I think those definitions — did and treatment on the treated — are not correct? Isn’t the Tot is the wald estimator — reduced form divided by first stage. Not compared to baseline. That is at least my understanding — which is why it corresponds to the LATE. But the did explanation isn’t right. It’s a 2x2 comparison between treatment and control on the long difference.
We may be using terms differently and I was just following Minton Mulligan on treatment on the treated. I’m not worried about compliance here so I’m not sure why I need to think about the Wald estimator. Here they use it to mean treated vs baseline. Again, that may not be the norm but it makes sense to me. I’ll defer to you.
I don’t think we differ on DiD either. I’m normalizing the pre treatment gap to zero. But check out their next figure and tell me if they and I am confused
Super clear explanation of why identification in markets is so slippery. The concert analogy is perfect, the kind of thing that makes you realize how often we confuse measuring substitution within a system versus measuring what happens when the whole system shifts. I ran into a version of this when trying to understand local labor market effects, where the control regions were obviously absorbing displaced workers, but the standard approach just treated that as noise. The Minton Mulligan framing of sepaating DiD from ToT from scale effects feels like it should be required reading beore anyone interprets a treatment study.
Well, I am sorry to report that these key differences were not covered or even mentioned in either of my econometrics courses as an econ major at UCI 😤
This really clicked for me, especially the concert analogy. It nails what’s been bothering me about how DiD results get casually scaled up. IMO, what we often treat as “impact” is really reallocation: who lost relative to whom, not what happened to the system overall.
The “missing intercept” framing helps explain why these debates so often go nowhere. We argue over slopes while the level shift, the part that hits everyone through prices, wages, and policy responses, quietly disappears from view.
Once markets are this networked, clean control groups start to feel more like a convenience than a reality. Curious whether you see this as mostly a communication problem, how results get interpreted, or a deeper limitation in how we study market-wide shocks at all.
The invisible hand doesn’t always serve the common good. Markets need context, accountability, and human values to prevent self‑interest from turning into distortion.
I think those definitions — did and treatment on the treated — are not correct? Isn’t the Tot is the wald estimator — reduced form divided by first stage. Not compared to baseline. That is at least my understanding — which is why it corresponds to the LATE. But the did explanation isn’t right. It’s a 2x2 comparison between treatment and control on the long difference.
We may be using terms differently and I was just following Minton Mulligan on treatment on the treated. I’m not worried about compliance here so I’m not sure why I need to think about the Wald estimator. Here they use it to mean treated vs baseline. Again, that may not be the norm but it makes sense to me. I’ll defer to you.
I don’t think we differ on DiD either. I’m normalizing the pre treatment gap to zero. But check out their next figure and tell me if they and I am confused
Super clear explanation of why identification in markets is so slippery. The concert analogy is perfect, the kind of thing that makes you realize how often we confuse measuring substitution within a system versus measuring what happens when the whole system shifts. I ran into a version of this when trying to understand local labor market effects, where the control regions were obviously absorbing displaced workers, but the standard approach just treated that as noise. The Minton Mulligan framing of sepaating DiD from ToT from scale effects feels like it should be required reading beore anyone interprets a treatment study.
Well, I am sorry to report that these key differences were not covered or even mentioned in either of my econometrics courses as an econ major at UCI 😤
I didn’t understand this. But I WILL be pretending like I did to forward my Austrian priors.
This really clicked for me, especially the concert analogy. It nails what’s been bothering me about how DiD results get casually scaled up. IMO, what we often treat as “impact” is really reallocation: who lost relative to whom, not what happened to the system overall.
The “missing intercept” framing helps explain why these debates so often go nowhere. We argue over slopes while the level shift, the part that hits everyone through prices, wages, and policy responses, quietly disappears from view.
Once markets are this networked, clean control groups start to feel more like a convenience than a reality. Curious whether you see this as mostly a communication problem, how results get interpreted, or a deeper limitation in how we study market-wide shocks at all.
The invisible hand doesn’t always serve the common good. Markets need context, accountability, and human values to prevent self‑interest from turning into distortion.
Super interesting piece!
The geometric explanation with R1 to R6 really made clear how DiD, ToT, and scale effect differ from each other! 👍