What most activates a sense of purpose within you in your career, work, etc.? For me, it's ...
* working with others
* who have something to teach me
* while I use people/data/survey-science skills
* to help others
How about you?
@TableSlamStan
And it should. But it’s reasonable to inquire into the possibility that the chance of success here is better. And certainly you should not assume that a failure is “overdue”
Whatever your stance on journals, it’s *BIZARRE* that Elsevier (with a profit margin rivaling Apple’s) won’t pay for your peer review labor, gets content for free, and sells it back to your school for profit. Serious question—how can we do this differently? Alternatives?
I’m 6 years post PhD, a tenured associate prof, and happy. I still have recurring dreams that I did not finish a class in high school, and so all of my education is a fraud, and I will lose my job if anyone finds out. Does anyone else have this? Is this just imposter syndrome?
Academic tweeps: do you find that you essentially have to write and rewrite and rewrite (and rewrite) in order to get a really clear, strong manuscript? Or is it just me.
Was asked by a student this week: “but don’t you always *want* to reject the null hypothesis?” ... I responded: “not if it’s true!” and it was a revelation for the class. They need someone to tell them the goal is *truth,* not p < .05
@HeadSnapss
Unclear what’s untrustworthy in this example, but the meme seems to imply that the doctor is “overdue” for a bad outcome (which is false; that’s not how probability works)
This is a great book! Do you analyze survey data? Do you lack the stats training to address sampling biases or to use stratified designs? Were you trained in social science rather than stats? Do you want to up your quant game? Get this book. You won’t regret it!
#scilearn
@rustykitty_
Start with the chair or dean
> you did not use AI
> there is no proof that you did
> the tool in question is questionable (can you see the evidence it supposedly has?) and besides the score was low!
> the faculty admits they don’t care
~ a former prof + AI / ML practitioner
This book is great and has more value than you’d think. But the best value is chapter 8 on “career advice” … this has to be the best DS career chapter I’ve seen. Red flags in a job? Why data teams succeed if fail? Why you love (or hate) your data job and what to do? It’s here
Does anyone go back and forth between thinking (a) their paper is amazing and a great contribution and (b) their paper is obvious and makes not contribution? How do you keep the faith alive?
Please RT! Psych Tweeps: Started a google doc to curate an annotated list of R books, tutorials, and resources for folks wishing to "make the jump." Think: colleagues, grad students, and undergrads. Fully editable! Let's keep this going? Edit/share.
Does anyone else do this?
Me: ooh, look at this great resource I see on twitter. Let me email it to myself for later.
Also me: why did I send myself so many emails!?
@ColIegeStudent
Serious take: Major is different from career path. The major is about building a foundation. There are MANY more career paths than majors. Most people build satisfying careers and most use skills from their major, but not directly
Friends--I am thrilled to announce a major transition: I'm walking away from tenure and pursuing an impactful career around analytics/data science. It's missional work to me and I am thrilled for the next chapter. I start in January at Microsoft! Details here :)
Let’s talk about reproducible analyses!
Before I left academia, I started making reproducible R notebooks for every study. Walks user through every step, explains logic & matches paper. Folks—it’s as much work as the actual paper and the only way i see to make this work
1/n
STOP talking trash about different statistical techniques
Multilevel models are VERSATILE
t-tests are ICONIC
Regression is MAGNIFICENT
Factor analysis is USEFUL
ANOVA
Mixture models are CUTTING EDGE
Psych folks who know
#rstats
but are intimidated by SQL databases (i.e., industry work)--you already know the core concepts. Let me help you translate. 🧵
1. The "database" is analogue to your R environment. It can store many tables. (1/n)
After a year at Microsoft (and out of academia), I’m amazed at how much more organized my personal life is. Planning. Notes. Not just doing the thing that’s on fire. Such a good life change!
Psych folk--consider this for your department. We added a "data analytics" track in our psych major. Students can learn advanced R, databases/SQL, and beginning programming. Add to stats/methods and all the people insights and you are positioned for industry with your BA
Just drafted 25-page guide to scientific writing for undergraduates in psychology (focused on students in a Research Methods). Broke good science writing into four principles and three skills. Looking for feedback! Also might there be value in submitting?
Analyzing longitudinal data? This collection of tutorials is *AMAZING* ... beautiful markdown tutorials for every kind of analysis with code, data, everything
@g_mppa
@TableSlamStan
Also “regression to the prior.” Yes, it’s possible that this doctor’s average is worse than 100%. But that does not imply that the recent successes make the one “due” for a failure as the meme suggests (by my reading)
Most psych grads won't go into academia or even "psychology." They get jobs. To use stats skills, most workplaces emphasize excel, not R (against my own wishes, but it's reality). We need some place in the psych major where they can get minimally competent in excel
Do you simulate data in R? What tools do you use? What can you share? I'm building a resource list (and will share here when done)! Share code snippets, blog posts, R packages, etc. For example, just found 'truncnorm' that generates normal data within a given range. What else?
Want to up your regression game? Nominal outcomes? Ordered outcomes? Switch between R and python? I am finding this book by
@dr_keithmcnulty
to be quite delightful (free online!)! Great, fast read! Has a nice section on power analysis
#scilearn
Linguistic analysis of language from police body cameras during traffic stops. Analysis reveals a considerable difference between how white and nonwhite individuals are treated. White folk cautioned about safety; black folk told to keep hands on wheel. h/t
@MikeShor
for posting
Do you have a background in social science and want to use stats to solve real world problems? Did your grad training stop at multiple regression variants? Do you find advanced stats books inaccessible?
This book was written for you. Read it!
#scilearn
Ex-academic in tech, end of week 1. I never realized how much I felt “perpetually behind” in academia and how much of an emotional toll that took. We habituate to it but it still takes a toll.
I’m struggling to find words, so I will just share some personal news: I have been promoted to Associate Professor, starting in the fall! I can’t help but think that psych twitter has played a major community support role. Thank you!
@tcarpenter216
@adamjnafa
The original hard drives, you used to be able to add more platters to expand storage. Thus bad boy probably weighed 200lbs and stored 10MB
Ex-academic in tech, month 4. The best thing I got from my PhD was the confidence that I could learn anything to do a project. Learning happens at (or over) the edge of comfort zones. So new projects will be uncomfortable but growth WILL come. Embrace it. You’re a learn-it-all.
Data science truly is an amazing job for a methods/stats person. Training cool models + running cool analyses—much cooler methods than i used in academia. Building with the coolest toys. Solving big problems. Really impactful work. Living the dream!
If you analyze data and care about stats, read this book. It’s accessible and will convince you that much of what you know is wrong. Many statistical techniques are much more fragile (non-robust) than you were taught.
🧵Want a great industry job doing behavioral science, surveys, data analysis, data science, or other "psych research" skills? You might be surprised that technical skills are NOT the thing you should focus on (1/n)
Note to self: when randomizing items in Qualtrics, be sure to randomize the order of the *items* and not the order of the *likert scale points* ... dodged a disaster there!
For the first time, I had a grant accepted today. Many failures to get here, but thanks to psych twitter for your encouragement and the conversations that inspired the project.
Since adopting
#OpenScience
practices, I’ve fallen in love with research all over again. It feels more real—a quest to find truth, not just a publication game. Just another perk. Thanks open science folk!
Love this intuitive explanation for R^2 from
@dr_keithmcnulty
. How large are your residuals using the mean as a predictor? How large are they under your regression? The R^2 reflects the proportion of the (squared, summed) residuals that you have reduced.
Curious for thoughts. Why is it *SO HARD* for people to admit they're wrong? What if we created a culture that normalized being wrong--and then improving? Seems integral to everything: science, relationships, personal growth. Have to know I'm wrong today to be more right tomorrow
Leaving academia? It can be a big leap. Remember this, PhDs -- your whole game is that you learn whatever you need in order to get the job done. Don't know databases? You will learn. Don't know what a blob storage service is? You can figure it out. You're a "learn-it-all"
It’s also not a knife or a spoon or a comb. Or many other things it doesn’t try to be.
#Rstats
is excellent at functional programming with data. It’s not useful for much else. And that’s ok. Anyone complaining that it’s not “more than that” doesn’t understand R
In the past 24 hours, I have been called unskilled, less educated than a HS student, and held up as the example of all that is wrong with the PhD system. I’ve taken down my post and will not be tweeting about data science any time soon. It’s not worth it.
Kids, don’t plagiarize, cheat, or fake your work. We—your professors—know that the only thing that can come of it is a speaking circuit with big fees, a prestigious position at Duke, and a tv show with a character based on you. Not worth it.
I'm totally flabbergasted: At what point will there be some consequences for this type of behavior from Duke University, academic journals, or other institutions that ostensibly care about academic integrity?
What is the appropriate outcome in a situation like this?
More on this book (basics of modern stats, data science, ML). I’ve rarely seen a book THIS GOOD at teaching. Every sentence is worded to avoid confusion, lay foundation, or build deep understanding. Makes me miss teaching stats. Worth your time
#scilearn
Data visualization in psychology is .... lacking. One of the best books on
#dataviz
is this one by
@ClausWilke
-- and turns out it's free online! Read it, preach it, teach it. Reviewer 2 often told me to remove figures... still worth a shot!
I...I have no words
Non-randomized, n=63,
Control: staring at desk
Active intervention: caring for plant
Outcome: tiny difference in proxy of mental health
Headlines:
In my analytics consulting career, I discovered that this is how many organizations do inference today. Fun fact—a client once thought I was LESS skilled than another analyst bidding because I claimed needed a larger sample size to draw a conclusion
Reading a paper from 1905 which reports results as means & then says "Just take a look, the means are different". No analysis? And this is how I learned the t-test wasn't invented until 1908. People were just like, looking at numbers & going "yeah, looks different" before then.
I’ve spent my life assuming there are infinite moments. My PhD taught me that life comes “later” (bye 20s). Tenure took my early 30s.
The things that mattered were when I showed up in love and care for others. Advocacy. Marriage. Fostering. Parenting
Worth thinking about
I realized a couple years ago I stopped caring whether my hypotheses were supported...and started treating contrary evidence as a sign that we need to revise theory (vs some failure of mine). It improved my science and sense of well-being. Feeling grateful today for
#openscience
Ex-academic in tech, Day 2: I love my job, love my team, and love my role. I did good work today. Then, at 5 pm, I put it away. It's weird putting work away and not feeling like it's a compromise somehow. Looking forward to this.
Ex-academic in tech, day 8: all the stuff I used to do that felt like “not real work” (but I still had to do) was real work. If there aren’t enough hours in the week to do that stuff *and* teach *and* write then the problem is the work plan, not the execution
Many PhDs lack some fundamental math training that help them feel confident in the data science space. This is a fantastic teaching book! In addition, it does it all in python (a great language to learn in addition to
#rstats
)
#scilearn
Causal inference comes down to 5 rules familiar to “mediation” crowd:
1. Control for confounders
2. Don’t control for colliders
3. Controlling for mediators makes estimate from X a direct (vs total) effect
4. Magically know the DAG to decide ^^^
5. random assign when possible
Stats, psych, data twitter -- need your help! Industry is full of smart and ambitious people doing analytics who haven't had stats/methods training. Have regular need to direct people to the equivalent of intro stats, intro research methods, advanced iterations, etc. Where send?
@ling_sprenger
Let’s normalize this. I do this occasionally. Negative feedback is commonly shared and positive feedback rarely is. Let’s share the good!
Decided that p-values make much more sense to students when phrased as "% of the time." E.g., "If H0 were true, results this strong would still occur 34% of the time" ... Wondering how much confusion we'd avoid if we taught it that way more consistently
Has anyone seen something like a causal inference boot camp or online course? Eg an online course covering modern techniques for those who didn’t get much in grad school beyond regression and RCTs? I’m self-taught and want to improve my skills this year
Moved advanced stats class to be R focused (instead of SPSS). Realizing I have to go slower but that’s because students are learning deeper. It was too easy in SPSS to give a 30,000 foot view with no depth. Teaching with R forces understanding.
Academics: not sure who needs to hear this, but responding to student emails, scheduling, and all the other things that take up your day are "work" and if you spent a day doing them, you actually did "get work done."
Most 'data analysis' resources are technical =(e.g., coding in R or python, data engineering, etc). Wanting to help others up level into the THINKING process that analysts/scientists do naturally (problem --> helpful question --> test --> insight). What can I recommend to ppl?
As a data scientist who has worked exclusively in
#rstats
for many years, I’m really excited by python. This book has a fantastic boot camp for python and DS work in general. Snarky. Full of gems. Great teaching tool. Highly recommend.
#scilearn
Thesis: the magic of Twitter was never the platform. It was that it made it effortless to find a community of people you didn’t already know. I have yet to find that on any other platform.
One of the kindest things my advisor told me in grad school was that she would support my career goals, even if I wanted to decorate cakes. She’d teach me how to design experiments to make the perfect cake. Heard that in 2012 and I still find it inspiring
We need to move to a publishing/hiring/training model that emphasizes quality over quantity, teams over PIs, accuracy over prestige, integrity over speed, and importance over flash. For science to reach its potential, these should be nonnegotiable. 6/n
Power for interactions is scary. Correlation in group 1 = .50, correlation in group 2 is ~ .40-.45. How many people do you need *per condition* to determine whether condition moderates this correlation?
Bandwidth is a budget. You can't solve a budget by saying you will magically have more money. You can't solve a bandwidth problem by simply asking people to "do more." If needs go up, bandwidth must be found by reducing commitments elsewhere. Thank you for coming to my TED talk.