The number of upper and lower records in both directions of the series
is considered for testing the hypothesis that a sequence of random va
riables is independently and identically distributed, against the alte
rnative of a polynomial trend in location. The asymptotic relative eff
iciency of the records tests, originally suggested by Foster and Stuar
t (1954), is derived. It is shown that the efficiency of these tests i
s raised distinctly by splitting the series of observations and by wei
ghting the records according to their position in the series.