A general bias toward making oblique responses may produce spurious serial dependence in response-contingent analysis (
Fritsche, 2016;
Pascucci et al., 2019). To eliminate this potential confound, we applied a correction to the data as described above. To check that this correction was successful, we ran an alternate flip trial analysis (
Gallagher & Benton, 2022). This involves inverting the order of even-numbered trials to remove any temporal order effects. If an experiment consisted of 100 trials, trial 2 would be swapped with trial 100, trial 4 with trial 98, and so on. The oblique response bias can produce spurious serial dependence regardless of any real relationship between trials. We would therefore expect to still observe “serial dependence” in uncorrected flip trial data but not when the correction is applied.
Figure 5 shows the results of this analysis. Spurious serial dependence is evident in uncorrected response-contingent alternate flip data (non-oppositional stimulus-contingent bias = 0.56°,
p = 0.99,
g = 0.14; non-oppositional response-contingent bias = 2.18°,
p = 0.03,
g = 0.56; oppositional stimulus-contingent bias = 0.00°,
p = 0.55,
g < 0.01; oppositional response-contingent bias = 4.05°,
p = 0.04,
g = 0.53). Applying residualization removes this effect (non-oppositional stimulus-contingent bias = 0.56°,
p = 0.55,
g = 0.14; non-oppositional response-contingent bias = –0.61°,
p = 0.50,
g = 0.16; oppositional stimulus-contingent bias = 0.00°,
p = 0.99,
g < 0.01; oppositional response-contingent bias = –1.30°,
p = 0.39,
g = 0.21). As expected, these results confirm that residualization is necessary only for response-contingent analysis.