We spent a while talking about different models for this process that distinguish preference and availability. In other words, as a demographic group becomes more or less prevalent due to baby booms or differential mortality or other factors, how should that affect the number of relationships that involve that group. There are a variety of answers. One is that the number of partnerships stays the same because the contact rate (number of partners per individual in the group) goes up or down. Fewer young black men means they have more partners because the number of women who want young black male partners hasn't changed, so they have to start sharing. Another possibility is that the contact rate stays the same but the number of partnerships goes down. Fewer young black males means fewer partnerships that involve young black males, and a greater number of young black women with no partners. There are a range of possibilities in between but we didn't talk much about the details of them. I think these are the sorts of log-linear models which give a mixing parameter based on the odds ratio between the observed number of partnerships and the number expected under random mixing with any given distribution of the population.
The most curious thing about the Intellectual Ventures model is the way it handled concurrency. There were three yes/no characteristics that each person in the model has a probability to have, one for each relationship type, that indicated whether you would still join the queue for a new relationship if you were already in a relationship of that type. In other words, some people will pursue a new relationship if they are in a transitory relationship, but not if they are in an informal or marriage relationship. Others might pursue a relationship if they are in a marriage or transitory, but not in an informal. There was also a limit on the max number of concurrent relationships (I don't remember what it was but it was fairly high, at least 5 I think). I never got clarification on this, but if I understand it correctly, it does lead to potentially strange situations if one tries to interpret these characteristics and their proportions rather than just regarding them as parameters in the model. For example, it is not true that someone who is transitory-concurrent but not informal-concurrent or marriage-concurrent is just someone willing to play the field for a while before settling down - they can be in a transitory relationship and start a marriage while that transitory is still on-going! Now, because transitories are short, this state of affairs probably doesn't last long and then being married they can't start new transitory relationships. (This example actually is probably a pretty common occurrence in the real world if we consider that "marriage" here doesn't mean a legal marriage, just one that has the duration, sex frequency, exclusion, etc of a marriage and furthermore that any relationship that eventually becomes a marriage is classified as such from the beginning - though this does point out a potential problem. Just because you eventually get married to someone and don't use condoms doesn't mean that condom use is low at the beginning of that relationship.)
During the presentation, Martina commented that these factors were created "at birth" like some kind of genetic defect. (I may take her to task on this.) But in the networks meeting, I pointed out that this isn't really any different than our practice of "backdating" marriages as described above - determining immediately on formation whether a relationship is classified as a marriage or non-marriage rather than modelling a process whereby non-marriage relationships transform into marriage relationships. And as the only distinction our models made for this distinction was in relationship duration, this completely doesn't matter. I think it may be also be true for their concurrency variables, as well. I also pointed out the similarity to the long-term partnership model Steve has sometimes used where for each person 10% of the population is randomly pre-specified as soulmates and whenever you form a relationship with a soulmate, it is a long-term relationship (more properly, a less-likely-to-dissolve-each-day relationship) and whenever you form a relationship with a person who isn't a soulmate, that's a short term relationship. Steve and I joke about how ridiculous this model is and yet it seems to be the way that so many people actually believe the world works. I'm not sure whether soulmate-ness is necessarily reciprocal in these models or if a relationship is LTR if either one is a soulmate for the other. I'd guess it is reciprocal and just formed with a static ERGM before running the real model.
Martina suggested that it was problematic to assume these concurrency-possible traits were stable over the life span and showed an interesting plot that I haven't previously seen. The figure had a row for each race and age categories along the x axis. Each age-race combination had a dot-and-whisker plot of the rate of concurrency in that subgroup. The interesting thing is that in whites, there is very high concurrency in young adulthood, but that drops quickly and then there is very, very little after that. In blacks, the rate dropped much more slowly and remained noticeable throughout the life span. There was another group (unsure if it was Hispanic or Asian) that seemed like a mix of the white and black patterns - started moderately high dropped slowly but eventually reached near zero.
I pointed out that this is consistent with the IV model so long as the concurrency types have race-specific probabilities. It could be that whites have high rates of transitory-concurrents, and because transitory relationships are rare beyond a certain age, so is concurrency in whites. Blacks on the other hand are more likely to be marriage-concurrent, thus preserving measurable concurrency rates later in life. It's an interesting model and I'll have to think more on it.