Distribution of q_IR as a function of redshift, split across increasing M_⋆ bins. In each panel, we compare the q_IR estimates of individual radio detections (black dots) with the median stacked values of non-detections (yellow squares) and with the weighted-average q_IR of detections and non-detections together (from Eq. (3), blue stars). Green upward arrows indicate the corresponding threshold q_IR, lim above which radio detections become inaccessible from an M_⋆-selected sample. We select relatively complete M_⋆ − z bins, in which at least 70% of radio detections have q_IR below the corresponding threshold. This criterion identifies 13 bins (in red square brackets). Within these bins, the peak of q_IR distribution (q_IR, peak) is indicated with red open circles. In the two highest M_⋆ bins, the best fitting trends with redshift are shown by the red dashed lines. Each panel reports the number of individual 3 GHz sources and their fraction with S/N_IR > 3. See Sect. 4.2.1 for details.