Bivariate and multivariable analyses
The internal reliability of the social distance scale was confirmed using Cronbach's alpha (α =.91) [see Additional file
3]
The social distance scale was linearly transformed to lie in the interval 0–100 (0 representing minimum social distance, 100 representing maximum social distance). Social distance scores occurred over the entire range of the scale, with a mean social distance of 30.4.
Bivariate analyses were conducted to examine the relationship between social distance and each of the independent variables and covariates separately, including the order of presentation (Table
1). In virtue of the clustered nature of the data, the bivariate analyses were performed using a maximum likelihood, multilevel (variance components) regression analysis (Table
1).
Table 1
Bivariate analysis of social distance, adjusting for the effect of repeated judgements within participants1
Order (base = First) | 352/1044 | | | | |
Second | | 4.81 | 1.7 | .005 | 1.47 – 8.15 |
Third | | 6.06 | 1.7 | .000 | 2.73 – 9.39 |
Age | 346/1026 | 0.00 | 1.05 | .997 | -2.06 – 2.06 |
Sex (base = Male) | 339/1005 | | | | |
Female | | -.78 | 1.81 | .666 | -4.34–2.77 |
Knowledge | 352/1044 | -.25 | .50 | .615 | -1.24 – 0.73 |
Disease (base = None) | 352/1044 | | | | |
Leukaemia | | -6.99 | 1.63 | .000 | -10.18 – -3.80 |
AIDS | | 7.39 | 1.63 | .000 | 4.20 – 10.58 |
Co-Characteristic (base = None) | 352/1044 | | | | |
BT | | .07 | 1.96 | .972 | -3.78 – 3.92 |
CBD | | 2.34 | 1.96 | .235 | -1.52 – 6.19 |
CS | | 19.06 | 1.98 | .000 | 15.17 – 22.94 |
IDU | | 25.13 | 1.96 | .000 | 21.29 – 28.97 |
The analysis showed the order of presentation to have a statistically significant effect on participants' judgments about social distance. As a percentage of the scale maximum, vignette two was, on average, 4.8% higher on the social distance than vignette one (p = .005), and vignette three was 6% higher (p < .001). That is, social distance increased significantly with the increasing order of presentation. None of the covariates (age, sex or HIV/AIDS knowledge) had a statistically significant effect on social distance.
The results were very different for the independent factors, Disease and Co-Characteristic. For the Disease factor, Leukaemia was significantly less stigmatising than no disease, resulting in an average reduction in social distance of 7% (p < .001). In contrast, AIDS was significantly more stigmatising than no disease resulting in an average increase in social distance of 7% (p < .001). As anticipated, having a blood transfusion did not significantly increase social distance above the base category. Likewise, commercial blood donation did not significantly increase social distance (p = .235). In contrast, having engaged in a commercial sex transaction increased social distance by an average of 19% of the scale maximum (p < .001), and having engaged in IDU increased social distance by an average of 25% (p < .001).
In the multivariable model of social distance the order of presentation was controlled for. Participants' sex, age, and knowledge of HIV/AIDS, however, were not included as covariates in virtue of their small and statistically non-significant association with social distance in the bivariate models. The focus of the multivariable model was the factors of Disease and Co-characteristic and their interaction effects. Table
2 shows the results of the maximum likelihood, multilevel (variance component) regression analysis, which takes account of the clustering effect of the repeated measures within participants. The intra-class correlation was .35 indicating that a substantial proportion of the variation in social distance was attributable to within participant effects.
Table 2
Multivariable analysis of social distance, adjusting for the effect of repeated judgements within participants and controlling for the order of vignette presentation
Order (base= First) | | | | |
Second | 4.91 | 1.30 | .000 | 2.37–7.44 |
Third | 5.73 | 1.29 | .000 | 3.20–8.26 |
Disease (base = None) | | | | |
Leukaemia (L) | 2.48 | 3.26 | .447 | -3.91 – 8.86 |
AIDS | 24.32 | 3.31 | .000 | 17.83 – 30.81 |
Co-Characteristic (base = None) | | | | |
BT | 5.00 | 3.27 | .126 | -1.40 – 11.41 |
CBD | 9.13 | 3.26 | .005 | 2.74 – 15.51 |
CS | 30.66 | 3.27 | .000 | 24.34 – 37.06 |
IDU | 43.84 | 3.25 | .000 | 37.48 – 50.20 |
Interaction Effects | | | | |
L × BT | -1.55 | 4.72 | .742 | -10.81 – 7.71 |
L × BS | -6.77 | 4.72 | .151 | -16.01 – 2.47 |
L × CS | -10.35 | 4.77 | .030 | -19.69 – -1.01 |
L × IDU | -25.57 | 4.73 | .000 | -34.84 – -16.30 |
AIDS × BT | -13.17 | 4.73 | .006 | -22.44 – -3.89 |
AIDS × CBD | -13.15 | 4.78 | .004 | -22.51 – -3.79 |
AIDS × CS | -25.13 | 4.76 | .000 | -34.46 – -15.79 |
AIDS × IDU | -31.64 | 4.73 | .000 | -40.91 – -22.38 |
In the multivariable analysis the two factors and the interaction were statistically significant. The number of tests (i.e. 14) that were conducted raises some issues about Type I errors in multiple comparisons. For reasons discussed by Cook and Farewell (1996), however, this was not regarded as problematic. The core questions revolved around Leukaemia, AIDS, IDU, CS, and their interactions. Most of the effects appeared to fall clearly one way or another. Even using as conservative a correction for multiple comparisons as the Bonferroni adjustment, the only questionable result would be the interaction effect between Leukaemia and CS – for which there was not, in any case, a specific a priori hypothesis.
In keeping with the literature, however, the pattern of results suggested that AIDS, IDU, CS, and to a lesser degree CBD, were each highly stigmatising. They increased the social distance score between 9% and 44% of the scale maximum. In contrast, BT in isolation had no statistically significant effect to the level of stigma, confirming its suitability as a control co-characteristic. The pattern of interaction also showed a moderating effect, as anticipated by Reidpath and Chan (2005). Specifically, the stigma associated with the co-occurrence of two highly stigmatised attributes was not simply a summation of the stigma of each; rather, the stigma appeared to lie generally between the stigma of one and the stigma of the sum.
In contrast to the bivariate analyses, leukaemia in isolation did not have a statistically significant moderating effect on levels of stigma, however, it had a statistically significant moderating effect on the stigma of IDU, but possibly not on the stigma of CS. In addition, commercial blood donation as a co-characteristic decreased the level of stigma associated with AIDS.
Following the approach described by Reidpath and Chan [
26], the layering of the stigmas between disease and co-characteristics was examined. Using the figures presented in Table
3, cell means (%SM) were calculated as the sum of the disease coefficient, co-characteristic coefficient, and the interaction effect (see Table
3).
Table 3
Model estimates (MLE) of social distance scores as a function of co-characteristics and disease conditions.
None | 0.00 | 2.48 | 24.32 |
BT | 5.00 | 5.93 | 16.17 |
CBD | 9.13 | 4.83 | 19.72 |
CS | 30.66 | 22.78 | 29.85 |
IDU | 43.84 | 20.75 | 36.52 |
Very different arrangements of layering were observed between AIDS and the various co-characteristics, each of which are discussed in turn. It should be noted, however, that the layering effects are based on estimates with often wide confidence intervals and this limits the extent to which one can interpret the results.