We analyze paradata from an online pilot of the Generations and Gender Programme (GGP) conducted in three countries (Croatia, Germany & Portugal) to understand the extent, timing and patterns in breakoffs during a long online survey. The GGS is notable as an online survey given that the median length of an interview is 52 minutes and the survey was initially designed for face-to-face. Paradata was collected for 3,378 web interviews. 17% of these interviews breakoff before the questionnaire was completed. The analysis uses Cox regression models to explore the timing of breakoff and the influence of contextual factors. The results indicate that the breakoff hazard does not increase or decrease across the length of the questionnaire. The risk of breakoff does vary considerably across countries, between genders, also by partnership status. Respondents are twice as likely to breakoff on a loop question and respondents completing the survey on a smartphone are 2.6 times as likely to breakoff as those using a tablet or PC. Respondents receiving a conditional incentive were 65% less likely to breakoff than those who didn't. The lessons from this work can help inform future strategies converting existing long, cross-national face to face studies into an online format.