Many embedded star clusters do not evolve into long-lived bound clusters. The
most popular explanation for this "infant mortality" of young (few Myrs)
clusters is the expulsion of natal gas by stellar winds and supernovae, which
perturbs the clusters' potential and leaves up to 90% of them unbound. A cluster
disruption model has recently been proposed in which this mass- independent
disruption of clusters proceeds for another Gyr after gas expulsion. In this
scenario, the survival chances of massive clusters are much smaller than in the
traditional mass-dependent disruption models. The most common way to study
cluster disruption is to use the cluster age distribution, which, however, can
be heavily affected by incompleteness. To avoid this pitfall we introduce a new
method of studying cluster disruption based on size-of-sample effects, namely
the relation between the most massive cluster, M_{max}, and the age range sampled.
Assuming that clusters are stochastically sampled from a power-law cluster
initial mass function, with index -2 and that the cluster formation rate is
constant, M_{max} scales with the age range sampled, such that the slope in a
log(M_{max}) vs. log(age) plot is equal to unity. This slope decreases if
mass-independent disruption is included. For 90% mass-independent cluster
disruption per age dex, the predicted slope is zero. For the solar
neighbourhood, SMC, LMC, M33, and M83, based on ages and masses taken from the
literature, we find slopes consistent with the expected size-of-sample
correlations for the first 100 Myr, hence ruling out the 90% mass-independent
cluster disruption scenario. For M51, however, the increase of log(M_{max}) with
log(age) is slightly shallower and for the Antennae galaxies it is flat. This
simple method shows that the formation and/or disruption of clusters in the
Antennae must have been very different from that of the other galaxies studied
here, so it should not be taken as a representative case.