Fig. 5

a shows actual and predicted linear trends in the monthly counts of fights (left) and assaults (right) at NDSP, before and after ND DOCR implemented changes to limit use of solitary confinement in response to rule violations and the established the Behavioral Intervention Unit (BIU). The vertical line denotes the intervention point of May 1, 2016, when these reforms were operating and was selected in consultation with ND DOCR leadership. The graph on the lefts suggests that monthly assaults increased from January 2010 to May of 2016 but did not reach statistical significance (Β = 0.06, p = 0.221, CI = − 0.03, 0.14). It also shows that during the early months after reforms, fights increased significantly (Β = 11.78, p > 0.0001, CI = 5.46, 18.11). However, the downward slope of line after the intervention point suggests a significant overall decrease in the monthly trend of fights at NDSP between May 2016 and the end of 2019 (Β = − 0.38, p > 0.0001, CI = − 0.60, − 0.16). The nearly flat slope of the pre-intervention line on the graph to the right that suggests monthly assaults did not change significantly between January 2010 and May of 2016, (Β = 0.004, p = 0.893, CI = − 0.54, 0.06). As with fights, this plot also shows that during the early months after reforms, assaults between incarcerated people increased significantly (Β = 8.80, p > 0.0001, CI = 4.57, 13.04). However, the downward slope of line after the intervention point suggests a significant overall decrease in the monthly trend of assaults at NDSP between May 2016 and the end of 2019 (Β = − 0.22, p > 0.001, CI = − 0.37, − 0.07). b shows actual and predicted linear trends in the monthly counts of fights (left) and assaults (right) at JRCC, before and after ND DOCR implemented changes to limit use of solitary confinement in response to rule violations and enhanced staffing and services in the Special Assistance Unit (SAU), a housing area for people with acute psychiatric needs. The vertical line denotes the intervention point of May 1, 2016, when these reforms were operating at JRCC and was selected in consultation with ND DOCR leadership. For the graph on the left, the upward slope of the prediction line, before the intervention, suggests that monthly fights increased significantly prior to May of 2016 (Β = 0.30, p < 0.0001, CI = 0.19,0.39). By contrast, the downward slope of line after the intervention point illustrates a significant decrease in the monthly trend of fights in the initial months, post-reforms (Β = − 10.32, p > 0.0001, CI = -0.69,-0.19), and a significant monthly decrease from May 2016 through December 2019 (Β = − 0.45, p > 0.01, CI = − 0.69, − 0.19). For the graph on the right, the downward slope of the line, before the intervention, suggests that monthly assaults decreased prior to reforms, but was not statistically significant ((Β = − 0.13, p = 0.377, CI = − 0.43, 0.02). Similarly, the downward slope of the line after the intervention point illustrates a small increase in assaults in the initial months post-intervention (Β = 1.34, CI = − 0.80, 3.48), followed by a decrease in the monthly trend of assaults in the post-reform period (Β = -0.21 p = 0.574, CI = − 0.9, 0.05) that was not statistically significant