An adequate theory of happiness or subjective well-being (SWB) needs to link at least three sets of variables: stable person characteristics (including personality traits), life events and measures of well-being (life satisfaction, positive affects) and ill-being (anxiety, depression, negative affects). It also needs to be based on long-term data in order to account for long-term change in SWB. By including personality measures in the 2005 survey, the German Socio-Economic Panel (SOEP) became the first available dataset to provide long-term evidence about personality and change in one key measure of SWB, namely life satisfaction. Using these data, the paper suggests major revisions to the set-point theory of SWB; revisions which seek to account for long-term change. Previously, theory focused on evidence that individuals have their own set-point of SWB and revert to that set-point once the psychological impact of major life events has dissipated. But the new SOEP panel data show that significant minorities record substantial and apparently permanent upward or downward changes in life satisfaction. The paper aims to explain why most people's SWB levels do not change, but why a minority do. The main new result, which must be regarded as tentative until replicated, is that the people most likely to record large changes in life satisfaction are those who score high on the personality traits of extraversion (E) and/or neuroticism (N). These people in a sense "roll the dice" more often than others and so have a higher than average probability of recording long-term changes. Data come from the 3130 SOEP respondents who rated their life satisfaction every year from 1985 onwards, among whom 2843 also completed a set of questions about their personality in 2005.