Older adults show increased chunking behavior with extended practice

Older adults seem to have a reduced ability to chunk keying sequences. However, it remains unclear whether they are unable to use chunking strategies or whether they are just slower in developing them. Using a discrete sequence production (DSP) task with extended practice, we here provide evidence supporting slower development. Across 3 to 4 hours of practice, older adults appeared to improve on multiple measures that are thought to be indicative of chunking behavior. A group of healthy older adults (n = 18, age 74 - 85) visited the lab on two days. For subject characterization we administered the Montreal Cognitive Assessment, a digit symbol substitution task and a visual working memory task. The DSP task was performed using a standard computer and keyboard. A sequence of 3 and a sequence of 6 elements were both practiced 432 times. The two sequences were presented in random order with a short break after every sequence. After the practice phase, an explicit knowledge questionnaire was administered. Then, a test phase with a block of 24 random trials and a block with 24 trials of learned sequences was performed. Chunking behavior is associated with a prolonged movement time for the first element of a sequence and shorter movement times for subsequential elements. The difference between the first and the average of the following movement times can be used as an index for the extent of chunking behavior. Our results show that this simple chunking index continued to increase over the extended practice period, with a steeper curve for the 3- than for the 6-element sequence. To control for general task learning, we also compared the magnitude of this chunking index between the random and learned sequences of the test phase. Based on this comparison, we find that our participants show more chunking behavior than participants in previous studies with less practice. Again, this effect was stronger in the 3- than in the 6-element sequence. As an additional test of chunking behavior in elderly, we analyzed our data using a Bayesian chunk inference algorithm that was recently published. First, we randomly shuffled the movement times within trials for all participants; this should remove most indications of chunking behavior while other properties like the mean, standard deviation and general learning effects remain. Then, we fitted the model to the original data and to the shuffled data for each participant. We found that the model fits the original data better than the shuffled data, indicating that chunking behavior at least partly explains the variance in reaction times in our data. Participants in our study continued to increase chunking behavior until the last phase of an extended practice period showing no ceiling effect. Hence, our results support the idea that that reduced chunking behavior in older adults results, at least partly, from slower chunk development.