Microsaccades Are an Index of Covert Attention

Horowitz, Fine, Fencsik, Yurgenson, and Wolfe (2007, this issue) investigated the relation between microsaccade congruency (MC, the congruency between the direction of the microsaccade and the location of the target) and reaction time (RT) in a spatial cuing paradigm and concluded that ‘‘fixational eye movements are not an index of covert attention.’’ We show that microsaccade direction is a reliable on-line measure of attention that potentially indexes effects beyond those reflected in RT. In Posner’s (1980) task, the spatial cue is the only objective marker of attention. Therefore, a cue-validity effect is a necessary property of any index of attention. Such an effect has been established for both RT (i.e., the RT cue-validity effect) and microsaccades (i.e., the MC effect; Engbert & Kliegl, 2003; Galfano, Betta, & Turatto, 2004; Hafed & Clark, 2002; Rolfs, Engbert, & Kliegl, 2004, 2005). Directional biases in microsaccades are not just an oculomotor reflex: If a task requires shifts in spatial attention, directional biases are elicited by circular color cues; if the task does not require attentional shifts, directional biases are not elicited even by arrow cues (Engbert & Kliegl, 2003; Laubrock, Engbert, & Kliegl, 2005). Horowitz et al. compared (a) RTs from validly cued trials with microsaccades oriented away from the target and (b) RTs from invalidly cued trials with target-congruent microsaccades, arguing that if microsaccades are useful as measures of attention, then RTs in the latter trials should be faster than RTs in the former trials. Here we show that this result can be obtained even if microsaccades and attention are substantially correlated. Let us assume that microsaccade direction is a valid but imperfect measure of attention and that the probability of a microsaccade being oriented toward the location where attention is deployed, p(microsaccade directionjattention), is .75. Pitting MC against cue validity assumes that attention does not always follow the cue (e.g., because subjects match probabilities; Brunswik, 1939). Let us therefore further assume that the probability that attention does follow the cue, p(attentionjcue), is .8. Given these assumptions, attention and microsaccades will both be directed opposite the target on 15% of the valid-cue trials (.2 .75 of all valid-cue trials), but attention will follow the cue toward the target despite a target-incongruent microsaccade on 20% of the valid-cue trials (.8 .25). Similarly, invalidly cued trials with target-congruent microsaccades will be a mixture containing ‘‘attentional error’’ trials on which microsaccades follow attention toward the target (.2 .75 5 .15 of all invalid-cue trials) and trials on which microsaccades are directed toward the target but attention is not (.8 .25 5 .20). Hence, the cue-validity effect on RT (i.e., benefits and costs of valid and invalid cues, respectively) will dominate the MC effect in the conditions that Horowitz et al. compared. Even if microsaccades are well correlated with attention, this selection of conditions imposes prior odds of 4:1 against finding a benefit of MC in RT. Therefore, showing that RT is slower for trials with valid cues and target-incongruent microsaccades than for trials with invalid cues and target-congruent microsaccades is not evidence against the direction of the microsaccade reflecting covert attention, but a consequence of the cue controlling attention.