Background

The 35-item Pediatric Symptom Checklist was designed to identify psychosocial dysfunction in children. Parent-report (PSC) and youth-report (Y-PSC) versions are widely used in the U.S., but their psychometric equivalence has not been well-established, especially in non-U.S. settings. To be used in combination, it is important that these forms function the same way regardless of reporter, including similar factor structure and item informativeness. This study compares the construct validity and psychometric properties of the PSC and Y-PSC among children affected by HIV in Botswana. In addition, this study uses Computerized Adaptive Testing (CAT) sessions to shorten the tool.

Methods

Children ages 7-17 years (n=1033) and their parents (n=1183) completed the PSC/Y-PSC in Gaborone, Francistown, and Maun, Botswana.

Exploratory and confirmatory item-factor models were performed using multidimensional Item Response Theory with expectation-maximization estimation. Oblimin rotation was used in exploratory models, and the number of factors was determined by the minimum average partial method combined with scree plot evaluation. Model fit was assessed using CFI, RMSEA, and SRMR. For the PSC, a bifactor model was used, and the general factor score was compared to the sum score.

For shortening the PSC, CAT sessions were simulated with the software Firestar. Three CAT sessions were simulated each for the PSC and Y-PSC. Standard errors of 0.3, 0.4, and 0.5 were set as stopping rules. 1000 respondents were simulated for each session.

Results

All methods revealed clear differences between forms, obviating fine-grained measurement invariance analyses. The optimal model for the PSC was a 2-factor, and a 1-factor model was best for the Y-PSC. The 1-factor Y-PSC model had acceptable fit (CFI=0.974, RMSEA=0.027, SRMR=0.043). The 2-factor model for the PSC was tested using a bifactor configuration to retain an overall score. This model was confirmed (CFI=0.967, RMSEA=0.033, SRMR=0.049).  

In order of descending standard error, the PSC was shortened to have 30 items, 19 items, or 13 items. In the same order, the Y-PSC was shortened to have 33 items, 27 items, or 18 items.

Conclusions

Clear violations of measurement invariance across forms suggest that scores from the PSC and Y-PSC should be interpreted differently.  The relatively low correlation of bifactor scores with sum scores suggests that multidimensionality should be accounted for when using the PSC in Botswana and potentially in other non-US settings. 

The PSC and Y-PSC can be shortened while retaining an acceptable level of standard error.