Revised Working Alliance Inventory

The Working Alliance Inventory (WAI, Horvath & Greenberg, 1989) has made possible tremendous contributions to psychotherapy research. Many studies support its reliability and validity, but some research suggests the 7-point response format and 3-factor structure of the client version WAI are problematic (Hatcher & Gillaspy, 2006). We used Rasch model item response theory (IRT) to analyze archival data from 1786 counseling center and community clients. Results suggest problems with crossed category thresholds among response points 2-4 (rarely, occasionally, and sometimes). IRT analyses identified a new scoring scheme that combines neighboring responses to create 5and 4-point scales. Although subscale variance was reduced by 11%-26%, rescoring yielded improved reliability and generally higher correlations with therapy process and outcome measures. IRT methods were also used to construct a new 16item, Brief Revised Working Alliance Inventory (br-WAI) with an 8-item Bond subscale and 8item Tasks/Goals subscale designed to maximize total test information and the “bandwidth” of item difficulty. Comparisons suggest the br-WAI performs much better than two other 12-item brief versions of the WAI (Hatcher & Gillaspy, 2006; Tracey & Kokotovic, 1989) at discriminating a group of clients suspected to have poor working alliances. Correlations with another measure of the therapy relationship (secure attachment to therapist), with therapy outcome (residual gain in symptom change), and with client-rated session depth were all higher for the br-WAI total score than either of the two brief versions, or the full scale WAI.