The Relationship Between Older Adults’ Subjective Age and Perceived Effort on Cognitive Tasks

ABSTRACT

Background

Although engagement in cognitively-demanding activities is beneficial for older adults, research suggests that older adults may be less motivated to engage in these types of activities because of the increased age-related costs associated with task engagement and their perceptions of the task demands.

Methods

Across three studies, we investigated if older adults’ subjective age predicted their perceptions of effort over the course of a working memory task. Younger and older adults reported their subjective age and then completed an increasingly difficult series of working memory trials, indicating perceived task demands and effort after each trial.

Results

Results from all three studies showed that there was no age difference in performance or in perceptions of task difficulty, contrary to previous results. Also, there was no significant association between older adults’ subjective age and perceived effort, suggesting that subjective age may not be a reliable predictor of perceptions of task demands in older adults.

Discussion

Participant characteristics and the testing environment may play a role in determining the relationship between subjective age and perceived effort.

Author Note

Meltem Karaca, Department of Psychology, University of Massachusetts Lowell, Lisa Geraci, Department of Psychology, University of Massachusetts Lowell, Robert Tirso, Student Affairs Planning, Assessment & Research, Texas A&M University, and Jonathan Aube, Department of Psychology, University of Massachusetts Lowell. Portions of these data were presented at the 19th Biennial Cognitive Aging Conference in Atlanta, Georgia, April 2022.

Acknowledgments

We thank Thomas Hess for providing us with the Letter-Number Sequencing Task.

Disclosure Statement

No potential conflict of interest was reported by the author(s).

Author Contributions

MK contributed to study design, data collection, data analyses, data interpretation, and writing of the manuscript. LG conceived of the research and contributed to study design and writing of the manuscript. RT contributed to data analyses, data interpretation, and provided critical feedback. JA contributed to study design, data processing, and provided critical revisions. All authors approved the final version of the manuscript.

Data Availability Statement

The data are available from the corresponding author at [email protected], [MK], upon request.

Ethics Approval for Research Participants Statement

These studies were approved by the University of Massachusetts Lowell Institutional Review Board (20-014-GER-EXM).

Notes

1. In Study 1, younger adults’ total response time in the working memory task (across 48 trials) was, on average, 659.97 seconds, whereas older adults’ was 606.17 seconds. In Study 2, younger adults’ total response time was, on average, 445.04 seconds, and older adults’ was 434.13 seconds. In Study 3, total response time was, on average, 480.22 seconds for older adults and 401.28 seconds for younger adults.

2. For all six sets of MLM models, we used multi-modal inference to determine whether estimating random intercepts and random slopes for both linear and quadratic terms of set size was a better fit compared to estimating only a random intercept, or a random intercept and a random slope just for the linear term. When we estimated a random effect for the quadratic slope, the model either failed to converge or the variance estimate for the quadratic slope was close to zero. Therefore, we did not estimate a random effect for the quadratic slope. When we compared the random intercept-only model to the random intercepts and random linear slope model, Chi-square difference test suggested that including random intercept and random slope for the linear term of set size produced an improved model fit compared to estimating only random intercepts (p < .001) for all models. Across six MLM models, both Bayesian Information Criterion and Akaike Information Criteria also favored the model which included random intercepts and random slope for the linear terms of set size (i.e., BIC and AIC were smaller). Thus, we estimated random intercept and random slope for the linear term of set size in the final model for all six scales.

3. We used multi-modal inference to determine whether estimating random intercepts and random slopes for linear term of set size was a better fit compared to estimating a random intercept-only. Chi-square difference test showed that including random intercepts and random slopes for linear term of set size produced an improved model fit (p < .001). In addition, both Bayesian Information Criterion and Akaike Information Criteria favored more complex model (i.e., BIC and AIC were smaller). We should note that when we estimated random effects for both linear and quadratic components of set size, we observed that the variance estimate for the quadratic random effect was close to zero (2.14). Thus, we removed the quadratic term of set size and only estimated linear term of set size for the random slope.

4. For all six sets of MLM models, we used multi-modal inference to determine whether estimating random intercepts and random slopes for both linear and quadratic terms of set size was a better fit compared to estimating only a random intercept, or a random intercept and a random slope just for the linear term. When we estimated a random effect for the quadratic slope, the model either failed to converge or the variance estimate for the quadratic slope was close to zero. Therefore, we did not estimate a random effect for the quadratic slope. When we compared random intercept-only model to random intercepts and random linear slope model, Chi-square difference test suggested that including random intercept and random slope for the linear term of set size produced an improved model fit compared to estimating only random intercepts (p < .001) for all models. Across six MLM models, both Bayesian Information Criterion and Akaike Information Criteria also favored the model which included random intercepts and random slope for the linear terms of set size (i.e., BIC and AIC were smaller). Thus, we estimated random intercept and random slope for the linear term of set size in the final model for all six scales.

5. Multi-modal inference was used to determine whether estimating random intercepts and random slopes for linear term of set size was a better fit compared to estimating a random intercept-only. Chi-square difference test showed that including random intercepts and random slopes for linear term of set size produced an improved model fit (p < .001). In addition, both Bayesian Information Criterion and Akaike Information Criteria favored more complex model (i.e., BIC and AIC were smaller). To note, we tried estimating random slopes for both linear and quadratic components of set size, but the model failed to converge, and the variance estimate for the quadratic random effect was close to zero (2.41). Thus, we removed the quadratic term of set size and only estimated linear term of set size for the random slope.

6. For all six sets of MLM models, multi-modal inference was used to determine whether estimating random intercepts and random slopes for both linear and quadratic terms of set size was a better fit compared to estimating only a random intercept, or a random intercept and a random slope just for the linear term. When we estimated a random effect for the quadratic slope, the model either failed to converge or the variance estimate for the quadratic slope was close to zero. Therefore, we did not estimate a random effect for the quadratic slope. When we compared random intercept-only model to random intercepts and random linear slope model, Chi-square difference test suggested that including random intercept and random slope for the linear term of set size produced an improved model fit compared to estimating only random intercepts (p < .001) for all models. Across six MLM models, both Bayesian Information Criterion and Akaike Information Criteria also favored the model which included random intercepts and random slope for the linear terms of set size (i.e., BIC and AIC were smaller). Thus, we estimated random intercept and random slope for the linear term of set size in the final model for all six scales.

7. Multi-modal inference was used to determine whether estimating random intercepts and random slopes for linear term of set size was a better fit compared to estimating a random intercept-only. Chi-square difference test showed that including random intercepts and random slopes for linear term of set size produced an improved model fit (p < .001). In addition, both Bayesian Information Criterion and Akaike Information Criteria favored more complex model (i.e., BIC and AIC were smaller). When we tried estimating random slopes for both linear and quadratic components of set size, the model failed to converge, and the variance estimate for the quadratic random effect was close to zero (1.29). Thus, we removed the quadratic term of set size for the random slope and only estimated linear term of set size for the random slope.