Validity and Reliability of the Positive Aspects of Caregiving (PAC) Scale and Development of Its Shorter Version (S-PAC) Among Family Caregivers of Older Adults

Abstract

Purpose of the Study:

To (a) assess the validity and reliability of the 9-item Positive Aspects of Caregiving (PAC) scale among a national sample of caregivers for older adults with functional limitations, (b) develop a shorter version (short-PAC [S-PAC] scale) and assess its psychometric properties, and (c) investigate both scales’ measurement equivalence/invariance (ME/I) across language of administration (Chinese/English/Malay).

Design and Methods:

Scale/item measurement property assessment, confirmatory factor analysis (CFA), testing the “original” 2-factor model (6 items: first factor; 3 items: second factor), and exploratory FA (EFA) of the 9-item PAC scale was done. Consequently, alternate CFA models were tested. The S-PAC was developed and subjected to CFA. For both scales, convergent (correlation with caregiver esteem) and divergent (correlation with caregiver depressive symptoms) validity, and language ME/I was assessed.

Results:

For the 9-item PAC scale, the “original” 2-factor CFA model had a poor fit; its EFA and scale/item measurement properties supported a single factor. Among alternate CFA models, a bi-factor model (all nine items: first factor [overall PAC]; six items: second factor [self-affirmation]; three items: third factor [outlook-on-life]) had the best fit. The bi-factor CFA model also had a good fit for the S-PAC scale, developed after eliminating 2 items from the 9-item PAC scale. Both scales demonstrated convergent and divergent validity, and partial ME/I across language of administration.

Implications:

Both the 9-item PAC and 7-item S-PAC scales can be used to assess positive feelings resulting from care provision among family caregivers of older adults with functional limitations.

Purpose of the Study

Caregiving, while often associated with stress and burden, can also be a fulfilling endeavor for family caregivers (Grant & Nolan, 1993). And, caregivers who perceive more benefits from caregiving experience greater subjective well-being and fewer depressive symptoms (Cohen, Colantonio, & Vernich, 2002). Thus, a focus on improving the positive aspects of caregiving to enhance the experience of care provision has been suggested (Cohen et al., 2002; Pinquart & Sörensen, 2003). This in turn calls for valid and reliable measurement of the positive aspects of caregiving.

One extensively used scale for measuring psychosocial benefits of caregiving among family caregivers is the Positive Aspects of Caregiving (PAC) scale (Tarlow et al., 2004). The validity and reliability of this Likert scale was established in 2004 among 1,229 caregivers of individuals with Alzheimer’s disease in the United States. Starting with 11 items, Tarlow and colleagues found psychometric support, through exploratory factor analysis (EFA) followed by confirmatory factor analysis (CFA), for nine items, loading on two factors (six items loading on one factor and three items on the other), to comprise the scale. And, they proposed the 9-item PAC scale, providing two subscale scores (Self-Affirmation [SA; six items]; Outlook-on-Life [OL; three items]) and an overall PAC score.

The PAC scale has since been used in several countries (Hilgeman, Allen, DeCoster, & Burgio, 2007; Hinojosa, Hinojosa, & Rittman, 2012; Kate, Grover, Kulhara, & Nehra, 2012; Las Hayas, López de Arroyabe, & Calvete, 2014; Liddle et al., 2012; Lou, Lau, & Cheung, 2015; Wakefield, Hayes, Boren, Pak, & Davis, 2012); many studies use the 9-item PAC scale, while others instead use all 11 items initially considered by Tarlow and colleagues. While used primarily among caregivers of patients with dementia (Liddle et al., 2012; Lou et al., 2015; Roth, Dilworth-Anderson, Huang, Gross, & Gitlin, 2015; Tan, Yap, Ng, & Luo, 2013; Yap, Luo, Ng, Chionh, Lim, & Goh, 2010), its use has extended beyond its intended caregiver population to caregivers of veterans with chronic disease (Wakefield et al., 2012), of individuals with acquired brain injury (Las Hayas, López de Arroyabe, & Calvete, 2015) and of stroke survivors (Hinojosa et al., 2012).

Given cross-cultural differences in question responses and measurement models, it is important to establish the validity and reliability of Likert scales prior to use in different countries (C. Chen, Lee, & Stevenson, 1995; Cheung & Rensvold, 2000). And, there is empirical evidence that scales developed among caregivers in one setting may not perform equivalently in other settings (Malhotra, Chan, Malhotra, & Østbye, 2012). However, despite its widespread use, globally and among various caregiver populations, the PAC scale’s validity and reliability, after its development among caregivers of individuals with Alzheimer’s disease in the United States, has only been established in two recent studies. One Spanish study, among caregivers for patients with acquired brain injury (Las Hayas et al., 2014) used the 9-item PAC scale, in Spanish (PACS). While the authors, through CFA, confirmed the factor structure of the PACS to be similar to that observed by Tarlow and colleagues, they only included a small, convenience sample of 141 caregivers. The other study, from Hong Kong among 456 caregivers for patients with dementia, proposed the Chinese PAC (C-PAC) in 2015 (Lou et al., 2015). Instead of considering the 9-item PAC scale, the authors used all the original 11 items but, unlike Tarlow and colleagues, they did not drop any of the 11 items, and observed two factors, with 6 items loading on one factor (“affirming self”) and 5 items on the other (“enriching life”), in EFA. However, the researchers refrained from using the two subscale scores (as they were highly correlated), rather they only utilized the overall score derived from all 11 items for further analyses. The variability in the psychometric properties and suggested scoring of the PAC scale across settings thus highlights the need to establish these aspects when using the scale in a different setting or population of caregivers.

The validity and reliability of the PAC scale has not yet been established in Singapore, a rapidly aging multiethnic Southeast Asian country. Family caregivers form the primary support for older Singaporeans (Ministry of Health, 2010). This stems from both cultural expectations of filial piety and the government’s encouragement for older Singaporeans to “age-in-place” (Mehta, 2006). However, shrinking family sizes and high workforce participation will likely make caring for older Singaporeans at home increasingly challenging for caregivers (Mehta & Vasoo, 2001). In such a scenario, augmentation of the positive perception of the caregiving experience will be beneficial (Cohen et al., 2002; Pinquart & Sörensen, 2003). As researchers and practitioners develop and implement interventions and services to maintain or improve such positive perceptions, validation of the PAC scale in Singapore will assist in their evaluation. An opportunity exists to assess the psychometric properties of the PAC scale through data gathered in the Singapore Survey on Informal Caregiving (SSIC), a national survey of family caregivers of older adults with functional limitations.

In this study, we capitalize on this opportunity and contribute to the global literature on family caregiving in several ways. First, this study is the first from Singapore to assess the psychometric properties of the PAC scale. Second, it is based on a national sample, of 1,132 caregivers, larger than previous studies (Las Hayas et al., 2014; Lou et al., 2015), allowing for a comprehensive psychometric analysis (detailed in Methods). Third, it speaks to a more general caregiver population, those providing care to home-dwelling older adults with functional limitations, compared with previous studies which were limited to narrower caregiver populations such as those caring for individuals with dementia, brain injury, or stroke (Hinojosa et al., 2012; Las Hayas et al., 2014, 2015; Liddle et al., 2012; Wakefield et al., 2012). Fourth, we gauge support for a shorter version of the PAC scale, driven by the following motivations: (a) In previous validation studies (Lou et al., 2015; Tarlow et al., 2004), its Cronbach’s alpha values are close to the threshold for item redundancy (Streiner, 2003; Tavakol & Dennick, 2011); and (b) a shorter version, taking less time to administer, will encourage health and social care professionals to integrate it into their daily practice and aid caregiving researchers reduce survey respondent burden. Development of shorter version of scales used among caregivers is common, for example, shorter versions of the Zarit Burden Interview (Bédard et al., 2001) and the Marwit-Meuser Caregiver Grief Inventory (Marwit & Meuser, 2005). Lastly, the multilingual composition of Singapore’s population allows assessment of measurement equivalence/invariance (ME/I) across language of administration.

Therefore, this study aimed to (a) comprehensively assess the validity and reliability of the 9-item PAC scale among a national sample of caregivers for older Singaporeans with functional limitations, (b) develop a shorter version (short-PAC [S-PAC] scale) and assess its psychometric properties, and (c) investigate the ME/I of the 9-item PAC and the S-PAC scales across language of administration (Chinese/English/Malay).

Design and Methods

Data Source

The SSIC was conducted in 2010–2011 by the Ministry of Social and Family Development, Singapore. It was a national survey of 1,190 dyads, each comprising an older care-recipient (home-dwelling Singaporean aged ≥75 years receiving human assistance for ≥1 activity of daily living—bathing, walking, dressing, standing up, toileting, and eating) and his/her primary family caregiver. Each dyad member was interviewed face-to-face, with informed consent, at home. Further details on the design of SSIC can be found elsewhere (Malhotra et al., 2012).

PAC Scale

The 9-item PAC scale (Tarlow, 2004) was administered to the caregivers. Each item (Table 1) is scored on a 5-point Likert scale: disagree a lot (1), disagree a little (2), neither agree nor disagree (3), agree a little (4), and agree a lot (5). Scores for the SA subscale, comprising Items 1–6 in Table 1, range from 6 to 30, and for the OL subscale, comprising Items 7–9 in Table 1, range from 3 to 15. The overall PAC score, comprising all nine items, ranges from 9 to 45—a higher score reflects a more positive perception of the caregiving experience.

The English version of the 9-item PAC scale was translated into Chinese and then back translated into English by another translator. A third translator then reconciled any inconsistencies to provide the final Chinese version. Malay and Tamil translations were similarly developed.

Center for Epidemiologic Studies Depression (CES-D) Scale

The 11-item version of the CES-D scale (Kohout, Berkman, Evans, & Cornoni-Huntley, 1993) gathered information on extent of depressive symptoms among the caregivers in the past week. Response options were 0 (none/rarely), 1 (sometimes), and 2 (often). A higher total score, the sum of item scores, reflects more depressive symptoms. The CES-D scale was used for assessing divergent validity of the 9-item PAC and the S-PAC scales.

Caregiver Reaction Assessment

The caregiver reaction assessment (CRA), validated in Singapore (Malhotra et al., 2012), comprising four subscales, was administered to the caregivers. Each item of the CRA is scored on a 5-point Likert scale ranging from 1 (strongly disagree) to 5 (strongly agree); subscale scores range from 1 to 5, an average of constituent items. The caregiver esteem (CE) subscale (comprising six items) was used for assessing convergent validity of the 9-item PAC and the S-PAC scales. A higher CE subscale score reflects greater perceived benefits from caregiving.

Statistical Analysis

The analytical sample comprised of 1,132 caregivers, excluding 58 caregivers who did not answer one or more items on the 9-item PAC scale.

Step 1: Scale/Item Measurement Properties of the 9-Item PAC Scale

Mean scores and Cronbach’s alpha of the 9-item PAC scale and its two subscales were calculated. Inter-item correlation (iic) of the nine items was assessed. The item-total correlation (itc: correlation between an item and the scale score, the latter excluding the item under consideration) was calculated for the 9-item PAC scale and its two subscales. The item-scale correlation (isc: correlation between an item on a particular subscale and the score on the other subscale) was calculated for items on the two subscales. An isc of >.4 suggests an overlap between subscales (Persson, Wennman‐Larsen, Sundin, & Gustavsson, 2008). Given the ordinal item response categories, all computed correlations were polychoric correlations.

Step 2: CFA of the 9-Item PAC Scale, Testing a Two-Factor Model

The sample was then randomly split into equal halves (Samples 1 and 2, each with n = 566). In Sample 1, CFA of the “original” two-factor model observed by Tarlow and colleagues (Items 1–6 on Factor 1 and Items 7–9 on Factor 2; Supplementary Figure 1, Panel A), using the weighted least squares method (Holgado–Tello, Chacón–Moscoso, Barbero–García, & Vila–Abad, 2010), was performed to test its fit. Model fit indices considered included (a) chi-square and its p-value (low chi-square and high p-value suggest a good fit), (b) chi-square/degrees of freedom ratio (value of <2 indicates a good fit), (c) standardized root mean square residual (SRMR), (d) root mean square error of approximation (RMSEA) (SRMR and RMSEA <.08 suggest a good fit and <.05 suggest an excellent fit), (e) comparative fit index (CFI), and (f) non-normed fit index (NNFI) (CFI and NNFI >.90 suggest a good fit and >.95 suggest an excellent fit; Hooper, Coughlan, & Mullen, 2008; O’Rourke, Psych, & Hatcher, 2013; Schermelleh-Engel, Moosbrugger, & Müller, 2003). Standardized factor loadings and their significance, and modification indices were examined.

Step 3: EFA of the Nine Items of the PAC Scale

In Sample 2, EFA of the 9 PAC scale items was conducted using the principal axis/factors method to extract factors, allowing for oblique rotation, without specifying the number of factors a priori. Value of the eigenvalues, scree plot, proportion of variance explained, and cumulative variance explained were examined—factors with eigenvalue >1 or factors that explained >10% of the proportion of variance or cumulative variance of >75% were retained (O’Rourke et al., 2013). Factor loadings were examined after determining the appropriate number of factors.

Step 4: CFA of the 9-Item PAC Scale Testing Alternate Factor Model Specifications

Informed by Steps 1–3, CFA of a one-factor model (all nine items loading on 1 factor; Supplementary Figure 1, Panel B) was conducted in Sample 1 and its fit indices examined. Based on its results, a bi-factor model (i.e., model with a common/general factor, and a number of group factors therein (Reise, Moore, & Haviland, 2010; Reise, Morizot, & Hays, 2007)—in our case, Items 1–9 loading on Factor 1 (common/general factor), Items 1–6 loading on Factor 2 (group factor), and Items 7–9 loading on Factor 3 (group factor); Supplementary Figure 1, Panel C) was subjected to CFA.

Step 5: Development of the S-PAC Scale, and Its Psychometric Analysis

After establishing the best fitting factor structure of the 9-item PAC scale, we developed a shorter version, that is, the S-PAC. First, the Cronbach’s alpha and itc of the 9-item PAC scale and its subscales, and the iic of the nine items were examined against item redundancy criteria [Cronbach’s alpha >.9 (Streiner, 2003; Tavakol & Dennick, 2011), itc >.8 (Alrubaiy, 2013; Walsh, 2013) and iic >.8 (Dykes, Hurley, Cashen, Bakken, & Duffy, 2007; Sepucha et al., 2012)]. Consequently, two items were deleted based on statistical and/or theoretical reasoning. Then, a bi-factor CFA model (seven items loading on Factor 1, five items loading on Factor 2, and two items loading Factor 3; Supplementary Figure 1, Panel D), which had the best fit for the 9-item PAC scale, was examined in both Samples 1 and 2 for the 7-item S-PAC scale. Finally, the S-PAC scale’s measurement properties were assessed.

Step 6: Convergent and Divergent Validity of the 9-Item PAC and the S-PAC Scales

Using the entire sample of 1,132, correlation of the 9-item PAC and the 7-item S-PAC scale scores with those on the CES-D scale (hypothesizing a negative correlation; for divergent validity) and the CE subscale of the CRA (hypothesizing a positive correlation; for convergent validity) was assessed.

Step 7: ME/I of the 9-Item PAC and the S-PAC Scales Across Language of Administration

Informed by our previous work (Malhotra et al., 2012), three language groups, based on language(s) used in the caregiver interview were created: (a) English (n = 547; English only or English and Chinese/Malay/Tamil); (b) Chinese (n = 324; Mandarin/Hokkien/Cantonese/Teochew only); and (c) Malay (n = 250; Malay only). The number interviewed in Tamil only (n = 11) was low for inclusion in ME/I analysis. Assessment of ME/I for the 9-item PAC and the S-PAC scales was conducted separately, in a step-wise manner, using the entire sample of 1,121 (1,132 minus 11), partitioned into the three language groups. First, equality of covariance, followed by configural invariance (the scale measures similar latent construct(s), i.e., same items load on the same factors, across the groups) was assessed by examining various CFA model fit indices against the thresholds specified in Step 2 earlier. For assessing configural invariance, a multigroup CFA was performed, specifying a bi-factor model and the same factor structure for the three groups; however the strength of the factor loadings was allowed to vary across the groups. Next, the strength of factor loadings were forced to be similar across the groups to assess for metric invariance—this is established when the decrease in CFI is <.010 and the increase in RMSEA and SRMR is <.015 and <.030, respectively between metric and configural invariance models (F. F. Chen, 2007). Its establishment allows for comparing regression slopes between groups. Finally, all CFA model parameters were forced to be similar across the groups to judge scalar invariance—this is established when the decrease in CFI is <.010 and the increase in RMSEA and SRMR is <.015 and <.010, respectively between scalar and metric invariance models (F. F. Chen, 2007). Its establishment suggests the intercept is similar across groups, allowing for comparing mean scores between groups (F. F. Chen, 2007).

Data analysis, using SAS version 9.4, conducted using de-identified data, was exempted from full review by the Institutional Review Board of the National University of Singapore.

Results

Table 1 provides the distribution of the responses provided to the 9 PAC scale items by the caregivers in the analytical sample. Items with the most and least endorsement were both from the SA subscale. Item 3 (made me feel needed) was the most endorsed, with 41.3% caregivers responding “Agree a lot,” and 85.4% responding “Agree a lot” or “Agree a little.” Other the hand, Item 2 (made me feel good about myself) was the least endorsed, with 33.3% caregivers responding “Agree a lot,” and 75.7% responding “Agree a lot” or “Agree a little.” The three OL subscale items were endorsed (“Agree a lot” or “Agree a little”) by ≥80% of the caregivers. Caregiver and care-recipient characteristics are summarized in Supplementary Table 1.

Panel A of Table 2 presents the mean score and indices reflecting scale/item measurement properties of the 9-item PAC scale. Cronbach’s alpha ranged from .85 (OL subscale) to .93 (overall scale). All items on a particular subscale had isc of >.40 for the other subscale. Table 3 presents iic for the nine items; values ranged from .64 (Items 9/1 and 9/5) to .86 (Items 7/8).

Table 4 lists fit indices for the various CFA models. The two-factor model assessed through CFA in Sample 1 (Supplementary Figure 1, Panel A) had subthreshold CFI and NNFI values (Table 4, Model 1). Modification indices, pointing to cross-loading of items, suggested an improvement in model fit if Items 7–9 (which the two-factor model purported to be exclusive to Factor 2) were allowed to load on Factor 1 and Items 5 and 6 (which the two-factor model professed to be limited to Factor 1) were allowed to load on Factor 2. All standardized factor loadings were >.80, and significant.

EFA, conducted in Sample 2, strongly suggested a single factor—the first factor had an eigenvalue of 6.44 and explained 95.8% of the variance. And, each of the nine factor loadings, in an EFA specifying a single factor, was ≥.75.

The results of the analyses done up to this point supported a one-factor model for the 9-item PAC scale: all items of the SA subscale had isc >.40 with the OL subscale and vice-versa; the two-factor CFA model fit was unsatisfactory, with modification indices pointing to item cross-loading; and EFA supported a single factor. Thus, a one-factor CFA model (Supplementary Figure 1, Panel B) was assessed in Sample 1. However, it had a poor fit (Table 4, Model 2). The largest error covariance values were observed between Items 7/8, 8/9, 3/5, 1/2, 5/6, 4/2, and 1/5, suggesting that specification of two more factors in the model, with Items 7–9 loading on one factor and Items 1–6 loading on another factor would improve the model fit. In other words, the results pointed to a bi-factor model (Reise et al., 2007, 2010), wherein all items load on a common/general factor, and some items load on one or more group factors. Thus, a bi-factor model, with Items 1–9 loading on Factor 1, Items 1–6 loading on Factor 2, and Items 7–9 loading on Factor 3 (Supplementary Figure 1, Panel C) was specified. And its CFA, again conducted in Sample 1, demonstrated an excellent fit (Table 4, Model 3).

While the 9-item PAC scale’s structural validity was established, high Cronbach’s alpha of the overall scale (.93) and SA subscale (.92) along with 5 of the iic and 14 of the itc values being >.8 suggested item redundancy, thus the scope for a shorter version. The iic values (Table 3) were examined against the suggested threshold of .80 (Dykes et al., 2007; Sepucha et al., 2012) to identify items for deletion. Items 7/8 had the highest iic (.86), and Item 8 was selected for removal. Given the very similar meaning of Items 7 and 8, the decision to remove Item 8, and not Item 7, was based on the value of the iic of each item with Item 9, the third item in the OL subscale: the iic for Items 8/9 was higher (.73) compared with that for Items 7/9 (.69), suggesting a greater overlap of Items 8/9 than of Items 7/9 in context of OL. The next highest iic values, >.80, were .82 (Items 2/1 and 2/4), .81 (Items 1/3), and .80 (Items 2/3). Given the similar iic values, we turned to the self-affirmation theory to determine which of these four SA subscale items could be deleted (Cohen & Sherman, 2014). Items 1 (made me feel more useful) and 3 (made me feel needed) capture goodness and efficacy, key aspects of the self-affirmation theory (Cohen & Sherman, 2014). While Item 2 also captures goodness, the wording. “…made me feel good about myself” does not parallel the self-affirmation theory, which suggests that “the motive for self-integrity,” central to the self-affirmation theory, “is not to esteem or praise oneself but rather to act in ways worthy of esteem or praise” (Cohen & Sherman, 2014). Thus, Item 2 does not entirely speak to the self-affirmation theory. Interestingly, as stated earlier, Item 2 was also the least endorsed item. While Item 4 (made me feel appreciated) was the next least endorsed item, it captures the extent to which the caregiver receives positive feedback from the care-recipient, or other family members—such positive feedback is postulated to play a role in increasing self-affirmation (Cohen & Sherman, 2014). We thus kept Items 1, 3, and 4, and removed Item 2 from the SA subscale. Consequently, we propose the S-PAC scale, which after excluding Items 2 and 8 of the 9-item PAC scale, comprises seven items (Items 1, 3–7, and 9) for the overall scale, and five items on the SA subscale (Items 1, and 3–6) and two items on the OL subscale (Items 7 and 9).

Panel B of Table 2 presents the mean score, and scale/item measurement properties of the 7-item S-PAC scale. Its Cronbach’s alpha ranged from .73 (OL subscale) to .91 (overall scale). Its CFA, testing the fit of a bi-factor model (Items 1, 3–7, and 9 loading on Factor 1, Items 1 and 3–6 loading on Factor 2, and Items 7 and 9 loading on Factor 3; Supplementary Figure 1, Panel D), demonstrated a good fit in Sample 1 (Table 4, Model 4), which was also reproduced in Sample 2 (Table 4, Model 5).

For both the 9-item PAC (Panel A of Supplementary Table 2) and the S-PAC (Panel B of Supplementary Table 2) scales, correlation between the scale scores and those on the CES-D and the CE subscale of CRA was in the hypothesized direction (p < .001). Further, the correlation coefficient magnitude was similar across the two PAC scale versions.

ME/I results are presented in Table 5. Equality of covariance matrices (Model 0) and configural invariance (Model 1) were established for both scales. However, metric invariance and scalar invariance were not supported for either as the difference between fit indices of Models 2 and 1 and Models 3 and 2 were above the suggested thresholds. Thus, there was partial, but not full ME/I across language of administration for both scales.

Implications

The current study, through comprehensive psychometric analyses, establishes the validity and reliability of the 9-item PAC scale, and a shorter version, the 7-item S-PAC scale, among a broad, important and growing population of caregivers, that of family caregivers of older adults with functional limitations. Either scale, based on the users’ preference, can be used to assess positive feelings resulting from care provision among such caregivers, providing an overall score for PAC and two sub-scores, those for SA and OL.

Two of the three studies that have previously assessed the psychometric properties of the PAC scale also recommended that it provided three scores—overall PAC score, SA subscale score, and OL subscale score—each of which could be considered independently in subsequent analyses (Las Hayas et al., 2014; Tarlow et al., 2004). Such a scoring recommendation is acceptable if the best-fitting CFA model is a bi-factor model—such a model justifies summing subscale scores to produce an overall score in addition to analyzing each subscale score individually (Reise et al., 2007, 2010). However, despite their scoring recommendation, these two studies did not empirically assess bi-factor models. Rather, Tarlow and colleagues found a good fit for a CFA model with two factors, corresponding to SA and OL, and recommended an overall PAC score only based on strong correlation between the two subscales. While Las Hayas and colleagues, did assess a higher-order model, wherein the SA and OL factors were explained by a secondary factor (PAC), it did not have the best fit indices; rather, the best-fitting CFA model for the PACS was the one with two factors (SA and OL) (Las Hayas et al., 2014). The third study did not conduct CFA, instead found two factors on EFA but used only the overall PAC score (Lou et al., 2015). Our analyses, firmly establishing a bi-factor CFA model for both the 9-item PAC and S-PAC scales, is thus the first to provide empirical evidence that supports scoring them in the “recommended” manner. Further, doing so provides researchers support for considering the three resulting scores independently in analyses as well as the possibility of employing only one of the subscales (SA or OL) in their work, if desired.

While previous studies (Lou et al., 2015; Tarlow et al., 2004) have observed the Cronbach’s alpha values of the PAC scale close to the threshold suggesting item redundancy (Streiner, 2003; Tavakol & Dennick, 2011), they have refrained from eliminating items. We observed the same, and chose to delete two of the items, one from each subscale, utilizing an empirical statistical basis (high iic values) as well as theoretical reasoning to arrive at the S-PAC. And, in subsequent analysis, we established the S-PAC’s structural, convergent, and divergent validity as well as internal consistency reliability.

ME/I analysis results, supporting partial but not full invariance, caution against directly comparing the PAC or S-PAC scale scores across the languages we considered. Related, though not fully (as those responding in English were from all four ethnic groups in Singapore—Chinese, Malay, Indian, and Others), caregiver ethnicity did correspond to a certain extent with the language of administration (those responding in Chinese and Malay were mostly of Chinese and Malay ethnicity, respectively). Thus, any direct comparison of the PAC or S-PAC scale scores across caregiver ethnicity should be interpreted keeping in mind that any observed differences could be due to ME/I.

Our study has its limitations. First, even the 7-item S-PAC exhibited item redundancy, overall, and specifically in the SA subscale (iic of .81 for items 1/3; itc of two items was >.8). However, eliminating more items may result in a subscale that is too narrow to measure important aspects of self-affirmation. Second, utilizing the opportunity given by existing data, we validated a scale originally developed in a North American population. While it has been subsequently validated in Europe (Spain) and Asia (Hong Kong), and used globally, including in Singapore, there may be additional dimensions of the benefits gained from caregiving which the scale does not capture, which may be relevant in the multiethnic Singapore context. Thus, future studies should endeavor to take a more emic perspective, gathering primary, qualitative data on such benefits and if needed, add more, culturally relevant items to the PAC scale for use in the Singapore or Asian context. Third, the scales we validate do not pertain to foreign domestic workers (FDWs), a group of caregivers relatively unique to some countries including Singapore. Hiring of live-in FDWs, mostly women from neighboring low-income countries, for eldercare is common in Singapore (Østbye, Malhotra, Malhotra, Arambepola, & Chan, 2013). The extent to which FDWs experience psychosocial gains from caregiving, and how to best measure such benefits should be explored in future research.

In addition to the five contributions detailed previously, this study has the following strengths. First, it provides a shorter version of the PAC scale, likely to benefit both respondents and consumers of the collected data (researchers/practitioners). While the deletion of only two items may result in only a marginal benefit in respondent burden, it has been suggested that a reduction in the number of items reduces missing responses, resulting in more complete data (McKnight, McKnight, Sidani, & Figueredo, 2007). Second, given the similar factor structure and estimates for convergent and divergent validity of the S-PAC and the 9-item PAC scales, the S-PAC scale captures all aspects of positive aspects of caregiving which the 9-item PAC scale does. Third, we utilized the recommended analytical approach for CFA of ordinal data, that is, using polychoric correlation and the weighted least squares method; most papers that perform CFA of ordinal data do not follow the recommended approach, using Pearson’s correlation and the maximum likelihood method instead (Holgado–Tello et al., 2010).

Conclusion

The 9-item PAC scale and its shorter version, the 7-item S-PAC scale are reliable and valid measures for assessing the positive benefits gained by the family caregivers when providing care to older family members with functional limitations. Their three resulting scores—overall PAC score, SA subscale score, and OL subscale—can be used independently in analyses. Further, if desired, only one of the subscales, SA or OL, can also be employed.

Supplementary Material

Supplementary data are available at The Gerontologist online.

Acknowledgment

The data utilized for this manuscript are from the Singapore Survey on Informal Caregiving, which was funded by the Ministry of Social and Family Development, Singapore. We acknowledge Mr Kwan Yu Heng for his comments on an earlier version of the article.

References

Author notes