Adam L
LightNews LightNews
banner
adam-lg.bsky.social
Adam L
@adam-lg.bsky.social
2.8K followers 370 following 3.9K posts
Data Scientist. Poli sci PhD. Cyclist (Gravel, MTB, road). Bayes, causal inference, etc. Also: cats, food opinions. Views & opinions my own. Not investment advice. He/him
Posts Replies Media Videos
adam-lg.bsky.social
"A Systematic Review of Spatio-Temporal Statistical Models: Theory, Structure, and Applications"
arxiv: arxiv.org/abs/2511.00422

Looks like a pretty solid review essay
#statssky
Data with spatial-temporal attributes are prevalent across many research fields, and statistical models for analyzing spatio-temporal relationships are widely used. Existing reviews focus either on specific domains or model types, creating a gap in comprehensive, cross-disciplinary overviews. To address this, we conducted a systematic literature review following the PRISMA guidelines, searched two databases for the years 2021-2025, and identified 83 publications that met our criteria. We propose a classification scheme for spatio-temporal model structures and highlight their application in the most common fields: epidemiology, ecology, public health, economics, and criminology. Although tasks vary by domain, many models share similarities. We found that hierarchical models are the most frequently used, and most models incorporate additive components to account for spatial-temporal dependencies. The preferred model structures differ among fields of application. We also observe that research efforts are concentrated in only a few specific disciplines, despite the broader relevance of spatio-temporal data. Furthermore, we notice that reproducibility remains limited. Our review, therefore, not only offers inspiration for comparing model structures in an interdisciplinary manner but also highlights opportunities for greater transparency, accessibility, and cross-domain knowledge transfer.
November 12, 2025 at 3:09 AM
adam-lg.bsky.social
I really liked the post; I didn’t even know it was a debate!

Also, as a fellow void owner, I appreciate the Percy and Ollie pictures. This is Stargazer
A black cat with a heart plushy
November 10, 2025 at 1:09 AM
adam-lg.bsky.social
Homemade cheesesteak with homemade cheese whiz
Cheesesteaks, two with very orange cheese oozing from them
November 9, 2025 at 11:07 PM
adam-lg.bsky.social
Singletrack in the NYC city limits, and the best bike shop in the city #bikesky
My mountain bike handlebars on a leaf-strewn trail My mountain bike in front of tread bike shop
November 9, 2025 at 5:12 PM
adam-lg.bsky.social
This would be his likely opponent: a normie center-left Dem (complimentary) who lives on the UWS & is skilled at working the brownstone/co-op crowd & is currently an Assembly member
Micah lasher’s Wikipedia page
November 8, 2025 at 5:18 PM
adam-lg.bsky.social
This is a very weird paper. It feels like it was written in 2011, not 2025. Even weirder, it doesn't cite the structural topic model of @bstewart.bsky.social, Roberts et al.

(Economists not citing political scientists? I'm as shocked as you are!)

link: andreaciccarone.github.io/FETM_draft.pdf
Fixed Effects Topic Model Dan Biderman1, David Blei2, Wei Cai2, Andrea Ciccarone3, Amir Feder3, and Andrea Prat2 1Stanford University 2Columbia University 3Hebrew University This version: October 13, 2025 Latest Version Abstract Social scientists wish to perform topic modeling on documents that are created by different authors in different contexts. However, the same broad topic may be expressed in different ways depending on the environment where the author operates. For example, one may wish to use employee reviews to identify broad corporate culture topics, but the language of reviews is influenced by industry-specific jargon. Existing methods attempt to control for these biases ex post, such as with traditional fixed effect regressions. But these methods cannot fully separate global themes from category-specific language within them. In this paper, we introduce the Fixed Effects Topic Model (FETM), a novel approach to disentangling broad topics from contextual influences by incorporating fixed effects directly into the generative process of language. We use the FETM to identify themes in a large corpus of Glassdoor job reviews. We show that it outperforms conventional topic models, both in interpretability and predictive accuracy.
November 7, 2025 at 2:16 AM
adam-lg.bsky.social
Overall, a good post. There's a missing part of the analysis tho. In the discussion of Cleveland & other northern cities, what drove the growth & fracturing to the suburbs was white flight. East Cleveland -> Euclid -> Mentor is an obvious case; you can see Mentor's population blew up in the 60s!
Compare Cleveland to Tallinn. In Cleveland, a single party controls the mayoralty and every seat in the council. The city proper covers only 17 percent of the metropolis. In Cleveland, as in nearly every US city, the vast majority of the metropolis is governed by a patchwork of small, separately incorporated suburbs that divide residents based on race, class, and culture—and thus, politics. In Tallinn, seven different parties hold seats in the city council. Tallinn covers 72 percent of Harju County—a representative cross-section of the city that yields lively local political competition.17 This is the norm internationally. In nearly every one of the non-US cities I’ve mentioned in this piece, the city proper governs a supermajority of the metropolis, from the traditional urban core to the suburbanizing periphery.18 mentor, ohio's population increased 747.8% between 1960 and 1970
November 7, 2025 at 2:00 AM
adam-lg.bsky.social
"Towards a Unified Framework for Statistical and Mathematical Modeling" Don't know that's it's super novel, but the idea that we should have a deductive mathematical model linking our verbal theory to data analysis is a good one.

Arxiv: arxiv.org/abs/2511.01960
#statssky #episky 📈📉
Ψ is the parameter space, which for the average causal effect of a binary outcome is [−1, 1]. The vertical dashed line indicates the null of no effect. Various bounds are shown Within the biological, physical, and social sciences, there are two broad quantitative traditions: statistical and mathematical modeling. Both traditions have the common pursuit of advancing our scientific knowledge, but these traditions have developed largely independently using distinct languages and inferential frameworks. This paper uses the notion of identification from causal inference, a field originating from the statistical modeling tradition, to develop a shared language. I first review foundational identification results for statistical models and then extend these ideas to mathematical models. Central to this framework is the use of bounds, ranges of plausible numerical values, to analyze both statistical and mathematical models. I discuss the implications of this perspective for the interpretation, comparison, and integration of different modeling approaches, and illustrate the framework with a simple pharmacodynamic model for hypertension. To conclude, I describe areas where the approach taken here should be extended in the future. By formalizing connections between statistical and mathematical modeling, this work contributes to a shared framework for quantitative science. My hope is that this work will advance interactions between these two traditions.
November 6, 2025 at 3:21 AM
adam-lg.bsky.social
This essay from @bschmidt.bsky.social on how history rejected computational methods, & so "quantitative history" ended up in the social sciences, & "digital humanities" in literature, with no historians doing computational work, is fascinating, & worth a read: dhdebates.gc.cuny.edu/read/computa...
The first is a working paper by three economists—Elliott Ash, Daniel Chen, and Suresh Naidu—from 2017. While the authors are economists, the actual contribution—summed up in a title that few historians would think debatable, “Ideas Have Consequences”—is about legal or intellectual history. It presents a powerful and discrete account of the transmission of ideas across social networks through textual analysis. The substance argues that privately funded Manne seminars in law and economics—which were attended by a substantial proportion of the federal judiciary—affected the language, decisions, and sentencing of federal justices who attended them and thus, by implication, allowed large-value conservative donors to capture the federal judiciary. The effect seems robust to a variety of covariates [...] Reading this paper was exciting, but looking through the tools and tricks and sources also made me feel like someone in a science fiction movie encountering an artifact sent back from a few decades in the future. The extraordinary quality of data that economists can obtain is almost unimaginable to humanists. It is not just a million or so circuit court votes and 300,000 opinions but also the institutional capacity to file Freedom of Information Act (FOIA) requests to get the exact years of attendance for every judge who went to the Manne program and the disciplinary capacity to casually use relatively new methods like word embeddings without spending pages slowly, gently analogizing them to some “simpler” concept. Humanists wandering through algorithms seem to have to justify using an algorithm by first identifying which Borges short story—whether about the Map of the Empire, the analytical language of John Wilkins, or Pierre Menard and the Quijote—it most closely resembles.
November 4, 2025 at 3:34 AM
adam-lg.bsky.social
Bodega cat!
Zohran petting a bodega cat
November 1, 2025 at 9:27 PM
adam-lg.bsky.social
Voted, got the Halloween voting sticker!
Halloween voter
October 31, 2025 at 1:31 PM
adam-lg.bsky.social
Building with a menacing aura
The building downtown with no windows
October 30, 2025 at 12:54 PM
adam-lg.bsky.social
We can torture the data until it screams out the answer we want
October 28, 2025 at 1:13 AM
adam-lg.bsky.social
The Predator handshake, with woke on one side, and python on the other
October 27, 2025 at 10:36 PM
adam-lg.bsky.social
If we're talking HEAT, Val Kilmer's description of how awesome making the movie was sticks with me
Well imagine being able to say, "Al and Bob" for the rest of your life. Not many people can do that. I have seen Bob, giggling like a school girl in a van in the middle of the night, because we have to be quiet cause they are filming outside... I have been hugged by Al Pacino in the middle of downtown LA like he was my older brother. I have shot live rounds from high powered assault weapons over Bob's head while rehearsing lines from our film. I got to kiss Ashley Judd. I sometimes lived at Michael Mann's house. I am in one of the greatest cops and robbers films in film history, has to be in the top 20. I am on the poster for goodness sake. What an honor! Priceless experience. Watching ALL the actors do their thing. We all work hard but when you are with the icons, you get really squared away. Loved every minute of it.
October 26, 2025 at 2:26 AM
adam-lg.bsky.social
The fall colors #bikesky
Leaves around a gravel road Leaves around a paved road Leaves around a paved road Fall colors
October 25, 2025 at 10:00 PM
adam-lg.bsky.social
The ride #bikesky
Strava screenshot: 57 miles, 7000 ft
October 25, 2025 at 8:00 PM
adam-lg.bsky.social
Mid-ride lunch of champions:
Chocolate chip pancakes and a full-sugar Mountain Dew
October 25, 2025 at 7:57 PM
adam-lg.bsky.social
You, normie: Levitsky and Ziblatt (2018)

Me, hipster: Levitsky and Way (2010)
Competitive Authoritarianism
October 23, 2025 at 3:41 PM
adam-lg.bsky.social
First day of work
Me, in a suit, no tie
October 20, 2025 at 1:01 PM
adam-lg.bsky.social
If you want to disable the AI in Firefox
set the following to false in about:config: browser.ml.chat.enabled browser.ml.chat.page.footerBadge browser.ml.chat.page.menuBadge browser.ml.chat.shortcuts browser.ml.chat.shortcuts.custom browser.ml.chat.sidebar browser.ml.checkForMemory browser.ml.enable browser.ml.linkPreview.shift
October 19, 2025 at 1:28 AM
adam-lg.bsky.social
Morningside Heights #nokings
A crowd A crowd A crowd
October 18, 2025 at 5:45 PM
adam-lg.bsky.social
The funny think about riding to Montauk is it’s *insanely* flat. Of the 2300 feet of climbing, 400 occur in the last 8 miles (the out and back from the town <> lighthouse).

If you can pace yourself, it’s not a hard ride, just long. The biggest issue is water (and LI traffic)

#bikesky
Strava of the time I rode to Montauk.
October 18, 2025 at 11:28 AM
adam-lg.bsky.social
🤨 this is what dropper posts are for

#bikesky
My bike in front of a sign that reads “Steep slope, walk bike”
October 15, 2025 at 5:09 PM
adam-lg.bsky.social
"News Media Reporting Patterns and our Biased Understanding of Global Unrest" - a cool paper that gets at something I've wondered: how much does a selection effect drive what we see in conflict datasets coded from news stories? Quite a bit, it turns out! esoc.princeton.edu/WP32
#polisky #statssky
The aforementioned examples are the tip of the iceberg: selection into media-based datasets can and does arise for many more reasons, raising fundamental questions about whether these datasets accurately reflect patterns of violence, or are systematically skewed. We identify a non- exhaustive list of sources of measurement error, including systematic under-reporting relating to violence intensity, electricity outages, weaponry employed, target, perpetrator, and deadly versus non-deadly violence used, among others. If the news media systematically under-reports particular types of violent events, then media-based conflict event datasets will do the same. Variables gener- ated from the datasets may thus suffer from significant biased sampling, undermining the validity of inferences using them. News reports of political violence are systematically compiled into large global conflict-event datasets used by academics, governments, and international organizations. These datasets present opportunities to examine the micro-dynamics of conflict but are often systematically skewed. We compare various news-report based datasets to high quality administrative records from Afghanistan, Iraq, the Philippines, South Africa, and Syria to identify sources of systematic missingness in the former. We identify under-reporting related to violence intensity, weaponry, target, perpetrator, and non-deadly violence. In a large replication exercise, we show that media- based data fail to uncover the results reported in leading economics/political science journal articles.
October 15, 2025 at 1:53 AM