george berry 🥥🥥🥥
@george_berry
Followers
2,555
Following
2,013
Media
931
Statuses
16,047
MGP Democrat. OLS aficionado. sociology fan. lofi jazz lover. card collector. he/him
jersey
Joined June 2011
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
#FayeYoko1stFMinMacau
• 129576 Tweets
FAYEYOKO MACAU FANMEET
• 127631 Tweets
Miami
• 105262 Tweets
FreeBuds6i x SKULLPANDA W CL
• 98120 Tweets
Knicks
• 93714 Tweets
भगत सिंह
• 77720 Tweets
Randle
• 74107 Tweets
mingyu
• 64013 Tweets
Donte
• 45461 Tweets
Wolves
• 43747 Tweets
Paola
• 39519 Tweets
ババコンガ
• 26291 Tweets
シリウスS
• 22113 Tweets
Virginia Tech
• 21218 Tweets
#AFLGF
• 18896 Tweets
#bhagatsinghjayanti
• 18358 Tweets
PERFECT SCENE x NORAWIT
• 15764 Tweets
Pinned Tweet
i turned my blog post on evaluating pres forecasts into a osf preprint
my basic argument is that it's likely we can evaluate political forecasts in a decade or less
this is a heuristic, but let me walk you through how i came up with this
1/
1
1
9
8
655
26K
extremely funny that nate silver is doing a better job than 538 covering the election
Today's update. Another good day for HARRIS, who now leads in our polling average in Arizona and is only 6/10ths of a point down in North Carolina.
427
778
6K
47
66
3K
propensity score matching is like
They both served one term in Congress in their 30s before deciding not to seek re-election
97
777
13K
2
116
1K
I get asked for stats/causal inference book recs a lot. Here are some I find useful.
10
93
868
doing machine learning engineering by reading the wikipedia page for ordinary least squares again
8
22
660
A really good rule of learning to program is: go where the best community is.
Tools can improve. Frameworks can get rewritten to be faster. Communities are difficult to change.
12
139
628
i got a phd and loved it. my time doing work-that-would-be-published-in-good-journals-but-at-a-company-so-it's-not-published has changed my mind on a lot of things in academia.
(thread)
2
24
325
As for what's next: I'm really excited to be joining
@reddit
(after a bit of a break) as a data scientist focused on experimentation.
Today is my last day working in Dem politics (for the moment, at least!).
It's been a great 18 months, I've learned tons of stuff—both technical stuff (high level of rigor!) and about US politics.
2
0
37
30
4
219
I’m getting a lot of q’s about being a “computational social scientist” in industry/tech working as a DS or ML eng.
Here are a few thoughts in case they’re helpful for anyone
4
48
197
I wrote a short and hopefully understandable intro to causal inference. I hope it'll be useful for infosci folks in particular.
PDF:
GitHub:
3
36
190
yes, if politicans do what you want and you do not then enthusiastically support them, they will then look elsewhere for votes
as someone with solidly progressive views, it has been frustrating to see people on the left continue to "hold out" for "leverage"
"Biden has been more solicitous of progressive interests than any Democratic president in decades, and it has earned him remarkably little slack or good will among actual progressives."
Yup.
199
314
1K
4
33
156
second, data is more important than models. you can't model your way out of bad data. endless "checks" by nitpicking reviewers are rarely worth the time. economics papers that are 60 and 70 pages on a single like 1500 observation dataset... i just shake my head lol.
2
6
137
Are you studying social relationships online? Are you using predictions for node attributes?
👇You may want to check out our new preprint👇
4
34
120
@evansiegfried
My perception could be wrong, but I do not think millennials are clamoring for less generous healthcare and lower corporate taxes.
6
6
103
TIL you can work with graphs in the tidyverse using tidygraph. It's wild.
g %>% activate(nodes) %>% 🤯
g %>% activate(edges) %>% 🤯
GitHub:
3
20
115
first off, i think studies are less authoritative than i used to. single studies almost never settle things. i wish academics would optimize more for quantity than "quality". but not quantity of re-analyzing the same datasets over and over.
1
1
115
I can't get over this graph
States where physicians are highly paid have lower Covid-19 mortality per capita.
853
109
583
14
10
107
All of these attempts to monitor students for cheating make me think: why do we care so much about what students know without any reference material. That’s... not how life is.
All the fun problems are ones where you have like 20 tabs open and you’re still scratching your head.
2
6
108
it is difficult for people to understand how wealthy the US is. we have big problems with deep poverty, but we also have a *lot* of people doing well.
"63% of Americans do not have $500 in the bank to pay for an emergency healthcare bill."
This is inaccurate.
Median household net worth is $192k, including $8k in checking accounts.
322
270
3K
2
7
102
third, so many questions are irrelevant in academic studies. the laser focus on "novelty" often overcomes what's important. sure, people have studied age voting gaps a ton. but that's because they are really important! there's still more to say. pile on!
1
3
101
i learned a new pattern in R, this is the modern tidyverse way to do regressions on subgroups
library(tidyverse)
library(broom)
df %>%
nest(dat=-grouping_col) %>%
mutate(
fit=map(dat, ~lm(y~x, data=.),
tidied=map(fit, tidy)
) %>%
unnest(tidied)
2
2
92
I read “The New Geography of Jobs” by Enrico Moretti after seeing
@Noahpinion
recommend it. Rather than just giving it effusive praise, here are three things if changed my mind about, and three beliefs it reinforced.
2
16
85
i sort of feel like this is true! the Serious Economists did the recovery after 08 and it sucked and now we have Meme Economists who are doing a better job lol.
The academic economists had their chance. Now we have a non-economist as fed chair and posters as economic advisors
19
48
479
2
7
86
judea pearl's work is a towering intellectual achievement wrapped in petty academic turf wars
6
3
84
one widely applicable lesson from FTX: you can
- get good grades
- have everyone tell you you're smart
- do things which publicly indicate intelligence
- convince yourself you're smarter than other people
and still be really bad at your job
1
11
83
@CringePosted
he's an above replacement level pundit i think, like to put it in NBA terms he has an epm of like 3
3
0
80
i see a ton of academics for example insisting the democrats should do X, and it's almost always wrong for reasons obvious to practitioners. living inside a set of constraints makes those constraints visible, and they are often invisible to outsiders (even smart outsiders!)
1
1
80
@evansiegfried
Just speaking from my experience, the healthcare/tax reform plans would prob make my life worse, and would certainly hurt many of my friends
3
2
65
i guess my final thought is that i hope for tigher academia/practitioner integration. academics can do long timeline stuff that industry people really can't, and that's their competitive advantage. but that also means feedback loops are slow.
1
1
79
fourth, there's just a lot of puzzlement in the social sciences about what to do with evidence from outside academia. i think the right answer is that we should default to thinking whatever practitioners in a competitive field are doing is well thought out.
1
3
76
I can’t believe my most liked tweet of all time is a picture of stats books. What a strange website this is.
2
1
72
I just realized
@mmolinamaydl
and
@ProfFilizGarip
's paper "Machine Learning for Sociology" is on
@socarxiv
:
If you're a sociologist who wants to know what all the fuss is about, check it out, it's a really good intro.
3
24
68
Happy our new paper is out, "Estimating Homophily in Social Networks Using Dyadic Predictions"
We address the question: when we use classifiers to estimate network measures, what properties does the classifier have to satisfy?
1
11
67
you really lack compassion and empathy for cash-strapped people if you think a 3-5% reduction of rent doesn't matter
YIMBYism has been successful enough that the criticism is apparently now "yeah but it doesn't lower rents 25%"
62
63
2K
2
3
59
I guess I should make this Twitter official, I switched jobs from Twitter to Civis. I was sad to leave some great colleagues at Twitter, but excited for new challenges.
8
0
62
one reason i am not bullish on GOP reforming from the inside is that they don't seem to understand why they lost the midterms
Ted Cruz on one of the reasons Democrats did better than expected: "They went hard left, they energized their base, they governed as left-wing lunatics, and their base showed up in big numbers."
157
66
448
3
7
56
Life update: I quit my job.
I’m thankful I got to work alongside so many great people.
I’m also really excited about having some time to finish up collaborations and think carefully about next steps.
5
1
55
it's probably not great that i'm a phd sociologist who has worked on misinfo and
@Sander_vdLinden
's debate with nate silver convinced me to be much more skeptical of misinfo research
The consensus of the US government and the scientific community is now that COVID origins are highly uncertain. Lab origin was incorrectly dismissed as misinformation and often as a conspiracy. Misinformation researchers would have more credibility if they acknowledged the error.
204
184
2K
10
7
57
in general academics can "steal" fast feedback loops by working with those in faster/competitive environments.
thanks for coming to my ted talk
2
1
52
my year in review: i built bayesian models which helped get joe biden elected while trying to avoid a literal pandemic
1
1
50
folks if you are predicting things in networks, and then analyzing those predictions, please read our paper here for some pitfalls to avoid
2
12
50
Are you getting a PhD? Want to spend a summer working on hard problems with smart people? You should apply to intern on the Computational Social Science team at
@facebook
.
2
23
49
folks it has happened
after nine years of programming
i have finally used recursion
6
1
48
Well, it finally happened. I burnt out at work. Got next week off to rest, which is nice.
I want to reflect a big on burnout in a fast paced DS environment.
2
2
49
it took me a long time (> 10 years) to realize it, but rhetoric like this is positively correlated with the status quo, and is often against increasing human welfare.
i think the two places we see it most are housing and climate.
1
0
46
one of my strong political beliefs is that if you advocate for X and X happens and you don't reward the party that delivered X with praise and votes, that party will pay less attention to you in the future and you will get less of X
0
8
46
I'm just a humble sociologist, but it seems to me that most news outlets are not adequately conveying the the scale of the coronavirus mismanagement in the US.
Very open to pushback on this. But this is my current perception.
3
1
44
this is classic SF policy: incredibly complex, unlikely to achieve the stated goal, and enriching someone connected
S.F. mayoral candidate: "I'm gonna jump-start housing production by cutting taxes on *already built* housing projects that my political cronies invested in."
Just, wow.
4
12
162
0
5
45
@evansiegfried
What does the party's rhetoric/platform offer millennials and minorities, though? Surely the policy has to be appealing to recruit.
1
1
37
I should just wake up every day and tweet "Pandas indexes are a terrible design choice"
4
0
44
this one is particularly funny because it's repeating a hacker news post quoting an AI fail. Ai-fail-inception
1
10
44
feels like it was a good choice to spend my last year trying to get dems elected
1
0
43
Today, I got a paper rejected.
That's OK! I learned a lot of stuff writing it and the reviewers had sensible comments. Looking forward to revising & improving it.
1
2
43
Hey social science folks, cool open job at Reddit as a DS focused on communities and moderation. Job is looking for people with experience. Happy to answer q’s if you’ve got em.
3
19
42
Im *so* happy our paper “The Opacity Problem in Social Contagion” is out in Social Networks!
Big point: contagion processes hand us data with error.
I’d recommend you read it if you study social contagion (for real).
2
10
40
Having a beer and thinking about random effects models AMA
7
1
41
I have thoughts about this (a thread).
@cristobalyoung5
is largely correct. The hype around ML is out of line with its usefulness for sociology. On the other hand, ML has two advantages that haven't been fully incorporated into sociology yet. 1/
2
16
42
I learned so much at
#ICWSM2019
! Here are 10 papers I thought were interesting (in no particular order).
1
13
40
A very weird thing about contemporary American life is
1) the unflinching exhortation to work
2) the genuine desire of most people to work
3) the large incentives to classify work as “non-work” or “not real work”
While this new policy increases the amount of guaranteed funding for certain cohorts in these divisions, it also caps total enrollment in individual programs and redefines graduate lectureships and TAships as “mentored teaching experiences.”
1
6
31
0
9
39
@NateSilver538
If you start with something it’s not supposed to answer, it will keep not answering. But if you re-ask in a different way you can get actual answers.
3
0
37
tariffs suck, trade is good, and we should ABSOLUTELY reduce tariffs with our close allies like japan and canada (assuming they do the same in reverse)
The US has moved away from tariffs over the last 200 years because they are counterproductive
111
193
1K
1
8
39
My last year working with people with extremely deep domain knowledge (but without PhDs) has convinced me that there's really a lot to be learned from listening to practitioners
This is not a sentiment I have encountered frequently in academic social science circles
TBH it's kind of astonishing you think that I (an econ grad, btw) wouldn't know that. And part of an annoying pattern of academics underestimating the knowledge of non-academics. Matt brought it up recently in the context of COVID and I thought it was appropriate to refer to him.
37
30
1K
2
3
37
(i'm generalizing here for sure, there is a ton of good work in academia! but i also eyeroll a lot these days.)
1
0
38
every time i procrastinate the thing i'm avoiding takes like 7 minutes
3
10
37
Pretty insane that people expect me to understand this code I wrote
3
3
37
Walking around NYC I’m just struck by how wildly inefficient cars are.
- Huge amount of space needed to transport small numbers of people, both in terms of roads and parking
- Discourage pedestrians and cyclists from using the same space
- Dangerous
- High carbon emissions
5
1
37
Today is my last day working in Dem politics (for the moment, at least!).
It's been a great 18 months, I've learned tons of stuff—both technical stuff (high level of rigor!) and about US politics.
2
0
37
it's just shocking how much easier i find it to read through code implementing statistical methods compared to the traditional math notation of the methods
am i the only one like this?
5
0
36
I’m an industry researcher with a sociological background. It’s wildly helpful.
The weird thing is a lot of companies don’t know they need sociologists (but they do).
0
10
36
I gave this a straight RT before but it bears additional praise.
@red_abebe
has seriously engaged with the sociology literature on inequality, and it informs her work on AI/ML, which is amazing. I can't speak highly enough about her process.
2
5
33
today I get to use the world's most powerful data science tool, the `lm` command in R
1
0
32
Check out my new Python package, Pandas2, which is identical to Pandas in every way except it calls .reset_index() after every operation.
2
3
30
folks we are hiring data scientists. if you're interested in progressive politics and hard DS problems, check out the job posting
feel free to DM with questions
0
6
33
Siri, show me the transit equivalent of throwing a GPU farm at a classification problem for a 2% boost over logistic regression.
I know Twitter is limited to NYC/DC concerns but I can't emphasize enough that this is a *real map* of what El*n M*sk is proposing for Chicago and it literally runs parallel to a train that costs $2.25 and leaves every 7 minutes.
I am not kidding, this is it. 1/3
335
4K
11K
0
7
33
a cumulative science shouldn't demand people do close readings of 150 year old texts, nobody in stats is like *reading* gauss
It is the only social science uninterested in its founders, so it can perpetually reinvent the same thing claiming it is something new.
16
150
944
1
2
32
We should do workflow demos.
Example: if you write a paper with a methods section, you should walk people through how you did the work, from data collection to plotting. Emphasis on the tools you use and how they fit together.
1
7
32
is it bad that I just googled "Python to COBOL transpiler"
1
1
31
1) If you’re (for instance) a sociologist, nobody knows what that is. They don’t know the set of skills you have. It’s very important to have a clear story about what you do.
1
0
32
Confession time: for years I had to look up Type I and Type II errors to remember which was which.
Please give your concepts descriptive names. 🙏🙏🙏
If a Type I error is a false positive, a Type II error is a false negative, and a Type III error is getting the right answer to the wrong question, is a Type IV error GIVING IMPORTANT CONCEPTS NUMBERS INSTEAD OF NAMES FOR NO GODDAMN REASON AND CONFUSING GENERATIONS OF STUDENTS
25
361
2K
5
2
30
I hate to be that “weekends increase your productivity” guy but I just drank coffee, grilled, and played video games for 4 hours and now I have a good work idea
0
0
31
i guess i'm now the "your confidence intervals are too tiny" guy
yikes
2
1
31
in a vacuum it's impossible to tell if you should care about the line or the noise around the line. in some domains this kind of finding would be amazing, in some it would be terrible. it depends!
the statistics heuristic i want to teach everyone is to LOOK AT THE SCATTERPLOT
i want people to be going around saying "SHOW ME THE SCATTERPLOT"
and then they show you some ridiculous noise with a line drawn thru it. you can literally just ignore the line
59
290
5K
0
3
31