1,950 minutes or 32.5 hours of Econometrics later...
Rebooting my brain and rethinking my assumptions
The note above is the last bit I wrote down from my Cross Sectional Data Analysis class….A reminder to use common sense and to predict things according to what really exists in the world and to understand what your data can and cannot tell you from those predictions! It is a simple takeaway, but a good reminder from the professor: use the simplest model first; remember that models create expectations and constraints; accept the limitations of your data; make good choices that would make sense in the world.
What a takeaway to have, sitting this far from home, seeing the Supreme Court take so many rights and civil protections that help foster what we might one day hope to call a multiracial democracy.
Make good choices that would make sense in the world.
Profound disappointment aside, I want to reflect on my week, albiet briefly. I am disappointed as I was hoping to have better writing to share - I have more snap in my sentences, generally, but between operating in a second language for daily basic needs, econometric-ing, and navigating, my words are spent.
For my first week of training here at the Barcelona School of Economics, I was enrolled in two different classes, each of which had their own lab. This meant the schedule (accounting for a prof’s flight delay) was 8:30 am-11:00 am, 11:30am -1:30pm, 3:00-4:15pm, 4:30-5:45 pm. My brain is a bit busted, but more so from trying to do this work while socializing, working out, and touristing. The socializing and working out were kiboshed on Sat and Sunday, as I’m regrouping. I even canceled a tennis thing, which if you know me, is really saying something.
Briefly, I found my world changed by Cross Sectional Data Analysis - there is nothing super tricky about it - in the sense that it’s some of the most basic ways to deal with the most basic data that social scientists often have: survey data. Not survey data over time, although a brief dynamic model was introduced, but what the survey data has to say for itself (with and without its limitations) and what it cannot tell you. With a binary discrete model, it’s about how to treat a yes/no outcome, predicting probability of behavior. Which model you chose depends on what your data tells you - and the professor for this class made it really clear what the trade offs would be.
Every model begins with a set of assumptions, and even if I cannot understand all of the elements of the derivation for each term, I understand what the derivations are. The question becomes picking which model to use and then understanding what limitations you are imposing on what it can tell you - and what limitations your data, in turn, imposes.
I also took a panel data class. I have more thoughts on this that I’m happy to share offline, but the reality is that this class was taught for someone who was way more capable of following matrix algebra than I was able to follow. My takeaway is that panel data used to make broad claims is fundamentally flawed - that the assumptions I just talked about above require so much more testing than I really understand - that there are literally dozens of tests for endogeneity and figuring out which ones to pick is its own form of art. I at least now trust fixed effects - or at least understand the assumptions baked in - but I have applied it in my own work in ways that I think are incorrect, and all too often we throw in a fixed effect and random effect set of dummy variables into a regression which might break the brains of a panel data econometrician, who would be seeing bright red.
But to step back even more: this is the first time I have been in an in-classroom setting taught by a professor since Spring 2009. I’ve taken tennis lessons, skating lessons, skiing lessons, horseback riding lessons - and started as a beginner in the last three sports I mentioned - so I have some humility when it comes to the difficulty of teaching, learning, and feeling fundamentally untalented. But wow, was it such a pleasure and… a relief … not to have to set up my own classroom and start to teach, and instead be taught.
I’ve been an autodidact for much of this whole Usher-quantitative turn, and it’s been way harder, I realize, to teach myself rather than be taught. Thanks, Captain Obvious (and the Mellon Foundation for knowing this)…. It has been a journey from the second I started taking Stats 200 at Illinois asynchronously in Spring 2021 and didn’t use most of this for a while but learned it WELL, such that it could be forgotten and relearned quickly. I spent from May to mid June re-teaching myself inferential stats and at least the formulaic assumptions/processes for basic Calc and limits/exponents/logs. It is so so much easier to have someone teach you, in person, and to be able to watch them answer other people’s questions and ask them spontaneously about questions that arise and to see them make mistakes.
Above: Captain Obvious in the style of Modernismo, Dall-E
My professors were deeply amused that they were teaching a professor and my classmates were also, I *think* deeply amused to be in a room with a real, tenured, American professor (I will note our higher ed system still is a crown jewel to many and an aspirational job market). I quickly disabused them of any notion that I might have about adulting was worth listening to or engaging with and am so thankful to have been invited out to dinners and social events.
I will admit that having a strong Google presence probably legitimized me to my Econometrics professors. One of the professors asked me what ethnography was — and it was a good reminder that to economists, qualitative data is… categorical data.
Why do Econ, many of you have asked? Well, fundamentally, I find questions of how to assess the health of local news and information ecosystems, democratic capacity, and the future of a commercial media system ones that are bound up in political economy- that there’s no getting around neoliberal capitalist assumptions nor little hope of dismantling the prevailing political order.
Economics, then, is one of scarcity, choice, decisions, and power - and yes, while I understand that we are not utility maximizing rational beings, I find the expectation that error be understood as part of the modeling process and formula derivation— what economics CANNOT tell you about individual behavior — as fundamentally more compelling than (what to me) seems like zero sum game of political science/political communication where ultimately influence and vote share are central preoccupations.
Do you know that the willingness to pay research in Journalism Studies is simply tiny? More often than not, based on my initial read, these WTP studies are about a binary discrete choice, and there is little attention paid yet to a more multi-nominal model that might consider the actual “subscription funnel.” How is that even a thing?
That I can find only a few studies that take advantage of the rich stores of survey data provided by Pew and RISJ (often those studies are done by RISJ folks or affiliates). Even less of that research brings in questions I am most interested in: how politics, corruption, social trust, and so forth implicate potential outcomes for predicting willingness to pay. My favorite term of the week was “protest zero” - the people who will simply never, ever, ever take a positive favorable action/become a yes - no matter what - these are dead on nos. We, as a field of Communication, need to wrestle with these protest zeros, not assume that they have infinite potential for moral suasion to care about tolerance, news and information, community, and beyond.
Till next week!