Thread by @NoahHaber

@economeager @statsepi Oh man there are SO MANY DIFFERENCES between econ and epi, I don't even know where to begin!!! (background: formal training mostly in econometrics, been doing crossover into the epi world for the last few years). Also, pinging @Emma_C_Clarke

THREAD TIME!

@economeager @statsepi @Emma_C_Clarke But before that DISCLAIMER TIME! I am going to be making overly broad statements about epi vs. econ. None of this is binary, and insert a mental "are more likely to" everywhere. I am also more econ than epi. That could be considered a perspective/bias/whatever.

@economeager @statsepi @Emma_C_Clarke At the root of the methodological differences between epi and econ is half cultural, half stats philosophical, and focus on causal inference (CI). While both agree on how CI goes wrong, it's hard to imagine how much more different the two can be in how to address it.

@economeager @statsepi @Emma_C_Clarke Epi: "Get as close to the 'right' answer as you can with the resources available for any given question."

Econ: "Only attempt to answer questions for which you have a reasonable chance of getting something close to the 'right' answer."

@economeager @statsepi @Emma_C_Clarke You can see how this creates an immediate rift. Economists prefer starting from a plausible position of strength (e.g. an IV, a rule to exploit, RCT, etc) and often the question of interest comes with it, and not doing so is point blank unacceptable.

@economeager @statsepi @Emma_C_Clarke After all, for any given question involving confounding/omitted variables, cyclical issues, selection, etc., how can you possibly control for everything? To attempt to do so is failing before you started. Better to just leave it be.

@economeager @statsepi @Emma_C_Clarke But epidemiologists have a flipped script. For any given question (remember, epi people tend to start with a question, and search for an identification strategy), there aren't many IVs, natural experiments, etc. available, and they'll all be bad.

@economeager @statsepi @Emma_C_Clarke So the best way to deal with that situation is to control for everything you can as much as humanly possible. Time varying confounding via marginal structural models, dealing more directly with missing data. And since that's the best you can do, that's what you do.

@economeager @statsepi @Emma_C_Clarke Boom. Epidemiologists distrust econ methods because if you start with a fixed question, the econ methods won't work. Economists don't trust epi methods because controlling for everything is a weak strategy to start.

@economeager @statsepi @Emma_C_Clarke There are also very salient failures of both fields which feature prominently. Let's not forget, the IVs in the 90's were WILD. And pretty much most epi studies you are likely to see in the news/social media is awful (heavy emphasis on IN THE NEWS/SOCIAL MEDIA).

@economeager @statsepi @Emma_C_Clarke There are also cultural differences in formalization. Epidemiologists prefer a heavily structured, fully formalized set of notation, rules, taxonomies of bias and error etc. Econ is much more wild west, in part due to a belief that strict formality isn't particularly useful.

@economeager @statsepi @Emma_C_Clarke CI in epi is all DAG all day, so epidemiologists think in DAGs, and are often baffled that economists often don't bother. There are some interesting reasons why economists don't (and why that might actually be preferable), but that's a thread for another day.

@economeager @statsepi @Emma_C_Clarke The publication models are also enormously different. There aren't many econ journals that really "count," the papers are several times longer, dryer, more jargony, spend years in pre-print, and vastly lower publication counts. They also aren't particularly "popular."

@economeager @statsepi @Emma_C_Clarke Compare with epi papers, which are usually made for semi-public consumption, are much shorter papers, and there are way more of them, and there are many more topic-specific outlets with a much wider ranging variety of cultures and standards.

@economeager @statsepi @Emma_C_Clarke Then there are just the regular old silo-ed fields disliking each other for general between-field cultural/political rivalries (though noting that there is definitely something to the perception of econ as aggressive in a counter productive way)

@economeager @statsepi @Emma_C_Clarke Will probably leave it here for now, but having been on both ends, the distrust between the fields is a real thing, and even many of the scientific reform-oriented folks seem prone to it (not naming names in this thread).

@economeager @statsepi @Emma_C_Clarke Pinging @BreskinEpi

@economeager @statsepi @Emma_C_Clarke SUPER META TIME! I am seeing a surprising variation on takes related to this thread, and people being really sure that I am saying one of the following:

1) Look at how differently we can all think!
2) Look at how wrong epi is!
3) Look at how wrong econ is!

@economeager @statsepi @Emma_C_Clarke Let me be absolutely explicit about where I come down on this: All three are at least somewhat true, but personally I feel that I am being slightly more critical of epi than of econ here.

The reason comes down to the costs of poorly identified health research.

@economeager @statsepi @Emma_C_Clarke The implicit comparison here is importance of question vs. ability to identify a robust answer to it. To make a "useful" analysis, you need both.

However, that doesn't mean that they are either equally important or independent of each other.

@economeager @statsepi @Emma_C_Clarke Failure in identification of a health question has HUGE consequences and costs to directly to health decisions, indirectly to things like credibility of health research and institutions.

Further, the more important the question, the bigger the costs of identification failures.

@economeager @statsepi @Emma_C_Clarke Good identification of less important questions has comparably low consequences, so I tend to favor the econ style.

Caveats: There are important reasons why epi and econ have these models that I am not discussing, and my formative training was econ, so I could be "biased"

@economeager @statsepi @Emma_C_Clarke I feel like this thread is good fodder for a blog post. What say you @BreskinEpi and @Emma_C_Clarke ?