Acceptability
The exploratory focus group confirmed the importance of racism as a form of oppression for Indigenous Australians, while the two confirmatory focus groups supported the conceptualization, clarity and completeness of the MIRE (see [
13] for details). No further modifications to the MIRE instrument were required as a result of these focus groups. The content validity of various MIRE items was also supported by the limited body of published research on the dimensions of racism experienced by Indigenous Australians [
34‐
36].
Q4 (race consciousness) was the only question queried during confirmatory focus groups, with one participant expressing confusion as to its meaning. Feedback from interviewers and participants in the population health study indicated that this question continued to cause some confusion during administration of the MIRE. However, given that only ten participants had missing data for this item and that it displayed plausible patterns of association with other elements of the MIRE (see below), it does not appear that this confusion unduly affected the validity of this item. Nonetheless, until the reasons underlying such interpretational difficulty are identified, this item should be interpreted with caution.
Psychometric Properties
The items in MIRE Q1a-i were designed to be mutually exclusive and comprehensive of the settings in which inter-personal racism may occur; as such, they are not based on one or more underlying theoretical constructs. The experience of inter-personal racism in any one of these settings does not necessarily increase the likelihood of experiencing inter-personal racism in any of the other settings. Therefore, although inter-dependence among these items is quite possible, internal consistency is not required for such a checklist and removal of items as a result of such consistency checking is not appropriate [
23]. Nonetheless, for study participants with at least one non-missing item in Q1a-i (n = 301), these items showed a good level of internal consistency (α = 0.83), suggesting that participants who reported experiencing inter-personal racism in one setting tended to report inter-personal racism in other settings. Furthermore, the addition of Q1j (intra-racial racism) did not change the internal consistency of this question, suggesting that the experiences of inter-personal racism from Indigenous and non-Indigenous perpetrators were related.
Among the 212 participants who reported experiencing some inter-personal racism and who had non-missing data, Q2a-f (responses to inter-personal racism) had an alpha coefficient of 0.48, suggesting more than one underlying theoretical construct for these items. As shown in Table
1, factor analysis of these items produced two factors. As Q2a-f items loaded on both of these factors, an oblique rather than an orthogonal rotation was utilized. Factor one, which explained 30% of item variance, had a strong negative loading on Q2a and Q2f and a strong positive loading on Q2c-e.
Table 1
Principle component factor analysis of MIRE Q2a-f (responses to inter-personal racism)
alpha coefficient = 0.48 |
Q2a | -0.36 | 0.46 | 37 |
Q2b | 0.06 | 0.76 | 57 |
Q2c | 0.39 | 0.71 | 59 |
Q2d | 0.78 | 0.28 | 63 |
Q2e | 0.79 | -0.05 | 63 |
Q2f | -0.46 | 0.55 | 57 |
Total variance explained (%) | 30 | 26 | - |
Both Q2a and Q2f assess passive maladaptive cognitive responses and hence would be expected to co-vary. Furthermore, the fact that Q2a and Q2f are non-adjacent items on the MIRE reduces the possibility that these findings are due to an order effect bias in which adjoining items are answered in a similar manner by respondents due to their proximity alone. The strong positive correlation of items Q2d and Q2e with factor suggests that this factor predominately captures outer-directed adaptive problem- and emotion-focused behavioural responses. As indicated by negative factor loadings for Q2a and Q2f, responses Q2d and Q2e are as theoretically different from Q2a and Q2f as possible, in that they are: (1) outer-directed (active) instead of passive; (2) adaptive instead of maladaptive; and (3) affective/behavioural instead of cognitive. Item Q2c assesses an inner-directed adaptive problem-focused behavioural response and, hence, is much more similar to Q2d and Q2e than to Q2a or Q2f.
Factor two, which explained 26% of item variance, displays the strongest positive loading on Q2b and Q2c as well as, to a lesser extent on Q2a and Q2f (see Table
1). As such, this factor predominately captures inner-directed adaptive problem-focused behavioural responses (assessed by both Q2b and Q2c). In contrast, the moderate positive loadings of factor two on the passive maladaptive cognitive responses (Q2a and Q2f) were unexpected. Although a small body of research suggests that trying to avoid racism (Q2b) or changing aspects of the self to prevent racism from occurring (Q2c) are adaptive [
21], a recent study indicates these responses may be maladaptive among Indigenous people with respect to mental health [
12]. If this is the case, it may explain the stronger than expected loading identified above.
For these same 212 participants, items Q2g-j (reactions to inter-personal racism) were characterized by a moderate internal consistency (α = 0.55). As shown in Table
2, a factor analysis of these items produced one factor which explained 45% of item variance and loaded highly on Q2g, Q2h and Q2j but poorly on Q2i. This factor explained less than 10% of the variance of Q2i, hence supporting the conceptual divergence of this item from items Q2g, Q2h and Q2j in that Q2i assesses an outer-directed empowered reaction to inter-personal racism whilst this factor captures disempowered reactions to racism (as assessed by Q2g, Q2h and Q2j).
Table 2
Principle component factor analysis of MIRE Q2g-j (reactions to inter-personal racism)
alpha coefficient = 0.55 |
Q2g | 0.75 | 56 |
Q2h | 0.68 | 47 |
Q2i | 0.31 | 9 |
Q2j | 0.82 | 67 |
Total variance explained (%) | 45 | - |
Responses from the 287 participants with non-missing items for Q3a-d (internalized racism) showed a very low degree of internal consistency (α = 0.23). As shown in Table
3, a factor analysis of these items produced two factors which explained 35% and 25% of the item variance, respectively. As these factors appeared to be orthogonal (i.e. no items appeared to load significantly on both factors), a varimax rotation was conducted. Results indicated that factor one primarily captured affective aspects of Indigenous identity (as assessed by Q3a and Q3d). As shown in Table
1, this factor also loaded to a reasonable degree on Q3b which assessed agreement with a statement of fact about the disadvantaged situation of Indigenous Australians. While factor one did not load to any significant degree on Q3c, this item was highly loaded on factor two with other items having weak loadings on this factor. Although there was no a priori reason to expect this, it appears that the factual item (Q3b) converges with the two affective items (Q3a, d) while the normative statement (Q3b) assesses a construct that was captured by factor two and was, hence, distinct from the other items in this subscale.
Table 3
Principle component factor analysis of MIRE Q3a-d (internalized racism)
alpha coefficient = 0.23 |
Q3a | 0.75 | 0.13 | 58 |
Q3b | 0.51 | -0.25 | 32 |
Q3c | 0.03 | 0.96 | 93 |
Q3d | 0.76 | 0.00 | 58 |
Total variance explained (%) | 35 | 25 | - |
Responses from the 288 participants with non-missing items in Q3e-g (systemic racism) also evinced a moderate degree of internal consistency (α = 0.47). As shown in Table
4, a with a factor analysis of these items producing a single factor which explained 50% of item variance and loaded adequately on all three items in this scale, supporting the unidimensional nature of the systemic racism measure.
Table 4
Principle component factor analysis of MIRE Q3e-g (systemic racism)
alpha coefficient = 47 |
Q3e | 0.57 | 32 |
Q3f | 0.84 | 71 |
Q3g | 0.69 | 47 |
Total variance explained (%) | 50 | - |
Construct Validity
Hypothesis one (that internalized and systemic racism measures would be inversely related) was supported in that internalized and systemic racism were statistically significantly inversely associated, with participants who had high levels of internalized racism being less likely to have high levels of systemic racism. Only 7 of 284 participants had both high internalized racism and very high systemic racism scores. Hypothesis two (that increased reporting of inter-personal racism would be related to higher levels of systemic racism and lower levels of internalized racism) was partially supported in that a reduced likelihood of agreeing that Indigenous people should try to think and act more like other Australians (Q3c) was associated with more settings of reported racism (p = 0.06). However, neither systemic/internalized racism scores nor any other items in Q3 were statistically significantly associated with levels of inter-personal racism.
In support of hypothesis three (that higher levels of inter-personal racism and systemic racism as well as lower levels of internalized racism would each be associated with heightened race-consciousness (Q4)), as reported experiences of inter-personal racism increased, the frequency with which participants thought about being Indigenous (Q4) also increased. Agreeing that 'Indigenous people have less opportunities than other Australians' (Q3b) was also significantly associated with increased race-consciousness (Q4). However, neither systemic/internalized racism scores nor any other items in Q3 were statistically significantly associated with race-consciousness. Hypothesis four (that reported experiences of racism from other Indigenous people would be associated with feeling less accepted by other Indigenous people (Q3a) and feeling less good about being an Indigenous person (Q3d)) was wholly supported in that reported racism from other Indigenous people was associated with feeling less accepted by other Indigenous people (Q3a) as well as not feeling good about being an Indigenous person (Q3d).
There was strong support in this study for hypotheses five and six (that more frequent use of adaptive responses to racism would be associated with higher levels of systemic racism and lower levels of internalized racism and vice versa for maladaptive responses to racism). The maladaptive response of ignoring, forgetting about or accepting racism as a fact of life (Q2a) was associated with higher internalized racism (p = 0.07) and the adaptive response of talking about or expressing racism experiences (Q2e) was associated with lower internalized racism and higher reported systemic racism in general, as well as with agreeing that 'there is hardly ever anything good about Indigenous people in the media' (Q3g) (p = 0.07).
Agreeing that 'Indigenous people should try to think and act more like other Australians' (Q3c) was associated with more frequent attempts to change the way one is or the things that one does to prevent racism (Q2c). Given that Q3c is an item of the internalized racism scale, this finding supports the possibility considered above that Q2c may be maladaptive rather than adaptive among indigenous people. Similarly, those who reported the maladaptive response of keeping racism to themselves (Q2f) were more likely to have higher levels of internalized racism in general and to agree that 'Indigenous people should try to think and act more like other Australians' (Q3c). Being angry, annoyed or frustrated (Q2h) as an adaptive reaction to inter-personal racism was associated with lower internalized racism in general (p = 0.06) as well as with agreement that 'Indigenous people have less opportunities than other Australians' (Q3a) (p = 0.07) and that 'other Australians think they are better than Indigenous people' (Q3f).
Q5 provides information on how well-known a respondent's Indigeneity was within their social group, while Q6 attempts to capture the extent to which a respondent 'looks' Indigenous to strangers. Participants who 'look' Indigenous to strangers (Q6) should also 'look' Indigenous to those in their social group (Q5). In support of this conjecture (i.e. hypothesis seven: that there would be a concordance in responses to Q5 and Q6), a cross-tabulation of Q5 and Q6 shows that all participants who reported that most/all of the people they meet for the first time knew they were Indigenous (Q6) also reported that most/all of those in their social group know they are Indigenous (Q5).
Convergent Validity
For those respondents with non-missing data for both the non-MIRE question on discrimination and the MIRE inter-personal racism score, there was a high degree of concordance between a positive response to the discrimination question and the inter-personal racism score. Only 6% of participants answered 'yes' to having been affected by discrimination while also reporting no inter-personal racism experiences. However, there was much less concordance between negative responses to these questions, with 64% of participants who reported some form of inter-personal racism in the MIRE also reported not being affected by discrimination. Put another way, only 25% of participants answered 'yes' to the discrimination question while about 70% of participants reported some experience of inter-personal racism across the nine settings included in the MIRE.
Reported discrimination and inter-personal racism evinced a similar pattern of association with other MIRE items/scales (given that the fewer 'yes' responses to the discrimination question reduced the statistical power to detect such associations). Participants who reported discrimination and/or inter-personal racism were more likely to report using adaptive behavioural responses (Q2d and Q2e) and to report experiencing all three disempowered affective (Q2g-i) and one somatic reaction to inter-personal racism (Q2k). Similarly, both discrimination and inter-personal racism were associated with race-consciousness (Q4). Furthermore, the discrimination item and MIRE Q1 items had similar patterns of association with health outcomes examined in the larger health study (see [
13]).
Such findings suggest that discordance between the discrimination and racism items is unlikely to result from a variation in underlying constructs. In fact, terminology differences between these questions means that respondents who experienced racism that didn't affect them, their family or friends would appropriately report racism in the MIRE but not in the discrimination item. This could explain the main source of discordance between these measures. It is also possible that, as noted in other research [
37,
38], the use of ethnoracial terminology in the MIRE introductory text [see Additional file
1] lead to increased reporting of racism. Given that the proportion of participants reporting some level of racism in any one setting in MIRE Q1 (25–45%) is comparable to the 25% of participants who answered 'yes' to the discrimination question, it is also possible that a single item may not prompt respondents to report the full extent of their racism experiences. That is, increased reporting of racism in the MIRE may have resulted from prompting respondents' recall via explicitly listing a range of possible settings in which racism may have occurred.
The fact that the discrimination item assessed exposure over the past year while the MIRE included no explicit timeframe may also have contributed to the discrepancy in reported prevalence. The higher prevalence of racism reported in the MIRE could be explained if most experiences of racism had taken place more than a year prior to the administration of the MIRE. It is notable that the prevalence of reported discrimination was lower than the prevalence of inter-personal racism despite the fact that the former related to experiences that affected family/friends as well as that of the respondents themselves. Given these results, it is evident that further testing of the MIRE's convergent validity against other existing psychometrically-validated instruments is warranted.
Acceptability
The proportion of missing responses for the 31 items in the MIRE ranged from 1% (Q5) to 10% (Q2g, h, j), with an average missing rate of 7% for the MIRE overall. In comparison, the missing response rate in the DRUID study was 15% for household income, 2% for marital status and 1% for self-assessed health status. This suggests that the MIRE had a moderate too high level of acceptability in relation to other variables measured in the DRUID study. As expected, across all MIRE items the rates of missing responses were 2–6 times higher for self- vs. interviewer administered questionnaires.