>> Peer-Olaf

Coming at your question from my AI and "long-time" perspective, I'm inclined to say that agent-based simulations are a SUBSET of micro-simulations in which the individuals are designed to have some slight degree of cognitive ability (eg, in your words, have a memory; capable of making decisions based on their own judgement; capable of learning). Of course this makes the simulations more complex to work with but arguably more powerful.

But I could believe that others on RG will give a quite different answer. 

And all a bit tribal, I feel...

Best

Jim

Hi there, this answer is exactly one paragraph of a text I am just writing (a chapter in a policy-based ABM). What I have so far:

According to Gilbert & Troitzsch, 2005, p.8 in Hamill & Gilbert, p. 4

"Microsimulation takes a set of data about a population – of people, households or firms – and applies rules to reflect changes, enabling the modeller to look at the overall impact. Such an approach is particularly useful for modelling policy changes, for example, to see who is made better or worse off by tax changes. However, although allowing for heterogeneity, microsimulation does not allow interaction. Only with the arrival of ABM did modelling interaction between agents become possible".

Crooks & Heppenstall, 2012, p. 92 argue that:

"However, crucially in contrast to ABM, MSM only models one-direction interactions: the impact of the policy on the individuals, but not the impact of individuals on the policy and interactions between individuals are not simulated. Furthermore MSM models do not have the behavioural modelling capability of ABM".

More importantly, I think, is the difference between ABM and Systems Dynamics (which for me are greater). And I was upset to find a text (by a famous author) that uses deterministic differential equations (one direction only) but calls it a ABM...

I hope it helps.

Best,

Hi. agent-based model (ABM) and microsimulation (MSM) look similar nowadays. In some fields, these two terms are not really separable. Actually, they were coming from different domain. MSM was from operations research while ABM was from computer science basically. However, if trace back the source of ABM, you will find multiple origins. In economics and other social science, there are fruitful models fitted the definition of ABM but earlier than the general simulation model. In terms of of implementation, ABM depends on object oriented programming, MSM depends on discrete event simulation, which are not necessarily OOP. The others just as Bernardo Alves said. 

For me, ABM is using OOP to tell specific stories.

Best,

ABS can be for heterogenous, while MS is ordinarily for homogenous

Traditional MSMs do not model communication among their micro-entities (except that they perhaps marry). They are indeed always data driven (starting with census or large sample data) whereas ABMs are most often concept driven and only very rarely start from empirical data. MSMs model the outcome of individual decisions, but its micro-entities never get to know what this outcome is like, hence they cannot take into account which effects their micro-behaviour has on the macro level --- whereas this is at the core of most ABMs.

Traditional MSMs do not model communication among their micro-entities (except that they perhaps marry). They are indeed always data driven (starting with census or large sample data) whereas ABMs are most often concept driven and only very rarely start from empirical data. MSMs model the outcome of individual decisions, but its micro-entities never get to know what this outcome is like, hence they cannot take into account which effects their micro-behaviour has on the macro level --- whereas this is at the core of most ABMs.

Agent based simulation is one of the micro scale models that help to model the behavior and interaction of different agents. This helps to forecast possible different scenarios and their effects before executed or applied in the real world.  

I largely agree with what Chu Chang Ku and Klaus G Troitzsch have said. The boundary between ABMs and Micro Simulations are indeed blurred; and they overlap a lot. Some empirical ABMs can be also called MS models; Some MS models can be also called ABMs. That said, the two are different in many perspectives.

>>> Are ABS models typically theory driven while MS models are typically data driven?

Yes, I would say so. Agents and behaviors in ABS could be very artificial and hypothetical; In MS, agents are usually parametrized from real data and/or the distributions.

>>> Do ABS models contain intelligent and non intelligent objects (have a memory; capable of making decisions based on their own judgement; capable of learning) while MS models only contain non intelligent objects (making decisions based on distributions derive from historic data)? What is the simulation execution algorithm they follow - do they differ?

This is hard to say. Agents in theoretical ABMs (e.g., Game of Life, Schelling’s Segregation model etc.) can be very simple and “non-intelligent”. Most of the agents don’t have the capability of learning in the current literature. In MS models, agents tend to follow certain fix rules (e.g., avoid collision, marry, migrate, search jobs etc.) and be responsive. To some extent, they are “non-intelligent”.

In addition, I also agree with Prof. Troitzsch: while interactions and feedback between micro & macro levels are the core of ABMs, they are not that critical in MS.

Another differences between the two: number of ABMs tends to be in medium size range, dozens to hundreds (maybe up to thousands); the number of MS is usually larger, up to a few millions or even larger (e.g., PENSIM model).  

This reference could be helpful for you:

Bae JW et al(2016).Combining Microsimulation and Agent-based Model for Micro-level Population Dynamics. Procedia Computer Science:https://ac.els-cdn.com/S1877050916307359/1-s2.0-S1877050916307359-main.pdf?_tid=26ecbb30-ae0a-11e7-8ba6-00000aab0f6b&acdnat=1507674608_a70ed273c6e492f37f5d792af9a73db2

I agree with Zhanli Sun and Klaus Troitzsch.

I would want to add the following reference, which contains a triangle of Microsimulation, ABM and Cellular Automata.

Williamson, P., 2007. The role of the International Journal of Microsimulation. International Journal of Microsimulation 1 (1), 1-2.

I don't fully agree with every aspect of the characterization from an ABM point of view, but found it interesting to see how the Microsimulation community viewed their relationship to ABM back in 2007. It certainly shows how both belong into a continuous spectrum of micro models.

Personally, despite the more specific definitions, I often use the term ABM to refer to simulations where the entities are individual entities (persons, firms, ...) that are interacting. If they are not interacting, I would call it microsimulation. In my experience, readers and reviewers will expect interaction if you call it an ABM. But this may be discipline-specific.

Article Combining Microsimulation and Agent-Based Model for Micro-Le...

https://ac.els-cdn.com/S1877050916307359/1-s2.0-S1877050916307359-main.pdf?_tid=e3f74b3c-c3bd-11e7-8caf-00000aab0f6c&acdnat=1510060780_172835e28efc267478fe2abe89c9a5f8

Dear Prof. Siebers,

I have worked in this for quite sometime and I am attaching a couple of recent papers which might answer your question in a somewhat novel method. Please see.

Soumitra K. Mallick

for RHMHM School

Article Multi-site modeling and prediction of annual and monthly pre...

@Zhanli Sun argues that ABMs are theory driven and microsimulations are data-driven. This is simply put wrong. While microsimulations are indeed always data-driven, as the agent-behaviour is always probabilistic and it makes little sense to endow them with random probabilistic behavior. Agent-based models on the other hand come as uncalibrated theoretical models as will as in calibrated thus data-driven models.

I think

Davoud Taghawi-Nejad

I think, ABMs focus on behavioral charecter of agent, and MS is based on output data of behavior .

ABMs are not restricted to "non intelligent" entities. Machine-learning can (and has) been integrated into the individual agents within models. The main limitations in doing so are the computational capabilities of the hardware the model is run on. Shameless plug...

Creating Intelligent Agents: Combining Agent-Based Modeling with Machine Learning. D Brearcliffe, A Crooks - 2020 Conference Paper Creating Intelligent Agents: Combining Agent-Based Modeling ...