| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||
Correspondence: Address correspondence to Catherine A. Sarkisian MD, MSPH, UCLA Department of Medicine, Division of Geriatrics, 10945 Le Conte Avenue, Suite 2339, Los Angeles, CA 90095-1687. E-mail: csarkisian{at}mednet.ucla.edu
 |
Abstract |
|---|
 TOP Abstract MethodsResultsDiscussionReferences |
|---|
65 years who were recruited through physicians; 429 individuals (73%) returned completed surveys. The mean age of participants was 77 years; 76% were White. In 2001, we surveyed 643 adults aged 65 years recruited at 14 senior centers. The mean age of participants was 78 years; 37% were Latino and 16% were African American. With the 1999 data, we selected items for the shorter version of the ERA-38 by using qualitative criteria and by evaluating the items' factor structure, internal consistency reliability of scales, and correlations with age and self-reported measures of health. Then, using the 2001 data, we evaluated the selected items with confirmatory factor analysis, and we reevaluated the internal consistency reliability and associations of the scales with age and self-reported measures of health. Results: The factor analyses of the ERA-12 on both samples provided support for three 4-item scales (expectations regarding physical health, expectations regarding mental health, and expectations regarding cognitive function), and one global expectations regarding aging scale combining all 12 items. In both samples, internal consistency reliability estimates for all scales exceeded 0.74, and the 12 items together explained over 88% of the variance in the ERA-38 total score. We found comparable associations of the ERA-12 scales with age and self-reported health measures in both samples. Implications: The ERA-12 demonstrated acceptable reliability and validity to estimate expectations regarding aging.
Key Words: Measurement • Psychometrics • Factor analysis • Behavioral science • Attitudes toward health
When one is measuring the health beliefs of older adults, there is great incentive to minimize response burden in order to increase participation rates and have time to measure other important domains of health. Though researchers with a primary aim of examining expectations for aging should use the full 10-domain ERA-38 whenever possible, because the ERA-38 takes 10–12 min to administer, it is not always practical to use if several other constructs are also being measured in the same study. For example, one might be interested in examining whether expectations regarding aging are an important effect modifier of patient-centered interventions, but adding 38 items to a survey that already includes the primary outcome measures plus covariates could compromise overall data quality. Survey length concerns have led to the development of short-form versions of other widely used health surveys (Mangione et al., 2001; Marshall & Hays, 1994; Ware, Kosinski, & Keller, 1996). The development of a shorter version of the ERA-38 would enhance the feasibility of measuring expectations regarding aging in studies of health behaviors, outcomes, and health services use.
The goal of this study is to develop a short version of the ERA-38 that preserves as much of the content, reliability, and validity of the full-length survey as possible. In this article we describe analyses designed to identify the best items to include in the shortened ERA survey.
|
Methods |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
1999 Data Collection
As described in detail elsewhere (Sarkisian, Hays, Berry, et al., 2002), we used focus groups and cognitive interviews to develop items for the ERA-38. We mailed the ERA-38 in self-administerable surveys to 588 community-residing older adults cared for by UCLA-affiliated primary care physicians by randomly sampling 40 patients aged greater than 65 years from each practice. Before surveys were mailed, physicians excluded patients who did not speak English or who were too physically or mentally impaired to complete a 30-min survey. Exclusion rates varied by practice from 3% to 60%, with dementia being the most common exclusion criterion. Of the 588 eligible patients who were mailed a survey, 429 (73%) completed and returned the survey.
We also administered several other items with the same mailing: (a) 7 demographic items; (b) the 12-item Short-Form Health Survey (SF-12; Ware et al., 1996); (c) the Charlson Comorbidity Scale modified for self-administration (Katz, Chang, Sangha, Fossel, & Bates, 1996); and (d) the 5-item Geriatric Depression Scale (GDS; Hoyl et al., 1999).
2001 Data Collection
The ERA-38 was translated forward and back into Mexican American–Central American Spanish by use of a professional translator service. We invited English- and Spanish-speaking adults aged 65 years and older at 14 Los Angeles County senior centers to complete a questionnaire that included the ERA-38 and the same sociodemographic items and health instruments already described. Eligibility criteria included being age 65 years or older and speaking English or Spanish. After making brief presentations about the project at senior center events such as bingo games and club meetings, we distributed questionnaires with $5 incentives; approximately 15% of seniors invited to participate refused. Between 8 and 96 seniors completed surveys at each site, resulting in a sample size of 636 participants; 56 participants completed the survey in Spanish. In order to calculate test–retest reliability, we mailed the same survey 2 weeks later to a random 20% of participants.
Item Selection and Scaling Evaluation
To select items and scales for the shortened ERA survey, we used the following criteria.
First, content validity assesses how well a measure represents the construct of interest (Hays, Anderson, & Revicki, 1998). To preserve the maximum amount of content validity, scales in the shortened survey should represent domains most highly valued by the focus groups conducted to identify content for the ERA survey (Sarkisian et al., 2001).
Second, the shortened ERA survey should capture 85% or more of the variance in the ERA-38 overall score.
Third, internal consistency reliability coefficients (Cronbach's coefficient alpha) for selected scales should exceed the 0.70 reliability standard for group level comparisons (Nunnally & Bernstein, 1994).
Fourth, items should correlate more strongly with items that measure the same thing than with other items (Nunnally & Bernstein, 1994). Thus, the number of dimensions (factors) underlying the items should correspond to the number of scales, and uniqueness of scales should be supported by a simple factor structure in which the number of factors in the solution equals the number of scales, and each set of items in a scale defines a single factor.
Fifth, our goal was to create a shortened ERA survey that would measure the same construct as the ERA-38, namely expectations regarding aging. If we accomplished this goal, we would expect the shortened ERA survey to correlate with age and other self-reported measures of health with the same direction and magnitude as the ERA-38.
Application of Criteria by Use of the 1999 Data
We accomplished item selection by using the following steps and the 1999 data: First, we conducted an exploratory principal components analysis of the ERA-38 to provide information on the number of dimensions (factors) underlying the 38 items. To address Criterion 1, we compared the content of each factor with the rankings from the previously conducted focus groups (Sarkisian et al., 2001), and we retained the dominant factors that overlapped with the five top-ranked domains of expectations regarding aging. To address Criterion 2, for the items loading on each retained factor, we used the MAXR option in the PROC REG function in SAS 8.0 (SAS Institute, 1999) to identify the 2, 3, 4, and 5-item (when applicable) subsets of items explaining the maximum amount of variation in the total ERA-38 score. To address Criterion 3, we estimated the internal consistency reliability for each of the item subsets. We selected final items to maximize the variation in ERA-38 score explained by each scale (Criterion 2) and the internal consistency reliability of each scale (Criterion 3).
To address Criterion 4, we performed a principal components analysis on the selected items, specifying the number of factors to extract, and we examined the structure of the rotated factor-loading matrix and communalities. To address Criterion 5, we measured associations between the shortened ERA survey with age and self-reported measures of health, and we compared these with those observed with the ERA-38.
We linearly transformed mean scores for the shortened ERA survey as well as each scale to a possible score of 0–100, with lower scores representative of expecting decline in health and functional status and higher scores more consistent with expecting aspects of the Rowe and Kahn model of successful aging (Rowe & Kahn, 1997). We computed the SF-12 Physical Component Summary (PCS-12) score and the Mental Component Summary (MCS-12) score (Ware, Koskinski, & Keller, 1995). We counted the total number of comorbidities reported on the modified Charlson Comorbidity questionnaire. We calculated the percentage of participants who scored 2 or greater on the five-item GDS score; this cutpoint has a sensitivity and specificity of 97% and 85%, respectively, for detecting clinical depression (Hoyl et al., 1999). We then computed product-moment correlations of the ERA-12 and ERA-38 overall scores with each of the following: age, number of medical comorbidities, PCS-12, MCS-12, and GDS score.
Testing the Shortened ERA Survey by Use of the 2001 Data
Using the 2001 data, we performed confirmatory factor analysis on the items selected for the shortened ERA survey, using the PROC CALIS command in SAS 8.0 (SAS Institute, 1999). We specified the hypothesized structure on the basis of the findings (and the resulting scales) from the exploratory factor analysis done on the 1999 data (already described), and we tested the fit between this structure and the data by using several goodness-of-fit indices. To further determine whether the instrument performs similarly in the different samples, we measured the amount of variation in ERA-38 score explained by each scale in the shortened ERA survey (Criterion 2) and measured the internal consistency reliability of each scale (Criterion 3). To ensure that results did not depend on inclusion of the Spanish-language surveys, we repeated these analyses by using only the English version (n = 587) of the 2001 survey.
|
Results |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
|
We selected 4 items for each scale because, on the basis of our previous work (Sarkisian, Hays, Berry, et al., 2002), 12 items was the maximum number that could be completed in less than 5 min by the majority of older adults surveyed. All 4 items in each selected scale maximized variance in the ERA-38 overall scale accounted for and internal consistency reliability with one exception: we retained 1 item from the ERA-38 pain scale in the expectations regarding physical health scale because pain was so frequently mentioned as an important domain of expectations regarding aging in our focus groups, as well as extensive literature documenting the importance of pain to older adults' quality of life (AGS Panel on Persistent Pain in Older Persons, 2002).
The shortened ERA survey, the ERA-12, consisted of 12 items representing three scales: expectations regarding physical health, expectations regarding mental health, and expectations regarding cognitive function (Appendix). To address Criterion 4, we conducted a principal components analysis on the 12 items with a three-factor promax rotation specified. We used promax rotation because we expected strong intercorrelations among the factors (Sarkisian, Hays, Berry, et al., 2002). Table 2 reports the rotated factor loadings, descriptive statistics, internal consistency reliability estimates, and percentages of variance in ERA-38 explained by each of the ERA-12 scales as well as the ERA-12 overall. The rotated factor loading matrix indicates a simple three-factor structure in which one factor is defined by the four expectations regarding physical health items, another is defined by the four expectations regarding mental health items, and the third factor is defined by the four expectations regarding cognitive functioning items. Product-moment correlations between factors were.52 for physical health with mental health,.59 for physical health with cognitive function, and.49 for mental health with cognitive function.
|
= 0.75 for each scale, and the overall ERA-12 itself had = 0.88. The expectations regarding physical health scale explained 70%, the expectations regarding mental health scale explained 62%, and the expectations regarding cognitive function scale explained 57% of the variance in the original ERA-38 score. The three scales together explained 88% of the variance in ERA-38 total score. Table 3 shows the results of the analyses we did to address Criterion 5—correlations between age and self-reported measures of health, the ERA-12, and the ERA-38. The ERA-12 correlated in the same direction and with similar magnitude as the ERA-38 with all constructs. Specifically, expectations regarding aging are associated moderately with physical and mental health-related quality of life, and inversely with depressive symptoms and age, but they are not (or very minimally, in the case of the ERA-38) associated with medical comorbidity. The ERA-12 expectations regarding physical health scale correlated most strongly with age (r = –.23, p <.001), whereas the expectations regarding mental health scale correlated most strongly with depressive symptoms (r = –.35, p <.001), and the expectations regarding cognitive function scale correlated most strongly with the MCS-12 (r =.19, p <.001).
|
In addition to the results from the 1999 data already described, Table 2 also shows the results of the confirmatory factor analysis performed with the 2001 data. Because the findings were essentially unchanged whether or not the 56 participants completing Spanish-language surveys were included, here we present the results from the analyses done on the entire sample of 643 participants. As with the 1999 data, the factor loading matrix indicates a three-factor structure in which one factor is defined by the four expectations regarding physical health items, another is defined by the four expectations regarding mental health items, and the final factor is defined by the four expectations regarding cognitive functioning items. In addition, each of these first-order factors loaded highly on the one second-order expectations factor, with loadings of.98,.83, and.94 for the expectations regarding physical health, expectations regarding mental health, and expectations regarding cognitive functioning factors, respectively. The model was rejectable statistically by the conservative chi-square test (
2 = 155.97, p <.0001, df = 51), but this test is well known to be sensitive to small departures from fit. The model fit the data quite well according to practical fit indices: adjusted goodness-of-fit index =.94, comparative fit index =.97, normed fit index =.95, root mean squared error of approximation =.06.
Cronbach's alpha exceeded = 0.75 for each scale, and, for the overall ERA-12 itself, = 0.89. The three scales together explained 89% of the variance in ERA-38 total score. Associations between each scale score, age, and each self-reported measure of health were also tested on the 2001 sample, and they were similar to the associations in the 1999 sample (data not shown).
Of the random 20% of participants who were mailed a follow-up survey 2 weeks after completing the original survey, 118 (90%) completed and returned the survey. The intraclass correlation coefficients were 0.78 for the expectations regarding physical health scale, 0.83 for the expectations regarding mental health scale, 0.81 for the expectations regarding cognitive function scale, and 0.94 for the total ERA-12 score.
|
Discussion |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
Some content is lost any time a survey is shortened, but it is not always practical to use the longer, original version. Therefore, to minimize the loss of content between the ERA-38 and the ERA-12, we selected scales and items that provide the most information about the domains of expectations regarding aging most highly valued by older adults. The correlations with age, physical and mental health-related quality of life, and depressive symptoms are remarkably similar to the correlations of the ERA-38 with each of these measures, thereby providing support for the construct validity of the ERA-12.
It is well documented that one's attitudes toward aging may influence one's future health (Leventhal & Prohaska, 1986; Levy et al., 2002; Rakowski & Hickey, 1992) and health behaviors (Goodwin et al., 1999). Unlike previous instruments that measure positive or negative attitudes, however, the ERA surveys do not attempt to label health beliefs (in this case, expectations regarding aging) as either positive or negative, but rather higher or lower along a scale without cutpoints for what is optimal. Measuring expectations regarding aging may contribute to our understanding of the mechanisms by which attitudes toward aging influence health. For example, the ERA-38 is currently being used by investigators to examine whether differences in expectations for aging contribute to differences in health services utilization across cultural subgroups of older adults. Other investigators are using the ERA-12 to examine whether expectations for aging influence preferences for care among older adults with chronic illness. The current wave of the Baltimore Longitudinal Study of Aging includes the ERA-12 and therefore will provide a rich opportunity to examine the relationship between expectations for aging and a wide range of health outcomes.
In our previous work, we found that older adults who had very low expectations regarding aging were more likely to report that it is not important to seek health care for modifiable conditions such as urinary incontinence and depression, in models controlling for many potential confounders including demographic characteristics and health status (Sarkisian, Hays, & Mangione, 2002). This suggests that expectations regarding aging may be a causal factor in older adults' missing out on health care likely to improve their quality of life, and it warrants further investigation. Investigators can use the ERA-12 (or the ERA-38) to examine the relationship between expectations regarding aging, health behaviors, service use, and subsequent health.
There are a number of limitations to this study. Though the ERA-12 performed well in two very different sociodemographic samples, participants were recruited from a single geographic region of the country (greater Los Angeles). The psychometric performance of the ERA-12 should be tested in a more geographically diverse population. Likewise, with so few Spanish-speaking participants, it is impossible in this study to disentangle the potential roles that ethnicity, language, and acculturation might have on survey responses; it would be premature to claim that this instrument had been adequately tested among Spanish-speaking Latinos. Though there is no a priori reason to speculate that the ERA-12 would perform differently when administered on its own instead of as part of the longer ERA-38, psychometric assessment should be repeated on the 12-item instrument when it is administered on its own, and not part of the larger instrument. Likewise, future studies could expand assessment of the construct validity of the ERA-12 by measuring the extent to which it correlates with previously tested instruments examining the different but related constructs of attitudes toward old people (Maxwell & Sullivan, 1980; Rosencranz & McNevin, 1969), ageism (Palmore, 2001), Rakowski's life-graph measure of future perceived health (Rakowski & Hickey, 1981), and locus of control (Wallston & Wallston, 1981). It is also important to point out that this report describes cross-sectional data; future studies should examine the ERA-12's ability to predict outcomes over time.
In summary, we have described the development of the ERA-12, a shorter version of the ERA-38, which demonstrated acceptable reliability and construct validity. Though much shorter, the ERA-12 captures most of the variation in the ERA-38. When using the ERA-38 is not practical, investigators should feel confident using the ERA-12 to examine expectations regarding aging among older adults.
|
|
Footnotes |
|---|
We very much appreciate the support of the UCLA Primary Care Network physicians, the Los Angeles County Area Agency on Aging, and the 1,072 older adults who took the time to participate in this study. We also thank Ms. Karen Spritzer for her programming assistance.
1 Division of Geriatrics, David Geffen School of Medicine at the University of California, Los Angeles.
2 Division of General Internal Medicine and Health Services Research, David Geffen School of Medicine at the University of California, Los Angeles.
Decision Editor: Linda S. Noelker, PhD
Received for publication January 30, 2004. Accepted for publication August 16, 2004.
|
References |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
| ||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
