Measurement invariance of the Research Attitudes Questionnaire in an age and ethnically diverse,community-dwelling sample

Abstract

Background

The Research Attitudes Questionnaire (RAQ), developed to predict individuals’ willingness to participate, is often used in Alzheimer's Disease and related dementia research.

Objective

The present investigation aimed to examine the suitability of the RAQ across age groups and three different racialized identities, i.e., to see whether the RAQ showed measurement invariance.

Methods

We administered the RAQ to six groups of participants: 457 younger and 594 older African Americans, 207 younger and 339 older American Indian/Alaska Native, and 173 younger and 679 older non-Hispanic White adults.

Results

Confirmatory factor analysis indicated that the best-fitting model was one-factor. All six groups fit the model well, with Comparative Fit Indices > 0.95. A series of cross-sample invariance tests using increasing constraints on factor loadings, means, and residuals revealed evidence of configural invariance, metric invariance, and partial scalar invariance.

Conclusions

These findings support the suitability of the RAQ for cross-cultural and/or age comparisons of willingness to engage in research in the groups and context studied.

Keywords

age Alzheimer's disease ethnicity measurement invariance Research Attitudes Questionnaire under-included groups

Introduction

Individuals’ attitudes toward biomedical research can affect research recruitment,^1–3 engagement,^3–7 and retention.^8,9 Given the disparity between the disproportionate burden of Alzheimer's disease (AD) and related dementia (ADRD) among minoritized (e.g., African Americans, American Indian/Alaska Natives, and Latinx) groups and their inclusion in dementia studies, especially clinical trials,^5,10–12 there has been increasing focus on understanding minoritized groups’ openness to engage in ADRD research. In addition to utilizing focus groups to explore under-included groups’ views regarding ADRD research,¹³ investigators have explored their attitudes toward research using self-report questionnaires.

Kim et al.¹⁴ developed the Research Attitudes Questionnaire (RAQ) to predict individuals’ willingness to participate in research. Several studies examined the scale's psychometric properties,^15–17 resulting in a 7-item version of the measure. The 7-item RAQ,¹⁷ the most frequently used version of the measure, is increasingly used in research and clinical settings to assess openness and willingness to engage in biomedical research and intervention. However, there has not been a systematic evaluation of its construct equivalence across groups differing in terms of age and ethnoracial identity. Instead, construct equivalence has been assumed, and investigators have reported findings based on comparisons across groups differing on sociodemographic factors. Notably, Rubright et al. encourage further research to evaluate RAQ's measurement invariance (MI) across groups stratified by age, gender, or racial identity.¹⁷ When examining willingness to engage in biomedical research and using the RAQ as a measure of one of the factors affecting willingness (i.e., attitudes toward research), it is essential to evaluate whether the RAQ has MI. MI assesses whether there is equivalence for a construct (via comparison of means, scores, factors, etc.) in the same way across the groups.^18,19 If there is MI, any observed differences reflect true differences between the groups and are not due to measurement bias and error.¹⁸ The demonstration of MI, therefore, is a critical construct validation process that involves the evaluation of the psychometric properties of a measure when applied to groups differing on one or more demographic and/or experiential variables.

The purpose of the present investigation was to evaluate the MI of the 7-item RAQ across different demographic variables. This study aimed to examine attitudes towards research among a group of diverse individuals in terms of age, gender, and ethnoracialized identity who were not recruited from a sample already engaged in research (i.e., registry participants). We sought to extend the literature by studying a larger sample, including respondents across the age range from 18 through 89, and purposefully recruiting participants from historically under-included and underrepresented communities. We hypothesized that RAQ scores would be lower amongst members of historically under-included and underrepresented communities who have a documented history of mistreatment by biomedical institutions (i.e., hospitals and physicians, universities, and research teams) compared to non-Hispanic White persons. We were interested in exploring whether there would be an age effect, whereby the older sample members would express more positive attitudes toward research within each ethnoracialized group. We did not expect there to be a difference in RAQ scores based on gender identity because, except for an investigation based on HIV-positive surgical patients,²⁰ a gender difference in RAQ scores has not been reported in the literature. We intend to fill the following gaps: absence of comparison of the RAQ by age and ethnoracial groups using Confirmatory Factor Analysis and MI analysis.

Methods

Participants

Approximately 2800 adult participants took the online questionnaire between September 1, 2022, and November 10, 2023. All participants were recruited through word-of-mouth, community engagement events, crowdsourcing, and efforts targeted at communities of color, e.g., advertisements in specialized community newspapers and outreach through a community advisory board. Individuals at least age 18 who could give informed consent, have normal or corrected-to-normal vision, and have access to a computer or mobile device with access to the web were eligible for inclusion. Individuals over 95 who volunteered and lacked reading proficiency in English were excluded from the study. Other exclusion criteria included guardianship status, self-reported diagnosis of a memory disorder, and/or unwillingness to indicate consent or to provide ethnoracialized group membership.

Procedure

This investigation was part of a larger project examining factors hypothesized to influence willingness to engage in biomarker testing. Qualtrics, a web-based survey and research software program, hosted the online questionnaire. Participants were given a link that brought them to an online consent form. After the consent form, participants were given a survey, which took approximately 30 to 45 min to complete. The items ascertained demographic characteristics as well as participants’ views regarding serious mental illness in general, schizophrenia, dementia, genetic testing, and biomarker testing. Several data integrity strategies were implemented, including reverse-scored items to identify straight-line responding, and “catch items”, i.e., infrequent items or items that would otherwise detect inattention or random response patterns.

We received 2812 surveys; 251 (8.9%) participants failed at least one of the three “catch” questions, or had too many missing responses, i.e., over 10% of the survey (N = 40; 1.4%). Two other participants were excluded due to their refusal to disclose their gender. Ten participants were excluded due to their nonbinary gender; their small number raised concerns regarding statistical power. (Data from nonbinary participants will be saved for later analyses when larger N's permit statistical robustness). This resulted in a final sample of 2549 (90.6%) participants.

The University of Wisconsin-Madison Educational and Social Sciences Human Subjects Board approved this research. All procedures followed were per the ethical standards of the Helsinki Declaration of 1975, as revised in 2000, and The Code of Ethics of the World Medical Association (Declaration of Helsinki) for experiments involving humans. All participants provided consent and were financially compensated for their participation in the study.

Research Attitudes Questionnaire (RAQ)

We examined participants’ attitudes toward biomedical research using the seven-item Research Attitudes Questionnaire (RAQ-7).¹⁷ The items in the RAQ-7 pertain to trust in science, the benefit of science, and feelings of altruism. Items are scored on a 5-point scale, ranging from 1 (strongly disagree) to 5 (strongly agree), so the total scores ranged from 7 to 35. Higher summed scores indicate more favorable attitudes toward biomedical research and intervention.¹⁷ To summarize the data for further group comparisons, responses to each RAQ item were then recoded into three categories (disagree, neutral, or agree) with answers of disagreement (i.e., “strongly disagree”, and “moderately disagree”) grouped and answers of agreement (i.e., “strongly agree” and “moderately agree”) grouped.

Statistical analysis

We used descriptive statistics to provide information on sample characteristics. Descriptive analyses were performed using SPSS (v27). We calculated means, standard deviations (SDs), and percentages to evaluate response patterns on survey items. Participants who missed more than 10% of the items were removed. After deleting these cases, we replaced missing values by substituting participant-level mean scores for each case, who were missing only 1 of 7 responses to the RAQ. Sixty-five participants (2.55%) were missing 1 of their responses to the RAQ. To examine the intraclass consistency, we calculated McDonald's omega.

We performed planned comparisons of the groups’ distribution of RAQ item responses using chi-square analyses. For most of the RAQ-7 items, we were interested in the percentage of participants in each group who endorsed (i.e., agreed with) the attitude. However, due to a priori predictions, we were interested in comparing groups regarding the proportion of respondents who agreed versus did not agree with the perception of medical research as safe (RAQ item 8).

We tested MI in a structural equation modeling- confirmatory factor analysis (SEM-CFA) framework using Mplus version 8.10.^21,22 A good fitting measurement model is a prerequisite to further interpretations.²³ Therefore, we conducted Exploratory Factor Analysis (EFA) followed by Confirmatory Factor Analysis (CFA) to identify a model solution across the six groups. This CFA model was fitted for each group separately to test for configural invariance, i.e., whether the same CFA was valid in each group. For the CFA, we reported the chi-square test. Still, we also evaluated the goodness-of-fit using the RMSEA and its confidence interval (% CI), comparative fit index (CFI), and the standardized root mean square residual (SRMR) because the χ² tests are sensitive to sample size and might yield misleading findings.¹⁸ After looking at these overall model fit measures, we evaluated MI by subsequently constraining loadings, means, and residuals to equality and examining whether the increased restrictions produced significant changes in the data-model fit. The test for configural invariance was also assessed using the overall fit measures described above. After assessing configural invariance by examining whether the structural model fit was the same in each group, other models needed to be estimated. To investigate the plausibility of the other models where restrictive conditions were compared, e.g., metric, scalar, and strict invariance, we also used relative fit measures, namely, the ΔCFI, ΔRMSEA, and ΔSRMR.²⁴ Metric invariance involved setting the factor loadings to be equal and allowing only the item intercepts to vary to assess whether the factor loadings are equivalent across the groups. Scalar invariance required testing whether the relationship between the latent construct and scale means was equivalent across groups by constraining the item intercepts to be equal. Strict invariance was tested by first constraining factor variances and then constraining error variances. Therefore, in order to examine the performance of the RAQ across age and ethnoracial identity, we evaluated goodness-of-fit statistics and change indices to determine whether conditions of MI were met to satisfy the criteria for varying levels, i.e., configural, metric, scalar, and strict MI.^18,24–29

Results

Participant characteristics

We classified participants into three ethnoracialized categories of Black/African American (AA, n = 1051), American Indian/Alaska Native (AI/AN, n = 646), and Non-Hispanic White (NHW, n = 852) based on their self-identification of their primary ethnicity. The sample was further divided into younger (18 to 29) and older (30 and older) groups, according to age. This age breakdown reflects the nature of our recruitment samples. Some of our recruitment samples (e.g., HBCU college campuses or special interest groups on PWI college campuses) focused on certain age groups, while others (e.g., health fairs, job resource centers) did not. It is noteworthy that most studies using the RAQ have mean ages over 50^4,30 or over 70.^7,9,17 This resulted in six groups: younger AA (n = 457) and older AA (n = 594); younger AI/AN (n = 307) and older AI/AN (n = 339); and younger (n = 173) and older NHW (n = 679) participants.

The participants were drawn from every state in the U.S., along with the District of Columbia. In order to protect the anonymity of the participants, they were asked to provide only the first three digits of their zip codes. Therefore, it was not possible to describe the sample in terms of the percentage of rural versus urban residency.

Table 1 provides the self-reported sociodemographic characteristics of the sample. The ages of the overall sample ranged from 18 to 89, with a mean age of 37.98 (± 14.6). When comparing the proportions of males and females across the three ethnoracialized categories, there were significantly more male (55.2%) respondents than females, χ²(2 )= 1129.17, p < 0.001. Looking more closely at the six groups, we observed significantly greater proportions of female participants in the older AA (p < 0.001) and younger AI/AN groups (p < 0.001) and considerably greater proportions of males in the younger NHW (p < 0.001) and older NHW (p < 0.001) groups. The respondents were heterogeneous in terms of their educational backgrounds. Although approximately 38% of the participants had completed four years of college, a significant proportion (36.0%) had some college education; many younger respondents were still enrolled. Similarly, the sample was diverse regarding marital status, with 42.4% of the respondents being single, never married, and nearly 41% being married.

Table 1.

Sociodemographic characteristics of the sample.

	Total	AA, younger	AA, older	AI/AN, younger	AI/AN, older	NHW, younger	NHW, older
Total N	2549	457	594	307	339	173	679
Age, y, mean ± SD	37.98 ± 14.6	24.32 ± 3.26	51.08 ± 13.3	23.78 ± 3.58	48.54 ± 13.6	25.42 ± 3.52	40.05 ± 8.63
Self-identified Female gender, n (%)	1141 (44.8)	187 (40.9)	329 (55.4)	211 (68.7)	161 (47.5)	51 (29.5)	202 (29.7)
Educational attainment
High school or less, n (%)	670 (26.3)	194 (42.5)	167 (28.1)	135 (44.0)	74 (21.8)	62 (35.8)	38 (5.6)
Some college (13–15), n (%)	918 (36.0)	170 (37.2)	243 (40.9)	129 (42.0)	150 (44.2)	53 (30.6)	173 (25.5)
College or higher, n (%)	961 (37.0)	93 (20.4)	184 (31.0)	43 (14.0)	115 (33.9)	58 (33.5)	468 (68.9)
Marital Status
Single, never married, n (%)	1081 (42.4)	395 (86.4)	206 (34.7)	235 (76.5)	86 (25.4)	124 (71.7)	35 (5.2)
Separated, n (%)	187 (7.3)	9 (2.0)	53 (8.9)	11 (3.6)	34 (10.0)	8 (4.6)	72 (10.6)
Married, n (%)	1046 (41.0)	42 (9.2)	223 (37.5)	56 (18.2)	147 (43.4)	37 (21.4)	541 (79.7)
Widowed, n (%)	62 (2.4)	3 (0.7)	27 (4.5)	1 (0.3)	23 (6.8)	1 (0.6)	7(1.0)
Divorced, n (%)	162 (6.4)	4 (0.9)	85 (14.3)	1 (0.3)	49 (14.5)	1 (6)	22 (3.2)
Other, n (%)	11 (0.4)	4 (0.9)	0 (0)	3 (1.0)	0 (0)	2 (1.2)	2 (0.3)

AA: NonHispanic Black/African American; AI/AN: American Indian/Alaskan Native; NHW: NonHispanic White.

Internal consistency of the RAQ

The RAQ's overall internal consistency was good (ω = 0.825). Table 2 provides the RAQ's intra class consistency (ICC), as measured by McDonald's omega coefficient, for each subgroup. McDonald's ω ranged from 0.82 to 0.85 for AA respondents, 0.82 to 0.85 for AI/AN respondents, and 0.78 to 0.79 for NHW respondents.

Table 2.

Descriptive statistics for RAQ.

	Overall Sample	Younger AA	Older AA	Younger AI/AN	Older AI/AN	Younger NHW	Older NHW
N	2549	457	594	307	339	173	679
RAQ Item 1	3.96 ± 0.89	3.97 ± 0.91	4.01 ± 0.88	4.01 ± 0.88	3.80 ± 0.96	3.98 ± 0.86	3.96 ± 0.86
RAQ Item 3	3.83 ± 0.96	3.73 ± 0.97	3.79 ± 0.90	3.80 ± 0.98	3.73 ± 1.0	3.95 ± 0.97	3.97 ± 0.93
RAQ Item 4	3.66 ± 0.98	3.58 ± 1.1	3.57 ± 1.0	3.62 ± 0.99	3.52 ± 0.97	3.82 ± 0.89	3.83 ± 0.89
RAQ Item 6	4.01 ± 0.90	4.00 ± 0.95	4.05 ± 0.84	4.03 ± 0.93	3.93 ± 0.90	4.15 ± 0.83	3.97 ± 0.92
RAQ Item 8	3.63 ± 0.93	3.52 ± 1.0	3.62 ± 0.89	3.44 ± 0.92	3.53 ± 0.88	3.62 ± 0.97	3.86 ± 0.89
RAQ Item 9	3.75 ± 1.0	3.61 ± 1.1	3.66 ± 1.0	3.78 ± 1.0	3.62 ± 1.0	3.86 ± 0.99	3.94 ± 0.88
RAQ Item 11	3.89 ± 0.89	3.81 ± 0.99	3.84 ± 0.86	3.87 ± 0.90	3.78 ± 0.90	4.02 ± 0.83	4.01 ± 0.82
RAQ Total score
Mean	26.73	26.21	26.55	26.54	25.91	27.40	27.54
± SD	4.57	4.89	4.61	4.60	4.85	4.18	4.11
ICCω	0.825	0.824	0.849	0.822	0.853	0.783	0.786

RAQ: Research Attitudes Questionnaire.¹⁷ ICCω: intra-class consistency as measured by McDonald's omega coefficient.

Attitudes about research

Figure 1 depicts the distribution of the total RAQ scores across the entire sample. Across the sample, the overall mean score was 26.73 (± 4.57), a median of 28.00, and a mode of 28. Histograms of the total RQ scores for each group can be found in the Supplemental Material. Table 2 provides the mean scores (and standard deviations) for each RAQ item by group, along with the mean total RAQ scores and standard deviations for each group.

Figure 1.

Distribution of total RAQ scores.

We observed no significant findings when we computed a correlation between age and total RAQ scores on the sample of 2549 participants, r = −0.024, n.s. Between -group comparisons of the total RAQ scores revealed significant group differences. The older NHW group had significantly higher total RAQ scores than the younger AA (p < 0.001), older AA (p < 0.001), younger AI/AN (p < 0.001), and older AI/AN (p < 0.001) groups. The younger NHW group had significantly higher total RAQ scores than the younger AA (p = 0.002), younger AI/AN (p = 0.019) and older AI/AN (p < 0.001) groups. No other group differences reached statistical significance. It is noteworthy that parametric testing revealed the same pattern of significant differences, i.e., a one-way ANOVA of total RAQ scores revealed significant group differences, F(5, 2543) = 8.71, p < 0.0001, eta-squared = 0.017. The less powerful method of nonparametric testing was chosen due to the skewed distribution of the RAQ scores.

Distribution of research attitude responses by ethnoracial groups

Table 3 provides the distribution of RAQ item responses for the entire sample and for each ethnoracial group. The groups differed significantly in terms of the proportion of their endorsement of RAQ item 1, “I have a positive view about medical research in general”, χ² (10) = 34.36, p < 0.001. They also differed significantly in terms of their endorsement of RAQ item 3, “Medical researchers can be trusted to protect the interests of people who take part in their research studies, χ² (10) = 51.17, p < 0.001. Regardless of age, more of the NHW participants (over 70%) agreed with the statement compared to the AI/AN groups (62–65% agreement) and AA groups (62–64% agreement).

Table 3.

Distribution of RAQ Responses by Group^a.

Level of Agreement
RAQ Item^b	Response^c	Overall Sample (N = 2549)	AA Younger (N = 457)	AA Older (N = 594)	AI/AN Younger (N = 307)	AI/AN Older (N = 339)	NHW Younger (N = 173)	NHW Older (N = 679)
1. I have a positive view about medical research in general.	Disagree	5.6	4.4	4.5	5.5	7.1	6.9	6.5
	Neutral	18.6	23.0	18.7	17.9	25.1	16.8	13.1
	Agree	75.8	72.7	76.8	76.5	67.8	76.3	80.4
3. Medical researchers can be trusted to protect the interests of people who take part in their research studies.	Disagree	8.0	9.4	6.4	9.8	10.3	7.5	6.6
	Neutral	24.3	28.0	29.5	24.8	27.4	22.0	16.2
	Agree	67.7	62.6	64.1	65.5	62.2	70.5	77.2
4. We all have some responsibility to help others by volunteering for medical research.	Disagree	11.1	15.1	12.6	11.7	13.0	5.8	7.2
	Neutral	27.4	28.0	32.0	32.9	31.6	23.7	19.4
	Agree	61.5	56.9	55.4	55.4	55.5	70.5	73.3
6. Society needs to devote more resources to medical research.	Disagree	5.4	6.8	3.9	6.8	5.0	4.0	5.7
	Neutral	17.9	17.7	17.2	15.3	23.9	15.6	17.4
	Agree	76.7	75.5	79.0	77.9	71.1	80.3	76.9
8. Participating in medical research is generally safe.	Disagree	9.8	13.8	7.6	14.0	10.3	10.4	6.6
	Neutral	32.6	35.4	37.5	38.4	36.6	32.4	21.8
	Agree	57.6	50.8	54.9	47.0	53.1	57.2	71.6
9. If I volunteer for medical research, I know my personal information will be kept private and confidential.	Disagree	10.1	14.2	10.8	9.4	12.7	10.4	5.7
	Neutral	24.7	27.6	29.1	26.7	26.8	22.0	17.5
	Agree	65.2	58.2	60.1	63.8	60.5	67.6	76.7
11. Medical research will find cures for many major diseases during my lifetime.	Disagree	5.9	8.8	4.1	5.2	6.8	4.6	5.2
	Neutral	21.9	24.7	27.1	25.7	27.1	16.2	12.4
	Agree	72.3	66.5	68.2	69.1	66.1	79.2	82.5

Attitudes toward research as measured by the Research Attitudes Questionnaire (RAQ).¹⁷

AA: Black/African American; AI/AN: American Indian/Alaska Native; NHW: NonHispanic White.

RAQ item numbers pertain to the item numbers on the original measure. The RAQ is scored on a 5-point Likert-type scale; strongly disagree to strongly agree.

Percent of respondents who disagreed, were neutral, or agreed with the RAQ items. “Strongly disagree” and “disagree” were collapsed, and “Strongly agree” and “agree” were collapsed.

RAQ item 4, “We all have some responsibility to help others by volunteering for medical research” also elicited differential responses from the three ethnoracialized groups, χ² (10) = 76.20, p < 0.001. More of the younger (70.5%) and older (77.2%) NHW participants agreed with RAQ item 4 compared to the AA participants (63–64%) and AI/AN participants (62–66%). Overall, most (77%) of the respondents agreed with the sentiment expressed in RAQ item 6 namely, “Society needs to devote more resources to medical research.” The ethnoracial groups did not differ significantly on this RAQ item, χ² (10) = 17.16, p > 0.05.

The groups responded differentially to RAQ item 8, “Participating in medical research is generally safe”, χ² (10) = 91.95, p < 0.001. More (71.6%) older NHW participants agreed about the safety of medical research than any of the other groups, where the proportion of participants who expressed agreement ranged from 47.0–57.2 percent. Collapsing the “neutral” and “disagree” responses into a “does not agree” category, we observed a statistically significant between-group difference in terms of the proportion of participants who agreed with RAQ item 8 and those who did not agree, χ² (5) = 80.36, p < 0.001.

We also observed group differences in response to RAQ item 9, which examined trust that one's personal information would be kept private and confidential, χ² (10) = 66.16, p < 0.001. RAQ item 11 probed respondents’ optimism regarding the outcome of medical research. The groups displayed an overall difference in their response to the statement “Medical research will find cures for many major diseases during my lifetime”, χ² (10) = 72.34, p < 0.001. More of the younger and older NHWs expressed optimism than the other participants, though over 65% of each participant group endorsed the attitude.

EFA and CFA for the whole sample

First, we performed an exploratory factor analysis (EFA) with oblique geomin rotation on the entire sample of 2549 participants with a focus on identifying the number of factors. Table 4 provides the results of the EFA; we found acceptable one-factor and two-factor solutions. Using the criteria of good model fit, model solution with clear factor structure, and model parsimony,³¹ we chose the one-factor solution. This factor structure was next confirmed by confirmatory factor analysis (CFA), as evidenced by χ² (14) = 125.03, RMSEA = 0.056 (90% CI: 0.047 to 0.065), CFI = 0.977, and SRMR = 0.023. These results indicated a good fit, with the RMSEA less than 0.06, CFI greater than 0.95, and SRMR less than 0.05.^26–28,32

Table 4.

Factor loadings, χ2 test, and model fit indices for the exploratory factor analysis with geomin rotation for the total sample (N = 2549).

	One Factor 1	Two Factor 1	2
Factor loadings
RAQ1	0.628*	0.676*	0.003
RAQ3	0.711*	0.484*	0.275*
RAQ4	0.642*	0.300*	0.392*
RAQ6	0.538*	0.599*	−0.024
RAQ8	0.643*	−0.002	0.735*
RAQ9	0.660*	0.280*	0.430*
RAQ11	0.613*	0.397*	0.258*
χ² test
χ²	125.030	51.831
df	14	8
p-value	<0.001	<0.001
Model fit indices
CFI	0.997	0.991
RMSEA	0.056	0.046
(90%CI)	(0.047, 0.065)	(0.035, 0.059)
SRMR	0.029	0.018

Factor loadings > 0.50 are in boldface. CFI: comparative fit index; RMSEA: root mean squared error of approximation; SRMR = standardized root mean squared residual. Model fit is considered adequate by meeting the following criteria: CFI ≥ 0.95, RMSEA ≥ 0.08, and SRMR ≥ 0.08.

Configural invariance across all six groups: CFA for each group

Next, we evaluated the model in all six groups. That is, we repeated the CFA process to evaluate invariance across age and ethnoracial groups by testing the one-factor model separately for each of the six groups. The results are provided in Table 5. The fit of the model was very good to excellent for the younger AA group and younger NHW groups, because the chi-square/df values were under 3.00, RMSEA values less than 0.05, and SRMR values less than 0.05. with the RMSEA less than 0.06, CFI greater than 0.95, and SRMR, an absolute difference between the observed correlation and the model-predicted correlation, was less than 0.05.^27,29 The model fit was relatively good for the younger AI/AN group, because the chi-square/df value was under 3.00, the RMSEA was less than 0.08, the CFI value was greater than 0.95, and the SRMR value was less than 0.05.

Table 5.

Results of CFA for each of the six groups.

	Younger AA	Older AA	Younger AI/AN	Older AI/AN	Younger NHW	Older NHW
n	457	594	307	339	173	679
χ² test
χ²	22.989	64.058	31.462	37.201	19.799	60.642
df	14	14	14	14	14	14
p	0.0604	0.0000	0.0048	0.0007	0.1366	0.0000
Model fit indices
CFI	0.990	0.965	0.970	0.971	0.976	0.954
RMSEA	0.037	0.078	0.064	0.070	0.049	0.070
(90% CI)	(0.000–0.064)	(0.059–0.097)	(0.034–0.094)	(0.042–0.104)	(0.000–0.095)	(0.053–0.089)
SRMR	0.022	0.031	0.033	0.031	0.040	0.033

The results of the confirmatory factor analysis (CFA) for each of the following groups: younger and older African-Americans (AA); younger and older American Indian/Alaska Native (AI/AN); and younger and older nonHispanic White (NHW) respondents. Note: The degrees of freedom (df) for each of the groups was 14. The goodness-of-fit statistics are as follows: chi-square statistic (χ²) including degrees of freedom (df) and level of statistical significance (p-value); comparative fit index (CFI); root-mean-square error approximation (RMSEA) with accompanying confidence interval (90% CI); and SRMR = standardized root mean square residual.

Model fit is considered adequate if it meets the following criteria: CFI ≥ 0.95, RMSEA ≤ 0.08, and SRMR ≤ 0.08.

The data-model fit appeared reasonably good for the older AI/AN group, with the chi-square/df values under 3.00, RMSEA less than 0.08, CFI greater than 0.95 and the SRMR less than 0.05. Based on the fit criteria, the model fit for the older AA and older NHW groups was not a close fit, though it was acceptable. For both groups, the chi-square/df values exceeded 3.00 and the value of the RMSEA was closer to 0.08 than that of any of the other comparison groups. However, like the other four groups, the SRMR was less than 0.05, suggesting an overall good fit. The data indicate that configural invariance was supported across the groups.

Two-group CFAs for testing measurement invariance across age

Next, we examined whether there was MI in terms of age by performing a series of comparisons of older versus younger participants within each ethnoracial classification. That is, for each of the NHW, AA, and AI/AN groups, we examined whether there was configural, metric, scalar, and strict invariance within each of those groups across age. Table 6 presents the results of the two-group invariance test of age for the NHW participants. Nested models were tested to determine whether components in the factorial structure of the RAQ were operating equivalently across age. Since the sample was too large for the Chi-square tests to be insignificant, we used fitness indicator difference tests.^19,24,33 The model (M1) showed good to acceptable fit, suggesting that configural invariance was supported: Chi-square/df = 2.87, CFI greater than 0.95, RMSEA less than 0.08 and SRMR less than 0.05. Configural invariance is important to establish a comparative baseline for subsequent models. The test for metric invariance (Model 2) examined whether the factor loadings were equivalent across the groups. The model was further constrained, i.e., setting the factor loadings to be equal and allowing only the item intercepts to vary. When the difference in the fitness indicators is smaller than 0.01, it indicates no significant difference. When the difference is between 0.01 and 0.02, it indicates there is a slight difference. If the difference in fitness indicators is larger than 0.02, it indicates a significant difference. As can be seen in Table 6, ΔCFI, ΔRMSEA, and ΔSRMR between the constrained (M2) and unconstrained model (M1) all met the threshold, indicating that metric invariance was supported. Scalar invariance (Model 3) was tested by constraining item intercepts to equality and then comparing the resultant fit statistics in Model 3 to those obtained in Model 2. As we constrained more and more parameters, moving from Model 2 to Model 3, the fitness indices only changed slightly. Therefore, scalar invariance was supported. Strict invariance was evaluated at two levels, namely, at the level of factor variance and at the level of residual error invariance. We found evidence to support MI at the level of factor invariance: the change indices were all below the threshold values for noninvariance. However, the changes in the fitness indices (Δ CFI > 0.02 and Δ SRMR > 0.02) ruled out invariance at the level of residual error.

Table 6.

Tests of measurement invariance: analysis of younger and older NHW participants.

Model	χ2/df (df)	CFI	RMSEA	RMSEA (90% CI)	SRMR	ΔCFI	ΔRMSEA	ΔSRMR
M1.Configural invariance^a	2.87 (28)	0.959	0.066	(0.050–0.084)	0.035	—	—	—
M2. Metric invariance^b	2.59 (34)	0.957	0.061	(0.046–0.077)	0.049	0.002	0.005	0.014
M3. Scalar invariance^c	2.69 (41)	0.945	0.063	(0.049–0.077)	0.061	0.012	0.002	0.012
M4. Strict invariance^d
M4a. Factor invariance	2.63 (42)	0.946	0.062	(0.048 −0.076)	0.061	0.001	0.001	0.000
M4b. Residual error invariance	2.41 (49)	0.946	0.057	(0.044–0.071)	0.087	0.025	0.004	0.032

N = 852; 173 younger and 679 older NHW participants.

For testing measurement invariance, we tested for configural (form) invariance (Model 1).

We tested for metric (weak factorial) invariance by constraining the factor loadings to be equal across groups (Model 2).

We tested for scalar (strong factorial) invariance by comparing the latent means, and constraining the item intercepts to

be the same across groups (Model 3).

We restricted the error variances to equal to test for strict (factor and residual or invariant uniqueness) invariance (Models 4a

and 4b).

Goodness of fit indexes: χ²= chi-square statistic and df = degrees of freedom; CFI = Comparative Fit Index;

RMSEA = Root Mean Square Error of Approximation;

RMSEA (90% CI) RMSEA 90% confidence interval; SRMR = standardized root mean square residual.

Change indices are as follows: ΔCFI = change in Comparative Fit Index

ΔRMSEA = change in Root Mean Square Error of approximation; and ΔSRMR = change in standardized root mean square residual.

Table 7 presents the results of the two-group invariance test of age for the AA participants. We followed the same procedure as before, i.e., using nested models to determine whether components in the factorial structure of the RAQ were operating equivalently across age in the AA group. The model (M1) showed good to acceptable fit, suggesting that configural invariance was supported: Chi-square/df = 3.11, CFI greater than 0.95, RMSEA less than 0.08 and SRMR less than 0.05. The tests for metric invariance (Model 2) and scalar invariance (Model 3) were considered acceptable. Based on the change in the statistics from the scalar model to the model of factor invariance (M4a), there was evidence to support factor invariance. There was a significant difference in the fitness statistics as we moved from Model 4a to Model 4b, indicating that there was no longer evidence of MI at the level of residual error.

Table 7.

Tests of measurement invariance: analysis of younger and older AA participants.

Model	χ2/df	CFI	RMSEA	RMSEA (90% CI)	SRMR	ΔCFI	ΔRMSEA	ΔSRMR
M1.Configural invariance^a	3.11	0.974	0.063	(0.049–0.079)	0.027	—	—	—
M2. Metric invariance^b	2.75	0.974	0.058	(0.044–0.072)	0.040	0.000	0.005	0.013
M3. Scalar invariance^c	2.38	0.975	0.051	(0.038–0.064)	0.041	0.001	0.007	0.001
M4. Strict invariance^d
M4a. Factor invariance	2.34	0.976	0.050	(0.038–0.064)	0.050	0.001	0.001	0.009
M4b. Residual error invariance	3.54	0.946	0.070	(0.058–0.081)	0.121	0.03	0.02	0.071

N = 1051; 457 younger and 594 older AA participants.

For testing measurement invariance, we tested for configural (form) invariance (Model 1).

We tested for metric (weak factorial) invariance by constraining the factor loadings to be equal across groups (Model 2).

We tested for scalar (strong factorial) invariance by comparing the latent means, and constraining the item intercepts to

be the same across groups (Model 3).

and repeating the process, i.e., comparing the latent means and constraining the item intercepts to be the same across groups (Model 3₁).

We restricted the error variances to equal to test for strict (factor and residual or invariant uniqueness) invariance (Models 4a and 4b).

Goodness of fit indexes: χ2 = chi-square statistic and df = degrees of freedom; CFI = Comparative Fit Index;

RMSEA = Root Mean Square Error of Approximation;

RMSEA (90% CI) RMSEA 90% confidence interval; SRMR = standardized root mean square residual.

Change indices are as follows: ΔCFI = change in Comparative Fit Index;

ΔRMSEA = change in Root Mean Square Error of approximation; and ΔSRMR = change in standardized root mean square residual.

The results of the two-group invariance test of age for the AI/AN participants are presented in Table 8. First, in Model 1, we tested whether there was evidence of configural invariance. As shown by the goodness-of-fit statistics, there was evidence for configural invariance across age in the AI/AAN group. Moving through nested models with increasing constraints, we found evidence for metric and scalar invariance as well. In the AI/AN participant group, when we restricted the error variances to equal, we found support for both factor invariance and residual error invariance. There were no significant differences in the magnitude of the change observed in CFI, RMSEA, and SRMR, as we moved from Model 4a to Model 4b.

Table 8.

Tests of measurement invariance: analysis of younger and older AI/AN participants.

Model	χ2/df	CFI	RMSEA	RMSEA (90% CI)	SRMR	ΔCFI	ΔRMSEA	ΔSRMR
M1.Configural invariance^a	2.45	0.9710	0.067	(0.047–0.087)	0.032	—	—	—
M2. Metric invariance^b	2.19	0.971	0.061	(0.042–0.079)	0.045	0.000	0.006	0.013
M3. Scalar invariance^c	2.31	0.962	0.064	(0.047–0.081)	0.058	0.009	0.003	0.013
M4. Strict invariance^d
M4a. Factor invariance	2.28	0.962	0.063	(0.046–0.080)	0.072	0.000	0.001	0.014
M4b. Residual error invariance	2.29	0.955	0.063	(0.048–0.079)	0.073	0.007	0.000	0.001

Note. N = 646; 307 younger and 339 older AI/AN participants.

For testing measurement invariance, we tested for configural (form) invariance (Model 1).

We tested for metric (weak factorial) invariance by constraining the factor loadings to be equal across groups (Model 2).

We tested for scalar (strong factorial) invariance by comparing the latent means, and constraining the item intercepts to

be the same across groups (Model 3).

We restricted the error variances to equal to test for strict (factor and residual or invariant uniqueness) invariance (Models 4a

and 4b).

Goodness of fit indexes: χ² =chi-square statistic and df = degrees of freedom; CFI = Comparative Fit Index;

RMSEA = Root Mean Square Error of Approximation; RMSEA (90% CI) RMSEA 90% confidence interval; SRMR = standardized root mean square residual.

Change indices are as follows: ΔCFI = change in Comparative Fit Index;

ΔRMSEA = change in Root Mean Square Error of approximation; and ΔSRMR = change in standardized root mean square residual.

Three-group CFAs for testing measurement invariance across ethnoracial identity

Following the results of the two-groups CFAs, three-group CFAs were next tested for factorial invariance across race/ethnicity. In this set of analyses, younger and older participants are combined for each of the three ethnoracialized groups. Results are summarized in Table 9. The analytic strategy for evaluating invariance testing involved the four different levels forming the nested hierarchy used previously (i.e., configural, metric, scalar, and strict invariance). We did not consider the chi-square test for this series of analyses, which included the total sample of 2549 participants, because the chi-square test is not suitable for evaluating the hypothesis of configural invariance in the case of large sample sizes.^24,33 Additionally, for this set of analyses, given the overall larger sample size, the criteria for goodness-of-fit and criteria for change indices were somewhat more liberal; for SRMR, the cutoff for SRMR ≤ 0.05, and for ΔSRMR, the cutoff criterion was set at no larger than 0.030.^24,28

Table 9.

Tests of measurement invariance across ethnoracial group identity.

Model	χ2/df (df)	CFI	RMSEA	RMSEA (90% CI)	SRMR	ΔCFI	ΔRMSEA	ΔSRMR
M1.Configural invariance^a	5.90 (49)	0.951	0.076	(0.068–0.085)	0.050	—	—	—
M2. Metric invariance^b	3.63 (54)	0.971	0.056	(0.047–0.064)	0.039	0.02	0.02	0.011
M3. Scalar invariance^c	4.83 (68)	0.947	0.067	(0.060- 0.074)	0.068	0.024	0.011	0.029
M3₁. Partial Scalar invariance	4.50 (49)	0.955	0.064	(0.056–0.073)	0.069	0.016	0.008	0.030
M4. Strict invariance^d
M4a. Factor variance	4.96 (70)	0.943	0.068	(0.061–0.076)	0.119	0.004	0.001	0.051
M4b. Residual error variance	4.84 (84)	0.934	0.067	(0.061–0.074)	0.134	0.009	0.001	0.015

N = 2549; three groups.

For testing measurement invariance, we tested for configural (form) invariance (Model 1).

We tested for metric (weak factorial) invariance by constraining the factor loadings to be equal across groups (Model 2).

We tested for scalar (strong factorial) invariance by comparing the latent means, and constraining the item intercepts to

be the same across groups (Model 3). We subsequently tested for partial scalar invariance by omitting an item (RAQ item 8),

and repeating the process, i.e., comparing the latent means and constraining the item intercepts to be the same across groups (Model 3₁).

We restricted the error variances to equal to test for strict (factor and residual or invariant uniqueness) invariance (Models 4a

and 4b).

Goodness of fit indexes: χ² = chi-square statistic and df = degrees of freedom; CFI = Comparative Fit Index;

RMSEA = Root Mean Square Error of Approximation; RMSEA (90% CI) RMSEA 90% confidence interval; SRMR = standardized root mean square residual. Change indices are as follows: ΔCFI = change in Comparative Fit Index

ΔRMSEA = change in Root Mean Square Error of approximation; and ΔSRMR = change in standardized root mean square residual.

The initial step in testing for configural invariance (Model 1) required that no equality constraints be imposed on the parameters. CFA estimation of this model met the following goodness-of-fit criteria: CFI ≥ 0.95, RMSEA < 0.08, and SRMR ≤ 0.05. Thus, the findings supporting invariance were acceptable. The test for metric invariance (Model 2) was further constrained, i.e., setting the factor loadings to be equal and allowing only the item intercepts to vary. The results of the CFA provided good evidence of multi-group model fit: CFI > 0.95, RMSEA < 0.06, and SRMR < 0.05. The change in the fit indices from Model 1 to Model 2 were at threshold for significance for CFI and RMSEA, though there was no significant difference for change in SRMR. These results were interpreted as plausible evidence for metric invariance across ethnoracial identity.

Then we tested Model 3 (scalar invariance) to check whether it would be appropriate to compare the groups’ latent mean scores.¹⁸ Scalar invariance was tested by constraining item intercepts to equality and then comparing the resultant fit statistics in Model 3 to those obtained in Model 2. The goodness-of-fit indicators and the model fit comparison (based on change indices) between Model 3 to 2 suggested noninvariance. Although RMSEA was less than 0.08, CFI was less than 0.95, and SRMR was considerably greater than 0.05. The change in CFI exceeded the cutoff of 0.02. Because the overall model fit was worse in the scalar invariance model (M3) compared to the metric invariance model (M2) we suspected that at least one item intercept differed across the groups.¹⁹

Given the earlier findings of group discrepancies on RAQ item 8, we suspected that this item 8 was the source of noninvariant intercepts. We omitted item 8 and retested the scalar invariance model as a potential alternative (Model 3₁). The comparison between Model 3₁ and Model 2 indicated acceptable goodness-of-fit: CFI > 0.95 and RMSEA < 0.08, though the SRMR remained greater than 0.05, suggesting a less-than-ideal fit. The change indices in Model 3₁ were in the good-to-acceptable range: ΔCFI < 0.02, ΔRMSEA < 0.01, ΔSRMR ≤ 0.03. Thus, the Model 3₁ -Model 2 comparison supported evidence for partial scalar invariance across ethnoracial identity.

Strict invariance was evaluated at the level of factor invariance (Model 4a) and residual error invariance (Model 4b) by constraining factor variances and error variances, respectively. In both cases, the values of CFI and SRMR were beyond acceptability. Strict invariance across ethnoracial identity was not supported.

Discussion

The RAQ has been used extensively in ADRD research. The RAQ has been administered to various types of samples, including community elders,^7,11,30 relatives of individuals with ADRD,⁸ relatives of individuals with Down syndrome,^34,35 registry participants,⁹ randomized survey panel participants,^4,14 lay participants attending scientific dementia conferences,³⁶ and individuals receiving outpatient medical care.³⁷

However, the interpretation of the research has heretofore been limited by several factors. First, most investigations of study willingness have under-included minoritized (e.g., non-Hispanic Black, American Indian/Alaska Native, Asian, and Latinx) communities. Another limitation is that the extant studies have relied upon registry or convenience samples (i.e., individuals already engaged in research, either by being a relative of an enrolled participant or being a participant themselves). Finally, while research has focused on ethnoracial differences based on the comparison of total RAQ scores or relative endorsements of RAQ items, there has been less attention to the suitability of the RAQ to adequately assess key components of willingness to engage in biomedical research.

Until approximately 2015, the RAQ had been administered predominantly to non-Hispanic White samples in the community and in research settings. Increasingly, investigators have administered the RAQ to more ethnoracially diverse samples. Until now, the appropriateness of the RAQ for community-based contexts with under-included groups in research was assumed but had not been empirically tested. The aim of the present investigation was to examine the MI of the RAQ, i.e., the suitability of the RAQ across age groups and different racialized identities. Using Confirmatory Factor Analysis and MI analysis, we sought to address the gaps in the extant research by comparing historically under-included groups on the 7-item RAQ.

Most of the comparisons regarding RAQ scores^7,11,36,37 consisted of relatively small samples of 250 or fewer respondents. Larger samples have enrolled predominantly NonHispanic White^34,35 or African American⁴ samples. There has been a dearth of survey research on attitudes toward research using the RAQ that included more than a few American Indian/Alaska Native participants. The present study is the largest and most representative examination of the RAQ in community-based individuals, including a relatively large group of American Indian/Alaska Native participants. The individuals surveyed varied in terms of age, gender, racialized identities, educational attainment, marital status, and geographical location.

Few extant studies^35,37 include a sample extending over a broad age range of 18 and above. In the Courtright et al.³⁷ investigation, most of the sample was older than 50. Although one group³⁵ observed an age effect, whereby higher mean RAQ scores were associated with increasing age of respondents, only 16 percent of the participants were under 45.

The present study included a larger proportion of respondents across a broader age range than prior studies. Our a priori hypothesis was generally confirmed that RAQ scores would be lower amongst members of historically under-included and underrepresented communities, i.e., AA and AI/AN participants. Overall, the NHW groups had significantly higher total RAQ scores than the others.

Comparison with prior findings

The psychometric characteristics of RAQ in this large, diverse sample were consistent with those reported in the original validation report.¹⁷ The reported internal consistency for the 7-item RAQ was good (Cronbach's α = 0.81;¹⁷ similarly, in the present sample, the internal consistency for the measure was good (McDonald's ω = 0.825; Cronbach's α = 0.825). Overall, the RAQ findings from the present study are consistent with those from earlier studies (e.g.,^11,36), indicating that data privacy and confidentiality concerns are salient for at least one-third of RAQ respondents. A significant proportion (35.1–37.4%) of our AA participants responded in a negative or neutral pattern when queried about trusting medical researchers, consistent with other investigations of minoritized samples.¹¹ We extended this area of research by observing that a comparable proportion of (34.6–37.7%) AI/AN respondents also shared mistrust of medical researchers.

While there were also significant concerns regarding the safety of participating in medical research, this concern was observed in younger NHW participants, along with younger and older AA participants and younger and older AI/AN participants. We observed that 42.8 to 52.4% of the younger and older members of the AA and AI/AN groups responded neutrally or negatively to the RAQ item 8, consistent with the findings of Neugroschl and colleagues,¹¹ whose sample consisted entirely of Latino or African American participants with a mean age of 72.6 years. Our observation of the historically marginalized groups’ lack of expressed confidence in the safety of medical research participation, regardless of age, is consistent with a shared cultural mistrust of biomedical research, at least partly reflecting knowledge of past misdeeds against their community. The relatively high proportion of the younger NHW group's disbelief in the general safety of medical research participation may reflect their growing awareness of the differential treatment of various society members in research, e.g., learning about the Tuskegee Untreated Syphilis Study in high school or college.

Measurement invariance across age and ethnoracial identity

MI was confirmed across age for the NHW and AA groups at the scalar invariance level. For the AI/AN group, MI was confirmed across age at the strict invariance level. Partial invariance was found across ethnoracial identity, highlighting the potential challenges inherent in measuring attitudes and beliefs about research participation in individuals with different cultural backgrounds and historical experiences. We can confidently assert that the 7-item RAQ items are measuring the same constructs across groups, and that the same items have the same factor loadings. However, we advise investigators to check the intercepts across groups, particularly in terms of the items referring to the perceived safety of medical research participation.

One of the strengths of the present study is the relatively large number of younger participants who were administered the RAQ; this may be one of the largest samples of younger participants administered the RAQ to date. We believe that this is significant for two reasons. First, the younger respondents may play a role in influencing whether their older relatives choose to participate in ADRD research, particularly if the younger respondents are involved in a caretaking role. Secondly, learning about the younger participants’ attitudes toward biomedical research may help provide an earlier period for educational intervention and attitudinal change to occur.

Limitations

Despite its many strengths, the present study has limitations. First, we had unbalanced, unequal groups for our invariance testing. By evaluating multiple test indices, including overall model fit indices, we sought to yield the most robust findings possible. For this investigation, we relied solely upon web-based survey administration. Thus, our sample was restricted to individuals who were comfortable taking online surveys and who were able to access the internet. Another limitation of our investigation is that we did not obtain data regarding respondents’ prior involvement in clinical research studies. As indicated by others,^34,36 survey respondents with prior research participation had higher average RAQ scores compared to research-naïve respondents. However, we note that the impetus for the umbrella project (of which the present investigation is an offshoot) was to explore underlying motivations for underincluded groups’ willingness to engage in research. Many of the survey respondents included in our sample are likely research naïve. Nonetheless, we conducted the investigation with people who were sufficiently open to the idea of research to answer our paid survey. It is possible that other members of the groups we sampled from may have been less willing to participate and/or had less positive attitudes toward research.

Our groups had unbalanced ns, in terms of gender. This precluded our testing of MI across gender identity. We recommend increasing efforts to recruit more participants from various gender identities in future investigations. The present sample consisted solely of participants who self-identified as being primarily nonHispanic White, Black/African American, or American Indian/Alaska Native. A limitation of the present investigation is our failure to include any measures of cultural accommodation or assimilation in the online assessment. We acknowledge that none of these groups are monolithic. It would be presumptuous to make any assumptions about the within-group homogeneity of any of the racialized groups that we studied. Unfortunately, the present study does not permit us to lend insights regarding intersectionality between ethnoracial identity and acculturation, or other factors such as employment, and their potential impact on attitudes toward research.

This online investigation was conducted in the United States. These findings may not generalize to individuals living in low- or middle-income countries. Furthermore, we cannot extrapolate about the suitability of the RAQ for other ethnoracial groups not surveyed in the present investigation, such as nonHispanic Asians, or Latine/Hispanic individuals. Sewell et al.⁷ examined responses to the RAQ in two homogeneous samples of elderly Latine community members. A comparable study of the RAQ, including a diverse sample of Latine community members, is needed; ideally, the measure would be made available in Spanish and English to facilitate the assessment of attitudes toward research. It is noteworthy that Sewell et al.⁷ translated the RAQ, and it is now available in both Spanish and English.

Future directions

Further areas for research would include investigating more factors that may affect MI, such as immigration status, acculturation, and literacy level, in addition to ethnoracial identity and age. Additionally, it would be interesting to assess whether there would be an interaction effect between the means of recruitment (e.g., physician referral, word-of-mouth, community-based) and the ethnoracial identity of the target sample in terms of how a given measure is interpreted, and how the items load onto the factors, i.e., will there be MI under different conditions of intersectionality?

What does it mean to say that the RAQ-7 is invariant across different groups? If the RAQ-7 is invariant, i.e., has equivalent meaning across groups, then differences in scores are less likely to be attributable to particular groups’ interpreting the items differently. This is important to acknowledge because it suggests that mean group differences on total scale scores and/or on particular items focusing on trust in biomedical research and feelings of safety are valid differences, rather than reflecting response biases. Indeed, we believe that ethnoracial group differences in terms of the RAQ likely reflect differential cultural histories and lived experiences between the groups. We noted the largest group differences in terms of RAQ item 8, which pertained to issues of safety while participating in medical research.

We also view these findings as motivation to further explore the nuanced experiences of feeling safe in biomedical research across community-based groups and to work towards addressing the underlying issues inherent therein. This is a particularly salient issue as further advances in preclinical biomarker testing for ADRD and clinical trials continue. It is imperative that further understanding of historically underrepresented groups’ attitudes towards research is gained to mitigate the barriers to participation. There will be a continued push to be more inclusive in our research to ensure that all groups benefit from the advances in knowledge and treatment.

Conclusions

In summary, when we compared across age groups and 3 ethnoracialized groups, we observed evidence of configural invariance, metric invariance, and partial scalar invariance. We found little evidence of strict invariance (also known as residual or invariant uniqueness) for the RAQ-7. There is a consensus that demonstrating configural, metric, and scalar invariance is sufficient for establishing MI.^18,19,29 Therefore, we can assert that RAQ-7 shows evidence of MI across nonHispanic White, nonHispanic Black/African American, and American Indian/Alaskan Native individuals living in the United States.

Supplemental Material

sj-docx-1-alr-10.1177_25424823251361513 - Supplemental material for Measurement invariance of the Research Attitudes Questionnaire in an age and ethnically diverse, community-dwelling sample

Supplemental material, sj-docx-1-alr-10.1177_25424823251361513 for Measurement invariance of the Research Attitudes Questionnaire in an age and ethnically diverse, community-dwelling sample by Diane Carol Gooding, C Malik Boykin, Carol Ann Van Hulle, Shenikqua Bouges, Jordan P Lewis, Susan Flowers Benton, Fabu P Carter and Carey E Gleason in Journal of Alzheimer's Disease Reports

Footnotes

Acknowledgements

We wish to thank all the individuals who participated in the UBIGR project. We also appreciate the following groups and individuals who assisted us in publicizing our study in their communities and/or at their institutions: the Black Leaders for Brain Health, and Drs. Emre Umucu, Neelum Aggarwal, and Yuri B. Saalmann. Thanks also to Drs. Jonathan Gallimore, Rebecca Addington, Madeline Pflum, and Chelsea Andrews for their assistance with study recruitment. Some of these data were presented at the 2024 Alzheimer's Association International Conference.

ORCID iDs

Diane Carol Gooding

C Malik Boykin

Shenikqua Bouges

Susan Flowers Benton

Carey E Gleason

Ethical considerations

The study was approved by the University of Wisconsin-Madison Educational and Social Sciences Human Subjects Board (approval #2021-1500) on 1/21/2022 and (approval #2022-0336) on 3/30/2022.

Consent to participate

All human subjects provided consent to participate.

Author contributions

Diane Carol Gooding (Conceptualization; Data curation; Formal analysis; Funding acquisition; Investigation; Methodology; Project administration; Resources; Software; Writing – original draft); C Malik Boykin (Formal analysis; Methodology; Software; Validation; Writing – review & editing); Carol Ann Van Hulle (Writing – review & editing); Shenikqua Bouges (Writing – review & editing); Jordan P Lewis (Writing – review & editing); Susan Flowers Benton (Writing – review & editing); Fabu P Carter (Project administration; Resources); Carey E Gleason (Conceptualization; Funding acquisition; Supervision; Writing – review & editing).

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was supported by the Department of Medicine Intramural Faculty Competition Grant (CEG, DCG) and the Leon Epstein Faculty Research Award (DCG). Additional funding was provided by NIH-NIA R01AG054059 (CEG, PI).

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data availability statement

The data supporting the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.

Supplemental material

Supplemental material for this article is available online.

References

Zhou

Elashoff

Kremen

, et al. African Americans are less likely to enroll in preclinical Alzheimer’s disease clinical trials. Alzheimers Dement (N Y) 2017; 3: 57–64.

Gilmore-Bykovskyi

Jin

Gleason

, et al. Recruitment and retention of underrepresented populations in Alzheimer’s disease research: a systematic review. Alzheimers Dement (N Y) 2019; 5: 751–770.

Gooding

Carter

Umucu

, et al. Factors affecting the willingness of African American and American Indian/Alaska native communities to engage in genetic and biomarker research: the UBIGR study. Biomark Neuropsychiatr 2024; 10: 100090.

Langford

Larkin

Resnicow

, et al. Understanding the role of message frames on African American willingness to participate in a hypothetical diabetes prevention study. J Health Comm 2017; 22: 647–656.

Salazar

Hoang

Gillen

, et al. Racial and ethnic differences in older adults’ willingness to be contacted about Alzheimer’s disease research participation. Alzheimers Dement (N Y) 2020; 6: e12023.

Chan

Acharya

, et al. Research attitude and interest among cancer survivors with or without cognitive impairment. Cancer 2023; 15: 3409.

Sewell

Neugroschl

Umpierre

, et al. Research attitudes and interest among elderly Latinx: the impact of a collaborative video and community peers. J Alzheimers Dis 2021; 82: 771–779.

Stites

Turner

Gill

, et al. Alzheimer’s disease cooperative study research attitudes questionnaire scores predict Alzheimer’s disease clinical trial dropout. Clin Trials 2021; 18: 237–244.

Witbracht

Morgan

, et al. Research attitudes questionnaire scores and retention in a recruitment registry. J Alz Dis 2025; 103: 406–418.

10.

Mayeda

Glymour

Quesenberry

, et al. Inequalities in dementia incidence between six racial and ethnic groups over 14 years. J Alzheimers Dis 2016; 12: 216–224.

11.

Neugroschl

Sewell

De La Fuente

, et al. Attitudes and perceptions of research in aging and dementia in an urban minority population. J Alzheimers Dis 2016; 53: 69–72.

12.

Lincoln

Chow

Gaines

, et al. Fundamental causes of barriers to participation in Alzheimer’s clinical research among African Americans. Ethn Health 2021; 26: 585–599.

13.

Portacolone

Palmer

Lichtenberg

, et al. Earning the trust of African American communities to increase representation in Dementia research. Ethn Dis 2020; 30: 719–734.

14.

Kim

McCallum

, et al.

What do people at risk for Alzheimer's disease think about surrogate consent for research?

Neurology 2005; 65: 789–792.

15.

Karlawish

Rubright

Cassrett

, et al. Older adults’ attitudes toward noncompetent subjects participating in Alzheimer’s research. Am J Psychiatry 2009; 166: 182–188.

16.

Muroff

Hoerauf

Kim

SYH

. Is psychiatric research stigmatized? An experimental survey of the public. Schizo Bull 2006; 32: 129–136.

17.

Rubright

Cary

Karlawish

, et al. Measuring how people view biomedical research: reliability and validity analysis of the research attitudes questionnaire. J Empir Res Hum Res Ethics 2011; 6: 63–68.

18.

Milfont

Fischer

. Testing measurement invariance across groups: applications in cross-cultural research. Int J Psychol Res 2010; 3: 111–121.

19.

Putnick

Bornstein

. Measurement invariance conventions and reporting: the state of the art and future directions for psychological research. Dev Rev 2016; 41: 71–90.

20.

Moodley

Madiba

. General attitudes toward research: pilot survey of HIV-positive surgical patients. S Afr J Surg 2019; 57: 28–31.

21.

Muthén

. Mplus user’s guide. Eighth Edition. Los Angeles, CA: Muthén & Muthén, 1998–2017.

22.

Muthén

. Mplus. The comprehensive modelling program for applied researchers. User's Guide 2012; 5.

23.

Kenny

. Measuring model fit. https://davidakenny.net/cm/fit.htm. (2023, accessed January 11, 2024).

24.

Rutkowski

Svetina

. Assessing the hypothesis of measurement invariance in the context of large-scale international surveys. Educ Psychol Meas 2014; 74: 31–57.

25.

Bentler

Bonett

. Significance tests and goodness of fit in the analysis of covariance structures. Psychol Bull 1980; 88: 588–606.

26.

MacCallum

Browne

Sugawara

. Power analysis and determination of sample size for covariance structure modeling. Psychol Methods 1996; 1: 130–149.

27.

Bentler

. Cutoff criteria for fit indexes in covariance structure analysis: conventional criteria versus new alternatives. Struct Equ Modeling 1999; 6: 1–55.

28.

Chen

. Sensitivity of goodness of fit indexes to lack of measurement invariance. Struc Equ Modeling 2007; 14: 464–504.

29.

Bialosiewicz

Murphy

Berry

. Do our measures measure up? The critical role of measurement invariance. In: An introduction to measurement invariance testing: Resource packet for participants. Claremont, CA, October 2013.

30.

Caban-Holt

Cuccaro

Lloyd

, et al. Attitudes and perceptions about brain donation among African Americans: implications for recruitment into Alzheimer's disease research. J Alzheimers Dis 2024; 973: 1621–1627.

31.

Fabrigar

Wegener

. Exploratory factor analysis. New York: Oxford University Press, 2012.

32.

Xia

Yang

. RMSEA, CFI, and TLI in structural equation modeling with ordered categorical data: the story they tell depends on the estimation methods. Behav Res Methods 2019; 51: 409–428.

33.

Cheung

Rensvold

. Evaluating goodness-of-fit indexes for testing measurement invariance. Struc Equ Modeling 2002; 9: 233–255.

34.

Lott

Kirby

Doran

, et al. Research attitudes in families of individuals with Down syndrome: importance for clinical trials. Alzheimers Res Ther 2022; 14: 176.

35.

Campos

Leader

, et al. Comparing research attitudes in Down syndrome and non-Down syndrome research decision-makers. Alzheimers Dement (N Y) 2024; 10: e12478.

36.

Walter

Kim

Flors

, et al.

Including general audiences in a virtual scientific dementia conference: will they get anything from it?

J Alzheimers Dis 2022; 990: 1001–1009.

37.

Courtright

Halpern

Joffe

, et al. Willingness to participate in pragmatic dialysis trials: the importance of physician decisional autonomy and consent approach. Trials 2017; 18: 474.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.21 MB