1
$\begingroup$

I'm trying to simulate a person moving through a household using a Markov chain. Each state would be a room in the house. The issue I'm running into is that I have no existing data telling me what a typical person's activity pattern is (goal is to generate completely synthetic data), and so I have no starting point to compute state transition probabilities. How do I handle this? Is it sound to randomly assign these probabilities?

I found a paper (http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.212.2548&rep=rep1&type=pdf) that I believe describes something similar. But I don't understand where they get the values in the diagrams they present in part 4.

Working with R.

$\endgroup$
  • 5
    $\begingroup$ In a nutshell, you appear to be asking "is it ok if I just make up all my data?" I suppose that depends on what the simulation is for. If it's an educational exercise, then go and have fun. But if you hope to derive any information from it about activity patterns, then this looks like the archetype of GIGO. $\endgroup$ – whuber Jun 24 '16 at 20:41
  • 2
    $\begingroup$ Agree w/ whuber. Just want to add that, if you're going to make up the the transition probabilities, then you at least have to respect the constraints of the physical system. For example, it shouldn't be possible to teleport between rooms, meaning that transition probabilities can only be nonzero between adjacent rooms. You also can't get 'stuck' in a room (hopefully!) so the Markov chain shouldn't contain any absorbing states. $\endgroup$ – user20160 Jun 24 '16 at 23:34
0
$\begingroup$

Partially answered in comments:

In a nutshell, you appear to be asking "is it ok if I just make up all my data?" I suppose that depends on what the simulation is for. If it's an educational exercise, then go and have fun. But if you hope to derive any information from it about activity patterns, then this looks like the archetype of https://en.wikipedia.org/wiki/Garbage_in,_garbage_out. – whuber

Agree w/ whuber. Just want to add that, if you're going to make up the the transition probabilities, then you at least have to respect the constraints of the physical system. For example, it shouldn't be possible to teleport between rooms, meaning that transition probabilities can only be nonzero between adjacent rooms. You also can't get 'stuck' in a room (hopefully!) so the Markov chain shouldn't contain any absorbing states. – user20160

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.