Window size

The window size is related to the timescale at which the system dynamics evolve and may affect results (Dakos et al. 2012). Therefore, sensitivity analyses were conducted to investigate the effects of windows of one, two or three weeks. We ensured that each window included an equal number of weekend days, thereby negating their effects on the results (Bayani et al. 2017). The sensitivity analyses indicated small differences for different window sizes (see Additional file 1), but results were largely robust. Therefore, based on suggestions on how bipolar dynamics develop in patients with high mood instability (Bonsall et al. 2012), we opted for a window size of two weeks (i.e., 70 observations). This means that, for a transition that happened at the 150th observation, 150–70 = 80 windows could be fitted, hence yielding 80 estimates of autocorrelations/standard deviations. The average transition happened later in time, and therefore, there were on average 168 windows (SD = 99, range = 35–415) per transition.