Eye position data files were analyzed retrospectively. Eye position was converted to retinal coordinates, and the retinal location corresponding to the center of the fixation target for each recorded fixation was plotted on the retinal image (
Fig. 1). Fixation stability was recorded for each trial by measuring the area of an ellipse that encompasses 68% of fixation positions (bivariate contour ellipse area [BCEA]).
19
An analysis adopted from spatial statistics was used to assess whether subjects used the same regions of the retina for fixating points and fixating words. The data arising from this experiment are a random variable from a spatial point process; therefore, an appropriate method from the catalog of methods for analyzing spatial statistics is required. In this analysis, the fixations recorded in two-dimensional space are assumed to be values generated from a multivariate spatial point process, with each point being 1 of 2 qualitatively distinguishable types. Spatial segregation is assumed to occur if, within some area, particular types of points predominate rather than being randomly intermingled. A technique modified from Diggle et al.
20 and implemented in the spatialkernel and splancs packages (Barry Rowlingson, Peter Diggle, adapted, packaged for R by Roger Bivand, pxp functions by Giovanni Petris, and goodness of fit by Stephen Eglen. Splancs: spatial and space-time point pattern analysis, R package version 2.01-24.
http://www.maths.lancs.ac.uk/∼rowlings/Splancs/.) using the open-source statistical programming environment R
21 –23 was used to assess this segregation of the fixation points. The purpose-written R script is given as
Supplementary Material, and is illustrated using data from the two control subjects and two patients. In short, this method gives a probability value against the null hypothesis that the fixation coordinates from the different tasks are yielded from the same spatial point process and are, therefore, not segregated. Some modifications from the original method were made. First, several fixations fall on exactly the same location (with the same coordinates) because of the measurement precision of the instrumentation. Moreover, the technique is very sensitive to small levels of segregation especially when the sample of points is large, as in the case with these data. To overcome these limitations, random Gaussian noise was first added to every recorded (
x,
y) fixation for both controls and patients. This also has the effect of randomly “mixing” the two spatial processes further. For the purpose of this analysis, it is assumed that if a statistically significant effect remains after this mixing, it provides clear evidence that the fixations for the two tasks are indeed generated from a different spatial process. The method uses a Monte-Carlo technique to derive probability values against the null hypothesis.