Caregiver Mental Health and Potentially Harmful Caregiving Behavior: The Central Role of Caregiver Anger

Gordon MacNeil, PhD; Jordan I. Kosberg, PhD; Daniel W. Durkin, MSW; W. Keith Dooley, PhD; Jamie DeCoster, PhD; Gail M. Williamson, PhD


Gerontologist. 2010;50(1):76-86. 

In This Article



The analyses presented in this article are based on the first wave of interviews from the second FRILL2 Project, a multisite longitudinal study of informal care provided by caregivers of community-dwelling elderly care recipients with whom they coresided. The voluntary sample was recruited from Athens, GA, Pittsburgh, PA, Tuscaloosa, AL, and surrounding areas. To be eligible for the study, caregivers had to be primarily responsible for the care of a cognitively or physically impaired care recipient more than 60 years of age with whom they coresided and they had to provide unpaid help with at least one basic activity of daily living or two instrumental activities of daily living.

A primary goal of FRILL2 was to oversample African American caregiving dyads in order to obtain data sufficient to address issues conspicuously missing in previous research on the quality of informal elderly care (e.g., longitudinal comparisons between White and African American caregivers). Within these constraints, we attempted to obtain as representative a sample as possible. These efforts began with random-digit dialing (RDD) in the data collection sites. We then narrowed our search to age-targeted RDD (e.g., individuals aged 60 years and older according to U.S. Census data). To increase the number of African American participants, we used community-based snowball referral methods at the Georgia site in which enrolled African American dyads were recontacted and asked to provide the names and telephone numbers of other potentially eligible dyads. In initial screening, RDD methods identified 877 potential dyads; snowballing methods produced 95 potential dyads. Our recruitment efforts produced 765 eligible dyads, 321 (42%) of whom declined participation, resulting in a sample of 444 caregivers (58% participation rate). Of these, 27 caregivers had enough missing data to be excluded from the current analyses, resulting in a sample size of 417. These methods resulted in a sample that was 55% White and 43% African American.

Face-to-face structured interviews, lasting between 1.5 and 2 hr, were conducted in respondents' homes by pairs of carefully trained interviewers. Each participant was paid $25 for a completed interview. To prevent data contamination, caregivers and care recipients were interviewed separately and simultaneously by two interviewers. The consent form for the study contained a sentence advising participants that suspected cases of abuse would be reported to the proper authorities. No reportable cases were observed, and no participants refused to be interviewed after receiving information informing them of our obligation to report suspected cases of abuse.


Similar to national estimates (e.g., National Alliance for Caregiving & the American Association of Retired Persons, 1997; Select Committee on Aging, U.S. House of Representatives, 1987), 69% of caregivers and 52% of care recipients in the study were women. Mean caregiver age was 60 years (SD = 15, range 18–98), and care recipients were, on average, 75 years (SD = 7.8, range 60–102). Slightly more than half of the caregivers (52%) were spouses of their care recipients. African Americans were oversampled in order to obtain a sample with roughly equivalent numbers of Whites and African Americans. The ethnic background of our final sample was 55% White, 43% African American, and 2% other ethnic heritage. Twenty-seven percent of care recipients had a medical diagnosis of dementia or another disorder that causes memory problems (e.g., Alzheimer's disease, Parkinson's disease). Duration of care provision at the time of the interviews ranged from less than 1 year to more than 50 years (M = 6.9 years, SD = 7.8, range <1–58). Median household income for these families was $20,000–$29,000 (range <$5,000–>$100,000).


Caregiver Depressed Affect.—Caregiver depression was assessed using the 20-item Center for Epidemiologic Studies Depression Scale (CES-D) (Radloff, 1977), which asks about the frequency with which respondents have experienced depressive symptoms within the last week (e.g., "I was bothered by things that usually don't bother me," "I felt that everything I did was an effort"). Total scores range from 0 to 60, with higher scores indicating elevated levels of depressive symptoms. The mean CES-D score was 11.9 (SD = 9.9, range 0–47). Individuals with scores of 16 or above are considered to be at risk for clinical depression; 27% of our caregivers were in this category. Cronbach's α for our sample was .89.

Caregiver Anger.—A short form of the Spielberger's State-Trait Anger Scale (Spielberger, 1983) was used to assess how often respondents "generally" (trait) and "in the past week" (state) have felt mad, furious, etc. Higher scores on the scale indicate greater anger. This instrument has been used extensively in other studies, demonstrating good internal consistency (Spielberger, Jacobs, Russell, & Crane, 1983). It has shown concurrent validity with measures of hostility, neuroticism, and anxiety (Spielberger et al.) and has been used successfully to identify anger and hostility among elderly caregivers (Vitaliano, Young, Russo, Romano, & Magana-Amato, 1993). Cronbach's α for our sample was .89.

Caregiver Resentment.—We operationalized caregiver resentment as the sum of responses to a 17-item instrument adapted from previous research. Prior analyses have shown that combining 10 items from a resentment scale devised by Thompson, Medvene, and Freedman (1995) with 7 items from the Caregiver Burden Scale (Zarit, Reever, & Bach-Peterson, 1980) results in a measure of caregiver resentment that is psychometrically sound (Williamson et al., 2000, 2005). Caregivers indicated how often (1 = never, 5 = almost always) they felt resentful about such circumstances as not having enough time for themselves, having to give up plans for the future, or that their care recipients were overly dependent or made unreasonable demands. As in previous studies, Cronbach's α in this sample was more than .90 (Martin-Cook, Remakel-Davis, Svetlik, Hynan, & Weiner, 2003; Williamson et al., 2000, 2005).

Caregiver Anxiety.—Anxiety was measured using the Spielberger's State-Trait Anxiety Inventory (STAI; Spielberger, Gorsuch, & Lushene, 1970). The measure consists of two subscales, one measuring enduring (trait) and the other transient (state) aspects of anxiety. We used the global score in our analyses. The STAI has high internal consistency and predictive validity in adult samples (Spielberger et al.) and is both reliable and valid in elderly populations (Nagatomo, Nomaguchi, & Takigawa, 1995). Cronbach's α for our sample was .89.

Potentially Harmful behavior.—Due to the voluntary nature of our sample, we did not expect to identify dyads in which caregivers were guilty of severe forms of abuse and neglect. Rather, we assessed PHB (e.g., Beach et al., 2005; Dooley et al., 2007; Miller et al., 2006; Shaffer et al., 2007; Williamson et al., 2001) defined as adverse care that is not so severe that social or legal intervention is required. PHB was measured using a 10-item instrument developed from the Conflict Tactics Scale (Straus, 1979), work by Steinmetz (1988) and Pillemer and Suitor (1992), and our own analyses of the elderly abuse and neglect literature. The resulting instrument has adequate psychometric properties (e.g., Williamson et al.) and comprises five indicators of psychological mistreatment (e.g., caregiver screams and yells at care recipient) and five indicators of physical mistreatment (e.g., caregiver hits, slaps, or handles care recipient roughly). These questions were placed near the end of the structured interview with instructions carefully worded to decrease respondent reactance. Specifically, caregivers were asked to rate how often (0 = never and 4 = all of the time) they employed "methods that caregivers use when care recipients won't follow a doctor's orders or do what caregivers feel they should do." Responses were summed to create a measure of PHB, with higher scores representing more frequent PHB, and an acceptable Cronbach's α of .62.

As has been found in samples representing the general population (Cooper et al., 2009; Laumann, Leitsch, & Waite, 2008), PHB was, on average, infrequent in this sample (M = 1.1), but there was adequate variability (SD = 1.9, range 0–15). Most frequently cited indicators of PHB were psychological in nature; that is, caregivers reported at least occasionally screaming and yelling (34.9%); swearing at, insulting, or calling the care recipient names (19.9%); and threatening to abandon the care recipient (5.9%) or to send him or her to a nursing home (8.5%). However, 58% of our caregivers admitted to at least one of the following: shaking, hitting, slapping, handling the care recipient roughly, and/or to feeling afraid that they might hit or try to hurt the care recipient.

Analysis Plan

The model examined by our analyses is presented in Figure 1. We think that mental health predictors influence caregiver anger (Path A), which in turn influences the occurrence of PHB (Path B). This suggests that anger mediates the relation between mental health predictors and PHB (Hypothesis 1). In addition, mental health predictors directly influence PHB above and beyond the effect of anger (Path C). Finally, we think that the strength of the relation between anger and PHB is influenced by the mental health predictors (Path D). This suggests that mental health predictors moderate the relation between anger and PHB (Hypothesis 2).

Figure 1.

Model tested.

The strength of the mediated effect of a mental health behavior going through anger can be determined by multiplying together the coefficient for Path A and the coefficient for Path B (Sobel, 1982). If a variable changes the value of this product then it means that it is moderating the strength of the mediation effect (Preacher, Rucker, & Hayes, 2007). Our proposal that mental health predictors moderate the relation between anger and PHB (Path D) suggests that the value of B will depend on the values of the mental health predictors. Mathematically, we know that when B changes, the value of A × B also changes. This means that the presence of Path D suggests that the strength of the mediated effect will depend on the values of the mental health predictors. Thus, it is reasoned that the values of the mental health predictors will moderate the strength of the mediating path going through anger (Hypothesis 3).

To assess the hypothesized paths by which mental health variables might influence PHB, we undertook correlational, mediational, and moderational analyses, leading us to develop moderated mediation tests of our model. The addition of the moderated mediation analyses enables us to explain both how and when a given effect occurs. Preacher and colleagues (2007) propose that moderated mediation effects may occur in multiple ways. In this study, we specifically propose that the independent variables also function as a moderator of the path from the mediator variable to the dependent variable (see Figure 1, Path B).


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.
Post as: