A change in pairwise spike train correlations is a well-established neural correlate of a shift in brain state, and as such has been the topic of intense study. However, while the explosion in population recording has provided a wealth of data to measure spike train correlations and their state dependence, there continues to be ambiguity in the biological mechanisms that mediate changes in correlated activity. One reason for this lack of understanding is the vast set of possible circuit and cellular realities that can support correlated spiking activity. To provide insight into how circuit and cellular neuroscience manifest spike train correlations, we analyze in vivo high-speed voltage imaging data of a population of L1 inhibitory neurons (~300 neurons) in mouse S1 and separate the behavior into active (running or whisking) and passive (no running or no whisking) states. We find that, for a given neuron pair, spiking activity often shows stronger correlations than the corresponding subthreshold activity; we label this observation correlation expansion. Correlation expansion is more prominent during active states and is associated with a rise in the dimensionality from subthreshold to spiking population activity. These results prompt a phenomenological model of neuronal transfer that identifies two requirements for correlation expansion. First, spiking activity results from a thresholded mapping of subthreshold activity. Second, the subthreshold activity must be non-Gaussian, with strong yet rare events that produce high output spiking correlations. Our framework points to clear features that circuit models must consider when tackling how spiking activity is correlated in neuronal populations.