Compare the same network as a node-link drawing and an adjacency matrix. Hover over a node to highlight its links, or over an edge to highlight that edge and both endpoints. Click node labels to select them. In the matrix, hover over any cell to shade its entire row and column.
Red: selected nodes and their linksGreen: current hoverMatrix hover uses 50% transparent green bands. Filled and empty cells remain visible.
Untimed practiceNo time limit
Enter labels in order, separated by commas, or click labels to append them. Start and end at the specified nodes. Any valid path is accepted; it need not be shortest. Nodes cannot repeat.
Your completed trials
Responses stay in this page. Export the CSV file before closing it. Practice and timed responses are labelled separately.
The original 36 participants each saw nine graphs in both representations and answered seven questions per graph: 126 questions per participant. Graph sizes were 20, 50 and 100 nodes, with three edge-count levels.
The seven tasks were estimating node count, estimating edge count, finding the most connected node, finding a named node, checking a direct link, finding a common neighbour, and finding a path.
Participants first practised on two demonstration graphs. Representation order was randomised. Graph order was randomised within a group of three easier graphs and a group of six other graphs. Endpoints were selected in advance for the three questions involving two nodes.
Each question had a 45-second limit. Participants could skip. The session included two five-minute breaks and a ten-minute break between representations. Both interfaces supported red selection and green hover highlighting, including both endpoints when hovering an edge.
Scope of this teaching reconstruction
This page implements two tasks and a short four-trial session. Each representation appears in two consecutive trials; which representation comes first is random. Practice can show both views. Timed trials show one view and lock the graph and task controls.
New graphs and layouts are bundled. The six-node example is the constructed graph on slide 9. Larger graphs use fixed seeds and exact edge counts m = round(d²n²), following the paper’s printed d = √(m/n²), with d = 0.2, 0.4 or 0.6. The displayed proportion m / [n(n−1)/2] uses the conventional definition for undirected graphs.
The original graph files and trial records were not located. These new graphs use Fruchterman–Reingold positions rather than the original neato positions. The original highest-degree boost for the most-connected-node question is omitted because that question is not implemented here.
Question targets may differ between representations. The path-entry interface, side-by-side inspection, translucent matrix row/column hover bands, local CSV export and feedback after each answer are teaching additions. The paper does not fully specify its response-entry and path-scoring rules; this page accepts any valid simple path, or a correct “no path” response.
The page records new responses. It does not recreate the original participants’ accuracy or response times, or establish a reliable ranking from four trials.