I would like to know if there is an automated way to transform probabilities (segregated into age cohorts) into a synthetic population of agents with the same statistical traits. This is: transforming data like “45% males, 55% females, 20% females smoke, 35% males between 15 – 30 years old have diabetes, etc" Into a population with, let’s say, 100 agents where 45 are males, 55 are females, 11 of these smoke, etc.
You need ,llink) equation) age sex . Strata parameters gender by age
Or use population sim like nemo or simupop where you set the intial probabilities and evolve the populations any way you like.
If these characteristics are independent why don't you just use these probabilities directly to initialize your agents; for example, using a uniformly distributed random number in [0,1], if the number
You need one of knowledge representation methods I recommend propegate logic
Dear Marti,
Do you mean making conclusions regarding the underlying population based on sampled data?
Dear all,
thank you very much for all you answers! I write a common "thank you" since I do not know how to add comments / thanks to everyone. I guess the answer from Richard is the most appropiate, though I have to go over the sofware. And we cannot do what Reiner suggests since some characterstics are not independent. And YES, try to make conclusions regarding the underlying population BUT not based on sampled data but synthetically generated data.
Why do you not simulate 100 random variables with these characteristics? In "R" you have the option "sample"
I think Gheorghita is moving in teh right direction here. Assuming you have an agent model with some kind of behavioral parameters (such as probabilities of changing state), you want to choose values of those parameters so that the average behavior of your population matches your population statistics measurements.
If that's what you're looking for, what you're doing is systems identification, a part of control theory (see electrical engineering). And what Gheorghita is describing is part of a genetic algorithm, where you randomly seed parameters, simulate, evaluate fitness of each result in some way, then cross-breed the "best" parameter sets to start anew. Repeat until you've identified the best / most consistent parameter values.
But as I read your data, it seems like you need a bit more. Do you have an age distribution as well? Do you have any hypotheses or a model for individuals getting diabetes, as a function of smoking and age? (If you do but just don't have parameter values, the method above coudl be used to get them, and the parameter values would become the scientific output / prediction to try to validate.)
Not sure if this helps ... best of luck -- Paul
Hi Paul,
many thanks. Yes, we need a bit more. We have an age distribution, but we do not have a model! :(
Ah, I see. Sorry, I misunderstood the synthetic agents to mean an agent-based model.
Maybe you will find interesting our publication https://www.researchgate.net/publication/49830883_Extinction_threshold_for_spatial_forest_dynamics_with_height_structure
We classify a tree population in a lattice in terms of (occupied by) mature tree, (occupied by) immature tree or empty site.
Article Extinction threshold for spatial forest dynamics with height structure
Hi Marti,
If you are to use a Monte Carlo/stochastic type sampling system you can produce a valid national level micro-population from aggregated data reasonably easily, dependent on the data you have available to you. Here is New Zealand we have quite good national level census' and based on that it is possible to great a representative population. I have had a go at this for a New Zealand population in relation to CVD characteristics using a combination of Monte Carlo and multiple imputation in R.
I would be happy to send you some R scripts for what I have done regarding population generation if you feel it would be useful, if so please feel free to send me an email on [email protected] . As it is quite source data specific so it would not be a 'drop in' type replacement but you may find it useful.
Regards,
Josh
Great Josh! Thank you very much! If you could be so kind to send me the R scripts, I'll have a look at them and see how they can be translated to our case and data! Thank you very much!
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis