(PS9-D86) Confirmatory Factor Analyses of the Interpersonal Needs Questionnaire Revised 15-item: An Examination of the Role of Item Sequences

The Interpersonal Needs Questionnaire (INQ; Van Orden et al., 2012) is a 15-item self-report measure that assesses perceived burdensomeness (i.e., PB; 6 item subscale) and thwarted belongingness (i.e., TB; 9 item subscale), conceptualized as proximal risk factors for suicide ideation. The INQ has strong psychometric properties across samples (e.g., Van Orden et al., 2012). However, PB has demonstrated more robust associations with suicide ideation than TB (Chu et al., 2017), and several items have been problematic in factor analyses. This may be influenced by the INQ subscales being presented with the PB item-block first and then the TB item-block. Therefore, to assess the possible influence of item sequencing, participants (i.e., Adult U.S. MTurk Workers) were randomly assigned into three conditions with different INQ item sequences (PB-TB [standard order], n = 272; TB-PB [flipped order], n = 276; random order of items, n = 279).

Confirmatory factor analyses (CFA) were conducted to examine the two-factor structure of INQ (TB/PB; Van Orden et al., 2012) for the different item sequence conditions. Across conditions, items 9, 11, and 12 were consistently problematic and caused model fit concerns, according to modification indices and consistent with prior literature (Hill et al., 2015; Van Orden et al., 2012). Correlating the residuals and moving these items to a different factor did not improve model fit. However, removing these items improved the model fit for the CFAs for all conditions.

The original model based in the standard order sample showed good fit according to CFI (.95) but poor fit according to RMSEA (.19) and SRMR (0.11). After removal of problematic items, the final standard order model indicated good fit according to CFI (0.99), SRMR (0.03), and RMSEA (0.08). In the original model for the flipped order sample, CFI (.95) and SRMR (.08) also indicated good model fit but RMSEA (0.18) showed poor model fit. The final model indicated good fit according to CFI (.99), SRMR (0.03), and RMSEA (.08). Finally, the original model for the random order scale showed adequate model fit according to CFI (.93) and poor model fit according to RMSEA (.19) and SRMR (.09). The final random order model exhibited good fit according to CFI (.99), SRMR (0.03), and RMSEA (.08).

After item removal, both PB and TB subscales showed good internal consistency (Cronbach’s alphas: PB = .95–.97; TB = .91–.92). Of note, the correlations between the TB and PB factors (i.e., standard, r = .33, p < .001; flipped, r = .36, p < .001; random, r = .36, p < .001) were reduced compared to previous literature (e.g., Van Orden et al., 2008). Thus, the resulting factors may be more distinct and could resolve issues of multicollinearity that influence the testing of the interpersonal theory of suicide (Mitchell et al., 2020)

Our study demonstrated regardless of where TB and PB items appeared in the item sequence, items 9, 11, and 12 were problematic, indicating participant interpretation of these items may be poor or inconsistent. These findings suggest that removing these items could optimize the factor structure of the INQ, which may improve the predictive validity of the INQ. Beyond the issues with these items, item sequencing does not significant impact the INQ factor structure.