Hi Catherine,
Your regression model is
Y_i = beta_0 + beta_1 X_i + e_i,
where i is the subject index ranging from 1 to 12, Y_i = % signal change, X_i = reaction time, e_i = random error.
The problem with the current model is that you only have one sample for each specific reaction time, in which case regression analysis is not possible. To get around the problem, you may consider breaking the whole reaction time range into several time bins. For example, consider 4 or 6 subintervals for [600, 1800]. In the case of 4 intervals you have (600, 900], (900, 1200], (1200, 1500], and (1500, 1800]. Within each interval you use the mean (for example, 750 for (600, 900]) to replace whatever original value (for example, 657). This way you will avoid the sample size problem.
Gang