Tuesday, March 03, 2015

Killer app for (Bayesian) regression analysis

"Wow."  That was the first comment today for Jonah Sol Gabry's announcement of shinyStan on Andrew Gelman's Statistical Modeling, Causal Inference, and Social Science.  I've tried it, and that's pretty close to my reaction, too.  Try out the demo here

In case you don't know Stan yet, it's a probabilistic programming language focused on Markov chain Monte Carlo sampling.  You can run Stan from the command line, or you can use its interfaces from R, Python, MATLAB, or Julia.

Check out the excellent language manual to learn more, and then dive in! Kudos to Jonah, Michael Andreae, Yuanjun Gao, and Dongying Song, the developers!

Labels: , , ,

Wednesday, May 14, 2014

Confidence intervals straight from the horse's mouth

Do you know what a confidence interval is?  Are you sure?

As a check, write down a brief definition, and then check out economics professor Dave Giles' Interpreting Confidence Intervals.  Is that what you thought?

While you're there, be sure to check out what Jerzy Neyman, the developer of confidence intervals, told Milton Friedman about confidence intervals.

Labels: ,

Sunday, October 27, 2013

How do you organize your analytical workflow?

If you're in an analyst role some or all of the time, you've no doubt faced the challenge of organizing your workflow.  I've been increasingly organizing my workflow with Org mode.  Rather than explaining that myself, I'll let Professor John Kitchin explain how he uses Org mode to solve three big problems he faces in his work.

While he uses Python in his work, Org mode works quite well with other languages, too, including R, C, Matlab, and others.  You can even use multiple languages to process the same data, if that simplifies your work.

Labels: , , , , ,

Tuesday, October 01, 2013

Want to raise the level of discourse on the ACA?

Want to try to raise the level of discourse on the US Affordable Care Act?  Here's a homework assignment: do the challenge in section 5.5.1 of John Sterman's Business Dynamics.  You should be familiar with chapters 3 and 4 and the first part of chapter 5 to do the work.

You are allowed encouraged to work in groups.

Did you discover anything useful?

Labels: , , , , , , , , ,

Friday, September 06, 2013

Sustainable computing

One aspect of sustainability has to do with not creating more waste than we can manage.  That's why I perked up when I read about sustainability on the Why Open Computing Web site.  They're not simply complaining or exhorting; they're doing something about it.  They're selling laptops intended to be disassembled and repaired.  They're selling laptops and desktops with software intended to be supportable over the long term; their page says "Remain up-to-date for a decade!  Computers that last, at competitive prices.  A radical, different philosophy!"  Later they write, "We're convinced: the era of throw-away products is coming to an end."

They're not the first to offer computers with many of these features, but they are the first I've seen that have coupled the hardware and software features with an obvious intent to focus on sustainability and customer experience.  Check them out.

Thanks to PCtipp.ch for the pointer and a preview of what promises to be a positive review of their laptop offering soon.

Labels: , , ,

Sunday, September 01, 2013

Feynman and skepticism

I've written about skepticism before.  Recently I found a transcript of The Relation of Science and Religion, a talk on the subject written by Richard Feynman. Perhaps it will speak to some of you, as well.

Labels: , , , , ,

Thursday, August 08, 2013

Celebrating beginnings: David Woolley

I've worked online for years, using asynchronous and synchronous means, and I've found that to be quite a productive environment.

I met David Woolley online years ago, thanks to a connection through Nancy White. Among other things, David hosts an excellent Web Conferencing Review reference site for systems we can use to connect at a distance.

It was only by chance that I discovered yesterday the extent to which David contributed to the early days of Web conferencing by reading PLATO Notes Released 40 Years Ago Today.

Thanks, David, for guiding the start of this work and for staying with it!

Labels: ,

Tuesday, August 06, 2013

Regression Analysis Tips

Part of my work involves regression analyses. There are myriad articles and books giving advice on good regression analysis practice, not all of which is consistent. In case you're looking for suggestions, here are three of the sources I find useful, in increasing order of length.

First, Andrew Gelman recently posted What are the key assumptions of linear regression? prioritizing five rules of thumb as to what is important.  Check out the discussion, too.  Hint: in general, don't sweat equal variance or normality of errors.

Second, David Hogg posted a link to his 55-page Data analysis recipes: Fitting a model to data, written with Jo Bovy and Dustin Lang.

Finally, for a more complete reference, see Data Analysis Using Regression and Multilevel/Hierarchical Models by Gelman and Jennifer Hill.

What would you add to that list?

Labels: , , , ,

Sunday, July 21, 2013

It's about time to adopt modern Bayesian data analysis

That's the message of an open letter from John Kruschke to "Editors of journals, Chairs of departments, Directors of funding programs, Directors of graduate training, Reviewers of grants and manuscripts, Researchers, Teachers, and Students" to encourage the adoption of Bayesian data analysis (BDA) methods.  Read it, and give BDA a try.  You might be surprised at how much more you can do with how much less effort.

Labels: , , , , , ,

Tuesday, July 09, 2013

Better Questions

I've written before of the power of questions to guide work.  Now I've discovered an article by Donella Meadows called Resources, Technology, Environment, What Is the Question?  With one exception, I find it a great exploration of how we can go about asking better questions to guide our work.  Read it, and see what you think.

The one exception?  On the first page, she describes a view of science that I used to hold but that I increasingly find less satisfying.  I'm increasingly sympathetic to ideas Andrew Gelman and Cosma Shalizi express.

If you found that article interesting, here are two more free resources.  First, if you want to read more of her work, you can view The Donella Meadows Archive: Voice of a Global Citizen, a collection of 15 years of very readable Global Citizen essays.  Second, you can also see the archive for Dynamica, the first major system dynamics journal, where I found Meadows' article.

Labels: , , , , ,

Tuesday, June 25, 2013

The Climate Leader from Climate Interactive

I received an email from Climate Interactive today announcing a series of online courses starting this fall "to share some time-tested insights into navigating the complex world of taking action on climate."  To learn more, see their announcement, and sign up for future updates.

Labels: , , , , , , ,

Sunday, December 02, 2012

Geoff Coyle

Geoff Coyle passed away November 19.  As Tom Fiddaman described on his blog, Geoff was a pioneer in system dynamics and certainly seemed to qualify for the term irrascible, as Tom noted, for he had his own views and a clear way of stating them.

He was very willing to share his insights.  He and I met on the old system dynamics mailing list. Through private emails, we discovered that we shared the belief that the engineering approach calls for varying levels of models, ranging from rules of thumb through the simple, highly-aggregated qualitative models all the way, in the appropriate cases, to carefully done simulation models.  It's not the case that every problem demands detailed simulations, nor is it the case that a simple causal loop diagram (or influence diagram, as he preferred) always suffices.  It was the same when I was an analog circuit designer: sometimes I would design a circuit based on simple rules of thumb, while other circuits called out for detailed analysis, and that had been his experience in his world, as well.

I learned a lot from Geoff: the rigorous use of influence diagrams, the use of QPID, the Analytic Hierarchy Process for decision making, and improved modeling skills at all levels of rigor.  I read his Management Systems Dynamics, and I read and taught from his System Dynamics Modelling: A Practical Approach.

In recent years, we wrote less, for he was more focused on his latest book on mining, but I'm glad to have known him.  I invite anyone who is interested in system dynamics to check out Tom's list of Geoff's writings. I certainly learned from them; perhaps you will, too.

Thank you, Geoff, for sharing your insights and engaging in dialog.

Labels: , , , ,

Friday, November 02, 2012

More Exploratory Data Analysis

I missed two items in my last posting on EDA.  First, there's another tool I use: J.  Over time, I seem to oscillate between R and J.  J, at least when I don't let myself get rusty in expressing ideas in J, is a powerful and concise way of thinking.  R has an enormous library collection.  I won't advocate for one over the other here, but you should try both.

Second, some tend to think of EDA as model-free statistics, but that's not quite right.  To get a better explanation of what that means, see Andrew Gelman's A Bayesian Formulation of Exploratory Data Analysis and Goodness-of-fit Testing.I'd go past what he wrote and note that using models in EDA extends to work in system dynamics.  To make sense of a dynamic (time-varying) situation, often trying to craft a model that approximates the situation is a great way to get started in making sense of the situation.

Labels: , , , ,

Saturday, October 27, 2012

Exploratory Data Analysis

I've used exploratory data analysis (EDA) in my work and in my thinking about situations ever since discovering it years ago.  I've used paper and pencil, XLISP-STAT, R, ggobi, and other tools.  I've read a number of the books.

When suggesting EDA to others, I've been puzzled: do I recommend they read one of the books, which is likely more than they want to start?  Do I tell them about particular techniques, which miss the flavor and contribution of EDA?

A year or two ago, I discovered John Behrens' Principles and Procedures of Exploratory Data Analysis, which seems to contain principles and some of the common procedures. I'm noting this here, both for you to find and for me to re-find more easily in the future.

In searching for this article, I ran across Chong-ho Yu's site on Exploratory data analysis and Data visualization, which seems to provide a good introduction to the ideas of EDA as expressed by Behrens.  That doesn't seem surprising, as Yu seems to have studied under Behrens at ASU.

Labels: , , ,