We study analytically the correlations between the positions of tagged part
icles in the random average process, an interacting particle system in one
dimension. We show that in the steady state, the mean-squared autofluctuati
on of a tracer particle grows subdiffusively sigma (2)(0)(t)similar tot(1/2
) for large time t in the absence of external bias but grows diffusively si
gma (2)(0)(t)similar tot in the presence of a nonzero bias. The prefactors
of the subdiffusive and diffusive growths, as well as the universal scaling
function describing the crossover between them, are computed exactly. We a
lso compute sigma (2)(r)(t), the mean-squared fluctuation in the position d
ifference of two tagged particles separated by a fixed tag shift r in the s
teady state and show that the external bias has a dramatic effect on the ti
me dependence of sigma (2)(r)(t). For fixed r,sigma (2)(r)(t) increases mon
otonically with t in the absence of bias, but has a nonmonotonic dependence
on t in the presence of bias. Similarities and differences with the simple
exclusion process are also discussed.