Perhaps the smartest kids do see the benefit of being pushed."
Maybe you are in that "highly skilled" group. Unfortunately, average student have different behavior.
I don't know if that necessary corresponds to being highly skilled or not.
But I like being pushed hard in a class - it brings out the best in me. Easy A's just engender boredom and contempt.
I remember once being pushed to enter an athletic competition where the field was not so good. "It'll be an easy win for you" I was told. That made me completely uninterested. I entered one where I was told "You'll be in way over your head." and I was. I came in last - but there was never a more satisfying one to me, because it was my best job ever, and just being on the field with those other excellent competitors was exciting.
An analogy would be just qualifying for the Olympic team, even if you have no hope of getting a gold.
> students for whom "getting out of here with high grades" is their motivation
What a sad way to live, and I wouldn't hire one of them, either.
I found the course catalog at college to be exciting reading, and was frustrated to be only able to take a small fraction of what was available.
I remember several "aha!" moments where suddenly I understood something that had always baffled me, and felt like I'd been given the keys to the universe.
Notice that if you buy into the signaling theory of education this behavior is instantly explained. The primary purpose of college is achieved by the time you get an offer letter. "We certify that you have a high enough IQ and conscientiousness to get into our school." Spiffy, job guaranteed! Now you get four years to find yourself and drink a lot. But for some reason you are constantly beset by atheist monks who think you should be praying more.
At the end of the semester, students who enjoyed the course should just tip their professors. Adding a real free-market incentive would increase class size and teaching quality, right? Let's let the invisible hand of the market decide
He said sarcastically. Ah well, I guess that's a lesson learned re: snark and downvotes. There's a lot to say about the 'signaling' point made upthread, and I agree with that.
Anyway, the important issue here is that course evaluations can serve two purposes. One purpose is the structural elimination of "bad" instructors, by linking promotion and retention decisions to course evaluations at the departmental and/or university level. I can tell you from my perspective, course evaluations have never realistically served this purpose at the two institutions where I taught.
The other purpose is to provide concrete feedback to the instructor so that they can improve their teaching the next semester. Official course evaluations are rarely useful for this purpose either, and in my experience ask the wrong questions (and at the wrong time - wouldn't it be better to get feedback from your students after three weeks so you can make course corrections?) So good professors tend to supplement official feedback forms with feedback instruments of their own design.
Poor professors, getting arbitrary ratings from people they meet once or twice a week for a few weeks, which determine their professional future. Sounds like being a student or something. Go suck on a lemon, professors.
The way things work at my university, at least the science faculty, is that word of good or bad teachers just travels trough the grapevine. And it is usually quite accurate.
Then we have (a active, and paid) representation in the faculties decision making body, which leads to, in my experience, the faculty actually dealing with bad professors.
Not everything has to be boxed in by numerics, sometimes simply speaking up and listening is the easiest and best solution.
So here is a fun fact about professor performance. When I was a grad student, various calculus classes had a shared final. We had some professors with great evaluations who devoted their life to teaching. We had some Germans and Chinese who students perceived to barely speak English [1] and hated. Student evals were fairly predictable - hard exams would reduce scores, jokes and sympathy would increase them.
Everyone's students followed the exact same normal curve.
I've been told that the only way one can make an observable difference in group final scores is to schedule a class at the same time as sports practice or remedial education. You need better students, not a better teacher.
[1] I've noticed that either students are unable/unwilling to listen to a foreign accent, or perhaps I am unable to notice them. Recently a friend of mine said my secretary had a heavy accent - I barely notice it.
Or they are all equally good, or they all have exactly the same wrinkle on their nose, as the very point of this article is that student evaluations don't measure what we think they do.
First, an 18 year old has less experience deciphering a thick accent than someone with years of experience in IT. I don't blink at accents that would have baffled me a decade ago.
Second, if there is cognitive load devoted to understanding the words, that is energy not devoted to understanding the concepts. Friends in those classes in college spent extra time outside of class with tutors and study groups than those with good teachers because they knew the tests didn't change.
Third, the lack of variance means the teachers don't make a difference in general. Perhaps there is a greater question to ask there.
concerning accents, I've had a lot of experience "dealing with"[1] people who speak English with a foreign accent, so I thought I reached a point where I could pretty much decipher anything automatically.
a year or so ago I went to a research conference, with a lot of people from Europe presenting. Generally, everyone there had a good handle on the language (as you might when presenting a bunch of stuff in English all the time), but about halfway through an American came on stage to present her work.
The instant she started talking I realised that listening to her was so much easier than listening to anyone else talk. I "confirmed" this from other talks by americans compared to european talks.
Not the most scientific thing but generally I think there is extra work happening in the brain when listening to a foreigner talk, and that it can be the point that makes something hard to deal with.
[1] I mean this with as little animosity as possible. Having the courage to learn a foreign language and then present in it is extremely commendable.
When evaluating universities for their CS programs before choosing one, accent was a factor in my choosing. One university seemed to only hire grad students to teach since they did not want to pay competitively for professors from the field. Theses foreign grad students did not have a love for teaching (that I could observe).
The university I finally chose did pay competitively to the field, most loved teaching and their research. Yes there were still a few with an accent which pronounced my name wrong, but they've had at least some speech therapy on the side to get better.
I think that foreign professors that teach should take part in a speech therapy program for the benefit of the university students. It truly is a cognitive load, and I already have difficulty in hearing and comprehending words as it is without any particular accent.
I TAed for a group of students who yelled and complained on the first day that they couldn't understand a word their professor was saying. She had the mildest of eastern European accents, and my students were all just being babies. Five weeks into the course I asked them, "Is the accent thing still a problem?" and they all agreed, "Oh, no not at all we got used to it." I will never listen to a student who complains about an accent ever again.
I understand the spirit of your comment, and the statistical difficulties in doing ratings in a meaningful way as described in TFA, but it is absolute truth that there are teachers who are absolutely _wretched_ and when I was an undergrad it drove many of us nuts to have no way to identify them. The problem might have been especially severe when I was an undergrad, since I went to one of the most populous schools in the US, before the proliferation of social medial and the like made a lot of this information gathering easier. But it was a real problem. The fact that in the case you mention all the students had the same normal curve (by which I assume you mean they had the same means and also the same shape) is interesting, but there's more to being a good or bad teacher than the numeric outcome of the class, which ought to be a familiar argument by this time.
Another problem that you mention, and that was also very real, was that some of the teachers literally could not be understood. I use 'literally' literally -- I took an automata theory class from a guy that seemed to have been speaking his own native tongue and not English. That class was an absolute delight, as you might imagine. Most of my teachers had strong accents, and I could deal with it, but some were incomprehensible and had no business teaching in am American university.
Acting like these are problems that don't exist, and that complaints are just small-mindedness or excuse-making, is just as bad as going the other way and throwing a fit when you have to 'endure' an unfamiliar accent. Everything's a matter of degree.
...there are teachers who are absolutely _wretched_...
Are there? We thought the same thing about our "worst" professors/TAs. Yet students ability to differentiate/integrate seemed unaffected.
Now this suggests one of two things. First, that there really are no "worst" professors/TAs. Second, that students have the ability to replace a $3-4,000 class with reading a $50-150 book (+ Khan academy, etc).
In the former case it suggests teacher quality don't matter much. In the latter case it suggests teachers are completely unnecessary and our usage of them should be reduced. This experiment at MIT is promising:
Are there? We thought the same thing about our "worst" professors/TAs. Yet students ability to differentiate/integrate seemed unaffected.
That's a data point worth paying attention to. However, I'll submit that the intellectual world is larger than integration and differentiation. I'll further suggest that there is a world of difference between understanding enough calculus to integrate and differentiate, and understanding calculus in sufficient depth to apply its principles to problems and in settings that do not afford simple template matching -- I wrote an integration/differentiation package in Scheme my freshman year, but would not claim that it understands calculus.
This is one of the ways in which a good teacher and a bad teacher can vary by an order of magnitude. You might counter that your tests were so rigorous that any such differences would fall out in the data. Perhaps you have achieved this apex of testing methodology, but I'm suspicious.
Now this suggests one of two things. First, that there really are no "worst" professors/TAs. Second, that students have the ability to replace a $3-4,000 class with reading a $50-150 book (+ Khan academy, etc).
Or that your measurements are not sufficiently granular, as described above.
This is a familiar platitude. I've never seen an actual argument that didn't appeal to invisible dragons.
Many parents produce children who survive long enough to successfully reproduce themselves. By one of the most fundamental biological heuristics that's a rip-roaring success, and yet presumably you would aspire to more with your own children. How's that for an invisible dragon?
I'll submit that the intellectual world is larger than integration and differentiation.
Calculus 1 and 2 are not. I don't see a reason why the non-simple problems you describe can't be put on an exam, apart from the fact that it's not part of Calc 1. I've also never observed a group of students from any professor who could reliably do it, so I suspect that your "order of magnitude" better teachers are a rarity.
How's that for an invisible dragon?
Are you seriously asserting that outcomes beyond reproduction are not measurable?
What about taking into account each student's performance in the class? In other words, partition the evaluations into sets based on the students' performance. Then you'd end up with a spectrum of instructor ratings, from F to A, not an average.
If the institution wants to make a choice of instructors for an introductory course, they might want to select the instructor with the higher low-grade rating. The intuition here is that well-performing students in an introductory course might have prior knowledge, so their ratings add noise.
If the institution wants to compare two instructors overall, they can compare their spectra after subtracting the skew that results from low-performers evaluating poorly and high-performers evaluating highly. What remains is some picture of how the instructor handles students of different strengths and weaknesses, which (I think) is the best measure of a teacher.
"If the institution wants to make a choice of instructors for an introductory course, they might want to select the instructor with the higher low-grade rating. The intuition here is that well-performing students in an introductory course might have prior knowledge, so their ratings add noise."
The complain in the article is that with exception of well-performing students, students rating does not measure how much they learned (measured be their performance in subsequent courses). Lower performing students tend to give lower marks to those teachers that pushed them to learn more.
Your plan maximizes satisfaction of low performing students in introductory course. It does not maximizes how much they will learn, quite the opposite. University should strive for the best teaching, not for the highest immediate satisfaction. It has importance too, but should not be more important then actual learning.
UChicago's rating system was quite useful, actually. The survey is given before the final, so exam difficulty doesn't matter. Professors who deliver insights you really learn from get the high ratings and those who don't (aka: they dwell in detail instead of creating understanding, can't communicate, read slides, make you fall asleep, or teach outdated material lose out). Over the years, professor ratings and course bidding points became a much better indicator of the quality of a course section than the description of a class.
From all of my experiences and everything every faculty member and career panel has ever told me (about research university positions in mathematics and computer science), teaching does not matter at all. Hiring boards don't look at your teaching statement even though they require one. Nobody reads a letter of recommendation if it is about teaching. Nobody reads your course evaluations. And in particular, the more you neglect teaching (and the more time you spend on research) the easier it will be to get tenure.
As sad as it sounds, that's how my field is. So is this report specific to other fields?
55 comments
[ 5.3 ms ] story [ 2388 ms ] threadBecause grades received in later courses isn't actually a meaningful metric. There was a theory that you could measure teacher quality this way using something called Value Added Metrics (VAM), but it's now been completely debunked. C.f.
http://www.epi.org/publication/bp278/
http://www.nber.org/papers/w14442
http://annenberginstitute.org/sites/default/files/product/21...
http://ies.ed.gov/ncee/pubs/20104004/pdf/20104004.pdf
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1142237
http://educationnext.org/files/ednext20022_10.pdf
Your first (accurately) says "measurements are noisy" and the inaccurately says "and other things matter for learning too". Unless I missed something, it doesn't provide a reason to expect bias in the other things.
The second link says that some VAM models assume student assignment is random when it is not, and also that measured educational outputs don't matter much for outcomes anyway. The logical conclusion of this is simply that VAM needs modification, and maybe we should reduce education. Claiming this debunks the idea of VAM is like saying "hah, this paper proves OLS doesn't always work well, therefore regression is debunked".
The third is really long and I couldn't figure out it's point. The fourth one says VAM is noisy.
If you believe VAM is unable to measure the effects of teachers, are you willing to follow through with the logical conclusions? Namely, the effect teachers have on students is so small it can't be measured and we should therefore focus on other things (e.g. reducing cost) instead?
"VAM’s instability can result from differences in the characteristics of students assigned to particular teachers in a particular year, from small samples of students (made even less representative in schools serving disadvantaged students by high rates of student mobility), from other influences on student learning both inside and outside school, and from tests that are poorly lined up with the curriculum teachers are expected to cover, or that do not measure the full range of achievement of students in the class."
> If you believe VAM is unable to measure the effects of teachers, are you willing to follow through with the logical conclusions? Namely, the effect teachers have on students is so small it can't be measured ...
We already knew that the contributions of teachers to student achievement was relatively small long before VAM was invented. The only people who seem to think that teachers have a very large impact are the pro-VAM folks (e.g. Bill Gates and Malcolm Gladwell), who all cite the same flawed study.
That said, the reasons why VAM can't measure teacher performance isn't because the contributions of teachers are small per se, but rather because there are all sorts of problems with all of the proposed measurement systems.
In order to create a VAM system that actually worked, assuming such a thing is possible, you'd have to make a number of changes to the way that the school system is designed... e.g. assigning students to teachers at random, designing standardized tests to measure teacher quality, implementing standardized tests in all subject areas, etc. But implementing all of those changes may well be worse for student learning than firing all the bad teachers or whatever. Not to mention that there is currently a major crisis in teacher retention, with half of all teachers either getting fired or quitting in their first five years. (After spending two years and god knows how much money getting their masters degrees to go into the profession.)
Your quote concerns variance, not bias.
(The main exception is the bit about "full range of achievement". It's quite true that teachers of upper class Asian students will likely have lower VAM scores than they deserve since the test caps out at 100%.)
We already knew that the contributions of teachers to student achievement was relatively small long before VAM was invented.
You dodged the question. Do you think policymakers and politicians should stop trying to improve teachers, and instead simply try to reduce costs? I.e., pay teachers less, increase their turnover (so that we don't have to pay expensive but ineffective senior ones), etc?
What were you originally referencing in terms of bias?
> Do you think policymakers and politicians should stop trying to improve teachers, and instead simply try to reduce costs? I.e., pay teachers less, increase their turnover (so that we don't have to pay expensive but ineffective senior ones), etc?
I mean the answer to any of those types of questions really depend on what kind of school system you want to create as a whole, you could go either way on any of them and have either a good or a bad school system depending on the overall design choices. But my point was that I don't think that VAM has much to say about any of these questions, because whether or not it's theoretically possible to measure teacher performance doesn't say much about whether or not good and bad teachers actually exist and whether or not they have an impact on students learning/lives. My point was that the reason we can't measure teacher performance isn't because the difference between good and bad teachers is small, even though it is relatively small in the grand scheme of things.
http://en.wikipedia.org/wiki/Variance
Alex, I'd suggest you refrain from opining that standard statistical methods fail in education until after you master basic statistics.
E.g. if the scores have no stability so who gets fired is basically just a coin toss (as these links suggest) then that's really important, but if these tests systematically overestimate or underestimate teacher performance then that doesn't really matter.
"As expected, the level of uncertainty is higher when only one year of test results are used [...] as against three years of data [...]. But in both cases, the average range of value-added estimates is very wide. For example, for all teachers of math, and using all years of available data, which provides the most precise measures possible, the average confidence interval width is about 34 points (i.e., from the 46th to 80th percentile). When looking at only one year of math results, the average width increases to 61 percentile points. That is to say, the average teacher had a range of value-added estimates that might extend from, for example, the 30th to the 91st percentile. The average level of uncertainty is higher still in ELA [English Language Arts]. For all teachers and years, the average confidence interval width is 44 points. With one year of data, this rises to 66 points."
"The Mathematica models, which apply to teachers in the upper elementary grades, are based on two standard approaches to value-added modeling, with the key elements of each calibrated with data on typical test score gains, class sizes, and the number of teachers in a typical school or district. Specifically, the authors find that if the goal is to distinguish relatively high or relatively low performing teachers from those with average performance within a district, the error rate is about 26% when three years of data are used for each teacher. This means that in a typical performance measurement system, more than one in four teachers who are in fact teachers of average quality would be misclassified as either outstanding or poor teachers, and more than one in four teachers who should be singled out for special treatment would be misclassified as teachers of average quality. If only one year of data is available, the error rate increases to 36%. To reduce it to 12% would require 10 years of data for each teacher."
So even with three years of data, the scores are fairly unstable, and this isn't even taking into account the fact that even if they are stable they aren't necessarily correct -- there are many more sources of error that would still be stable over time.
Furthermore, short term student gains correlate poorly with longterm student gains, so if you wanted to measure longterm student gains (presumably the most important metric) then you'd have to wait an addition couple of years after each grade before testing. This means that, in addition to the 3+ years of data you'd need to have even a semi-stable variance, it would take at least 5 years total to have even decent data:
Economist Jesse Rothstein found that students' learning decays over time, and that the teachers who had the biggest added value gains at the end of the first year didn't necessarily have the biggest gains at the end of years two or three when the same students were retested. According to the Rothstein paper, "Three aspects of the results are of note. First, there is much more variation in fourth grade teachers' effects on fourth grade scores than in those same teachers' effects on fifth grade scores. [...] Second, the average persistence of fourth grade teachers' effects one year later is only around 0.3, again consistent with recent evidence. Third, the data are not even approximately consistent with the notion that this persistence rate is uniform across teachers: The correlation between teachers' first-year effects and their two year cumulative effects is much less than one, ranging between .33 and .51 depending on the model and subject. Three-year cumulative effects show a similar pattern, correlated around .4 with the immediate effect. Even if we assume that the VAM-based estimates can be treated as causal, a teacher's first-year effect is a poor proxy for his or her longer-run impact. [...] Only about a third of teachers in the top quintile of the distribution of two year cumulative effects are also in the top quintile of the o...
Sounds to me like "teachers don't matter, lets try to cut costs". Do you disagree with this?
I agree that he should be more critical of his own recommendations. It's not clear that faculty would be free of the "gender, and ethnicity, and attractiveness" biases studies suggest students have -- I was hoping he'd have a clever way to blind evaluations.
The alternative means of evaluating teachers mentioned in the article are also quite reasonable. Both peer review and content analysis of instructor-prepared materials can identify better teachers with different sources of bias in the evaluation, allowing a triangulation with student ratings.
Here on Hacker News, from another participant, I learned about a more rigorous method of student ratings of teachers[2] that ought to be applied (with appropriate adaptations) to higher education teaching. It appears to work will in K-12 teaching.
[1] http://www.sciencedirect.com/science/article/pii/S0272775714...
[2] https://news.ycombinator.com/item?id=4559682
It's funny how most of the good universities in the US started out as being explicitly Christian, but over the last couple centuries the Christianity has been pretty much wholesale replaced with numerology -- student grades, teacher evaluations, impact factors, statistical significance, etc. The entire system is built upon the worship of numbers, and every phenomenon is assigned some sort of numerical value (to of course be aggregated and statistically 'analyzed') even if it's plainly non-quantitative in nature.
You've got to wonder whether this is some sort of temporary societal blind spot, or whether this ideology is so toxic and so subtle and complex to refute that it will just be around forever as the endpoint of scientific history.
Humanity will just have to build more complex models to quantify complex, dynamic objects, such as courses. There is nothing special in human interaction data that inherently prevents its quantification. We just need more data and more efficient algorithms.
Sometimes even negative comments would give me information that I should take a course. I remember one detailed negative review that complained that the lectures were generally too low on covering information needed to successfully complete the assignments, mostly referring to aspects of C++ (it was a "data structures" course using C++). Instead, the review complained, the lectures were too high-level and theoretical and not helpful in writing code. Which is exactly how I like my lectures: I want lectures to give me a map of the landscape to help orient me towards what's important, how things connect, and where I should look for more information. And I really hate the kind of class that spends all the class time belaboring details of what I should type into a C++ source file, which I can look up myself. If it had been a numerical rating, this very useful piece of positive information (for me) would've been treated negatively!
I have taught at two higher-ed institutions. In one, each student was given a 20-question evaluation. Of those twenty questions, two were of any importance to the instructor: overall instructor and overall course. The other eighteen were frequently ignored by administration and had absolutely zero influence on tenure, promotion, raises or retention. Those two examined scores were averaged over all course over the entire year, but not weighted - you'd have a six-person course count as much toward your final average as a sixty-person course. The scores were from 1 (bad) to 5 (good). You needed an overall average of 3 on each scale to avoid being fired. Higher scores yielded no tangible (or intangible) rewards. Every faculty member taught enough grad-level courses (PhD in particular tend to yield all 5s) that nobody ever really worried about their evaluations from an enforcement perspective. Those of us who cared about actually improving our practice had better ways to get honest reflections from their students. Personally, I tended to get low-ish evaluations from students expecting an easy A ("Professor __ expects his students to work really hard and learn" "He thinks he's the King of Science" etc etc), and high evaluations from students who wanted to learn the stuff the course was about.
Of course, I also got better over time and so did my course evaluations. I also got better at gaming the system. Add too much reading to the syllabus, and remove an assignment or two so you can be "responsive to student needs", for example.
At the other institution, course evaluations were ignored entirely. They played no role whatsoever in anything. On the other hand, I had a better time teaching those students, and according to my evaluations that had no impact on my career, I was doing a better job of it to boot.
Of course, anecdotes vs data, etc etc. Still, I think tips are the way to go. We already know mean course evaluations can be predicted by the grades students expect to get [1], so let's just flip that around a little.
[1]: https://stat.duke.edu/~dalene/chance/chanceweb/153.johnson.p...
First of all, broke college students certainly aren't going to 'tip' their professors. I guess perhaps you're thinking part of tuition will be used for this?
Still, what students 'enjoy' is not necessarily what is most beneficial for them in a college course. I think you're setting up the incentives wrong.
Exactly. The majority of them just want the highest grades with the least effort, and the professors who will quite naturally want to get tipped more, will optimise toward that. It'll create a situation where teaching next-to-nothing and giving everyone perfect scores gets the highest rewards; i.e. not much of an education.
I wasn't interested in paying lots of money for nothing. I also wasn't interested in attending a university known for easy grading.
Maybe you are in that "highly skilled" group. Unfortunately, average student have different behavior.
But I like being pushed hard in a class - it brings out the best in me. Easy A's just engender boredom and contempt.
I remember once being pushed to enter an athletic competition where the field was not so good. "It'll be an easy win for you" I was told. That made me completely uninterested. I entered one where I was told "You'll be in way over your head." and I was. I came in last - but there was never a more satisfying one to me, because it was my best job ever, and just being on the field with those other excellent competitors was exciting.
An analogy would be just qualifying for the Olympic team, even if you have no hope of getting a gold.
On HN, I imagine a significant number of us here are sufficiently motivated to learning even without anyone to remind them of its importance.
What a sad way to live, and I wouldn't hire one of them, either.
I found the course catalog at college to be exciting reading, and was frustrated to be only able to take a small fraction of what was available.
I remember several "aha!" moments where suddenly I understood something that had always baffled me, and felt like I'd been given the keys to the universe.
This is already overwhelmingly the case in an attempt to secure more and more student loan funds.
He said sarcastically. Ah well, I guess that's a lesson learned re: snark and downvotes. There's a lot to say about the 'signaling' point made upthread, and I agree with that.
Anyway, the important issue here is that course evaluations can serve two purposes. One purpose is the structural elimination of "bad" instructors, by linking promotion and retention decisions to course evaluations at the departmental and/or university level. I can tell you from my perspective, course evaluations have never realistically served this purpose at the two institutions where I taught.
The other purpose is to provide concrete feedback to the instructor so that they can improve their teaching the next semester. Official course evaluations are rarely useful for this purpose either, and in my experience ask the wrong questions (and at the wrong time - wouldn't it be better to get feedback from your students after three weeks so you can make course corrections?) So good professors tend to supplement official feedback forms with feedback instruments of their own design.
Then we have (a active, and paid) representation in the faculties decision making body, which leads to, in my experience, the faculty actually dealing with bad professors.
Not everything has to be boxed in by numerics, sometimes simply speaking up and listening is the easiest and best solution.
How do you know? How would you know?
Everyone's students followed the exact same normal curve.
I've been told that the only way one can make an observable difference in group final scores is to schedule a class at the same time as sports practice or remedial education. You need better students, not a better teacher.
[1] I've noticed that either students are unable/unwilling to listen to a foreign accent, or perhaps I am unable to notice them. Recently a friend of mine said my secretary had a heavy accent - I barely notice it.
Second, if there is cognitive load devoted to understanding the words, that is energy not devoted to understanding the concepts. Friends in those classes in college spent extra time outside of class with tutors and study groups than those with good teachers because they knew the tests didn't change.
Third, the lack of variance means the teachers don't make a difference in general. Perhaps there is a greater question to ask there.
a year or so ago I went to a research conference, with a lot of people from Europe presenting. Generally, everyone there had a good handle on the language (as you might when presenting a bunch of stuff in English all the time), but about halfway through an American came on stage to present her work.
The instant she started talking I realised that listening to her was so much easier than listening to anyone else talk. I "confirmed" this from other talks by americans compared to european talks.
Not the most scientific thing but generally I think there is extra work happening in the brain when listening to a foreigner talk, and that it can be the point that makes something hard to deal with.
[1] I mean this with as little animosity as possible. Having the courage to learn a foreign language and then present in it is extremely commendable.
The university I finally chose did pay competitively to the field, most loved teaching and their research. Yes there were still a few with an accent which pronounced my name wrong, but they've had at least some speech therapy on the side to get better.
I think that foreign professors that teach should take part in a speech therapy program for the benefit of the university students. It truly is a cognitive load, and I already have difficulty in hearing and comprehending words as it is without any particular accent.
Another problem that you mention, and that was also very real, was that some of the teachers literally could not be understood. I use 'literally' literally -- I took an automata theory class from a guy that seemed to have been speaking his own native tongue and not English. That class was an absolute delight, as you might imagine. Most of my teachers had strong accents, and I could deal with it, but some were incomprehensible and had no business teaching in am American university.
Acting like these are problems that don't exist, and that complaints are just small-mindedness or excuse-making, is just as bad as going the other way and throwing a fit when you have to 'endure' an unfamiliar accent. Everything's a matter of degree.
Are there? We thought the same thing about our "worst" professors/TAs. Yet students ability to differentiate/integrate seemed unaffected.
Now this suggests one of two things. First, that there really are no "worst" professors/TAs. Second, that students have the ability to replace a $3-4,000 class with reading a $50-150 book (+ Khan academy, etc).
In the former case it suggests teacher quality don't matter much. In the latter case it suggests teachers are completely unnecessary and our usage of them should be reduced. This experiment at MIT is promising:
http://phys.org/news/2014-09-online-classes.html
...there's more to being a good or bad teacher than the numeric outcome of the class, which ought to be a familiar argument by this time.
This is a familiar platitude. I've never seen an actual argument that didn't appeal to invisible dragons.
http://www.godlessgeeks.com/LINKS/Dragon.htm
That's a data point worth paying attention to. However, I'll submit that the intellectual world is larger than integration and differentiation. I'll further suggest that there is a world of difference between understanding enough calculus to integrate and differentiate, and understanding calculus in sufficient depth to apply its principles to problems and in settings that do not afford simple template matching -- I wrote an integration/differentiation package in Scheme my freshman year, but would not claim that it understands calculus.
This is one of the ways in which a good teacher and a bad teacher can vary by an order of magnitude. You might counter that your tests were so rigorous that any such differences would fall out in the data. Perhaps you have achieved this apex of testing methodology, but I'm suspicious.
Now this suggests one of two things. First, that there really are no "worst" professors/TAs. Second, that students have the ability to replace a $3-4,000 class with reading a $50-150 book (+ Khan academy, etc).
Or that your measurements are not sufficiently granular, as described above.
This is a familiar platitude. I've never seen an actual argument that didn't appeal to invisible dragons.
Many parents produce children who survive long enough to successfully reproduce themselves. By one of the most fundamental biological heuristics that's a rip-roaring success, and yet presumably you would aspire to more with your own children. How's that for an invisible dragon?
EDIT: figured out how to make quotes italicized.
These two claims contradict each other.
If you want to argue that the exams couldn't measure anything at all, go ahead. You can find examples here: http://cims.nyu.edu/~stucchio/classes/fall2008/index.html
I'll submit that the intellectual world is larger than integration and differentiation.
Calculus 1 and 2 are not. I don't see a reason why the non-simple problems you describe can't be put on an exam, apart from the fact that it's not part of Calc 1. I've also never observed a group of students from any professor who could reliably do it, so I suspect that your "order of magnitude" better teachers are a rarity.
How's that for an invisible dragon?
Are you seriously asserting that outcomes beyond reproduction are not measurable?
If the institution wants to make a choice of instructors for an introductory course, they might want to select the instructor with the higher low-grade rating. The intuition here is that well-performing students in an introductory course might have prior knowledge, so their ratings add noise.
If the institution wants to compare two instructors overall, they can compare their spectra after subtracting the skew that results from low-performers evaluating poorly and high-performers evaluating highly. What remains is some picture of how the instructor handles students of different strengths and weaknesses, which (I think) is the best measure of a teacher.
The complain in the article is that with exception of well-performing students, students rating does not measure how much they learned (measured be their performance in subsequent courses). Lower performing students tend to give lower marks to those teachers that pushed them to learn more.
Your plan maximizes satisfaction of low performing students in introductory course. It does not maximizes how much they will learn, quite the opposite. University should strive for the best teaching, not for the highest immediate satisfaction. It has importance too, but should not be more important then actual learning.
As sad as it sounds, that's how my field is. So is this report specific to other fields?