. Each gene at the locus j is represented as a real
valued variable 
. Mutation occurs by adding a 
distributed
random variable 
to the current value of the gene:
. The pairwise
recombination rate is 0.5. The phenotype is represented as a Euclidian vector of two quantitative characters 
and
. The genotypic values of the characters are calculated from the
genic values using a linear mapping function (Wagner, 1989):
. As in the deterministic models, we
assume symmetrical effects on both characters: 
and
if j is odd and 
if j is even. In all
simulations the number of genes is 50, and the average effect of the
mutations is adjusted to keep the muational variance for both characters
at 0.005 (Lynch, 1988). The adjustment of the mutational effect is done by scaling the b-coefficients. The phenotypic values are calculated by
adding to the genotypic values a 
distributed random variable e,
the environmental effect. The fitness is calculated according to Eq. (2).
Selection occurs by viability selection. The survival probability of an
individual with a given phenotype is determined by dividing its fitness
value by 
where 
is the phenotypic standard deviation of the first character. This rescaling
is done to obtain values between 0 and 1, as required for a probability.
For each randomly chosen pair, one offspring is produced, and then another
pair is chosen until enough survivors have been produced for staffing the
next parental generation.
Each simulation is initialized with a population that has reached a quasi equilibrium under stabilizing selection. The population then evolves in the corridor for 1000 generations. Subsequently, the simulation run lasts 1500 generations and statistics are taken after every 100th generation to decrease autocorrelation between observations. Each parameter combination is tried in 50 simulations. The statistical estimates are thus based on 750 data points each. Statistics are taken from offspring prior to selection. The effective population size is estimated from the variance of family size by the formula
where 
is the number of parents and 
is the
variance in family size (see Falconer, 1980). The selection intensity is
measured by the standarized selection differential.