Description
Test whether mapped point locations such as taps, clicks or fixations are clustered, regular or random, and at what distance scale.
Heatmaps show where points are dense but not whether that density is more than chance or at what scale it occurs. Point-pattern analysis compares the observed locations with complete spatial randomness. Quadrat counts give a variance-to-mean ratio (1 under randomness). The Clark-Evans ratio compares mean nearest-neighbour distance with its expectation. Ripley's K function counts, with edge corrections, the expected number of further points within distance t of a typical point; its transform L(t) is straight under randomness, above for clustering and below for regularity. Stoyan and Stoyan add the pair correlation function, a non-cumulative alternative, and mark correlation for points carrying values such as dwell time. Significance comes from Monte Carlo envelopes: simulate 99 random patterns with the same number of points and see where the data curve leaves the band. If not random, cluster or inhibition models can be fitted.
Attributes
- Discipline
- Statistics & Modelling
- Loss type
- Unmigrated
- Stage
- Evaluation, Post-launch
- Data
- Quantitative
- Effort
- Days
Steps
- Define the study region precisely (screen, page area or map) and record every point; note its area and boundary.
- Optionally count points in quadrats and compute the variance-to-mean ratio as a first check.
- Compute nearest-neighbour distances and the edge-corrected Clark-Evans ratio (above 1 regular, below 1 clustered).
- Estimate K(t) with edge-correction weights and plot L(t) = square root of K(t)/pi against t; optionally estimate the pair correlation function.
- Simulate 99 (alpha 0.05) or 999 (alpha 0.01) random patterns with the same number of points in the same region, compute the same curves, and plot upper and lower envelopes; the p-value is the data's rank among simulations over m + 1.
- Read the distances at which the data leave the envelopes as the scale of clustering or regularity; for marked points, test marks by random relabelling.
- If the pattern is clustered because intensity varies across the region (for example near the fold), model the varying intensity before concluding that points attract each other.
Sources
- Spatial Statistics (Ripley, 1981)
- Fractals, Random Shapes and Point Fields (Stoyan & Stoyan, 1994)
- The Flying Bomb and the Actuary (Shaw & Shaw, 2019)
Also known as
- Point pattern analysis
- Spatial point pattern analysis
- Complete spatial randomness test
- CSR test
- Ripley's K function
- L function
- Pair correlation function
- Clark-Evans nearest-neighbour test
- Empty-space function
- Simulation envelopes
- Quadrat count analysis
Categories
- Analytics & Behaviour
- Statistics & Modelling