Dean Eckles @deaneckles Twitter profile

Pinned Tweet

Dean Eckles

1 year

People in some places in the US have more long ties — ties between people without mutual contacts. This is a map from our newly published paper: Why does this matter? How do people end up with these different network structures? 🧵

10

113

475

Last Seen Profiles

@TIGR27

@benhaticeniz

@EliPropper

@steveabson

@cri

@bokeplokalmalam

@TheNothingMan78

@Martinboth

@DMRegister

@jizake_douraku

@celik71287

@SimplyAllegra

@artquint1

@sergid_

@imrayneb

@tradewithumar

@__Shakshi_19

@goncalveskis

@my_yy2000

@SteveGreenHL

@MukeshSiso517

@TimelabApp

@JeseniaWhyte

@kazarigiri

@buzz_g

@imrayneb

@AiZgmund

@stw_pdg

@ycn_beatbox

@slytherjennie

@ElDonte22

@heoiheoei

@Itzz_dk_

@AminataNamasia

@suzuyatorihiki

@sarcole

Dean Eckles

@deaneckles

7 years

Everything is obvious — once you know the answer @duncanjwatts

12

473

767

Dean Eckles

@deaneckles

1 year

Why not require 10,000 hours? Anyone arguing for less than 10,000 hours has blood on their hands. It can be so hard to reverse escalating training and safety regulations.

More Perfect Union

@MorePerfectUS

1 year

Sen. Kyrsten Sinema is holding up a major bill unless it includes her proposal to require less training for airline pilots. Sen. Tammy Duckworth, an ex-pilot, says Sinema will have 'blood on her hands' if it passes. Why would anyone argue for less training for pilots? Thread.

321

2K

7K

28

18

558

Dean Eckles

@deaneckles

4 years

A rare negative p-value spotted in the wild

13

12

470

Dean Eckles

@deaneckles

6 years

When asked to draw a regression line, my experience is people draw something closer to the PCA line.

Lionel Page

@page_eco

6 years

Excellent comparison of regression and PCA in a visually pleasing gif. By @andre_quentin , ht @HernanBruno

5

448

1K

5

93

448

Dean Eckles

@deaneckles

2 years

You love to see it

1

21

449

Dean Eckles

@deaneckles

8 years

Once you start looking for confirmation bias you see it everywhere.

9

259

413

Dean Eckles

@deaneckles

2 years

Some intuitions for matching and diff-in-diff

4

40

359

Dean Eckles

@deaneckles

6 years

Foundations of Agnostic Statistics is available for pre-order It is an accessible, lucid & rigorous introduction to key tools of statistics, econometrics & data science — while relying only on minimal assumptions.

7

46

356

Dean Eckles

@deaneckles

1 year

I will teach design & analysis of experiments again at @MITSloan in the fall. This was the 2019 syllabus: So already looking to collect new materials — new methods developments, new learning resources & new example experiments.

syllabus Experimental Design & Analysis PhD 2019

15.838 Experimental Design & Analysis Research Seminar in Marketing MIT Sloan School of Management Spring 2019 Professor Dean Eckles Wednesdays 10am – 1pm, E51-057 Randomized experiments are import...

docs.google.com

8

58

353

Dean Eckles

@deaneckles

1 year

A new profile of me, covering research on peer influence, causal inference, and where network structure comes from

Advancing social studies at MIT Sloan

MIT Sloan Associate Professor Dean Eckles studies social networks and the effects they have on people’s lives. He examines questions such as: “How do social networks form and evolve? And what are the...

news.mit.edu

4

25

264

Dean Eckles

@deaneckles

3 years

That WhatsApp is down (along with FB & Instagram) is hugely impactful globally, disrupting a lot of one-to-one and group messaging, but also payments and calling. This is a clear illustration of the value of decentralization and robustness in key communication infrastructure.

2

40

260

Dean Eckles

@deaneckles

3 years

UNC-Chapel Hill

@UNC

3 years

First sips ✔️ #UNC

2K

199

1K

7

35

241

Dean Eckles

@deaneckles

2 years

When should we match prior to doing diff-in-diff? Highlights when to match on covariates but also pre-treatment outcomes

5

38

222

Dean Eckles

@deaneckles

2 years

@panzer_michael @AlecStapp Way less risky in the sense that they just predictably, continuously harm people's health?

2

0

203

Dean Eckles

@deaneckles

6 years

Happy to be promoted to associate professor without tenure alongside the others in this group!

MIT IDSS

@mitidss

6 years

IDSS congratulates Hamsa Balakrishnan, Guy Bresler, Tamara Broderick, @deaneckles , Stefanie Jegelka, and Suvrit Sra ( @optiML ) on reaching new career milestones. @MITAeroAstro @MITEECS

1

3

40

23

7

208

Dean Eckles

@deaneckles

4 years

Across many countries, most people say they will accept a COVID-19 vaccine. Giving people accurate info about others' acceptance intentions increases vaccine acceptance — according to our new randomized experiment involving 300k people in 23 countries.

Providing normative information increases intentions to accept a COVID-19 vaccine

Despite the availability of multiple safe vaccines, vaccine hesitancy may present a challenge to successful control of the COVID-19 pandemic. As with many human behaviors, people's vaccine acceptance...

osf.io

3

77

210

Dean Eckles

@deaneckles

4 years

That's exactly what a cat pretending to be a lawyer would say

Lawrence Hurley

@lawrencehurley

4 years

“I’m here live, I’m not a cat,” says lawyer after Zoom filter mishap “I can see that,” responds judge

6K

90K

334K

5

29

206

Dean Eckles

@deaneckles

7 years

Courses on experimental design, especially in politics — syllabus repository @johnholbein1

Syllabi

Shared with Dropbox

www.dropbox.com

1

55

201

Dean Eckles

@deaneckles

7 years

Efron & Hastie's history of statistics in a simplex

3

96

193

Dean Eckles

@deaneckles

4 years

Most people intend to accept COVID-19 vaccines. And saying this further increases vaccine acceptance. Advice for public health messaging and journalism from our research:

Op-Ed: Why public health messaging should emphasize vaccine acceptance, not hesitancy

Overstating or overemphasizing vaccine hesitancy can keep others from deciding to get the COVID-19 shots.

www.latimes.com

6

91

192

Dean Eckles

@deaneckles

4 years

What places are best to reopen, given demand & danger due to density? New working paper from @sbenzell @avi_collis & @CNicolaides tries to quantify this using surveys of Americans & pre- #COVID mobile phone data about density at different kinds of places.

26

71

192

Dean Eckles

@deaneckles

7 years

Syllabus from PhD course I taught last year on design and analysis of experiments, especially field experiments

1

54

189

Dean Eckles

@deaneckles

4 years

🎯How can we learn to target interventions when we care about outcomes that are only observed after a delay?🗓️ This comes up in settings as varied as clinical trials (5 yr all-cause mortality) and marketing (customer lifetime value). Our new paper:

7

49

184

Dean Eckles

@deaneckles

6 months

To explain away smoking & lung cancer link association, you need a confounder that: - is associated with a 20x increase in lung cancer - is 20x as prevalent among smokers What's the equivalent for social media & mental health?

Paul Novosad

@paulnovosad

6 months

People dunking on this too quickly, "correlation is not causation", "we don't vote count studies", etc. *Not every important question can be answered with a perfect RCT* For decades, the cigarette-cancer link was similarly correlational, based on case studies, etc. 🧵 1/N

14

65

465

12

22

182

Dean Eckles

@deaneckles

4 years

I've recently become an associate editor for Management Science in the new-ish "Big Data Analytics" department. The name makes me chuckle, but I think it could be a great place for boundary-crossing quantitative & methods work. Send us your papers!

5

26

181

Dean Eckles

@deaneckles

7 years

How can you test for spillovers from a treatment in a network? Final version of our paper now available in JASA @Susan_Athey

1

77

178

Dean Eckles

@deaneckles

4 years

Andrew Gelman finally has an outlet to share his perspectives on causal inference — on Tuesday.

6

34

178

Dean Eckles

@deaneckles

1 year

Which of these diff-in-diff scenarios yields the most credible estimate of the effect of treatment? Or are they all the same to you? (Here treatment is applied in period 1.) Some more discussion here

16

29

178

Dean Eckles

@deaneckles

3 years

Multicollinearity as an excuse for dropping a bunch of relevant covariates

15

18

175

Dean Eckles

@deaneckles

3 years

How does algorithmic ranking in social media work? How can we assess algorithmic impact & amplification? What can policy-makers do here? My written testimony from today's Senate hearing:

Dean Eckles

@deaneckles

3 years

In case you're interested in social media & algorithms, I'll be participating in this hearing before the Senate Subcommittee on Communications, Media & Broadband at 10:45am Eastern this morning.

2

13

90

3

35

172

Dean Eckles

@deaneckles

5 years

How can we learn about how effects of treatments spill over in space, groups, and networks? Here's a new review, summarizing & illustrating (with an R package my coauthors wrote) how to analyze experiments to this end @cdsamii @szonszein

5

54

173

Dean Eckles

@deaneckles

5 years

I'm psyched for this all-star #datascience #econtwitter #epitwitter discussion of causal inference 5:45pm Friday at @MIT with @Eytan Bakshy, Guido Imbens & @EpiEllie Murray. This "fireside" panel is the one part of CODE @MIT that's open to the (non-registered) public. #CODECON19

3

27

170

Dean Eckles

@deaneckles

4 years

🎲Noise-induced randomization in regression discontinuity designs📈 A new approach to inference that directly uses the "local randomization" — from measurement error or other noise — that is often used to informally justify RDDs

6

50

166

Dean Eckles

@deaneckles

2 years

Neat illustration of the bias–variance tradeoff in analysis of a regression discontinuity... But then the variance turns into bias with file drawer bias

3

26

172

Dean Eckles

@deaneckles

1 year

A good reminder of how actively misleading "mediation analysis" is — and alternative approaches to designing experiments to learn about mechanisms

The Failings of Conventional Mediation Analysis and a Design-Based Alternative - John G. Bullock,...

Scholars routinely test mediation claims by using some form of measurement-of-mediation analysis whereby outcomes are regressed on treatments and mediators to a...

journals.sagepub.com

3

53

166

Dean Eckles

@deaneckles

6 years

I am teaching a doctoral seminar at @MIT on design and analysis of experiments, with emphasis on field experiments, randomization inference & policy learning. Hence the query below... Draft syllabus:

syllabus Experimental Design & Analysis PhD 2019

15.838 Experimental Design & Analysis Research Seminar in Marketing MIT Sloan School of Management Spring 2019 Professor Dean Eckles Wednesdays 10am – 1pm, E51-057 Randomized experiments are import...

docs.google.com

Dean Eckles

@deaneckles

6 years

What's the best PhD-level introduction to counterfactual policy learning, dynamic treatment regimes, and the like? Some options I've considered follow...

3

8

86

4

37

165

Dean Eckles

@deaneckles

6 years

Python library for causal inference. It explicitly represents identified estimands and assumptions. Does sensitivity analysis. @amt_shrma @emrek

1

71

164

Dean Eckles

@deaneckles

5 years

Headlines like this from @MIT @techreview are irresponsible and embarrassing

6

15

158

Dean Eckles

@deaneckles

6 years

How informative is rejecting the null? A further rebuke of the idea that you learn nothing when you retain the null

4

45

150

Dean Eckles

@deaneckles

5 years

Which products are purchased together can reveal a lot about their relationship. But so can which products are viewed together, but not co-purchased. Our ( @madhavkumar2005 & @sinanaral ) new working paper leverages both types of baskets...

3

41

153

Dean Eckles

@deaneckles

2 years

TIL that Poisson misspelled Rev. Thomas Bayes' name throughout his work in the 1830s

10

20

155

Dean Eckles

@deaneckles

4 years

Design of experiments: Notes from course by Art Owen, who has worked a lot on Monte Carlo & Quasi-MC, but also consulted for Google

3

10

141

Dean Eckles

@deaneckles

3 years

What a fundamental misunderstanding in some widely-diffusing work claiming that (eg) “how to hit a woman so no one knows” had been typed into Google over 160 million times during the pandemic...

7

22

143

Dean Eckles

@deaneckles

4 years

Nice example of ridiculous observational causal inference... Come for the obvious confounding (even after adjusting for a couple coarse covariates of course), stay for "moderated mediation"

Geoffrey Supran

@GeoffreySupran

4 years

"The @GretaThunberg Effect" is now an empirically demonstrated, peer-reviewed phenomenon: "We find that those who are more familiar with Greta Thunberg have higher intentions of taking collective actions to reduce global warming." Open access:

69

535

2K

9

12

143

Dean Eckles

@deaneckles

1 year

Does the rollout of Facebook across college campuses in 2004-5 provide clear evidence of negative effects of social media on mental health? I wrote some comments on this now widely-cited study

7

31

137

Dean Eckles

@deaneckles

4 years

Randomized experiment with evidence against compensatory behavior (at least on average)

Thomas Talhelm

@ThomasTalhelm

4 years

@OlegUrminsky There was a neat study about this with masks. They tested whether wearing masks made people stand closer together out of a sense of security. It did the OPPOSITE. People stood farther apart.

3

22

77

5

37

138

Dean Eckles

@deaneckles

6 years

What evidence is there for the effectiveness of Cambridge-Analytica-style targeting on Facebook? Here we describe confounds in a widely-cited PNAS paper @garjoh_canuck

3

61

133

Dean Eckles

@deaneckles

4 years

I'll be sharing our new work on regression discontinuity designs — Tuesday at 11:30am Eastern Excited about how this approach reconciles intuitions about RD with formal basis for identification, estimation & inference

3

22

129

Dean Eckles

@deaneckles

7 years

How credible are observational studies of peer effects? We compare against an experimental "gold standard"

Bias and high-dimensional adjustment in observational studies of...

Peer effects, in which the behavior of an individual is affected by the behavior of their peers, are posited by multiple theories in the social sciences. Other processes can also produce behaviors...

arxiv.org

5

76

126

Dean Eckles

@deaneckles

3 years

For decision theory, I turn to (and would recommend) Berger's Statistical Decision Theory and Bayesian Analysis. What else is a good reference here, especially any more recent developments?

12

17

130

Dean Eckles

@deaneckles

6 years

How are biological and social contagion affected by changes to network structure? Recent work has claimed a "weakness of long ties" for social contagions, unlike biological contagions. Our new paper substantially revises this conclusion.

3

53

122

Dean Eckles

@deaneckles

5 years

Econometrics of networked data

0

34

125

Dean Eckles

@deaneckles

3 years

Are "behavioral" phenomena — such as cognitive biases — just something that crops up in low-stakes settings? This new paper adds to the evidence that often the answer is "no". Here "high incentives" amount to more than a participant's monthly income:

4

28

125

Dean Eckles

@deaneckles

3 years

On understanding why some methodologists don't like balance tests in randomized experiments — and perhaps some consensus This is my first post at @StatModeling

Andrew Gelman et al.

@StatModeling

3 years

Does the “Table 1 fallacy” apply if it is Table S1 instead?

1

8

41

9

19

126

Dean Eckles

@deaneckles

1 year

A good rule of thumb is to not believe anything involving this kind of "SEM" practiced in (social) psych and related areas, perhaps better known as "multiple regression with pictures"

7

11

125

Dean Eckles

@deaneckles

4 years

What questions should you ask in a survey with limited time? Answers to the questions are correlated, so you can potentially actively learn which to ask and fill in the rest using matrix factorization Just noticed this paper is now published:

3

21

120

Dean Eckles

@deaneckles

4 years

Our new working paper examines spillovers of distancing policies via both geographic and social networks in the US. We estimate that these spillovers are large and that social peers' mobility behaviors substantially affect individual's compliance.

The Cost of Uncoordinated Responses to COVID-19 - MIT Initiative on the Digital Economy

Groundbreaking study measures the surprising impact of uncoordinated responses to COVID-19.

ide.mit.edu

Sinan Aral

@sinanaral

4 years

We combined daily, county-level policy data w/ movement from over 27M mobile devices, social networks among 220M FB users, daily weather from 62K weather stations & county-level census data to estimate geographic & social network spillovers in regional policies across the US. 3/

2

6

29

3

32

120

Dean Eckles

@deaneckles

5 years

Now available: @Eytan Bakshy, Guido Imbens & @EpiEllie Murray in conversation on the promise and limits of observational causal inference at CODE @MIT earlier this month @CODEConference #CODECON19

2019 MIT Conference on Digital Experimentation (CODE@MIT): Fireside...

Promise and Limits of Observational Causal Inference. Featuring Eytan Bakshy, Facebook. Guido Imbens, Stanford. Eleanor Murray, Boston University. Dean Eckle...

www.youtube.com

2

41

115

Dean Eckles

@deaneckles

2 years

Yes, tech companies are hiring econ PhDs, but I think this misses the bigger picture here: Quantitative social scientists of many stripes working at tech companies. 🧵 For example: I think Meta employs more political science PhDs than economics PhDs.

Jaume Vives-i-Bastida

@jaumevivesb

2 years

Excited to be featured on the Economist! Why are economists so valuable to tech companies? @steve_tadelis @Econ_4_Everyone Why economists are flocking to Silicon Valley

4

43

209

5

15

116

Dean Eckles

@deaneckles

4 years

What does this person's choice of mask (and reading material) say about the causal inferences they are drawing? #bookofwhy @yudapearl

2

5

110

Dean Eckles

@deaneckles

1 year

Cross-cultural encounters

6

8

108

Dean Eckles

@deaneckles

1 year

Is it possible for experiments with 23,000 subjects to be underpowered for plausible, potentially important effect sizes? Yes @testingham

How Much has Social Media affected Polarization? | Tom Cunningham

tecunningham.github.io

2

28

107

Dean Eckles

@deaneckles

7 years

That's why I use LastFace, which autogenerates a unique face for each new account.

anildash.com

@anildash

7 years

Be sure to change your face every 90 days, the longer the better, and avoid having a common face or using the same face as on Facebook.

55

930

3K

4

23

108

Dean Eckles

@deaneckles

4 years

@dfreelon I also think bundling "no" and "unsure" people is a mistake — both scientifically (these are quite different attitudes!) and as a public health message that makes people overestimate hesitancy

Op-Ed: Why public health messaging should emphasize vaccine acceptance, not hesitancy

Overstating or overemphasizing vaccine hesitancy can keep others from deciding to get the COVID-19 shots.

www.yahoo.com

2

5

106

Dean Eckles

@deaneckles

4 years

Given that there is sophisticated removal of people (and change of wall material) behind him, I wonder if this triggers (e.g.) Twitter & Facebook's rules about "manipulated media"

Enforcing Against Manipulated Media | Meta

We're strengthening our policy toward misleading manipulated videos that have been identified as deepfakes.

about.fb.com

Philip Melanchthon Wegmann

@PhilipWegmann

4 years

About that new Trump ad alleging that Biden is “Alone. Hiding. Diminished.” The clip at the four second mark is edited to make it look like he is sitting on the floor of his basement alone. But it’s from an Iowa event last December. The room was full. See below:

810

7K

13K

5

37

100

Dean Eckles

@deaneckles

4 years

Please stop getting your COVID info from Feigl-Ding, who has been spreading misinfo and rumors for months. I still keep seeing people resharing his stuff. Obviously he isn't always wrong, but that's the nature of bullshitting. At least he has lost his Harvard affiliation.

zeynep tufekci

@zeynep

4 years

First tweet in thread, with 4000+ retweets and quote tweets. "Good and bad news." (What bad news? It was all-around excellent news). Finally, 22nd tweet with a clarification. Only 37 retweets and quote tweets. Don't do this people. Also don't get your information this way.

19

108

490

7

17

104

Dean Eckles

@deaneckles

4 years

Why settle for just plotting 95% confidence intervals? Tech giants' experimentation tools frequently plot many — below are 90, 95 & 99% CIs from an A/B test at Facebook. When actually making decisions, it's useful to easily see the strength of evidence.

Dean Eckles

@deaneckles

4 years

@ashdgandhi Gelman & Hill (2007) has a lot of examples of plotting multiple intervals. As does paper with the plot below. Experimentation tools at tech firms often show multiple intervals in varied lightness (eg 90, 95, 99% CIs)

2

3

21

7

14

105

Dean Eckles

@deaneckles

7 years

Headline: At Yale, we conducted an experiment to turn conservatives into liberals. Reality: At Yale's Bargh lab we negligently continue our poor research practices + hype.

Dean Eckles

@deaneckles

7 years

@dinapomeranz @SylvainCF Sorry but this study is not at all credible. Seems like noise mining. Study 1 p=.044. Study 2 swaps the outcome because primary outcome wasn't significant, still just p=.016.

2

1

11

2

22

98

Dean Eckles

@deaneckles

7 years

Learning a predictive model from observational data may not make for a good policy. A nice example:

Ronny Kohavi

@ronnyk

7 years

@vaughanbell . @deaneckles Gwern asks for examples. Rich Caruana etal in is good. See section on Pneumonia dataset

1

15

36

5

58

101

Dean Eckles

@deaneckles

4 years

Curve fitting, experimentation & causal inference — together at last @seanjtaylor

CDSM20 – Keynote 1

Keynote by Sean Taylor at the Causal Data Science Meeting 2020.

www.youtube.com

3

26

103

Dean Eckles

@deaneckles

7 months

When your t-statistic is 4, your 95% confidence interval goes from half off your estimate to 150% of it. A short post on relative confidence interval width:

0

20

103

Dean Eckles

@deaneckles

3 years

Here's Nobel laureate Guido Imbens — congrats Guido! — on the leaky machines that helped motivate the "credibility revolution" in economics

4

11

102

Dean Eckles

@deaneckles

5 years

100% of the people who made this chart have tried pot in the last 10 minutes. — @alex_peys

Dorsa Amir

@DorsaAmir

5 years

Easily the funniest data viz I've ever seen. (1) First of all, it's a pie chart. (2) The total is greater than 100%. (3) The relevant categories are today, last year, and... the year 1997? (4) The margin of error (MOE) is listed as the source.

625

7K

30K

1

13

99

Dean Eckles

@deaneckles

4 years

I don't want baseless anti-vax views to spread. But also: Do people really want Facebook deciding whether to allow critical commentary by scientists about some world government pushing a COVID vaccine? The ubiquity of "do something" ideology...

Axios

@axios

4 years

Mark Zuckerberg defends Facebook’s policy to not remove anti-vax content: “If someone is pointing out a case where a vaccine caused harm or that they're worried about it, that's a difficult thing to say from my perspective that you shouldn't be allowed to express at all.”

131

114

204

4

10

97

Dean Eckles

@deaneckles

6 years

Excited to receive an Amazon Research Award, alongside a number of impressive researchers! Will enable continuing and intensifying some work with @aminrahimian at @MITSloan & @mitidss .

7

5

99

Dean Eckles

@deaneckles

4 years

Social-media-induced #FOMO and social distancing: Whether people are going out and about is affected by what their (often distant) friends are doing

Interdependence and the cost of uncoordinated responses to COVID-19 | PNAS

Social distancing is the core policy response to coronavirus disease 2019 (COVID-19). But, as federal, state and local governments begin opening bu...

www.pnas.org

1

37

97

Dean Eckles

@deaneckles

6 years

This animation illustrates L1-regularization much better than similar static plots of the contours.

Pierre Ablin

@PierreAblin

6 years

Illustration of the Lasso and its path in 2D: for t small enough, the solution is sparse!

6

97

367

1

18

100

Dean Eckles

@deaneckles

3 years

Sometimes I get riled up about how much published work in the social sciences is bad. And then I get reminded how bad the medical literature is also.

1

99

Dean Eckles

@deaneckles

2 years

For people tempted to study social media via browser extensions: To a first approximation, no one consumes social media on desktop computers.

12

8

95

Dean Eckles

@deaneckles

2 years

I'm thrilled to start my first sabbatical shortly. So I'm interested in others' experiences, advice, etc. What did or didn't work on your sabbatical? Or what did or didn't work when your advisor, spouse, etc. took their sabbatical?

17

3

95

Dean Eckles

@deaneckles

7 years

Truth is much too complicated to allow anything but approximations. - von Neumann

3

31

96

Dean Eckles

@deaneckles

5 years

On the birth of Gladwellian statistics

2

16

95

Dean Eckles

@deaneckles

4 years

The value of communication technology in quarantine. Ad is from 1910 — not related to the "Spanish flu" but perhaps referring to other quarantines, such as scarlet fever

3

22

93

Dean Eckles

@deaneckles

5 years

Enjoyed searching this new archival history of computing at MIT (1950-1962) with my colleague John Little just now. We found this: Receiving 5 hours of machine time in 1954 for his dissertation. @DigitalHumMIT

3

17

93

Dean Eckles

@deaneckles

4 years

This blog post about replicability requires some key context: - The data are prices in a prediction market, not yet replications. - The sampled studies are not representative of these fields (eg economics as a whole is not mainly lab experiments).

Alvaro de Menard

@AlvaroDeMenard

4 years

Economics is (predictably) the strongest field. Education was a positive surprise (large samples, RCTs). Criminology, marketing, management are in a terrible state. EvoPsych seems to have absorbed the worst tendencies of social psych.

16

86

382

3

16

94

Dean Eckles

@deaneckles

5 years

We are launching a new virtual seminar series for quantitative research in marketing. Topics will range from customer relationship management to obesity and self-control. More info and email list here:

Virtual Quant Marketing Seminar

This is an open online academic seminar focusing on topics related to quantitative marketing.

sites.google.com

2

28

93

Dean Eckles

@deaneckles

3 years

Looking for a new diff-in-diff (or even triple diff) example? How about the effect of the introduction of wolves on deer–vehicle collisions:

nxthompson

@nxthompson

3 years

Loads of people die or are injured when their cars collide with deer. What can be done? Warning signs? 👎🏻 Overpasses? 👎🏻 Introduce wolves? 🤛

16

32

103

3

12

93

Dean Eckles

@deaneckles

7 years

Why limit A/B testing to randomizing users? Physicists at Netflix discover the power of within-subjects designs: One reason PlanOut supports arbitrary units.

2

38

92

Dean Eckles

@deaneckles

6 years

The difference between statistically significant and not is often not statistically significant.

Tavneet Suri

@SuriTavneet

6 years

A pet research peeve of mine: when comparing 2 regression coefficients (or 2 treatments in an RCT) people often report one is significantly diff from zero & one is not & imply this means something about a comparison of the two. It does not. Please test the two directly. #FTest

13

40

335

2

16

90

Dean Eckles

@deaneckles

6 years

I hope I'm still taking classes now and then when I'm 49.

1

12

91

Dean Eckles

@deaneckles

3 years

In case you're interested in social media & algorithms, I'll be participating in this hearing before the Senate Subcommittee on Communications, Media & Broadband at 10:45am Eastern this morning.

Subcommittee: Disrupting Dangerous Algorithms: Addressing the Harms of Persuasive Technology

WASHINGTON, D.C. — U.S. Senator Ben Ray Luján (D-NM), Chair of the Subcommittee on Communications, Media, and Broadband will convene a hearing titled, “Disrupting Dangerous Algorithms: Addressing the...

www.commerce.senate.gov

2

13

90

Dean Eckles

@deaneckles

3 years

How different are causal estimation and decision-making? eg: estimating a treatment effect vs. deciding whether to treat a unit Some comments on this contrast over at @StatModeling :

2

15

89

Dean Eckles

@deaneckles

1 year

Can you target interventions in a network without observing the whole network? There are heuristics (using the friendship paradox), but they don't work in all networks. So we developed algorithms with guarantees under standard contagion models:

Amin Rahimian

@aminrahimian

3 years

Starting from n*rho random initial nodes, our “PROBE” algorithm approximately simulates T cascades using subsampling and stopping constraints that allow us to upper bound total edge queries - collecting enough information to choose seeds with a performance guarantee

1

0

4

1

18

90

Dean Eckles

@deaneckles

6 years

Facebook using phone numbers provided for two-factor authentication for advertising purposes. This seems like a substantial breach of user trust to me. Important reporting by @kashhill based on research by academics including @sapiezynski

5

56

87

Dean Eckles

@deaneckles

3 years

A PhD in a management field can be a good way to do impactful, interdisciplinary social science — & often with more reliable career prospects. Please share this program with your promising undergrads, especially from underrepresented groups:

1

40

90

Dean Eckles

@deaneckles

4 years

We're surveying people (>400k respondents so far) in 67 countries to learn about beliefs, behaviors & norms related to #COVID19 . Check out the public data & dashboard here: Researchers can request access to the microdata.

1

46

87

Dean Eckles

@deaneckles

1 year

So I just got an email that goes, "I would like to introduce you to ****, a motivated 9th-grade student... He is seeking an opportunity to collaborate with a professor..."

Stephanie M. Lee

@stephaniemlee

1 year

fascinating investigation by @propublica into a brand-new industry that charges high-school students for the privilege of publishing research (or..."research"), making scientific literature even more of a mess than it already is

2

33

110

3

86

Dean Eckles

@deaneckles

3 years

The best instrumental variables are randomized experiments

Yiqing Xu

@xuyiqing

3 years

Their discrepancy is negatively correlated with the strength of the IVs in studies where the IVs are not experimentally generated, suggesting potential violations of the exclusion restriction; such a relationship is weaker with experimental studies.

2

4

32

5

4

86

Dean Eckles

@deaneckles

3 years

Responses to the impressive mask RCT in Bangladesh are great reminders that confident claims you encounter about statistics and data analysis are often wrong, eg:

8

6

87