COVID-19 vaccine hesitancy has previously been modelled using data on intentions – expressed prior to vaccine availability. Once vaccines became widely available, it became possible to model hesitancy using actual vaccination uptake data. This paper estimates the determinants of the joint distribution of COVID-19 vaccination intentions (declared before the release of any vaccine) and actual vaccination take-up (when it was widely available across the age distribution). We use high quality longitudinal data (UK Household Longitudinal Study) collected during the pandemic in the UK, merged to a wide variety of individual characteristics collected prior to the COVID-19 pandemic. Our estimation draws on pre-Covid values of variables for a sample that includes 10,073 observations from the September 2021 wave. The contribution of this paper is to model hesitancy and uptake jointly. The work shows that people who might be regarded as marginalised in society (measured, before the pandemic began) are less likely to say that they intend to be vaccinated and they go on to also be more likely to actually remain unvaccinated. It also shows that there is a large positive correlation between the unobservable determinants of intention and of uptake. This high positive correlation has an important implication - that information campaigns can be reasonably well profiled to target specific groups on the basis of intention data alone. We also show that changing one's mind is not correlated with observable data. This is consistent with two explanations. Firstly, the new information available on the arrival of vaccines, that they are safe and effective, may be more optimistic than was originally assumed. Secondly, individuals may have been more pessimistic about the effects associated with infection before vaccines became available.