Dr. Brown is affiliated with the department of occupational therapy education at the University of Kansas Medical Center in Kansas City, with which Ms. Bothwell was with when the research was conducted. Dr. Rempfer is with the department of research service at the Veteran's Administration Medical Center in Kansas City, Missouri. Dr. Hamera is with the University of Kansas Medical Center School of Nursing in Kansas City. Send correspondence to Dr. Brown at the University of Kansas Medical Center, 3901 Rainbow Boulevard, Kansas City, Kansas 66160 (e-mail, tbrown@kumc.edu).

Understanding the mechanisms of impairment associated with functional deficits is essential for designing effective interventions. Although several studies document the role of cognition in predicting community functioning among persons with serious mental illness (1), cognition accounts for only a modest portion of the variance. Therefore, it is important to identify other mechanisms or mediators that lie between the basic process of cognition and the performance of skills. Identifying mediators may help to explain how or why cognition influences outcomes for people with serious mental illness.

Mediation analyses are used to explore how a third variable (a mediator) accounts for the relationship between outcome and predictor variables. The identification of mediators can be useful in determining which underlying mechanisms influence outcomes, thereby identifying the relevant targets of intervention. The most common method for determining mediation requires four specific relationships between the variables of interest: a relationship between the predictor and the outcome variables; a relationship between the predictor and mediator; a relationship between the mediator and outcome variable when the analysis controls for the predictor; and a significant reduction in the strength of the relationship between the predictor and outcome variables when the analysis controls for the mediator (2,3,4).

Previous work documented a relationship between cognition and the performance of grocery shopping skills (5,6). The purpose of this study was to determine whether knowledge of grocery shopping skills acts as a mediator or mechanism through which cognition influences real-world performance. We hypothesized that cognition plays an indirect role in the performance of grocery shopping skills. With regard to the skill of grocery shopping, foundational cognitive abilities (for example, memory, attention, and problem solving) allow an individual to use the act of grocery shopping in a way that increases knowledge about the skill; these cognitive abilities are needed to effectively recognize the systematic manner in which stores are organized or to create strategies for locating cheaper items. This study answers the research question, Does knowledge of grocery shopping skills mediate the relationship between cognition and performance?

This study was part of a larger project examining the efficacy of a grocery shopping intervention. The study was approved by the institutional review board of the University of Kansas Medical Center, and data were collected from 2001 to 2003. Sixty-six individuals from five community programs participated in the study. Individuals were recruited through flyers, announcements at meetings, and case managers. Of those 66 individuals, complete data were available for the mediational analysis for 51 participants.

After the study was described to the participants, informed consent was obtained. The participants completed a battery of cognitive tests, the Knowledge of Grocery Shopping Skills (KOGSS) to measure knowledge of skills, and the Test of Grocery Shopping Skills (TOGSS) to measure performance of skills. Trained graduate-level research assistants administered the measures. The research assistant who administered the cognitive measures was blind to the results of the grocery shopping measures, and the research assistant who administered the grocery shopping measures was blind to the results of the cognitive measures. All measures were administered within a one-month period and were baseline measures for the larger study.

In the mediational analysis, the focus was not on specific cognitive measures but on cognition as a compound construct. Consequently, a cognitive composite was formed by summing z scores from a battery of cognitive measures selected because of their relationship to community function in other studies (1) and their relevance to grocery shopping (5,6). The following is a list of the cognitive measures used in the study. The d2 Test of Attention concentration (7). Possible scores range from 0 to 300, with higher scores indicating better selective attention. The total number of correct answers on the Months Ordering Test of Working Memory (8). Possible scores range from 0 to 20, with higher scores indicating better working memory. The total number of words recalled for three trials on the Rey Auditory Verbal Learning Test (9). Possible scores range from 0 to 45, with higher scores indicating better verbal memory. The digit span backwards subtest on the Wechsler Memory Scale. Possible scores range from 0 to 14, with higher scores indicating better working memory. The time taken to complete trails b from the Trail Making Test (10). A higher score indicates the individual took more time and consequently had worse alternating attention. The Controlled Oral Word Association Test (11) requires the individual to say as many words he or she can think of that begin with a certain letter. In this case, the FAS version was used. Higher scores indicate better verbal fluency. The total number of cards sorted correctly on the 64-card version of the Wisconsin Card Sorting Test. Possible scores range from 0 to 64, with higher scores indicating better executive function.

The KOGSS is a 31-item self-administered measure of a person's grocery shopping knowledge, such as strategies to find the lowest-priced item and to determine how items are typically grouped. For example, pictures of different brands are presented, and participants are asked to identify the lowest-priced item. The items are in a multiple choice or matching format. In a reliability estimate of the KOGGS, coefficient alpha was .72. Possible scores range from 0 to 39, with higher scores indicating more grocery shopping knowledge.

The TOGSS assesses grocery shopping performance in the natural context of a grocery store. From a list of ten grocery store items, participants select the correct size and lowest price for each item to give an accuracy score. Validity and reliability of the TOGSS is described elsewhere (12). Possible scores range from 0 to 30, with higher scores indicating better grocery shopping performance.

The Structured Clinical Interview for DSM-IV confirmed DSM-IV diagnoses, which included bipolar disorder (22 participants, or 43 percent), major depression (14 participants, or 27 percent), and schizophrenia or schizoaffective disorder (15 participants, or 29 percent). Participants included 36 women (71 percent) and 15 men (29 percent). The majority were white (41 participants, or 80 percent); three were African American (6 percent), three were American Indian (6 percent), and four were from other racial or ethnic groups (8 percent). The mean±SD age was 43±10 years (range of 18 to 61 years).

For the d2 Test of Attention concentration (7), participants had a mean score of 100.5±42.8, range of 12 to 232. For the Months Ordering Test of Working Memory, participants had a mean of 7.6±3.5 correct answers, range of 0 to 14. For the three trials of the Rey Auditory Verbal Learning Test, participants had a mean of 20.2±6.6 correct answers, range of 6 to 33. For the digit span backwards subtest of the Wechsler Memory Scale, participants received a mean score of 4.7±1.9, range of 0 to 9. For the trails b on the Trail Making Test, participants had a mean time of 126.8±33.7 seconds, range of 43 to 580 seconds. On the Controlled Oral Word Association Test, participants received a mean score of 28.3±9.4, range of 13 to 50. For the 64-card version of the Wisconsin Card Sorting Test, participants scored mean of 38.3±11.1 total number of cards sorted correctly, range of 16 to 58. The mean score on the KOGSS was 31.9±3.6, with a range of 17 to 37, and the mean score on the TOGSS was 24.2±4.1, with a range of 12 to 30.

Following the mediation model described in the introduction (2,3,4), four tests were performed by using multiple regression to estimate the paths between variables. (More information about mediation analysis may be found at http://davidakenny.net/cm/mediate.htm.) The cognitive composite score acted as the predictor, the KOGSS score acted as the mediator, and the TOGSS score acted as the outcome.

For the first test, there must be a significant relationship between the predictor and the outcome. This test was performed by regressing performance of grocery shopping skills (outcome) on cognition (predictor) (Figure 1). The unstandardized regression coefficient was statistically significant (R=.40, B=.32, df=1, 49, p=.004).

For the second test, the predictor must be related to the proposed mediator. This test was performed by regressing knowledge of grocery shopping skills (mediator) on cognition (predictor). The unstandardized regression coefficient was statistically significant (R=.57, B=.42, df=1, 49, p<.001)

For the third test, the outcome must covary with the mediator, even after the analysis controls for the predictor. And finally, for the fourth test, the relationship between the predictor and outcome variables should either disappear (complete mediation) or become substantially reduced (partial mediation) when the analysis controls for the mediator. Both tests were conducted in the same analysis by regressing performance of grocery shopping skills (outcome) simultaneously on cognition (predictor) and knowledge of skill (mediator). For the third test, the unstandardized regression coefficient for the relationship between the knowledge and outcome was statistically significant when the analysis controlled for cognition (R=.71, B=.84, df=2, 48, p<.001). For the fourth test, almost complete mediation was achieved (B was reduced from .32 to .03; p=.74).

The results indicate that knowledge of grocery shopping skills mediates the relationship between cognition and performance of such skills. These results may explain some of the inconsistencies and modest relationships found in previous studies of cognition and community outcome. Various measurement approaches are used to assess community outcomes in serious mental illness; the results of this study suggest that distinctions should be made between outcome measures of knowledge and performance of skills.

The findings also have implications for interventions. Cognitive rehabilitation is criticized because improvements in cognition do not typically generalize to daily life. The results of this study suggest an intermediate intervention strategy which involves providing knowledge of skills as the bridge between cognition and performance of skills. This finding is consistent with social skills training approaches, in which instruction includes training to increase knowledge about the skill.

Finally, it is likely that other mediators exist, for example, learning potential (1), which should be considered when determining predictors of community function. Future research is needed that identifies mediating variables and the efficacy of interventions that target these variables.

This initial study suggests that measures of knowledge and performance should be distinguished in the realm of community functioning. Although knowledge of grocery shopping skills appears to mediate the relationship between cognition and performance, other areas of community functioning and other mediators of outcome should be explored.