On dependence assumption in p-value based multiple test procedures