Cover photo
Shravan Vasishth
Works at Potsdam
Attended School of Mathematics and Statistics, University of Sheffield
Lived in new delhi
139 followers|88,988 views


Shravan Vasishth

Shared publicly  - 
Our first ever paper using Stan (possibly the first paper in psycholinguistics using Stan) is now accepted in PLoS ONE:

And our tutorial on fitting linear mixed models in JAGS and Stan is now under review in PLoS ONE too:
Titus von der Malsburg's profile photo

Shravan Vasishth

Shared publicly  - 

I think the misuse of statistics in general is a huge problem and is producing lots of bad science. The misuse of p-values is just one of many issues. The lack of proper control for multiple testing is another major problem; some fields are quite aware of multiplicity adjustment, but others don’t understand it or [don’t] do it nearly as often as they should.

Trying many statistical procedures until one obtains the desired answer is another problem. Keep working on ``cleaning'' the data; keep working on the models; keep trying out different tests; keep altering things until you finally get a p-value that is less than 0.05 and declare victory! The wonderful paper [Simmons, J. P., L. D. Nelson, and U. Simonsohn. 2011. Psychological Science 22:1359–1366] shows how one can, in this way, obtain ``significant evidence'' that listening to the song “When I’m Sixty-Four” by the Beatles can reduce a listener’s actual physical age by 1.5 years.
Titus von der Malsburg's profile photoShravan Vasishth's profile photo
It's ironic that ASA does not make its journals open access. Statisticians can talk the talk, it's time they walked the walk!

He's just complaining about ignorance. I think ignorance is a large part of the problem. For example, top US labs publish results with positive conclusions like "there is no effect of X" with low power studies. That's not driven by the p-value, it simply reflects a lack of understanding about what the null result means in a low power setting.

Shravan Vasishth

Shared publicly  - 
Titus likens conference abstract-writing to programming with punch cards:

Recently, I wrote some guidelines for my students on abstract writing:

A CUNY and AMLaP abstract usually allows you to use about 500 words. Here, I provide some guidance on how to write an abstract that is likely to be successful. These guidelines are also the minimal requirements I have when I receive an abstract from you for comments and revisions.

1. Eliminate typos from your abstract. If you use emacs, enable spell check. If you use any LATEX software or Textmate, spelling correction is usually automatically enabled. In my own text editors (TeXShop and Textmate), the word is underlined if there is a typo. In particular, do not mis-spell researchers’ names; they are likely to be your reviewer and will not look at your work kindly if you can’t even spell their name correctly.
2. Learn from your grammar mistakes and never make the same mistake again. There is no credit assigned for being a non-native speaker of English. For better or worse, English is the language of science; this is unlikely to change. You have to find a way to construct grammatical sentences.
3. Make words count. Don’t use more words than you need.
4. The abstract has to have a relatively fixed structure. It is possible to deviate from this pattern presented below, but this requires experience, which takes time. For now, you should stick to this “house style”:
(a) An introductory paragraph which (a) establishes common ground, (b) shakes up the common ground by showing the gap in the topic you will address; and (c) explains the (novel) contribution and its importance to the topic. This section is not the same as the start of a paper; you have very little space, and you have to get to the point right away. You can’t start expansively the way you could in a paper. You should not have extensive references here. The sentences in this section need to have a logical connection that the reviewer will agree with.


A mechanism of predictions implies that the parser has to build up potential upcoming continuations, store them, and then evaluate them given the actual continuation (bottom-up evidence). This raises the question: do individual differences affect the capacity for building-up and maintaining predictions, and for recovering from unpredicted continuations?

No, this mechanism of predictions doesn’t “raise” this question! What the author is really trying to say is that they are going to study prediction at the individual level. The author is taking a novel perspective on the issue, not following up on an implication of previous work.

(b) The description of what you did must have enough detail that the reviewer gets all the relevant information. How to decide what is “enough detail”? Put yourself in the place of the reviewer and ask yourself: if this were an abstract written by someone else, would you be able to figure out what was done? If you have developed a new computational model, you have to motivate its novel aspects. If you have a new experimental design, you have to explain what it achieves. New terms have to be defined.


The similarity difference of matching only a single feature ... to a full match (an item with both features) was estimated ....

Here, the reader has no idea what “similarity difference of matching” means; this was never defined. So, for a reader not willing to think hard about your abstract (which be the case in 100% of reviewers), this sentence means nothing; and yet, it is an important component of the abstract.

(c) Any claims made in the results and discussion sections should be supported by the data. Some common mistakes are (a) to set up an expectation at the start of an abstract that is not met; (b) to make claims that are not substantiated by the data, or worse which are the opposite of what the data in the abstract show. In this section, you should not mention new terms and new ideas that have not been defined before; treating new terms and new ideas as if they are in the common ground is a frequently-occurring mistake.


Numerical difference between the conditions. The trend confirms the results from the literature where the ....

A statistician might buy this claim, but you are writing for psycholinguists, who consider a p-value of less than 0.05 a “true effect”, and anything bigger than 0.05 as “no effect.” You cannot say that you confirmed anything given a null result. Another example:
This suggests that in some cases low cognitive control readers may fail to inhibit the predicted gender form.

In the abstract, there was nothing about “some cases”. The claim is unsupported by data.

(d) The conclusion section should end with a summary statement about the novel finding, and explain the significance of the work. For example, saying things like “More research is needed” does not add any useful information. Another example is to end weakly with something that doesn’t even address the main finding:


Regarding the lack of interactions between condition and either X or Y, we speculate that the sentences were too simple and short in order to capture differential effects.

5. In some abstracts I see fonts changing from one line to the next. Don’t allow that to happen.
Shravan Vasishth's profile photo
You can also publicly comment on this tutorial on Andrew Gelman's blog:

Unfortunately, the first two blog comments (which will eventually get deleted, I imagine) are advertising links.

Shravan Vasishth

Shared publicly  - 
Gelman and Hill 2007 on fitting linear mixed models (p. 549):

"Don’t get hung up on whether a coefficient “should” vary by group. Just allow it to vary in the model, and then, if the estimated scale of variation is small (as with the varying slopes for the radon model in Section 13.1), maybe you can ignore it if that would be more convenient.

"Practical concerns sometimes limit the feasible complexity of a model—for example, we might fit a varying-intercept model first, then allow slopes to vary, then add group-level predictors, and so forth. Generally, however, it is only the difficulties of fitting and, especially, understanding the models that keeps us from adding even more complexity, more varying coefficients, and more interactions."

Shravan Vasishth

Shared publicly  - 
It is usual and convenient for experimenters to take-5 per cent. as a standard level of significance, in the sense that they are prepared to ignore all results which fail to reach this standard, and, by this means, to elimina...
Have him in circles
139 people
Johanna Thieke's profile photo
Jiří Semecký's profile photo
Frank Burchert's profile photo
Gisbert Fanselow's profile photo

Shravan Vasishth

Shared publicly  - 
Andrew Gelman in "the monkey cage" blog:

"... in my own work I’ve introduced major errors into the analyses on occasion. Sometimes it happens when I’m copying numbers from a printed source into a typed document, or even when editing a document. So I could definitely see how minus signs could disappear."

Maybe we should formulate an adaptation of Cromwell's rule as follows: "I beseech you, in the bowels of Christ, to use literate programming tools like Sweave and Knitr."

Of course, errors will still creep in, but at least they won't be copy-and-paste errors.

Gelman's post is at:
An influential study claimed that global warming had positive economic effects. Now that claim has been revised.

Shravan Vasishth

Shared publicly  - 
Here's a tip if you want to do a PhD with me: don't start your email like this; I will read no further.

"Dear, Proffesor."

Shravan Vasishth

Shared publicly  - 
From my talk in April at Osaka, Japan.

Shravan Vasishth

Shared publicly  - 
Dan Everett is apparently coming to town. Here is a piece on him from the New Yorker:
Reinhold Kliegl's profile photo
This is a very interesting piece. 
Have him in circles
139 people
Johanna Thieke's profile photo
Jiří Semecký's profile photo
Frank Burchert's profile photo
Gisbert Fanselow's profile photo
professor of linguistics
linguistics, computing, statistical and computational modeling, Japanese-to-English patent translation
  • Potsdam
    university professor, 2008 - present
  • Potsdam
    assistant professor, 2004 - 2008
  • Saarbruecken
    postdoc, 2002 - 2004
  • Hara Kenzo Patent Law Firm
    patent translator, 1990 - 1992
Map of the places this user has livedMap of the places this user has livedMap of the places this user has lived
new delhi - osaka (minoo) - nishinomiya - columbus, ohio - saarbruecken - berlin
Other profiles
Contributor to
There is no point in doing an experiment and gathering data if you already know what you are going to conclude from the data. Save yourself some time and money, just write up the result, skip the data-gathering step.
I'm professor of linguistics at the University of Potsdam, Germany. Before that I was a patent translator in Japan. 

See my home page for more details.
Bragging rights
Survived many attempts by nature to kill me
  • School of Mathematics and Statistics, University of Sheffield
    graduate certificate in statistics, 2011 - 2012
  • Dept. of Linguistics, Ohio State University
    PhD linguistics, 1997 - 2002
  • Dept. of Computer and Information Science, Ohio State University
    MS computer and information science, 2000 - 2002
  • Graduate School of Language and Culture, Osaka University
    PhD student (transferred to OSU), linguistics, 1995 - 1997
  • Dept. of Linguistics, Jawaharlal Nehru University
    MA, linguistics, 1992 - 1994
  • Center for Japanese Studies, Jawaharlal Nehru University
    BA (Hons), japanese, 1986 - 1989
  • Alliance Francaise de Delhi
    french, 1986 - 1989
  • Hans Raj College, University of Delhi
    BA program (dropped out after first year, due to illness), economics, 1983 - 1984
  • St. Columba's School
    Indian School Certificate, 1971 - 1983
Basic Information
Other names
ワシスト シュラワン, 和髭須土 修羅椀