Response Clustering of Reverse-Keyed, Adversely Framed Items in the Human Rights-Based Well-Being Checklist: An Exploratory Network Analysis

The Convention on the Rights of Persons with Disabilities (CRPD) enumerates a broad, heterogeneous set of rights and requires their monitoring. The Human Rights-Based Well-Being Checklist (HRWC) is a 39-item instrument aligned with the CRPD; it covers this heterogeneous content, and no subscales were defined at its development. Using exploratory network analysis, we examined the response structure of the HRWC to clarify its descriptive interpretation and to assess whether empirically detected item communities were sufficiently stable to justify subscale construction or item removal. The focal case was a set of eight items that are reverse-keyed in the scoring scheme and framed in terms of adverse rights-related experiences (Items 1-4, 12, 14, 17, and 18). This set was defined by the scoring scheme independently of the network results; the decision to focus on it was post hoc.

This secondary analysis used registry data from adults with disabilities in Muntinlupa, Philippines (N = 1,014). The HRWC is a self-report instrument whose items describe day-to-day situations corresponding to CRPD rights; each item is scored 0-4, giving a possible total of 0-156. Regularized partial-correlation networks, in which each edge represents the association between two items after accounting for all the other items, were estimated under two parallel specifications: rank-based Spearman correlations, which use only the rank order of the responses, and polychoric correlations, which treat each response as a coarsened observation of an underlying continuous variable. Item communities, that is, groups of items more densely interconnected with one another than with the rest of the network, were detected with the Walktrap algorithm, and their reproducibility was examined in 1,000 bootstrap resamples per specification.

Under Spearman estimation, the eight items formed one community containing no other items (27 of 28 possible within-set edges positive and nonzero; one absent), were retained across algorithms, regularization settings, and the nonparanormal network, and were reproduced in 885 of the 1,000 resamples (88.5%). Under polychoric estimation, the Walktrap point estimate divided the eight items along a shared stem (Items 1-4 versus 12, 14, 17, 18), although three alternative detection algorithms recovered that single community; the same eight-item-only community recurred under Walktrap in 333 of the 1,000 resamples (33.3%). Whether the eight items formed one community or two subgroups, thus, varied with the correlation method, the detection algorithm, and the resample.

Item-level profiles and the existing unweighted total serve different descriptive functions. Item-level profiles preserve information on the distinct CRPD-related conditions represented by individual items, whereas the unweighted total can provide a concise summary across all 39 responses. Substantive content could not be separated from keying, shared stems, item position, or response distributions, and the findings provide no empirical support from these analyses for constructing subscales or removing items on the basis of the observed network communities.