Economic research usually studies cooperation behavior in situations where cooperation in the social optimum is also beneficial to the individual (e.g. public goods games). In many situations however, cooperation is defined by helping a member of a group who -- by no fault of her own -- is in need of support. Helping in these situations is costly for the contributing individuals and may only be beneficial at the group level. Furthermore, the cost of helping is often endogenously determined by how many others also help. The experiment studies such a situation by introducing a novel experimental game in differently-sized groups (groups of eight vs. groups of four). In both group settings, a subject is need of help because of an exogenous shock to her income. Other group members then decide to help by giving up parts of their income. The cost a helper has to pay is decreasing in the total number of other helpers. Helping rates significantly drop in group size. Moreover, the experiment also studies helping behavior in situations when groups are combined. Often individuals cooperate in groups but group sizes change over time (e.g. teams are combined, firms merge, etc.). Group norms persist over time when two small groups of four subjects are randomly combined to a big group of eight subjects in the second part of the experiment. Subjects in combined groups help more than subjects in big groups. In particular, high solidarity groups continue to help while low solidarity groups withhold cooperation from old group members. However, they are more prone to help subjects in need that have been in another group before.