Based on the conceptual framework of stress and coping (Dunkel-Schetter et al
), social support is typically defined in terms of various functional domains and refers to any process through which social relationships might promote health and well-being (Cohen et al
). The types of social support dimensions were derived from past empirical work (Broadhead et al
; Krause 2002
; Sherbourne and Stewart 1991
) and focus group data (Wong et al
). The initial development and testing of this instrument is described in detail elsewhere (Wong et al
).Functional support refers to the degree to which interpersonal relationships serve particular functions (Sherbourne and Stewart 1991
). The functions most often cited are: emotional, instrumental aid or tangible, information/advice, companionship, and validation (Cohen et al
). Tangible support refers to the provision of assistance with activities of daily living such as transportation, and assistance with cooking, cleaning, and shopping. Financial support refers to assistance through monetary means. While House and Landis (1988
) refers to tangible and financial support as instrumental support, in our focus groups with older Chinese and Koreans, there was a clear separation between these two domains. Emotional support refers to having a person express sympathy, caring, and acceptance of the individual. Someone who provides information about resources or gives advice is giving the individual informational support. Companionship refers to having a person with whom to share activities such as going to movies, eating together, and shopping. A person who provides validation gives the individual feedback about him/herself.
Nature and sources of survey respondents
As part of a larger study, eligible participants included African American, Latino, Asian (mainly Chinese), and non-Latino White women, between 50 years and 80 years of age, fluent in either English and Spanish, or Chinese (Cantonese or Mandarin), and had seen the same clinician for at least one visit in the previous 2 years (Kim et al
). Women were recruited from three clinical sites in San Francisco, screened by telephone, and invited for an extensive face-to-face interview lasting up to 90 min. The clinical sites for recruitment were four primary care practices (two general internal medicine, family medicine, and women’s health) at the University of California, San Francisco (UCSF) Medical Center, a community-based clinic in Chinatown, and the Community Health Network Clinics affiliated with the Department of Health and San Francisco General Hospital.
Study procedures The research protocol was approved by the UCSF Committee on Human Research, the Institutional Review Board at San Francisco General Hospital (SFGH), and the Chinatown Clinic Director. We generated a list of potentially eligible women based on available administrative data at each site. Names of patients were submitted to the appropriate clinician listed and permission requested to contact their patients by mail to inform them about the project. Clinicians were asked to return the list with their approval and indicate names of patients not to be contacted (e.g. no longer a patient, cognitively impaired, dead). Personalized letters were mailed to each patient, in English, or English and Spanish or Chinese, informing her about the study and requesting that a postcard be returned by checking a “Not interested, do not call” box or a “Interested, please call me” box. If no postcard was returned within 2 weeks, trained interviewers contacted women to complete a 20-min screening questionnaire in English, Spanish, or Chinese (Cantonese or Mandarin). After 8 months of limited success recruiting Asian women, we began screening and recruiting these women (mostly Chinese) using face-to-face methods at the community-based Chinatown clinic. Our goal was to recruit similar numbers of African American, Asian, Latina, and non-Latino White women and a significant number of limited English proficiency participants.Women participating in the baseline telephone survey then were asked to participate in a face-to-face survey in the language of their choice. Each face-to-face survey lasted an average of 50–90 min, with the longer survey times occurring for those conducted in Spanish and Chinese. Interviews were conducted at a UCSF research office, a clinical practice site, or their home. Women received $20 at the end of the interview in appreciation for their time.
Survey The final survey included items derived from standard questions used in previous surveys and from formative focus groups and individual interviews completed as part of the larger study. The questionnaire was developed simultaneously in English, Spanish, and Chinese using bilingual experts and then pre-tested in each of the four ethnic groups, specifically testing the cultural, linguistic, and literacy appropriateness. Data collected from the surveys included socio-demographic characteristics (e.g. age, years of education, household income, language use, birthplace), personal and family history of cancer, use of cancer screening tests, risk perceptions of getting cervical, breast, or colon cancer, health status, social support, and interpersonal processes of care.
Social support instrument
The original 27-item social support measure in this survey was first developed for use with older Koreans and Chinese; a review of existing measures revealed no one instrument that captured all domains identified focus groups of older Chinese and Koreans (Wong et al
) or reported psychometric properties in these groups. Responses were a 1–5 likert scale, “none of the time”, “a little of time”, “some of the time”, “most of the time”, “all of the time.” Items for each subscale were summed and divided by the number of items in the subscale such that each participant had a score for each subscale. This instrument was tested in a sample of older Koreans and Chinese using exploratory factor analytic (EFA) techniques (Wong et al
). The results provided evidence that social support was a reliable and valid multidimensional construct measuring support in the following areas: language (5-items); informational (4-items); financial (4-items); emotional/companionship (8-items).Tangible support (6-items) is a considered a standard dimension of social support, therefore, we also included this subscale in the current survey and subsequent analyses. For the purposes of this study, we used 22-items that included all the social support dimensions, except for language support because these questions were only answered by people who answered the survey in Spanish or Chinese.
We identified and received consent from physicians to send recruitment letters to 4,523 women. We were unable to reach 906 women due to wrong telephone numbers or addresses and 871 were ineligible due to factors such as illness or having left the physician’s practice. We were able to contact 2,786 women. Screening data were collected from a total of 1,137 African American (n
159), Latino (n
236), Chinese (n
353), other Asian (n
63), and non-Latino White women (n
326). Other Asian women were excluded due to small sample size. Our sample final size for this study was 1,074 women for women completed the telephone screener and face-to-face interview.
Table displays the socio-demographic characteristics. Women’s mean age was 61 years with Latinas being significantly older (63.6 years). Over 80% of the women had at least one child and over half of all women were married or living with a partner, with Chinese (73%) being the most likely to be married or living with someone, and African Americans (30%) and Latinas (39%) the least likely to be married or living with a partner. The majority of Chinese and Latinas (68%, 59% respectively) reported having significantly less than a high school education compared to non-Latino Whites. Similarly, the many African Americans, Chinese, and Latinas (39%, 54%, 50%, respectively) all reported a significantly lower household income compared to non-Latino Whites. Over one-third of African Americans (31%) reported being on disability, most Chinese (38%) were not working, and 34% of Latinas were retired compared to over one-third of non-Latino whites (36%) who were still working full-time. African American, Chinese, and Latinas were significantly less likely to have private insurance and more likely to report being in poor health status (p
.001) compared to non-Latino Whites. Not surprisingly, most Chinese and Latinas were not born in the U.S.
Sociodemographic Characteristics of Women
Using a confirmatory factor analytic (CFA) framework, we sought to test whether the instrument’s psychometric properties were invariant across each of the four ethnic groups (African American, Chinese, Latino, and non-Latino White). That is, we assessed whether four common factors (Tangible, Information, Financial, Emotional/Companionship) were present among the 22-items of the social support measure within the total sample and within each ethnic group. Four common factors were identified by their specified items (no cross-loadings), common factor variances and co-variances were freely estimated, and all item error variances were constrained to be uncorrelated. None of the models had acceptable fit. Thus a revised empirical model was pursued.
Multitrait scaling was used to examine whether each item within a hypothesized dimension was linearly related (r
0.30, corrected for overlap) to the total score for that dimension (item convergence
) and if each item was correlated two or more standard errors higher with its hypothesized domain than with other domains (item discrimination
) (Hays and Hayashi 1990; Stewart et al
. 1992). We first eliminated items not meeting the item convergence criterion across all in at least three race/ethnic groups. We then eliminated items not meeting the item discrimination criterion using the same approach.
Data from the remaining items were simultaneously modeled with multiple-group CFA. These analyses were used to assess the unidimensionality of items within each originally hypothesized subdomain across the four ethnic groups. Items were dropped via a process of backward elimination, either because they did not have salient loadings (<0.40) on the hypothesized factor, or loaded highly on more than one factor.
Next, we conducted a series of nested models to assess the level of factorial invariance across the four ethnic groups. We assessed configural invariance to examine whether each common factor was associated with identical item sets across the ethnic groups. A multi-group model specified four factors with identical item sets across the four ethnic groups. All parameters were freely estimated. We then examined metric invariance which constrained corresponding factor loadings to be equal across groups. Finally, we assessed strong factorial invariance, also known as scalar invariance, which requires that regression of items onto their associated common factors yields a vector of intercept terms that is invariant across the ethnic groups (Gregorich 2006
). That is, the corresponding item intercepts were also held equal across groups in addition to the constraints of metric invariance.
All CFA models were fit using LISREL 8.5 software (Jöreskog and Sörbom 1998
). Goodness of fit was assessed by examining model chi-squares and degrees of freedom, the root mean square error of approximation (RMSEA, Steiger 1990
), and the comparative fit index (CFI, Bentler 1990
). Significant chi-squares and chi-square difference tests indicate lack of exact fit
. It is known that the chi-square goodness-of-fit test can too often reject correct models (Type I error) with finite samples of nonnormal data. Therefore the root-mean-square error of approximation (RMSEA), the comparative fit index (CFI), and the expected cross-validation index (ECVI) were also used to assess model fit. Values of RMSEA <0.05 and values of CFI
0.95 suggest approximate fit
(Browne and Cudeck 1993
; Hu and Bentler 1999
). In a series of nested models, the ECVI will obtain relatively lower values for models with higher expectation of replication in subsequent samples of the same size. Ninety percent confidence intervals were included to supplement the point estimates of RMSEA and ECVI (Browne and Cudeck 1993
). Selection of cross-group equality constraints to relax was based on LISRL’s 8.5 modification indices.
Missing data (<5% of data points) were dealt with using the linear interpolation algorithm of SPSS version 16.0 (SPSS Inc, 2008) which uses nearby data points to estimate missing values. The skew and kurtosis of each item were investigated to assess normality of the distributions. Provided that each social support dimension met the criteria for configural, metric, and scalar invariance, we then compared the means using the appropriate statistical test, analysis of variance with post-hoc comparisons.