models (Caswell 1989, Silvertown et al. 1993). In most cases, the majority of 

 seeds probably germinate without a dormant period (see Results), and we have 

 used matrices with reproductive transition and recruitment columns combined to 

 calculate X. We calculated separate elasticities for reproductive transitions 

 and recruitment by dividing the reproductive+recruitment elasticities 

 proportionately between their two components. 



We used the ratio of new recruits to survivors to compare annual 

 recruitment rates among populations. Growth was measured as the ratio of 

 plants in each population that grew into a larger size class to those that 

 remained in the same class or became smaller. We examined differences in age 

 at maturity by comparing ratios of plants that flowered during the first two 

 years to those that flowered later. Differences in recruitment, growth, new 

 recruit survival, survival of bolting plants, age at maturity and proportion 

 of bolting plants were assessed with an overall chi-square goodness of fit 

 test. If a 3 X 3 test showed a significant result, I used 2X2 tests to 

 determine which pairs of sites were different. Probability values were not 

 corrected for multiple tests. 



We compared survivorship of the uneven-age sample population present in 

 1989 and the 1990 cohort among the sites using the nonparametric logrank test 

 (Pyke and Thompson 1986, Hutchings et al. 1991). Survivorship curves were 

 constructed following methods outlined in Hutchings et al. (1991). 

 Probability values were not adjusted for multiple tests. 



The effects of site (population), year and bolting on number of fruits 

 per plant and number of seeds per fruit were analyzed using analysis of 

 variance (ANOVA) followed by contrast tests. Dependent variables were log- 

 transformed prior to analysis. The effects of treatment, site and bolting on 

 the arcsine-transformed proportion of germinating seed were also analyzed by 

 ANOVA. 



