Each column shows one step in our analysis procedure, and each row shows the comparison of two features. Top row shows the comparison of location-modulation (x-axis) with time-modulation (y-axis). Left-hand panel: probability density derived from combining the independent KDE-distributions of location-modulation and time-modulation. Middle-left panel: real joint probability distribution (red-scale). One point is a single neuron’s modulation after z-scoring per recording. Middle-right panel: the difference in probability density between the real and null-hypothesis distributions. Blue indicates fewer neurons in the real joint-distribution than in the null-hypothesis distribution. The upper-right quadrant is mostly red, indicating the real distribution contains more neurons that encode both location and time than expected by chance. Right-hand panel: we sampled 10,000 synthetic neuronal populations from the null-hypothesis probability distribution, and calculated for each synthetic population the number of neurons that fell in the upper-right quadrant (URQ): that is, the number of neurons that show both above-average location-modulation as well as above-average time-modulation. The histogram shows the number of neurons in the URQ for each of the 10,000 populations, and the blue vertical line shows that the real number of neurons in the URQ is much higher than expected by chance (z = 3.812, p = 1.144 × 10–4). Middle row: as top row, but now comparing time- and mismatch-modulation. The URQ-analysis showed that time- and mismatch-modulation were not more often seen in the same neurons than expected by chance if the two features are independently encoded (z = −0.059, p = 0.953 (n.s.)). Bottom row: location on the virtual track and visuomotor mismatch are less likely to be encoded by the same neuron than if the two features were randomly distributed (z = −4.310, p = 2.019 × 10–5). This shows that neurons show an encoding specialization for either spatial location or visuomotor mismatch, but not both. Analyzing the Pearson correlation per recording we find qualitatively similar results. Mean Pearson correlation between time-modulation and location-modulation across recordings: n = 7, mean r = 0.2272; one-sample t-test vs. 0: p = 0.0164. Mean correlation between time- and mismatch-modulation: r = 0.1267, p = 0.2829 (n.s.). Mean correlation between location- and mismatch-modulation: r = −0.2249, p = 0.0414.