Use It—and Keep It,Longer,Probably: A Reply to Salthouse (2006)

SALTHOUSE'S CRITICISMS OF THE SSES LONGITUDINAL SURVEY RESEARCH PROGRAM

A major goal of the SSES longitudinal research program (Caplan & Schooler, in press; Kohn & Schooler, 1983; Schooler & Mulatu, 2001; Schooler et al., 1999, 2004) has been to disentangle the effects of cognitively demanding environments on psychological functioning from the selection effects of psychological variables on individuals' environments. To do this, we have used reciprocal-effects structural equation modeling (SEM) to simultaneously estimate the psychological effects of the environmental conditions in question on the person and of the person's characteristics on the likelihood that he or she will be subjected to these environmental conditions. Our findings indicate that although individuals with relatively high levels of cognitive functioning are more likely than others to be selected into cognitively demanding environments, the reciprocal effect also holds: Exposure to such environments, regardless of age, leads to better cognitive functioning. Using longitudinal data and SEM, we have shown that these findings hold true for paid work (Kohn & Schooler, 1983; Schooler et al., 1999, 2004) and leisure-time activities (Schooler & Mulatu, 2001) and have recently extended these longitudinal findings to women's and men's housework (Caplan & Schooler, in press). If anything, positive effects of complex environments are greater for older than for younger people (Schooler et al., 1999).

Salthouse expressed several significant reservations about our findings. His primary reservation was that our measure of intellectual functioning is apparently not negatively related to age and that, consequently, we may not be measuring a type of intellectual functioning relevant to the mental-exercise-and-aging hypothesis.

Salthouse was correct that our correlations between age and intellectual functioning in 1964, 1974, and 1994–1995 (published in Schooler et al., 1999, Table C7) were very low. It should be noted, however, that these correlations were for a select subsample—men who were working at all three time points (n = 166). It should also be noted that the tables in that article were based on regression factor scores, which, because they do not fully account for measurement error, tend to underestimate true correlations.

Elsewhere (Schooler & Mulatu, 2001, Table A1 in the appendix), we have provided SEM-based estimated correlations (i.e., correlations that do take measurement error into account) separately for those respondents (men and women combined) who were and were not working in 1974. Among those who were not working in 1974 (n = 190), the correlation between age and 1994–1995 intellectual flexibility (IF) was −.27 (p <.001); the correlation between age and 1974 IF was −.16 (p <.03). Among those who were working in 1974 (n = 516), the parallel correlations were −.20 (p <.0001) and .01 (not significant).

The correlations between IF and age for all of the men (n = 351) and all of the women (n = 355) in the 1994–1995 sample were as follows (Caplan & Schooler, in press). For men, the correlation was −.33 (p <.0001) in 1994–1995 and −.25 (p <.0001) in 1974. The parallel correlations for women were −.38 (p <.0001) and −.22 (p <.0001). Other analyses revealed that in the total population, the correlation between IF and age was −.34 (p <.0001) in 1994–1995 and −.22 (p <.0001) in 1974. Thus, in the total sample, as well as among men and women considered separately, there was a clear and significant negative correlation between age and our IF measure.

Taken across all of the various analyses, the correlation between age and IF is reduced to the degree that there is evidence of a relatively continuous work history in the particular subsample being examined. As noted, the correlation between age and IF was very low among the subsample of men selected as working in 1964, 1974, and 1994–1995. It was higher among the subsample of men and women who were not working in 1994–1995 than among the subsample that was. The negative correlations between age and intellectual functioning were distinctly higher when the total sample and the full samples of men and women were considered. It is plausible that these differences may have been due to some combination of the characteristics of those older individuals who stay in the workforce and of the way the cognitive demands of their work affect them. Nevertheless, even if the correlation between age and IF appears to be affected by how subsamples are selected in terms of work history, the negative correlation between age and IF seems to generally hold true. Not only was the negative correlation between age and IF relatively high in the full sample, but the correlations between age and IF tended to be higher the older the sample was at the time IF was measured (e.g., the correlation between age and 1994–1995 IF was higher than that between age and 1974 IF).

In addition, as we have noted (Schooler et al., 1999), the correlation between our IF measure and a latent factor based on more standard cognitive measures (immediate recall, category fluency, different uses, number series, verbal meaning, identical pictures) is very high (r =.87). It is also worth noting, given Salthouse's concerns with age-based interactions, that in Schooler et al. (1999), we reported and discussed a difference between the older and younger workers in the effect of substantively complex work on IF—the effect actually being greater for the older than for the younger workers.

Salthouse raised several other concerns about our research methods that are not justified. He stated that “the measure of intellectual flexibility was based on five variables, three measurements derived from ratings by the examiner and two objective measurements” (p. 77). Actually, only one of the indices of the latent IF factor (i.e., interviewer's rating of the respondent's alertness and estimated intelligence) was based on a subjective rating. The answers to the two other questions that Salthouse seems to be referring to as “subjective”—“What are all of the arguments you can think of for or against allowing cigarette commercials on TV?” and “What questions would you consider in deciding which of two locations offers a better business opportunity for opening a hamburger stand?”—were coded by trained coders using relatively objective ratings. For the cigarette-commercial question, the answers were coded in terms of whether the respondent could provide no argument, an argument for or against such cigarette commercials, or arguments for both sides (scores of 1, 2, and 3, respectively). Adequacy of response to the hamburger-stand question was scored according to whether the respondent's answer did not deal with the question, reflected a concern for either potential costs or potential sales, reflected a concern for both costs and sales, or reflected an explicit understanding that profits result from the difference between the two (scores of 1, 2, 3, and 4, respectively). The coding of neither the cigarette-commercial question nor the hamburger-stand question was more subjective or difficult than that used in many standard IQ subtests (e.g., Similarities subtest of the Wechsler Adult Intelligence Scale-Revised; Wechsler, 1981).

Salthouse combined all of his doubts to raise

the possibility that the meaning of the composite intellectual-flexibility variable formed by combining the individual variables may have changed from one measurement occasion to the next, and that some of what was interpreted as a change in intellectual flexibility may have actually reflected a change in what was being assessed. (p. 77)

The points discussed earlier address this concern. In addition, it is the case that the analyses reported in Schooler and Mulatu (2001), Schooler et al. (2004), and Caplan and Schooler (in press) were all based on models in which the factor loadings of the IF indicators were constrained to be equal over time, so that the relative contribution of each indicator to determining the latent factor remained the same across time points. Even when these equality constraints were included in the statistically more conservative full-information SEM approach in which the measurement and causal aspects of the model are estimated simultaneously (Schooler et al., 2004), we demonstrated a significant and substantial effect (β =.26) of complex, cognitively demanding work conditions on intellectual functioning.

Salthouse concluded his discussion of the SSES's longitudinal research program's findings by stating that it is “probably premature to summarize the findings of this project … by stating that ‘complex intellectual work increases the cognitive functioning of older workers’ … the available results are quite complex, and the extent to which they should be considered as providing support for the mental-exercise hypothesis is not yet clear” (p. 77). I would argue that although many questions remain about who can be helped how, how much, and under which conditions, our findings provide quite strong nonexperimental support for the use-it-or-lose-it mental-exercise hypothesis and its applicability to older people.

SALTHOUSE'S CRITIQUE OF OTHER RELEVANT NONEXPERIMENTAL RESEARCH

Although the SSES findings provide the strongest nonexperimental support for the use-it-or-lose-it hypothesis, there have been numerous studies (e.g., Avolio & Waldman 1990, 1994), including examples cited by Salthouse, that have shown that people who do, or have a history of having done, more cognitively demanding paid work tend to have higher levels of intellectual functioning than those who do, or have a history of having done, less cognitively demanding work. In addition, numerous studies have shown parallel findings for leisure-time activities (e.g., Bosma et al., 2002). These relationships tend to hold whether one focuses on “normal” cognitive functioning or dementia (for a review, see Schooler, in press). These studies should be credited as at least being congruent with the use-it-or-lose-it hypothesis. Nevertheless, Salthouse was correct that these studies provide no fully convincing proof of the hypothesis. This is so because they do not allow one to assess the relative importance of the two potential sources of covariance between the cognitive demands of the activities in question and the intellectual functioning of the people carrying them out: (a) the effects of these activities on the intellectual functioning of the people who carry them out and (b) the likelihood that people having a given level of intellectual functioning are in a position to, or are predisposed to, carry out such cognitively demanding activities.

A further argument that Salthouse used against the applicability of general activity questionnaires is more questionable. Arguing for an essentially subjective measure of cognitive difficulty, he noted, “Some researchers have relied on judges' ratings of cognitive demands …. However, because most of these judges were likely of high cognitive ability, they may not have an accurate perception of the difficulty of the activities for lower-ability individuals” (p.79). Leaving aside the issue that, contrary to Salthouse's implications, it is possible to develop relatively objective coding schemes to rate the cognitive difficulty of various environmental demands (e.g., the U.S. Department of Labor's, 1965, 1977, Dictionary of Occupational Titles codes for “complexity” of work with things, data, and people), there is no reason to necessarily define the intellectual demands of the task in question in terms of how complex it seems to the person carrying it out. In fact, the argument can be made that individuals' subjective judgments about how intellectually demanding their environments are do not necessarily represent a reasonably valid estimate of how cognitively complex and demanding their environments actually are (Schooler, 1987). For example, in the SSES sample, someone whose job it was to sort potatoes into grades A, B, C, and so forth, saw this task as intellectually demanding because of the number of decisions required, but this does not mean that the cognitive demand posed by these relatively simple decisions represents the level of cognitive challenge (i.e., proximal developmental scaffold; cf. Vygotsky, 1978) that could lead that individual to better cognitive functioning.

SALTHOUSE'S EMPIRICAL AND THEORETICAL CRITIQUE OF EXPERIMENTAL-DESIGN-BASED STUDIES

Salthouse's strong doubts about the existence of studies supporting the use-it-or-lose-it hypothesis extend beyond those based on nonexperimental surveys or comparisons of preexisting groups to include those that follow essentially experimental designs. His doubts that experimental studies support the use-it-or-lose-it hypothesis rest on two bases: empirical and theoretical. His empirical concerns stem from what he sees as the lack of generalizability from trained to untrained skills. However, given that psychologists have spent a fair amount of time and effort attempting to isolate theoretically and empirically independent psychological processes (e.g., Herrmann et al., 2001), it should not be too great a surprise that research designs that tend to focus on improving a carefully delimited cognitive function do not necessarily show much transfer or generalization to other functions. Nor is improvement in an even relatively delimited area of functioning either a useless functional gain or a clinching disproof of the use-it-or-lose-it hypothesis. Nevertheless, one could argue that cognitive interventions should include training in transferring the skills learned to a variety of tasks. Furthermore, given that many daily activities involve the functional integration of relatively independent psychological processes (Marsiske & Willis, 1995), it would seem desirable for cognitive-intervention protocols to include training in selecting from the armamentarium of cognitive processes those processes that can be most effectively integrated and brought to bear to deal with the problem at hand.

Interestingly, recent experimental studies that have provided evidence of generalization in older adults have focused on providing training applicable to a variety of tasks. They have done so either by promoting a particular strategy that is applicable in diverse tasks or by training people in a particular skill over a multiplicity of activities. Following the former approach, in an experiment using word-series, letter-series, and letter-sets tasks, Saczynski, Margrett, and Willis (2004) found that training older individuals in inductive reasoning led them to use effective strategies (e.g., underlining repeated letters or words in a series) that, in turn, significantly improved performance on all three tasks. The improvement on the letter-series and letter-sets tasks remained significant when assessed in a 3-month posttest. In another experiment with older adults, Dunlosky, Kubat-Silman, and Hertzog (2003) found that “training a monitoring skill—self-testing—can improve older adults' learning” (p. 340).

Following the approach of directly training people in a particular skill—speed of processing—through training sessions involving different types of tasks with different demands, Edwards et al. (2002), (2005) found that, although their speed-of-processing training did not generalize to psychometric measures of other cognitive domains, the training improved the performance of older adults not only on the Useful Field of View measure (Ball & Owsley, 1993), which tests rapidity of processing multiple stimuli across the visual field, but also on the Timed Instrumental Activities of Daily Living test (Owsley, Sloane, McGwin, & Ball, 2002), which involves laboratory measurement of timed tasks that emulate instrumental activities of daily living. The gains appear to be particularly reliable for individuals with initial processing-speed or processing difficulties. Roenker, Cissell, Ball, Wadley, and Edwards (2003) reported generally similar findings for the effect of the computerized speed-of-processing training protocol on driving. They found that although speed-of-processing training did not affect performance on a variety of generally used cognitive tests (e.g., Trails A and B or Stroop), such training not only positively affected function on a laboratory test constructed to mimic performance in real-life situations (i.e., the Road Sign Test administered in a driving simulator, Roenker et al., 2003), but also improved actual driving.

Evidence for the potentially relatively long-lasting effects of cognitive training is provided by a study conducted by the ACTIVE (Advanced Cognitive Training for Independent and Vital Elderly) Study Group (Ball et al., 2002). Participants were randomly assigned to one of four groups: a control group and three experimental groups. Each of the experimental groups received 10 sessions of group training in one area of cognitive functioning (i.e., memory, reasoning, or speed of processing). Although the significant training effects were limited to the cognitive function trained and did not seem to carry over to the measures of everyday function used, they endured through the 2-year follow-up—the effects being stronger among participants who received four-session booster training after 11 months than among those who received no booster training. The training effects were equal in magnitude to the amount of decline expected over 7- to 14-year intervals in elderly persons without dementia. As the authors observed, the notable ceiling effects due to the relatively high initial level of functioning of the experimental and control participants, taken together with the powerful practice effects resulting from the control subjects' 5 hr of practice on cognitive testing, may explain why there were no significant differences between the experimental and control groups in generalization to the daily-functioning measures.

The likelihood of finding further experimental evidence of some forms of learning generalization in older adults is increased by the findings of the SSES survey research program. As Ceci (1990) noted, the program's findings based on its 1964 and 1974 survey waves provided as strong evidence as then existed that a range of cognitive skills and lessons learned in meeting the cognitive demands of one's environment can be transferred to meeting the cognitive demands of other environments. As described earlier, further findings, based on analyses including the data from the 1994–1995 survey wave, indicate that such transfer occurs at least as much in older as in younger adults (e.g., Schooler et al., 1999).

Salthouse's strongest doubts about whether the experimental studies provide acceptable evidence for the use-it-or-lose-it hypothesis were essentially based on the same theoretical concerns as were his doubts about almost all of the other types of studies he discussed—the lack of a significant interaction indicating that the given intervention or experience reduces the rate of decline more for older than younger individuals. It is not as though Salthouse did not see the potential benefits of mental (or physical) exercise very clearly. As he stated:

Still another reason for an optimistic perspective on the role of mental exercise may be related to a particular conceptualization of mental functioning in which there is some absolute threshold for functioning …. If a threshold of this type exists, then anything, perhaps including increased mental activity, that will increase the distance of one's level of functioning from that threshold will likely prolong the interval until the critical level of performance is reached. For example, if an individual is 6 units away from the threshold and is declining at a rate of 3 units per year, then he or she will reach the threshold in 2 years. However, if the individual's level of functioning could be increased by 3 points, then even without affecting the rate of age-related decline, the time until the threshold is reached will be increased from 2 to 3 years. Under these circumstances, therefore, an increase in the individual's level of functioning would have the effect of slowing the progression to the critical level of functioning. Although an outcome of this type would not be considered evidence for the mental-exercise hypothesis according to the argument developed here, the practical benefits of this kind of finding could be enormous and might lead to an understandable lack of concern about the theoretical issue of whether there is an effect on the rate of age-related change. (p. 84)

More generally, it seems that if one spate of exercise increases an older person's level of intellectual functioning so that it would take that individual longer to decline to a given lower level of intellectual functioning, then, assuming that the asymptote of possible exercise-based improvement has not been reached, two exercise spates would mean it would take that individual even longer to reach that lower level than if he or she had engaged in only one spate of exercise.

The same logic that suggests that two spates of exercise would be better than one spate also suggests that three spates would be better than two spates. One must again bear in mind the possibility of reaching an asymptote where further exercise does not result in an improvement of function. Nevertheless, it seems reasonable to ask whether continuous exercise, with its possibly continuous bumping up of the individual's level of cognitive functioning, would eventually lead to a slowing of that individual's rate of decline compared with what it would have been had that individual not done the exercise. Long-term experimental studies testing this possibility have not been carried out. The findings of the longitudinal SSES studies on the cognitive effects of carrying out cognitively demanding paid work, housework, and leisure-time activities suggest that mental exercise does improve cognitive functioning. Salthouse's own data also suggest that carrying out relatively complex tasks may reduce the slope of cognitive decline. In the analysis of crossword-puzzle experience that he reported (Fig. 6), at the youngest age (around 25 years), the reasoning ability of individuals in the lowest quartile of crossword-puzzle experience was about the same as or possibly slightly higher than the reasoning ability of those in the highest quartile of crossword-puzzle experience. At the oldest age (around 75 years), individuals in the highest quartile of crossword-puzzle experience showed significantly greater levels of reasoning ability than those in the lowest quartile, and, in fact, the relative advantage of being in the highest quartile did seem to increase with age. Of course, older individuals with higher reasoning ability may be more likely to do crossword puzzles than those with lower reasoning ability, and one cannot specify how much such self-selection may have affected the results. Nevertheless, Salthouse's findings are congruent with the possibility that mental exercise can slow the rate of mental decline.

CONCLUSIONS

In common usage, the saying “use it or lose it” does not necessarily imply a change in rate of decline. All it implies is that people are more likely to “lose it” if they do not “use it.” Given that everyone will “lose it” in the end, at issue is whether a given person is likely to function at a higher cognitive level for a longer period of time if he or she exercises mentally (or physically). One cannot say that this is always the case, and most likely it is not. Nor can one confidently specify the extent to which the benefits of various particular types of mental exercise generalize across different types of cognitive functions. Nevertheless, the weight of the experimental and nonexperimental evidence strongly suggests that the hypothesis is generally correct—if a given person exercises mentally, that person is likely to function to some degree better for longer than if he or she had not done that exercise. Furthermore, taken together with all of the other evidence discussed here, even Salthouse's crossword-puzzle findings suggest that it is at the very least premature, and most probably wrong, to rule out, as Salthouse did, “the idea that the rate of mental aging is moderated by amount of mental activity” (p. 68).

Salthouse did recommend that “people should continue to engage in mentally stimulating activities because … there is no evidence that it has any harmful effects, … and [it] may contribute to a higher quality of life” (p. 84). In this he was correct, but the reasons for engaging in such activities go further. The available evidence distinctly points to the probability that for older individuals, mental exercise has a positive effect both on the level of cognitive functioning and on the probable rate of decline. It remains for further research to determine how much of an effect there is, who can benefit from it, and exactly which kinds of activities have specific kinds of effects. Even if the whole story is not yet known, in regard to cognitive function, at some level and to some degree, “using” it often delays the eventuality of “losing” it.