Nick HK
@nickchk
Followers
19,152
Following
341
Media
1,194
Statuses
17,178
Econ prof @SeattleU . Book The Effect out now! Check my pinned thread for all my projects. Substack
Joined October 2010
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
Chiefs
• 360550 Tweets
namjoon
• 317714 Tweets
Lamar
• 118766 Tweets
Adalet
• 103436 Tweets
Chester
• 74080 Tweets
For Asia
• 69655 Tweets
علي النبي
• 63715 Tweets
小泉進次郎
• 43568 Tweets
Hayırlı Cumalar
• 41909 Tweets
サブスク解禁
• 41801 Tweets
MUSINSA BEAUTY WITH KARINA
• 32796 Tweets
エイリアン
• 29814 Tweets
Engin Polat
• 27624 Tweets
#BAEKHYUN_PINEAPPLESLICE
• 27356 Tweets
PRETTY GIRL ERA MILKLOVE
• 26972 Tweets
小泉さん
• 24898 Tweets
#يوم_الجمعه
• 23935 Tweets
資さんうどん
• 22670 Tweets
TransisiSEJUK PRABOWOJKW
• 21980 Tweets
KitaKOMPAK KitaMENANG
• 21094 Tweets
知的レベル
• 20464 Tweets
KPLC
• 18797 Tweets
ちいかわ寿司
• 18677 Tweets
資さん買収
• 17391 Tweets
ベイマックス
• 13248 Tweets
Pinned Tweet
This is a thread of me QTing myself so I can pin all my posted-on-Twitter projects.
Causal inference animated graphs:
As requested, slower graphs! Also added a graph on collider bias, the webpage explanation helps there.
These graphs are intended to show what standard causal inference methods actually *do* to data, and how they work.
This is what controlling for a binary variable looks like:
26
458
2K
6
94
485
I've been getting used to gganimate and thought it would be useful to put together some illustrations of what various causal inference methods *actually do to data* and how they work. Here, for example, is what it means to control for a (binary) variable
43
1K
3K
In regression, there are several things that students are eternally concerned about but are actually Just Fine:
1. Your coefficients don't need to be significant
2. Your R^2 doesn't need to be huge
3. Your predictors can be correlated
4. Your variables don't need to be normal
51
362
3K
My students have been having trouble figuring out how to put together a regression model - what to include, what to interact, which variables are left-hand/right hand. So I made a flowchart.
23
380
2K
As requested, slower graphs! Also added a graph on collider bias, the webpage explanation helps there.
These graphs are intended to show what standard causal inference methods actually *do* to data, and how they work.
This is what controlling for a binary variable looks like:
26
458
2K
A graphical explanation of why you shouldn't use R^2 to pick your model.
28
328
1K
Over winter break I'm planning to automate my CV, so I can just update a spreadsheet with accomplishments and it will spit out an updated website and PDF CV. Is this something anyone else would use or should I write it just for myself?
86
55
1K
Today I'm releasing the first video designed to accompany my book The Effect, about research design and causal inference. There are ~70 videos planned for the series. These can be used to accompany the book, as classroom material, or just on their own.
10
231
1K
Today I finally get to share the full draft of my new book The Effect. I think it turned out really well. The Effect offers a highly accessible and intuitive approach to research design and causal inference with observational data. Find it here:
20
285
1K
My causal inference book The Effect is finally out today! It's a super intuitive look at causal research design, casual diagrams, and standard research methods. I'm super excited, and jealous of everyone getting their copies today before I get mine!
27
162
1K
I'm excited to show off something I've been working on for a few months now: the materials from a new class that I pinpoints a very different, and I think very promising way of teaching undergrad econometrics (thread)
20
221
887
Last night my husband offered every trick or treater (old enough to handle it) a choice between a candy or a "mystery box".
Every single child chose the mystery box. Trust levels high in my neighborhood.
22
23
829
I've written an undergraduate-friendly guide to using simulations to do power analysis. It's in R but the concepts are general. Suppose I could do a Stata version too.
10
148
855
Statistics terminology is truly unhinged. Different names for the same terms across fields. The same names for different things across fields. Common language terms applied in highly specialized ways. Two similar terms used to mean entirely opposite concepts.
35
70
749
I wrote a guide to common errors and mistakes I see beginning R students making in my data viz class. Stuff like "function not found" or "package was compiled with R version..." or "don't put numbers in quotes". Your students may find it handy too.
8
153
753
I have accumulated many Stata tips over the years, but none of them are as simple yet provoke as many jaw-dropping "I wasted *all that time*" responses as inlist().
(as for why inlist() only accepts ten arguments I do not know)
21
81
756
If you teach students to work with data, you're doing them a great disservice if you just teach them how to run models/analyses and not how to clean and manipulate data. If you haven't tested them directly on this, you'll be surprised how unintuitive this is to new users!
9
79
714
I am going to make a lil website that simply has correct definitions of "statistically significant", "confidence interval", and "p-value" and also a list of common incorrect definitions of these terms.
If you have a good incorrect definition, list it here, I may include it.
39
34
683
I have a new paper out in Economic Inquiry with a buncha coauthors: The Influence of Hidden Researcher Decisions in Applied Microeconomics. What's it about?...
15
200
610
I've updated my Data Wrangling in the Tidyverse course material and am uploading a 17-part video series. This assumes little previous R knowledge (although some). Covers tidying, manipulating variables, cleaning factors, dates, and strings. Enjoy!
5
108
574
Difference-in-differences and regression discontinuity estimates don't directly tell you the effect of a treatment. They tell you the effect of a *cutoff*. It's only our theoretical understanding of what that cutoff means that lets us interpret that as an effect of treatment.
6
37
534
I don't post about my personal life a lot but today we finally had our court date to finalize the adoption. I've found that academics, who make up a lot of my followers, are above-average curious about adoption. So here's a little about the process, and how it went.
14
22
529
If you look at autocorrelated data, pick any random day, and look for a break in linear trend on that day, you will find a statistically significant break a ludicrous amount of the time. In this random AR(1) data with delta = .8, I reject the null at 95% about half the time.
17
79
512
It's fully-featured and ready to go! The did Stata package allows Stata access to the Callaway and
@pedrohcgs
R package "did" for estimating difference-in-differences with staggered treatment and covariates. Check it out, and installation instructions, at
4
131
510
I have a new paper in the Journal of Causal Inference about IVs!
You can greatly improve the finite-sample performance of your IV by simply modeling first-stage effect heterogeneity. Super easy! Even if you do a bad job, it helps! Big error reductions.
10
103
507
Success! My package📦causaldata 📦 is available on CRAN (install.packages('causaldata')), ssc (ssc install causaldata), and PyPI (pip install causaldata).
causaldata contains data sets used in The Effect (by me), The Mixtape (
@causalinf
), and What If (
@_MiguelHernan
and Robins)
1
90
504
This fixed effects animation stands less well on its own (i.e. explains less), but if you already sorta know how FE works I think this works better and looks nicer
(if you wanna drop this in some slides, this is from the Wooldridge crime4 data set, GIF: )
10
97
493
I have a new package for R, Stata, and Python: causaldata. Data sets to run example code from causal inference books. Currently it just has mine, The Effect, but planning to add at least The Mixtape before uploading to CRAN/ssc/PyPI. Install instructions:
Today I finally get to share the full draft of my new book The Effect. I think it turned out really well. The Effect offers a highly accessible and intuitive approach to research design and causal inference with observational data. Find it here:
20
285
1K
8
77
489
Some neat tools I use with some regularity and IMO should be more visible:
1) TablesGenerator, to convert tables between Word / Excel / LaTeX / Text / HTML
4
112
482
I've been working on an intuitive and accessible intro textbook on causality and research design. It's going pretty well, Volume 1 almost done. I'll be posting chapters from the book, one per week. Link here, and I'll update this thread with chapters
5
96
477
Hey all! I'm excited to reveal my new R package. It fills a gap I've been sorely missing since I've been moving from Stata to R: a quick and easy VARIABLE BROWSER! One you can use Find-in-Page on, and variable labels! Introducing the package vtable()! (thread)
12
116
451
In private consulting, the econometrics tool that people have not heard about but most often gets them very excited when I suggest and explain it is Kitigawa-Oaxaca-Blinder decomposition
9
28
421
I just learned that you can scale your LaTeX table to the textwidth by wrapping the tabular in \resizebox{\textwidth}{!}{} and OMG I'm never going back
12
27
424
In addition to being a professor, I also do freelance statistical consulting (my rent is too damn high). I thought it would be interesting to go over the way that people think about statistics in this hidden little corner of the internet.
7
150
417
When the abstract says "we use an instrumental variables approach to establish causality" and doesn't say what the instrument is, you know you're in for a wild ride
11
19
417
I still find it hard to believe that OLS consistency only relies on X being uncorrelated with ε, not independent of ε, and I occasionally run a simulation to re-convince myself.
8
49
412
All the materials for my econometrics class are online (except assignments/exams). Check them out if you like.
Videos:
Slides:
Most readings are Bailey, but some are my book:
Swirl:
5
105
399
The econ job market is starting again. Just a minor advice thread, specifically aimed at people in the academic market who are not superstar candidates. IMO too much of the advice is aimed at people who are heading towards a job at an R1.
8
93
375
ⁿᵒᵗ ʳᵉᵃˡˡʸ ᵗʰᵉ ᵗᵒᵖᶦᶜ ᵒᶠ ᵗʰᵉ ᵐᵒᵐᵉⁿᵗ ᵇᵘᵗ ᶦ ᵍᵒᵗ ᵗᵉⁿᵘʳᵉ ᵃʸʸ, ˡᵉᵃᵛᶦⁿᵍ ᶠᵒʳ ˢᵉᵃᵗᵗˡᵉ ᵃⁿʸʷᵃʸ ᵗʰᵒ
23
0
354
scales, which formats numbers for presentation, is such an undersung package. I use it all the time. If you make anything in R that is intended for an audience to see - graphs, tables, RMarkdown/Quarto, it's perfect.
6
28
351
Speaking of DID by the way, if you missed it, there's now an R package for
@jmwooldridge
's extended two-way fixed effects estimator for staggered treatment cases, courtesy of
@grant_mcdermott
.
2
58
351
When I was in grad school, learning LaTeX was 100% worth my time. For current grad students I think you're probably better off learning how to write in Markdown and have something else do the LaTeX conversion for you when you want a PDF.
20
33
342
Slides available for my course on causal inference. Causal concepts and the whole deal with what identification actually is, plus methods like controlling for stuff, standard designs like DID, RDD, IV, and details like estimation and het. treatment fx!
5
76
339
Some neat R pipe %>% tricks you might not know:
1. Pipe to {} to do any sort of calculation you like, referring to the passed object with .
2. Pipe to `[[`() to apply the "[[" function (i.e. object[[index]]), or similarly `[` for object[index]
10
45
336
Diff-in-diff is interesting because you start with *such* a simple implementation - OLS with an interaction term and a fairly grokkable parallel trends assumption. And it's great! Then you realize that breaking that basic case at *all* makes you have to change *everything*
2
24
335
"OLS of Y on X is unbiased only if X is unrelated to the error e. If there's a term Z related to X in e, the sign of the bias is the product of the signs of Cov(X,Z) and Cov(Y, Z)"
???
"If Z hangs around X but OLS doesn't know about it, it'll give X all the credit for Z"
"oh ok"
5
21
331
The 20th page has been added to the Library of Statistical Techniques. I'll be posting new pages here as they come in. Please consider contributing! I'm planning to write one a week. Edit pages that are there or add your own. It's easy.
4
85
330
causaldata is complete in R, Stata, and Python! Or at least it now has all data sets from both The Effect by me and Causal Inference: The Mixtape by
@causalinf
(except for judge_fe, it's too big). Install instructions:
I have a new package for R, Stata, and Python: causaldata. Data sets to run example code from causal inference books. Currently it just has mine, The Effect, but planning to add at least The Mixtape before uploading to CRAN/ssc/PyPI. Install instructions:
8
77
489
2
57
323
without exaggeration, the most important skill for any data analyst working in any field is the ability to notice when something is wrong
7
47
316
Announcement and invitation! A new project that aims to improve the quality of research in applied microeconomics by examining researcher choices. I am hoping to recruit up to *200 researchers* of all kinds (with pay) and hope you will join me! (Thread)
12
140
319
Never been a fan of the way students are commonly taught about outliers as things to be precisely defined, detected, and destroyed. Sure, sometimes checking outliers past a certain cutoff lets you detect data errors, and dropping/fixing those makes some sense...
9
29
318
Updated my data viz course slides if you're looking for material (or want a free course). Note it's also a "motley assortment of a bunch of stuff you gotta be good at data" course, like data cleaning, communication, statistical intuition, etc.
7
40
310
The causaldata R, Stata, and Python packages, with data sets for running code examples in The Effect, Causal Inference: The Mixtape, and What If, have a minor update to v.0.1.3 on CRAN/ssc/PyPI, fixing some issues with Mixtape data sets.
1
43
308
Code now available in a repo:
13
29
292
I have a new paper! "Human Capital vs. Signaling is Empirically Unresolvable." The most unusual paper I've written. I try to identify the relative importance of HC and S in education returns, and come to the conclusion that this question is unanswerable.
9
56
285
People seemed to be into the idea, so I am launching the Library of Statistical Techniques, or LOST. LOST is a Wiki guide to doing things in statistical software/code. Instructions, examples, and a little Rosetta stone btwn languages.
6
86
283
Been working on a Stata package requiring ML and realized I could whip something up that would be pretty useful here. So it's MLRtime, a Stata package for running Machine Learning commands in R! So yes, now you can do random or causal forests in Stata, with mostly-Stata syntax.
3
68
284
Here's a link to a page with these graphs alongside DAGs and more explanatory detail, if you want something to link your students to. Also if you have any ideas for other methods I should animate let me know.
4
42
285
Very cool paper just out in AER. They actually find *negative* amounts of publication bias - review process selects *against* marginally sig. results, even though reviewers like significance. Observed bunching is driven by p-hacking before submission.
3
55
275
Conditional probability P(A|B) = P(A and B)/P(B) means "out of (denominator) all the B's, how many (numerator) are also A's?" and every time I see a textbook that doesn't use this intuition it annoys me
7
23
275
A clean way to interpret the univariate OLS slope, Cov(X,Y)/Var(X), is "out of all the variation in X (/Var(X)), how much of that is related to Y (Cov(X,Y))?"
Logic extends to multivariate (X'X)^(-1)X'Y if you remember that inverse = divides by = "out of"
9
24
271
I have a new short working paper: Linear Rescaling to Accurately Interpret Logarithms.
Do you use logs? Do you interpret increases in ln(X) in terms of percentage changes in X?
Did you know there are a lot of problems in the way we're told to do that?
5
54
265
Happy father's day to me, I relaxed and made this animation.
This is what it looks like to control for a continuous variable in a linear model.
8
46
259
been hearing plenty of arguments as to why I should start using = instead of <- in R. i'm only sorta convinced so i've started alternating between the two in the same code. this should make everyone happy
12
5
260
A video version of my workshop "Teaching Econometrics with R", targeted at faculty already familiar with another language, and focusing on getting used to R as well as relevant functions and packages for undergrad econ
1
52
260
Copilot is live in the most recent normal release of RStudio, btw. Tools -> Global Options -> Copilot
FYI, I haven't installed it myself and will probably just wait for the actual release, but the newest daily version of RStudio has GitHub Copilot integration, and I hear it works quite well!
5
12
100
5
35
253
A frequent error I see students make is thinking that they can account for "the effect of X on Y is different for group A vs group B" by adding group as a regression control variable. This is incorrect! So I made a graph.
6
40
248
Now with a cover! Get your preorders of The Effect in while it's still on sale. and as always you can check it out first at
6
40
251
the "use OLS for everything" vs. "don't" debate is neat because it really highlights how doing statistics well is much less about being right about anything than it is about making your wrongness as harmless as possible
7
17
251
Some neat ggplot2/adjacent tricks I didn't discover until surprisingly recently
- expand option in scale_*_discrete to add room around factor variables
- position_dodge() to line up geoms without position = 'dodge' with something dodged
- scales::label_wrap to word wrap labels
4
16
245
Finishing my econometrics prep. Gonna make a hard sell for my vtable package in your class if you're using R.
- Easily explore variable characteristics and values with vtable
- Super easy summary tables with sumtable
- Balance tables with sumtable
6
26
247
A demonstration of sampling variation in OLS with N = 2 (albeit resampling from a larger sample, not the population)
6
42
245
Having a tenure-track job where it is perfectly fine that I will never publish in a top 5 is actually pretty sweet, I recommend it to anyone with the same crossed wires I have where academia is endlessly fun
2
0
243
Still mildly annoyed by the RateMyProfessor that says my class is impossibly hard, and then at the end tacks on that if you want to pass you should try going to lecture or reading the book because they did neither.
10
9
238
Something that would be extremely useful would be a Wikipedia-style statistics practitioner cookbook. Something you can look up the method you're about to do and it will remind you of all the best practices. (thread)
11
32
235
I do a lot of statistical programming and econ work on upWork. Data cleaning, research design, editing, coding. If this is work you are capable of and you need another income stream bc of quarantine, give it a shot. Also let me know and I can probably send some clients your way
3
40
238
Reading a bunch of DID papers and it's nuts just how much more widely applicable DID is than any other CI method. Anything improving DID to make results more believable (and there have been a few recently) is maybe the highest marginal value an applied econometrician could offer.
7
23
234
Haha I'm never maintaining a repository of citation info ever again, this is so nice.
In the Rstudio/Quarto visual editor (which I think is soon to be not-just-in-Rstudio)
10
17
235
Clue is a board game about finding a dead body but being unsure about which room you found it in and unable to tell the difference between someone who was bonked over the head or shot
3
18
218
I've never been a huge fan of adjusted R^2, but it has now been pointed out to me that whether an additional variable makes adj R2 go up or down is whether its t-stat is above or below 1, and now I like it less.
8
27
231
At long last (and, of course, exactly one day after I sent out my one-email-a-month-I-swear update email), my book The Effect: An Introduction to Research Design and Causality is now available for preorder and on sale! Please check it out.
6
49
231
I'll never forget the time someone came to my house, saw the various dolls and princess castles and stuffies, and said "wow, you must really like toys" instead of correctly deducing that I have a child
8
5
225
Guess at how many observations you need to have 90% statistical power to detect an effect that increases a rate from 50% to 55% (not an inconsequential boost in many settings!). Assume perfect randomization, half treated/half control
Answer in third post.
21
40
225
Making the same package in R, Stata, and Python really drives home how nice the package development environment is in R. It's possible (likely) that I'm missing some neat tools in Python to make this easier but R is just on a different level for this task
7
13
224
I have not been able to get my hands on exact numbers, but from some back calculations it looks like The Effect has sold something like 3000 copies so far, which is pretty amazing, especially for a book also available for free! Thank you all very much.
9
12
226
One very cool thing about dbplyr and dtplyr is that they don't just run your code, but convert the dplyr syntax itself into SQL/data.table, and show you how to do what you wanted.
Are there any other packages like this? Could I type in dplyr or data.table and get pandas syntax?
6
23
225
I've just signed a book deal for The Effect with Chapman & Hall, who also publish the excellent What If by
@_MiguelHernan
and Robins. So a physical version will be available at some point. Building a little causal inference library over there!
11
16
216
What I have learned from academic twitter is that every statistical method is both the gold standard and also so completely worthless and unbelievable that it doesn't even justify explaining why it's bad
4
27
210
This is for a class I'm designing on programming and causal inference (h/t
@causalinf
) designed to go BEFORE the rest of the econometrics sequence. The idea is teaching concepts before methods. Notice that none of these graphs use regression! It's not necessary!
3
11
214
I correctly predicted the econ Nobel last year, but only because I kept predicting the same people every year until it happened. Strongly recommend this strategy.
9
6
215
This is the first time I've been excited about something new from Microsoft since like 2010
5
37
208