« Strategies for validating observational studies 2/2 | Main | Practical issues in observational studies »


Feed You can follow this conversation by subscribing to the comment feed for this post.

Michael Droy

I need more than a theory if you are going to demand I wear a mask.
I need more than a theory if you are going to demand school children <15 wear masks at school when we have data that the school children themselves are at negligible risk.
And I need a lot lot more than a theory if you are going to lock down societies. The costs in education are well defined (starting with a lot of data from a teachers strike in Brazil).

Quite simply I need to know that the risk can be quantified because the cost/annoyance/reduction in basic education can be.

The costs of masks and especially lockdowns are both simple theories backed up with data.


MD: I know you want to talk about masks. I will talk about that next - it's not as simple as lockdowns.

Eric Novik

Michael, I think you are right that you should ask for more than a theory to comply with a policy (not in general, but in this case), but that doesn't contradict what Kaiser is saying. His point is that you don't need data to show that you won't catch an infection if you don't have contact. Compliance, cost of intervention, and so on is an entirely different matter. Here is a less controversial example: I will guarantee you, without any data, that if you adopt the following diet, you will lose weight: drink three glasses of water per day and have one banana per day for one month. If you run a trial for this intervention, you'd better analyze it using the Intention To Treat (ITT) population, i.e., good luck with compliance.

David Norman

And this shows 1) the problem with using the word theory when something is well established (like germ "theory") and 2) that many people (like commenters here) don't understand what is meant by "theory"

Ron Kenett

The point here is the ability to generalize claims. Generalization is done by statistical inference, from sample to population, by mechanistic models and, yes, "intuition". In the later case no data is needed. See https://www.youtube.com/watch?v=ADs7fWIvuVk

Antonio Rinaldi

Another terrific post.

Just a few side notes.

We have a theory stating that lockdown and masking and vaccination work together to reduce spread: the swiss cheese model: https://en.wikipedia.org/wiki/Swiss_cheese_model; so it is our own theory that makes a measurement of each factor difficult|impossibile.

Unless something like an RCT emerges... like in the UK due to a /random/ accident: https://warwick.ac.uk/fac/soc/economics/research/centres/cage/publications/workingpapers/2020/does_contact_tracing_work_quasi_experimental_evidence_from_an_excel_error_in_england/

Theory - or structural model - in Bayes formula: isn't it represented by the likelihood function form?

The comments to this entry are closed.

Get new posts by email:
Kaiser Fung. Business analytics and data visualization expert. Author and Speaker.
Visit my website. Follow my Twitter. See my articles at Daily Beast, 538, HBR, Wired.

See my Youtube and Flickr.


  • only in Big Data
Numbers Rule Your World:
Amazon - Barnes&Noble

Amazon - Barnes&Noble

Junk Charts Blog

Link to junkcharts

Graphics design by Amanda Lee

Next Events

Jan: 10 NYPL Data Science Careers Talk, New York, NY

Past Events

Aug: 15 NYPL Analytics Resume Review Workshop, New York, NY

Apr: 2 Data Visualization Seminar, Pasadena, CA

Mar: 30 ASA DataFest, New York, NY

See more here

Principal Analytics Prep

Link to Principal Analytics Prep