st: Panel Data with Truncation and Gaps

[John Simpson <john.simpson@ualberta.ca>]

Dear Stata and statistics experts,

I am looking for a strategy to handle a large amount of panel data  
that features both truncation and gaps.  In particular I would like to  
know how I might go about fitting a model to the data I have on hand.   
Important features of the data are as follows:

1. It is population data generated from agent-based evolutionary  
simulations .  Each trial population has a series of observations  
associated with it over the length of time that it was being run.



2. To conserve memory and processing time two data collection  
shortcuts were used.

2a. Summary statistics from the population were collected on the  
initial creation of the population, after running it for one  
generation, and again after the second generation.  Following this the  
same statistics are collected every five generation until generation  
100 at which point the simulation of the population ends.  If the  
population drops below two members then no more information is  
collected either (There is no single-agent reproduction).

2b. If the population grew over 15000 members then summary statistics  
were collected in the generation in which this occurred and then the  
population was dropped.



3. There are a collection of variables that need to be taken into  
account.

3a. Some of these are fixed throughout the trial (These include things  
like the initial population size, the cost to live from generation to  
generation, and the cost to spawn with another agent).

3b. Others change throughout the course of each simulation and are  
randomly distributed at the beginning (These are the behaviours that  
the agents exhibit under certain conditions.  Over time as  
opportunities to express these behaviours present themselves agents  
with more good/useful beahviours get to spawn more, increasing the  
likelihood that these useful behaviours will become more prevalent in  
the population).


In particular I have two worries.  First, that as successful  
populations are truncated out, those that remain will bring down the  
mean.  Second, that a combination of successful population truncating  
out and unsuccessful populations having few members with highly  
similar behaviour sets will skew any investigation into which  
behaviours are successful.


Any suggestions regarding possible models or methods for handling this  
dataset or directions to possibly useful resources would be appreciated.

John Simpson
Department of Philosophy
University of Alberta, Canada
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/