JiscMail Logo
Email discussion lists for the UK Education and Research communities

Help for SIMSOC Archives


SIMSOC Archives

SIMSOC Archives


SIMSOC@JISCMAIL.AC.UK


View:

Message:

[

First

|

Previous

|

Next

|

Last

]

By Topic:

[

First

|

Previous

|

Next

|

Last

]

By Author:

[

First

|

Previous

|

Next

|

Last

]

Font:

Proportional Font

LISTSERV Archives

LISTSERV Archives

SIMSOC Home

SIMSOC Home

SIMSOC  September 2019

SIMSOC September 2019

Options

Subscribe or Unsubscribe

Subscribe or Unsubscribe

Log In

Log In

Get Password

Get Password

Subject:

Re: Identifiability analysis for agent-based models

From:

Ernesto Carrella <[log in to unmask]>

Reply-To:

Ernesto Carrella <[log in to unmask]>

Date:

Thu, 5 Sep 2019 16:12:33 +0100

Content-Type:

text/plain

Parts/Attachments:

Parts/Attachments

text/plain (25 lines)

I'd preface this by saying that identification (retrieve model parameters from data) is a hard problem for all complicated models. There is a famous survey on identification in macro-economics (strictly non-ABM): Canova and Sala (2009; 10.1016/j.jmoneco.2009.03.014) that ends like this:
"Liu (1960) and, twenty years later, Sims (1980) have argued that traditional models of simultaneous equations were hopelessly under-identified and that identification of an economic structure was often achieved not because there was sufficient information but because researchers wanted it to be so [...] Still, it appears that a large class of popular log-linearized DSGE structures is close to being under-identified; observational equivalence is widespread; and reasonable estimates are obtained not because the data is informative but because of a-priori restrictions"
Speaking of grim!
That's actually a great paper as a survey on the problem of under-identification more broadly.

ABM literature often ignores identification issues which is a problem, in my opinion, when you are building heavily data-driven models. 
Partially i think is that, without closed form solutions, the only real way to figure out if the model is under-identified is by cross-validation which is quite expensive and even then you can only test whether you have underidentification for the specific combination of data you have, model you have and estimation method you are using (which is a much weaker result than proving underidentification in a model for all possible data and estimation algorithms).

Some techniques make it relatively easy to see under-identification. If you are using an algorithm that produces full posteriors on the parameters to estimate then a flat posterior is an indication of underidentification.
For example, the famous JASSS paper by Thiele on estimation and Rnetlogo (http://jasss.soc.surrey.ac.uk/17/3/11.html) uses a bird model where one parameter (scout.prob) is actually not identifiable and you can tell, for example in figure 7, by looking at the very spread out posterior on scout.prob.
Unfortunately the authors do not mention this.

A full Approximate Bayesian Computation (ABC) study on an agent-based model that does deal with underidentification is Vaart et al (2015; https://doi.org/10.1016/j.ecolmodel.2015.05.020) which is a biological model of earthworms; they use a ABC specific way to test underidentification and they show that some parameters are well identified but not all.

In general you can always use cross-validation to check if you can estimate parameters successfully: use one model run as "real data" and try to retrieve its parameters, do this multiple times and see the difference between the estimated parameters versus just using the "mean" parameter. This is "predictivity". I have two examples of this in my work. 
One of them deals again with the bird model from Thiele, but I show that using more data fixes identification (https://arxiv.org/abs/1807.01579  ).
The other I am going to present at SSC2019 later this month, is here: http://carrknight.github.io/assets/nofreelunch.html 
I take 20 simulations, some are ABMs, and 9 estimation algorithms and check which estimates better by checking their predictivity and examine identification failures.
If you have some other ABMs you want to test under this framework, I'd love to collaborate!

########################################################################

To unsubscribe from the SIMSOC list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=SIMSOC&A=1

Top of Message | Previous Page | Permalink

JiscMail Tools


RSS Feeds and Sharing


Advanced Options


Archives

November 2020
October 2020
September 2020
August 2020
July 2020
June 2020
May 2020
April 2020
March 2020
February 2020
January 2020
December 2019
November 2019
October 2019
September 2019
August 2019
July 2019
June 2019
May 2019
April 2019
March 2019
February 2019
January 2019
December 2018
November 2018
October 2018
September 2018
August 2018
July 2018
June 2018
May 2018
April 2018
March 2018
February 2018
January 2018
December 2017
November 2017
October 2017
September 2017
August 2017
July 2017
June 2017
May 2017
April 2017
March 2017
February 2017
January 2017
December 2016
November 2016
October 2016
September 2016
August 2016
July 2016
June 2016
May 2016
April 2016
March 2016
February 2016
January 2016
December 2015
November 2015
October 2015
September 2015
August 2015
July 2015
June 2015
May 2015
April 2015
March 2015
February 2015
January 2015
December 2014
November 2014
October 2014
September 2014
August 2014
July 2014
June 2014
May 2014
April 2014
March 2014
February 2014
January 2014
December 2013
November 2013
October 2013
September 2013
August 2013
July 2013
June 2013
May 2013
April 2013
March 2013
February 2013
January 2013
December 2012
November 2012
October 2012
September 2012
August 2012
July 2012
June 2012
May 2012
April 2012
March 2012
February 2012
January 2012
December 2011
November 2011
October 2011
September 2011
August 2011
July 2011
June 2011
May 2011
April 2011
March 2011
February 2011
January 2011
December 2010
November 2010
October 2010
September 2010
August 2010
July 2010
June 2010
May 2010
April 2010
March 2010
February 2010
January 2010
December 2009
November 2009
October 2009
September 2009
August 2009
July 2009
June 2009
May 2009
April 2009
March 2009
February 2009
January 2009
December 2008
November 2008
October 2008
September 2008
August 2008
July 2008
June 2008
May 2008
April 2008
March 2008
February 2008
January 2008
December 2007
November 2007
October 2007
September 2007
August 2007
July 2007
June 2007
May 2007
April 2007
March 2007
February 2007
January 2007
2006
2005
2004
2003
2002
2001
2000
1999
1998


JiscMail is a Jisc service.

View our service policies at https://www.jiscmail.ac.uk/policyandsecurity/ and Jisc's privacy policy at https://www.jisc.ac.uk/website/privacy-notice

For help and support help@jisc.ac.uk

Secured by F-Secure Anti-Virus CataList Email List Search Powered by the LISTSERV Email List Manager