(A) fMRI version of Cattell task. On each trial (each row), participants select the odd-one-out from four panels with a single finger button-press (green circles). Condition blocks (30 seconds) alternate between easy vs. hard puzzles. (B) Behavioural age-related decline. Performance (correct minus incorrect in fMRI version of Cattell task) significantly declined linearly with age (upper). High reliability was observed between performance measures from the standard Cattell task and the modified version used for fMRI (lower). In the upper panel, the black line represents the fitted-regression estimates with shaded 95% confidence intervals. In the lower panel, the black line represents perfect correlation between the two Cattell versions. (C) Univariate task effect. Whole-brain voxel-wise activations for solving the puzzles in the hard, relative to easy, blocks, after threshold-free cluster enhanced (TFCE) correction (slices are labelled with z MNI coordinates).

Univariate analysis. (A) Whole-brain effects of age and performance. Age (green) and performance (red) positively predicted unique aspects of increased task activation, with their spatial overlap (yellow) being overlaid on a template MNI brain, using p < 0.05 TFCE with z MNI coordinate labels. (B) Intersection ROIs. A bilateral cuneal (magenta) and frontal cortex (brown) ROI were defined from voxels that showed a positive and unique effect of both age and performance (yellow map in Figure 2A). (C) ROI Activation. Activation (raw = left; RSFA-scaled = right) is plotted against behavioural performance based on a tertile split between three age groups (19-44, 45-63 & 64-87 years).

Standardised coefficients in multiple regression predicting fMRI activation (Hard - Easy) as a function of Age and Performance for the two ROIs identified in Figure 2.

Note that the p-values for the main effects of Age and Performance are biased by the selection of these voxels. RSFA = scaled by Resting-State Fluctuation Amplitudes (see text).

Multivariate Analysis. (A) MVB Decoding. Points represent the difference in log evidence per participant (for the real vs. shuffled model) for the joint model using activation patterns to decode the [Hard > Easy] contrast. (B) Boost Likelihood Model Comparison. Across age, a smoothed density estimate represents the likelihood that there was a boost (of log-evidence > 3; green) or no difference (grey) to model evidence per participant when decoding models included activation patterns from either of the compensation ROIs (Figure 2B) in addition to the MDN (Figure 1C), relative to a model that sampled only from the MDN. A significant positive linear effect of age on boost likelihood was observed for the cuneal (lower) but not frontal ROI (upper). (C) MVB Spread. Points represent the spread (standard deviation) of multivariate responses, showing a reduction with age in frontal cortex.

Standardised coefficients in MVB multiple logistic regression analyses predicting Boost likelihood as a function of age or the spread of voxel weights (with Sex and Mean Univariate Activation as covariates).