Migration as an adaptive strategy: application to the US-Mexico Corridor

Presenter: Andrew Bell

Part of the Session on Migration

Model source code: https://github.com/andrew-nyu/migration

1 Like

Thanks for the presentation. MIDAS seems indeed to have a lot of potential. It was helpful that you provided an ODD in the github so I get a better idea how the decisions are modeled. It seems that the basic model is a kind of social foraging model. Agents move to locations that gives them the highest utility and they exchange information with other agents in their social network. Do all agents derive the same utility from a specific location? Or do individual agents differ in preferences and skills? If there are no seasonal changes in the data will all agents end up at the same location? (I do not see a feedback from number of agents in a location on wage for example, which might not be needed for the current version with a small number of agents).

Thank you for the presentation. Interesting and highly relevant model. Very impressive set-up. I was wondering if the risk preference is strongly related to gender and age, as often young men are migrating. Interesting to explore deeper what the drivers are. Could you obtain data on network structures of individuals based on mobile data? Could be used in developing a social belongingness in the model.

Your remark on building upon each other’s work is a good one. Here’s a short note on our migration model that is focussing on the acculturation process once people have migrated. https://www.euractiv.com/section/justice-home-affairs/opinion/refugee-crisis-easy-solutions-cause-big-problems/

Cheers

Wander

Hi Marco! Thanks for the questions. I set up the present version model so that all agents in a specific place will derive the same utility from the same layer. The reason to do this was simply so that I could store a single array of utilities (layers * locations * timesteps) and improve the scalability of the model to larger n. However, the next step would be to give agents different coefficients on each layer, which could represent preferences or technical efficiency, so that you have individual variation.

At present, agents differ in their overall risk aversion, meaning that some agents will not mind high costs or losses, and some agents will have higher preferences for smooth incomes. Thus, I would not expect them all to end up in the same places - I would expect the ‘hardier’ agents to remain in riskier environments.

The next stage in research is to try to (cost-effectively) collect preference data, which would then merit further development of the decision model.

As a final note, the decision module itself is fully plug-and-play so that examining the relative importance of different technical efficiencies/preferences and comparing across decision structures will be simple.

Hi! Thanks for the question. I would guess you are right that gender and age will strongly shape the risk preferences. I think the network structure data is really important, as we can also then start to understand the scale at which migration decisions get made (i.e., examine the emergent meta-household across which things are shared and which in turn provide the safety net that enables migration).

In recent proposals we’ve considered the idea of doing snowball sampling of social networks from individual seeds drawn from larger random samples of individuals. Thus, you get some idea of the different structures you might expect to see, and as well can put them in the context of a representative sample. We’ve also been trying to get funding to do android-based data collection over time (rather than sit-down in-person surveys) so that you can capture the interaction structure over time.

Thanks for sending your summary as well - that’s an interesting model and it would be nice to have some overlap and model sharing. I have been thinking of a migration session for iEMS in Colorado, though the deadline is very soon. Had you considered attending that meeting?

And just a second reply on the last point - I set up the utility layer functions to be basic anonymous array functions, again just to try to have an engine that would move quickly even with a large number of agents. In the example I provided I’ve only used existing data for wage rates as the function, but it would be trivial (and within the existing framework) to add a scaling factor like (a + n )/ b where n is the number of agents occupying the layer, so that you get density dependent scaling of available wages as a feedback mechanism.

Interesting model, and just within a period when migration has brought so much controversy now. I really would like to understand the way risk preferences for migration. I think that the exercise that helps to capture shocks with the model is great and it would provide great tools to forecast sensible changes in migration. It probably will help to zoom in a little bit the sensitivity analysis to see the variables taken into consideration. (it is hard to read them)

Thanks for the comment - I’ve posted the slides here:

https://drive.google.com/file/d/0B3uzoIN9blCkdUZpLVRCWkcyTVE/view?usp=sharing

which might make it easier to tear into them.

I agree that migration is an important issue right now. With better data, I think we could use the same framework to capture responses to conflicts, environmental shocks, etc. I also think we could capture outcomes like internal displacement as a preferred path to leaving one’s home country, etc. This is where finding ways to bring big behavioral data into ABM will be really important. I have a review I wrote last year where I really pushed the idea of mobile phones-based data collection: http://www.sciencedirect.com/science/article/pii/S1364815216304078

I really like the framework you’ve created. Have you considered integrating other push factors, such as high crime rates or low social status, into risk preferences? My colleague and I built a model on detecting human trafficking among migration flows and I’d love to use your framework to give more granularity to the migrant agents’ behavior (https://www.openabm.org/model/5801/version/1/view). The structure your provide for economic migrants could be a wonderful basis for risk preferences with regard to trafficking factors such as dubious labor recruiters/transporters, etc. The International Organization for Migration will be releasing data in the coming weeks on human trafficking victims it and a few NGOs have collected worldwide over the years. Perhaps that will provide additional data useful for your purposes. I’d love to explore opportunities for collaboration!

Thanks for your note, apologies for the delay. I don’t get a ‘ping’ to my email so I’m slow.

Absolutely - I’m trying to make the ‘layer’ framework as flexible as possible. We could consider social status by having agent-specific costs of access to different layers, and as well, having agent-specific utility coefficients that capture the inability to derive the same utility from an experience as other more privileged agents.

I’ve downloaded your paper and will give a good read over the coming days. Look forward to talking sometime in the near future!