Density distribution of the spectral index α = dlog (λF_λ)/dlog λ. The dashed lines define the Class II, Flat spectrum, and Class I source locations. The number of objects for these three classes is indicated for each cluster. The density distribution is obtained using the kernel estimator presented in Silverman (1986). It is particularly well adapted to small samples since it does not depend on the first bin position and is oversampled.