Measures

Participants completed computerised versions of the LST (experimentally

manipulated to include Basic and DC items), symmetry complex span (SSPAN), and Raven’s Advanced Progressive Matrices (APM). Participants also completed the Arithmetic Chain Task though these results are not reported here. All tasks were programmed with Inquisit Lab

4 (Millisecond Software, 2014).

Latin-Square Task (LST)

Participants were presented 24 items adapted from Birney and Bowman (2009) across two blocks (basic and DC). All items used the same four shapes as element types (circle, triangle, square, cross) and each had a 2-minute time-limit (with a countdown displayed to participants). If the time expired, the item was recorded as incorrect and the next item presented. Bateman (2015) reported that only 0.2% of all LST items attempted were marked incorrect through timeout, indicating that 2 minutes is sufficient. Each item had an RC (RC=2D/3D/4D) and steps (steps=1S/2S) combination (e.g., 2D-1S). Each block had 12 items with an equal distribution of RC*steps combinations (two of each combination).

Participants completed the basic block and the DC block in a random order, with the blocks visually differentiated by screen colour (white for basic; green for DC). There were separate instructions related to each (along with practice items) presented at the beginning of the experiment.

LST Basic. Basic items included the matrix and response options in the centre of the

screen (see Figure 3.2). When a participant clicked a shape in the response options, the background of the selected shape would turn pink and reset/confirm buttons would appear. This gave participants a chance to confirm or change their response before moving on to the next item.

Figure 3.2. Example Basic item, as presented to participants.

LST Dynamic Completion (DC). DC items allowed participants to fill in the matrix

before solving the target cell. To fill a cell in the matrix, participants selected their desired shape from the options then clicked on an empty cell (see Figure 3.3). The shape would appear in the cell and the cell background would change to pink to indicate it was an interim shape they had inserted. Participants could fill as many cells as they wished, though the instructions asked they only fill as many cells as necessary. A ‘move’ was recorded as any time a cell was filled by the participant and moves continued to cumulate regardless of resets.

Participants indicated their solution by placing the desired shape into the target cell in the same way as an empty cell (in this way, the minimum moves for each item was one). Only when a shape was placed into the target cell would the reset/confirm buttons appear for them to confirm their answer.

Figure 3.3. Slide from the DC instructions, demonstrating an example item.

Symmetry Span

Participants were presented with alternating storage and processing tasks, as in Kane et al. (2004). For the storage task, participants viewed a 4x4 grid where a sequence of red squares would appear in one of the 16 potential locations. Two to five squares would appear in each set with each square appearing for 850ms. For the processing task, participants judged whether a displayed pattern was symmetrical along the vertical axis. Participants first viewed one square in the set, then completed a symmetry judgement, then viewed another square, then made another symmetry judgement, and so on until the entire storage set of squares had been displayed. After solving the last symmetry judgement for that set, participants would attempt to recall the squares in the order they were presented. The score analysed was the total number of correctly recalled squares across the task (2 x each set size), resulting in a

possible range of 0-28. This partial scoring was favoured over a ‘span’ score (number of recalled squares within correctly recalled sets only) as it captures more variance (Redick et al., 2012).

Raven’s Advanced Progressive Matrices (APM)

Fluid intelligence was measured using a shortened 20-item version (odd items + items 34 and 36) of set II of the APM (J. C. Raven, 1941). Participants had 20 minutes to solve as many items as possible. This 20-item version has shown excellent reliability as a shortened version of the APM as it is sufficient for participants to learn and apply the rules that govern APM items (Bui & Birney, 2014).

Although a single task defines a construct such as fluid intelligence narrowly, this task was also chosen for its important surface and structural similarities to the LST. Both tasks employ a visuo-spatial matrix layout, and both are based on relational integration. The LST differs in that there is a single rule known to participants (the defining rule that only one of each type of shape can appear in each row and each column), while APM involves several unknown rules (Carpenter et al., 1990) that the participant must induce. In Chapter II, rule induction was identified as a defining characteristic of abstract reasoning tasks which set them apart from relational integration tasks such as the LST. However, APM elements are also generally more complex. Where the LST involves the same set of shapes (circle, triangle, square, cross) each time, APM elements are complex, with each element in a cell composed of multiple features. For instance, lines may, inter alia, be straight, wavy, dotted, and/or differ in orientation; shapes may, inter alia, differ in size, shading, numerosity, and/or form. Element complexity such as this is necessary to ensure the rules are being generalized across features. In the LST, changing the elements between items is unnecessary because the rules are given each time. Thus, although element complexity can be absolved into rule

induction, it is important to consider that the difference in complexity between the two tasks could contribute to additional discrepancy in their intercorrelation.