Integrating a refreshment sample in a longitudinal survey: effects of different weighting methods over four waves