Scott Barkowski
@scottbarkowski.bsky.social
Health and labor economist, University at Buffalo (SUNY). Draws his graphs perfectly on the board, the first time, every time.
Website: https://sites.google.com/site/sbarkowski/
Website: https://sites.google.com/site/sbarkowski/
I am not familiar with this code so I could be wrong, but I think you need one more restriction.
January 17, 2026 at 4:25 PM
I am not familiar with this code so I could be wrong, but I think you need one more restriction.
I feel like one way to look at it is you can do a one sided test if you are willing to make a bet and sacrifice your power on the other side. But given the incentives, we should stick to two sided tests in this case unless we prespecify the use of a one sided test, so that we have to stick to it.
January 17, 2026 at 4:23 PM
I feel like one way to look at it is you can do a one sided test if you are willing to make a bet and sacrifice your power on the other side. But given the incentives, we should stick to two sided tests in this case unless we prespecify the use of a one sided test, so that we have to stick to it.
A ton of negative measurement error or just randomness.
January 17, 2026 at 4:18 PM
A ton of negative measurement error or just randomness.
I think you are close. You wanna re-paramaterize so that you have d1=b1-b2, d2=b1-b3, and d3=b2-b3, then you can do what Adam did and test H0: d1 = d2=d3=0 vs H1: d1<0, d2<0, d3<0 using that paper he posted.
January 17, 2026 at 4:15 PM
I think you are close. You wanna re-paramaterize so that you have d1=b1-b2, d2=b1-b3, and d3=b2-b3, then you can do what Adam did and test H0: d1 = d2=d3=0 vs H1: d1<0, d2<0, d3<0 using that paper he posted.
Yeah I can see that happening! And I also don’t mean it as a critique, seem like you made a reasonable approach. But if you find bhat>0, then arent you implying b1>0 and b2>0 and b3>0, not b1>0 and/or b2>0 and/or b3>0?That would seem to be a higher rejection threshold.
January 13, 2026 at 7:53 PM
Yeah I can see that happening! And I also don’t mean it as a critique, seem like you made a reasonable approach. But if you find bhat>0, then arent you implying b1>0 and b2>0 and b3>0, not b1>0 and/or b2>0 and/or b3>0?That would seem to be a higher rejection threshold.
This is a good suggestion. But is the alternative in the test you did the same as the one you wanted to test? Seems like it is a higher threshold, meaning you are less likely to reject/ less powerful.
January 13, 2026 at 6:58 PM
This is a good suggestion. But is the alternative in the test you did the same as the one you wanted to test? Seems like it is a higher threshold, meaning you are less likely to reject/ less powerful.
Alternatively maybe you can express your hypothesis as a conditional moment and use this?
users.ssc.wisc.edu/~xshi28/rese...
users.ssc.wisc.edu/~xshi28/rese...
users.ssc.wisc.edu
January 13, 2026 at 6:53 PM
Alternatively maybe you can express your hypothesis as a conditional moment and use this?
users.ssc.wisc.edu/~xshi28/rese...
users.ssc.wisc.edu/~xshi28/rese...
I thought about it some more and the distribution of restricted least squares with inequality constraints is not normal, so the test would be more complicsted than an f/Wald test.
This paper might help
www.jstor.org/stable/1912529
This paper might help
www.jstor.org/stable/1912529
Likelihood Ratio Test, Wald Test, and Kuhn-Tucker Test in Linear Models with Inequality Constraints on the Regression Parameters on JSTOR
, Alberto Holly, Alain Monfort, Likelihood Ratio Test, Wald Test, and Kuhn-Tucker Test in Linear Models with Inequality Constraints on the Regression Parameters, Econometrica, Vol. 50, No. 1 (Jan., 19...
www.jstor.org
January 13, 2026 at 6:53 PM
I thought about it some more and the distribution of restricted least squares with inequality constraints is not normal, so the test would be more complicsted than an f/Wald test.
This paper might help
www.jstor.org/stable/1912529
This paper might help
www.jstor.org/stable/1912529
One way to do this could be to do restricted least squares to estimate your restricted model.
January 13, 2026 at 5:17 PM
One way to do this could be to do restricted least squares to estimate your restricted model.
Can you embed these assumptions into your model to get a restricted version of your full model? If so, then you should be able to do a standard f test.
January 13, 2026 at 5:15 PM
Can you embed these assumptions into your model to get a restricted version of your full model? If so, then you should be able to do a standard f test.
Their numbers likely reflect their decades of exposure to recent finances since then, at least in part. So you’re overstating what they considered financially successful back then.
January 12, 2026 at 3:32 AM
Their numbers likely reflect their decades of exposure to recent finances since then, at least in part. So you’re overstating what they considered financially successful back then.
Seems like you’re comparing inflation adjusted (at least implicitly) numbers to ones that weren’t. Hard to imagine boomers recall exactly what they would have thought financially successfully meant in 1973 precisely when they were asked about it in 2024.
January 12, 2026 at 3:31 AM
Seems like you’re comparing inflation adjusted (at least implicitly) numbers to ones that weren’t. Hard to imagine boomers recall exactly what they would have thought financially successfully meant in 1973 precisely when they were asked about it in 2024.
I Would consider the issue with calling his work at McKinsey as peer reviewed as potentially a problem for a university president, but it would also be a problem for a Dean, too. So why wasnt it an issue before?
January 12, 2026 at 3:01 AM
I Would consider the issue with calling his work at McKinsey as peer reviewed as potentially a problem for a university president, but it would also be a problem for a Dean, too. So why wasnt it an issue before?
Apparently he was hired as the business school dean 10 years ago straight from McKinsey, and he did not have a PhD (or maybe just earned it at the time). So wasnt it totally apparent that the guy is not an academic? If so, is it still fair to critique the guy basically For not being an academic?
January 12, 2026 at 2:57 AM
Apparently he was hired as the business school dean 10 years ago straight from McKinsey, and he did not have a PhD (or maybe just earned it at the time). So wasnt it totally apparent that the guy is not an academic? If so, is it still fair to critique the guy basically For not being an academic?
Interesting experiment! That said, couldn’t some of the response have been fraudsters? And couldn’t there be more fraud that didn’t respond not respond at all? I don’t see why we should expect compilers to be a large portion of the population of fraudsters.
March 11, 2025 at 2:33 PM
Interesting experiment! That said, couldn’t some of the response have been fraudsters? And couldn’t there be more fraud that didn’t respond not respond at all? I don’t see why we should expect compilers to be a large portion of the population of fraudsters.
Sorry Mike that is *really far* something they’d consider. Lebron is still putting up 24/8/9.
Besides, that wouldn’t work under the cap. Salaries would be extremely far from being matched.
Besides, that wouldn’t work under the cap. Salaries would be extremely far from being matched.
February 4, 2025 at 12:48 AM
Sorry Mike that is *really far* something they’d consider. Lebron is still putting up 24/8/9.
Besides, that wouldn’t work under the cap. Salaries would be extremely far from being matched.
Besides, that wouldn’t work under the cap. Salaries would be extremely far from being matched.
Enrollment increasing doesn’t mean the FAFSA didn’t ramp down on college going, though. So don’t draw conclusions until we can estimate the counterfactual!
January 14, 2025 at 4:14 AM
Enrollment increasing doesn’t mean the FAFSA didn’t ramp down on college going, though. So don’t draw conclusions until we can estimate the counterfactual!
Which one? The one who was right about them, or the one who was wrong?
January 14, 2025 at 2:09 AM
Which one? The one who was right about them, or the one who was wrong?
you reduce options for low income renters. That because it’s all linked: wealthier renters can outbid the less wealthy. In the end, it’s not the wealthy who can’t find housing, it’s the poor. But all the renters are in less preferred housing. Everyone is worse off. Consider removing this article!
January 13, 2025 at 2:45 AM
you reduce options for low income renters. That because it’s all linked: wealthier renters can outbid the less wealthy. In the end, it’s not the wealthy who can’t find housing, it’s the poor. But all the renters are in less preferred housing. Everyone is worse off. Consider removing this article!
David, what are you doing here? You’re single-handedly reducing LA area housing supply when people desperately need housing. This literally makes everyone worse off: the renter who loses out on housing he or she would have been willing to pay for, the owner who would make extra money, and worse…
January 13, 2025 at 2:45 AM
David, what are you doing here? You’re single-handedly reducing LA area housing supply when people desperately need housing. This literally makes everyone worse off: the renter who loses out on housing he or she would have been willing to pay for, the owner who would make extra money, and worse…
Yes this is one way of interpreting it.
December 17, 2024 at 8:18 PM
Yes this is one way of interpreting it.
estimate all the covariances.
December 17, 2024 at 8:17 PM
estimate all the covariances.
I guess for you the safest thing to do would be to estimate separate models for each group and compare estimates that way. If you want to test the difference then you can do fully interacted models or, even better, SUR models that combine the groups into two totally separate regressions but also…
December 17, 2024 at 8:17 PM
I guess for you the safest thing to do would be to estimate separate models for each group and compare estimates that way. If you want to test the difference then you can do fully interacted models or, even better, SUR models that combine the groups into two totally separate regressions but also…