The results of more than 8-yr monitoring (1988-1997) of the Wolf-Rayet binary WR 147 (WN8(h)+OB) with the Westerbork Synthesis Radio Telescope (WSRT) are presented. When the strong winds of the Wolf-Rayet (WR) and OB binary components collide, they produce non-thermal excess radiation in the region where the two winds interact. The binary system, monitored at 1.4 and 5 GHz (21 and 6 cm), is not resolved by the WSRT, thus we observed the total flux density of the system. The time-averaged 5 and 1.4-GHz flux densities are 35.4+/-0.4 mJy and 26.4+/-0.3 mJy, respectively. These give a time-averaged spectral index of alpha5 -1.4 GHz ~ 0.23 +/- 0.04, where Snu ~nu alpha . The departure from the value expected for thermal radiation from a spherically symmetric stellar wind, alpha =0.6, can be attributed to non-thermal emission from a bow-shaped source to the north of the thermal source associated with the WN8 star. With a possible detection at 350 MHz of 16+/-4 mJy, in our separate study of the Cygnus region, the spectral energy distribution, after the contribution of the southern thermal source is subtracted, can be fitted by a synchrotron emission model which includes free-free absorption. The non-thermal emission originates in the region where the winds of the binary components collide. This region, therefore, contains a mixture of relativistic particles accelerated by shocks and thermal particles, responsible for the free-free absorption. We show, in a simplified model of the system, that additional free-free absorption may occur when the line of sight to the collision region passes through the radiophotosphere of the WR wind. The 1.4-GHz flux density of WR 147 varied between ~ 20 mJy and ~ 30 mJy. We attribute the irregular, stochastic variations with a typical timescale of about 60 days to inhomogeneities in the wind, with different mechanisms involved in the flux-density increase than in the flux-density decrease. A flux-density increase results when the inhomogeneities in the wind/clumps enter the wind collision region, fuelling the synchrotron emission. The typical timescale of the flux-density decrease is shorter than the timescale of synchrotron loss ( ~ 103 yr) or the Inverse-Compton lifetime (~4.5 yr), but of the order of the flow time in the colliding-wind region ( ~ 80 d). Therefore, we suggest that the flux-density decrease is due to plasma outflow from the system. Furthermore, variable free-free absorption due to large clumps passing the line of sight may also cause variations in the flux density. We observe a possible long-term flux-density variation on top of the stochastic variation. This variation is fitted with a sinusoid with a ~ 7.9-yr period, with a reduced chi 2 of 1.9. However, as the period of the sinusoid is too close to the monitoring time span, further monitoring is needed to confirm this long-term variation. Based on observations made with the Westerbork Synthesis Radio Telescope (WSRT), operated by the Netherlands Foundation for Research in Astronomy (NFRA).