To see the effect of burden on children on suicide rates. (I would expect suicide to become more evolutionarily attractive, obviously.)
Each cycle, parents cause their children to lose some health (basic implementation, obviously)
[2006-05-01 12:35] I've had difficulty getting suicide to evolve here. I've succeeded with latest attempt, but I've had to make the conditions very obviously favourable to suicide. Namely, I've made it so that agents can no longer reproduce after a certain age and they are a fairly heavy burden on their children from that point on. This is quite a far cry from the subtlety of the suicide simulations in my dissertation. Even still, and this is quite perplexing, suicide still doesn't evolve to have a probability of 1 (it's from 0.4-0.8). I would have thought suicide would become overwhelmingly good under these conditions, since the only thing the agent can do if it keeps living is to continually harm its children.
One thing I notice from the decision tree's I'm observing is that agent's settle into a pattern where the rules favourable to suicide simply aren't triggered very often. Of course, given that suicide causes the agent to disappear and thereby no longer trigger any rules, it could just be a side effect of that. That is, all the other rules have to trigger a lot because they are trigger throughout the agent's life, while the suicide rule's will only trigger once (or 2 or 3 times, depending on the suicide probability).
Ok, with the decision tree structure, "1 0.3 ( L 1 0.5 ( L L ) )", suicide evolves to have a probability of 1 after the 'menopausal' age. Now, I'm going to try disabling burden and seeing what happens.
[2006-05-02 17:13] Yesterday, I ran 2 sets of simulations that involved turning kin selection on/off. For both types, there was no burden but there was still a menopausal age. I found that suicide evolved strongly (usually p>0.6, often p=1) in the kin-selection simuation, and sometimes did/did not evolve in the no-kin-selection simulation. I came to think that the latter outcome might have been a result of drift, and a little later I realised that I could control for drift by making suicide pleiotropic :). What I did was this: when an agent suicides, they cause an energy hit to each of their children (equal to 1 piece of food). Therefore, in cases where suicide would otherwise be neutral, it should now be selected against. Of course, this may neutralise any positive effect to suicide as well, but hopefully not . . .
Anyway, I tested that version of suicide, and (after 1 simulation run) found the kin-selection simulation still evolved suicide strongly(p>0.9), and that the no-kin-selection simulation evolved suicide out completely (p=0).
The simulations, however, aren't all that stable (a lot of them go extinct early), so I'd like to produce more stable simulations and see if I can get a statistical summary of what I have. After that, I'd like to explore how robust/general the finding is.