無母數適測方法結合信心水準評定在閾值研究之應用

Abstract

在心理物理學中，無母數適測方法可應用於測量偵測與區辨閾值。在閾值研究中，多數研究使用兩選項強迫選擇作業(2-alternative forced-choice paradigm)，配合此作業，受試者必須從兩個選項之中強迫選出包含目標刺激者或刺激強度較大者。近來，有一些學者開始討論配合無母數適測方法進行實驗，且提供受試者更多的反應選項來收集受試者對反應的信心，透過收集反應的信心能夠增進閾值或參數估計的效率(Kaernbach, 2001; Klein, 2001; Garcia-Perez, 2010)。 在本篇研究中，將選用無母數適測方法中之「上下加權調整法」(weighted up-down procedure)進行實驗，並使用反應工具─鍵盤或視覺類比量尺(visual analogue scale)─配合兩、三或四點之反應信心量尺來估計區辨閾值，比較不同反應信心量尺之閾值估計效率。另一方面，本篇研究也使用預先選定刺激強度之非適測方法進行閾值估計，用以比較上下加權調整法與非適測方法兩者在不同反應信心量尺中閾值估計的一致性。 結果顯示，以非適測方法估計結果為基準，上下加權調整法在不同反應信心量尺中，閾值估計都有些微偏差，支持過去模擬研究之結論。另外，單看上下加權調整法在不同反應信心量尺中之閾值估計，結果顯示，三或四點之反應信心量尺在估計效率上些微勝過兩點。最後，視覺類比量尺作為反應工具需要配合較多點數之反應信心量尺，才能得到與鍵盤反應一致之估計結果。

In the psychophysical research field, detection or discrimination thresholds can be estimated by non-parametric adaptive procedures. The two-alternative forced-choice paradigm is commonly used for the experiment. In this paradigm, participants have to choose the alternative that contains the target stimulus. Recently, some researchers conducted psychophysical experiments using non-parametric adaptive procedures that incorporate the confidence rating of responses from participants. The efficiency of thresholds and parameters estimations was enhanced by the confidence rating of responses (Kaernbach, 2001; Klein, 2001; Garcia-Perez, 2010). In this thesis, the weighted up-down procedure, a non-parametric adaptive procedure, was used to conduct experiments. Two response tools - keyboard and visual analogue scale — and the confidence rating of two-, three-, and four-response options were used to collect responses, comparing the efficiency of discrimination threshold estimations among different options of the confidence rating. We also used a non-adaptive procedure to measure the thresholds and tested the consistency of threshold estimations between the non-adaptive and weighted up-down procedures. To conclude, threshold estimations by the weighted up-down procedure were biased compared to the estimations by the non-adaptive procedure. And this result was consistent with previous simulation studies. The results in the weighted up-down procedure showed that tasks with three-, and four-response options were more efficient methods on estimating thresholds than the task with the two-response option. Moreover, threshold estimations by the visual analogue scale were inconsistent with those with keyboard, unless the former was incorporated with more response options of confidence rating.