Capturing Dynamics of Biased Attention: Are New Attention Variability Measures the Way Forward?

Abstract

BACKGROUND: New indices, calculated on data from the widely used Dot Probe Task, were recently proposed to capture variability in biased attention allocation. We observed that it remains unclear which data pattern is meant to be indicative of dynamic bias and thus to be captured by these indices. Moreover, we hypothesized that the new indices are sensitive to SD differences at the response time (RT) level in the absence of bias. METHOD: Randomly generated datasets were analyzed to assess properties of the Attention Bias Variability (ABV) and Trial Level Bias Score (TL-BS) indices. Sensitivity to creating differences in 1) RT standard deviation, 2) mean RT, and 3) bias magnitude were assessed. In addition, two possible definitions of dynamic attention bias were explored by creating differences in 4) frequency of bias switching, and 5) bias magnitude in the presence of constant switching. RESULTS: ABV and TL-BS indices were found highly sensitive to increasing SD at the response time level, insensitive to increasing bias, linearly sensitive to increasing bias magnitude in the presence of bias switches, and non-linearly sensitive to increasing the frequency of bias switches. The ABV index was also found responsive to increasing mean response times in the absence of bias. CONCLUSION: Recently proposed DPT derived variability indices cannot uncouple measurement error from bias variability. Significant group differences may be observed even if there is no bias present in any individual dataset. This renders the new indices in their current form unfit for empirical purposes. Our discussion focuses on fostering debate and ideas for new research to validate the potentially very important notion of biased attention being dynamic.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 91 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.