Fig 1.
After registering for the Lumosity program (green), participants were eligible for assessment with a neuropsychological test battery (NCPT) at two time points separated by at least 10 weeks (blue). In parallel, participants freely played games from the Lumosity CT program (yellow), with the amount of gameplay varying between participants. Changes in performance on the NCPT between assessments were examined as a function of the amount of intervening gameplay.
Fig 2.
Population pyramid for fully analyzed participants reporting age, gender, and a classifiable level of education (87%, N = 93,562).
Table 1.
Age, gender, and education of fully analyzed participants.
Fig 3.
Change in GI from T1 to T2 as a function of the amount of gameplay between T1 and T2 and amount of gameplay before T1.
Amounts of pre-T1 and T1-to-T2 gameplay are shown respectively on the x- and y-axes. The numbers on each axis are the upper limits of 10 bins of participants partitioned by number of games played. The amount of change in GI from T1 to T2 for each combination pre-T1 and T1-to-T2 gameplay is indicated by color.
Fig 4.
Distribution of T1-to-T2 gameplays for fully analyzed participants.
Note that the distribution is truncated at the 98th percentile.
Fig 5.
Empirical D-R function (dots) and exponential fit (line) for NCPT Grand Index (GI) involving all fully analyzed participants.
Dots show means for 20 equal-sized bins of participants partitioned by number of games played between T1 and T2. The line shows an exponential approach to an asymptote (Eq 1) fit to the 107,005 points corresponding to individual participants. The heights of the blue and yellow areas indicate respectively the effect of repeated testing and the maximum effect of transfer from gameplay on NCPT performance.
Table 2.
Linear regression of change in NCPT Grand Index on demographic factors while controlling for number of games played.
Table 3.
Intercept, rate, and asymptote–intercept of fit exponential functions for each level of each demographic factor.
Fig 6.
Medians and 95% confidence intervals of intercept, rate, and asymptote–intercept of exponential functions fit to each age level.
Medians (center points) and CIs (upper and lower edges) were determined through a bootstrapping procedure.
Table 4.
Intercept, rate, and asymptote–intercept of fit exponential functions for each age level.
Fig 7.
Exponential approach to an asymptote fit to D-R function at each age level.
The intercept, rate, and asymptote—intercept are shown in Table 4.
Fig 8.
Empirical D-R function (dots) and exponential fit (line) for each individual NCPT subtest.
Dots show means for 20 equal-sized bins of participants partitioned by the number of games played between T1 and T2. Lines show exponential functions (Eq 1) fit to the 107,005 points corresponding to individual participants. AR = Arithmetic Reasoning; DSC = Digit-Symbol Coding; FVMS = Forward Visual Memory Span; RVMS = Reverse Visual Memory Span; TMA = Trail Making A; TMB = Trail Making B; GR = Grammatical Reasoning; PM = Progressive Matrices.
Table 5.
Intercept, rate, and asymptote–intercept of fit exponential functions for NCPT subtests.
Fig 9.
Medians, 95% CIs, and comparisons between parameter estimates for individual NCPT subtests.
Left: medians (center points) and 95% CIs (upper and lower edges) for intercept, rate, and asymptote-intercept of exponential functions fit to each subtest. Right: statistical significance for each comparison of parameter estimates between subtests. Medians, CIs, and p-values were determined through a bootstrapping procedure. The Benjamini-Hochberg correction (False Discovery Rate, FDR) was used to control for multiple comparisons: ** = FDR < 0.01, * = FDR < 0.05, ^ = uncorrected p < 0.05, blank = uncorrected p > 0.05.
Fig 10.
Medians, 95% CIs, and statistical significance of correlations between age level and each exponential parameter for each NCPT subtest.
The medians (center points), 95% CIs (upper and lower edges), and p-values of the correlations were determined through a bootstrapping procedure. The Benjamini-Hochberg correction (False Discovery Rate, FDR) was used to control for multiple tests: ** = FDR < 0.01, * = FDR < 0.05, ^ = uncorrected p < 0.05, blank = uncorrected p > 0.05.