Phase 1: instrument development
Of the 23 panel members who received the first list of 24 statements, 16 (70%) reviewed them, rank ordered them, and generated an additional 26 statements. We rearranged the combined list of 50 statements according to the proposed ranking ("essential," "interesting but not essential," and "irrelevant") and returned them to the 16 panel members for a second ranking. Based on their responses, six statements were eliminated and the remaining 44 were returned to the 16 panel members for a third ranking. Nine panel members returned responses after this iteration, and we eliminated nine additional items according to their recommendations. The remaining 35 items then constituted the first draft of the instrument we named the Healthcare Professionals' Attitude Toward the Homeless Inventory (HPATHI). Although seven panel members did not participate in the third iteration, their rankings in the second iteration corresponded to the final list of items.
Phase 2: pilot administration of the 35-item HPATHI
The sample population of third-year medical students completed a pilot administration of the 35-item HPATHI; the subset of students who responded two weeks later underwent a second administration of the same instrument. Table displays the means and standard deviations of student responses to both administrations of the HPATHI. Items 2, 5, 6, 11, 15, 16, 20, and 23 were reverse-coded for the analysis so that a higher total mean on the instrument would indicate a positive attitude toward the homeless.
Means and standard deviations for HPATHI* test and retest
Cronbach's alpha coefficient for the first administration was 0.87; the test-retest reliability coefficient (Pearson r) was 0.69. Through an item analysis we discarded 12 items that were either highly correlated with other items, and were thus considered repetitious, or that had item-scale correlations less than 0.20.
Phase 3: administration of the HPATHI to the target population
One hundred and sixty health-care professionals (24 primary care physicians, 15 primary care residents, 47 clinical medical students, 71 preclinical medical students, and 3 medical students who did not specify their education level) from one academic institution completed the HPATHI; 147 of them also completed the ATHI. Table displays the means and standard deviations for both instruments by gender, by level of training (primary care physician; primary care resident; clinical medical student; preclinical medical student), and by experience with the homeless (no experience; <1 month, >1 month but <1 year, and >1 year).
Comparison of the HPATHI and ATHI by gender, level of training, and experience
The exploratory factor analysis (principal components), using a Promax rotation to account for the relationship among the factors, yielded a three-factor structure that explained 39% of the variance of the data. Factor 1 consisted of items 12, 13, 14, 17, 18, 19, 20, 21, 22, and 23 and was labeled Personal Advocacy; factor 2 consisted of items 1, 3, 4, 7, 8, 9, and 15 and was labeled Social Advocacy; and factor 3 consisted of items 5, 6, 10, and 11 and was labeled Cynicism. Table presents the items with their loadings in each factor.
Factor loadings for the 23-item HPATHI
The HPATHI was further shortened to 19 items by the deletion of four more items, which either were not represented in the three-factor structure (items 2, 14, and 16) or had an adverse effect on the subscale's reliability (item 15). The three subscales were also significantly related to each other: factor 1 had Pearson's r correlations of 0.47 (n = 160; p < 0.001) with factor 2 and 0.43 (n = 160; p < 0.001) with factor 3; and factor 2 had a Pearson's r correlation of 0.48 (n = 160; p < 0.001) with factor 3. Table displays the descriptive statistics and measurement properties for the 19-item HPATHI total and subscales. These three factors, if taken as subscales for the HPATHI, showed satisfactory Cronbach's alpha coefficients: 0.75, 0.72, 0.72, and 0.84 respectively for factor 1 (mean = 3.86; sd = 0.47), factor 2 (mean = 4.06; sd = 0.46), factor 3 (mean = 4.06; sd = 0.50), and total scale (mean = 3.96; sd = 0.38).
Measurement properties for reduced HPATHI scale and subscales
The Pearson's correlation coefficient between the HPATHI and the ATHI was 0.68 for the HPATHI's total scale (concurrent validity) (Table ). For the extreme group comparisons, no significant difference was found between preclinical medical students and primary care physicians in their responses to the HPATHI (F = 1.05; df = 3, 156; p = 0.371). On the other hand, respondents who had more than one year of experience with the homeless scored significantly higher than those who had less than one month of experience (F = 6.19; df = 2, 157; p = 0.003) (Table ). When the individual hypothetical subscales were considered, all items were either moderately or strongly correlated with their respective subscales (range of Pearson's correlation coefficients, 0.38 to 0.68). However, when the entire instrument was considered, the item analysis showed that items 1, 2, 15, and 16 had low item-scale correlations (Pearson's r < 0.24).
Extreme group comparisons by level of training and experience with the homeless