Data on abundance, survival and productivity were collated from a variety of sources. Regular or sporadic counts were available from all sites, based on whole colony or plot counts. Productivity was available from all four SPAs for kittiwakes, and for European shags at two SPAs, otherwise data on demographic rates was limited to the Isle of May in the Forth Islands SPA. All models were fitted using a Bayesian approach in the software R/WinBUGS. Model fitting was in ‘state-space’ form, which allows for ‘observation error’ and environmental stochasticity simultaneously within the same model. Models forecasted the population size for each species at each SPA, for 25 years from 2016 to 2041. Adult survival was set to decline by one of a range of specified rates equating to offshore renewable effects, namely 0% (i.e. no change), 0.5%, 1%, 2% and 3%. Annual productivity was set to decline by 0%, 1%, 2%, 3% and 5%. Previous work has indicated that ratio PVA metrics are less sensitive than probabilistic PVA metrics. Accordingly, we tested the sensitivity of six PVA metrics, comprising two ratio metrics (median of the ratio of impacted to un-impacted annual growth rate; median of the ratio of impacted to un-impacted population size); two metrics related to the ratio metrics (median difference in impacted and un-impacted annual growth rates; median difference between impacted and un-impacted population size) and two probabilistic metrics (probability of a population decline exceeding 10%, 25% or 50%; centile for un-impacted population which matches the 50th centile for the impacted population). Sensitivity of the six PVA metrics was assessed in relation to mis-specification of input parameters. We considered adult mortality (the complement of survival, since survival is high in seabirds and % increases are limited by the constraint of lying below a survival rate of 1) and productivity to differ from those of the baseline by: -30%, -20%, -10%, 10%, 20% and 30%. We then assessed sensitivities in relation to population status, combining data from all species/SPAs for which we achieved model convergence. Finally, we assessed PVA sensitivities in relation to scenarios of change resulting from the renewables development (i.e. the effect size). The state-space modelling approach proved extremely powerful in forecasting population sizes, in particular where censuses were regular. Even in cases where censuses were sporadic, the models generally performed well, though for three species/SPA populations the models would not converge successfully. The two ratio metrics were least sensitive to mis-specification in input parameters. They performed well in populations of different status, and under different scenarios of change. The two difference metrics were not readily interpretable, but proved useful when growth rates or population size estimates were small. The probabilistic metrics were more sensitive to mis-specification to input parameters than the ratio PVA metrics. The ‘probability of a population decline’ metric has been widely used in assessments but proved highly sensitive to mis-specification. The metric representing the centile from the un-impacted population size equal to the 50th centile of the impacted population size at the end of the wind farm showed moderately low sensitivity to mis-specification of survival and productivity. It performed considerably better than the other probabilistic metric with markedly lower sensitivity to mis-specification, population status and renewables effect size. However, it was more sensitive than ratio metrics, and in some cases showed unstable sensitivity which was less apparent in ratio PVA metrics.