Aim Two
We next examined the regression relationships between the four topics and three predictors: NSSI history, emotion dysregulation, and sample. We report likelihood ratio tests (LRT) and Cox and Snell (1989) pseudo-R2 measures (R2CS) for tests of main effects and interactions (see Supplemental Table 2). Regression coefficients (b ) are reported on the natural logarithm scale (i.e., a log-link function was used) in Table 2.
First, the three-way interaction among sample, NSSI history, and emotion dysregulation was not statistically significant, LRT (4) = 9.05,p = .060, R2CS = .042.
Second, there was a significant overall interaction between NSSI history and sample, LRT (4) = 17.27, p = .002,R2CS = .08, when holding emotion dysregulation constant (see Figure 1). This interaction was primarily driven by a significant interaction between sample and NSSI history for usage of Topic 3, Z = 3.96, p < .001. In the undergraduate sample, the proportion of Topic 3 was significantly greater for participants with a history of NSSI (94%) than those without a history of NSSI (83%), Z = 3.87, p< .001, while there was not a significant difference in the community sample (14% for participants with or without a history of NSSI), Z = -0.99, p = .323. This interaction was not statistically significant for Topics 1, 2, and 4, all p> .10.
Third, there was a significant overall interaction between emotion dysregulation and sample, LRT (4) = 19.04, p = .001,R2CS = .09 (see Figure 2). This interaction was primarily driven by a significant interaction between emotion dysregulation and sample for usage of Topic 3, Z = -3.62,p < .001. In the undergraduate sample, the proportion of Topic 3 increased significantly as emotion dysregulation increased,Z = 4.30, p < .001, while there was no significant change in the community sample, Z = 1.30, p = .194. This interaction was not statistically significant for Topic 1, 2, and 4, all p > .50.
Fourth, there was a significant overall interaction between NSSI history and emotion dysregulation, LRT (4) = 15.34, p = .004,R2CS = .07, as shown in Figure 3. This interaction was primarily driven by a significant interaction between NSSI history and emotion dysregulation for usage of Topic 3,Z = 4.31, p < .001, and Topic 4, Z = 2.39, p = .017. For participants without a history of NSSI, the proportion of Topic 3 did not change significantly with emotion dysregulation, b = 0.001 (SE = 0.005), Z = 0.18,p = .855, 95% CI: [-0.009, 0.011]; for participants with a history of NSSI, the proportion of Topic 3 increased significantly as emotion dysregulation increased, b = 0.046 (SE = 0.009),Z = 5.00, p < .001, 95% CI: [0.028, 0.064]. For participants without a history of NSSI, the proportion of Topic 4 did not change significantly with emotion dysregulation, b = 0.0002 (SE = 0.011), Z = 0.02, p = .982, 95% CI: [-0.018, 0.004]; for participants with a history of NSSI, the proportion of Topic 4 increased significantly as emotion dysregulation increased, b = 0.013 (SE = 0.006), Z = 2.08,p = .037, 95% CI: [0.001, 0.026]. This interaction was not statistically significant for Topic 1 and or Topic 2, both p> .10.
As reflected above, there was a significant main effect of sample,LRT (4) = 158.34, p < .001,R2CS = .53, a significant main effect of NSSI history, LRT (4) = 14.16, p = .007,R2CS = .07, and a significant main effect of emotion dysregulation, LRT (4) = 29.44, p< .001, R2CS = .13. Because the main effects of sample and NSSI history were not robust in the presence of the interactions involving Topics 3 and 4 (as described above), we do not interpret them further. There was a significant marginal relationship between Topic 1 and emotion dysregulation,b = 0.012 (SE = 0.004), Z = 2.64, p = .008, 95% CI: [0.003, 0.021], suggesting that usage of Topic 1 increased significantly as emotion dysregulation increased, regardless of sample or NSSI history.