A criterion-shift model for enhanced discriminability in perceptual identification: A note on the counter model
JEROEN G. W. RAAIJMAKERS
0
1
2
3
0
University ofAmsterdam
,
Amsterdam, The Netherlands
1
LAEL J. SCHOOLER Pennsylvania State University, University Park
,
Pennsylvania
2
ERIC-JAN M. WAGENMAKERS and RENE ZEELENBERG University ofAmsterdam
,
Amsterdam, The Netherlands
3
Rene Zeelenberg was supported by a grant from the Foundation for Behavioral and Social Sciences of the Netherlands Organization for Scientific Research. We thank Rich Shiffrin for helpful discussions and this article can be addressed to E.-J. M. Wagenmakers or R. Zeelenberg, Department of Psychonomics
, Roetersstraat 15, 1018 WB Amsterdam,
The Netherlands (
A criterion-shift model for enhanced discriminability in perceptual identification: A note on the connter model The original version ofthe counter modeljor perceptual identification (Ratcliff & McKoon, 1997) assumed that word frequency and prior study act solely to bias the identification process (i.e., subjects have a tendency to prefer high-frequency and studied low-frequency words, irrespective of the presented word). In a recent study, using a two-alternative forced-choice paradigm, we showed an enhanced discriminability effect for high-frequency and studied low-frequency words (Wagenmakers, Zeelenberg, & Raaijmakers, 2000). These results have led to a fundamental modification of the counter model: Prior study and high frequency not only result in bias, but presumably also result in a higher rate offeature extraction (i. e., better perception). We demonstrate that a criterion-shift model, assuming limited perceptual information extracted from theflash as well as a reduced distance to an identification threshold for high-frequency and studied low-frequency words, can also account for enhanced discriminability.
-
sentation of the response alternatives LIED and DIED), Rat
cliff and McKoon claimed that subjects tended to prefer
the studied alternative, regardless of whether that alterna
tive had been flashed or not. Such a tendency would lead to
benefits when the target (e.g., LIED) had been studied, but to
costs when the foil (e.g., DIED) had been studied. Further,
they found that the size ofthe benefits about equaled the size
of the costs (e.g., Ratcliff, Allbritton, & McKoon, 1997;
Rouder, Ratcliff, & McKoon, 2000). Hence, the effect of
prior study was supposed to reflect a bias rather than some
kind of enhancedperceptualprocessing of the flashed word.I
Additional evidence for this idea came from the observation
that even if performance was at chance when neither alter
native was studied, effects of prior study were still present
and thus appeared to be independent of information ex
tracted from the flashed stimulus. Similarly, subjects had a
preference to choose a high-frequency (HF) alternative
such as MILE over a low-frequency (LF) alternative such as
TILE. Therefore, effects of word frequency were likewise at
tributed solely to bias. Importantly, the biases for prior
study and word frequency are supposedly mediated by dif
ferent mechanisms, an issue debated by Wagenmakers,
Zeelenberg, and Raaijmakers (2000). We will return to this
later. For ease of reference, we will term the original ver
sion of the counter model the counter model I.
The counter model I is one of the few models to pro
vide a quantitative account of repetition priming effects
in visual word identification. The model successfully ac
counted for the effects of prior study and word frequency
in three different word identification tasks: naming (or
free response identification), forced-choice identifica
tion, and yes-no identification. Recent studies by Bowers
(1999) and Wagenmakers et al. (2000), however, have
shown that the counter model made some incorrect pre
dictions. First, in a two-alternative forced choice para
digm, a choice between two HF alternatives was found to
be more accurate than a choice between two LF alterna
tives. Second, prior study of both alternatives improved
performance, albeit only for LF words. These results sug
gest problems for the counter model I, because it does not
predict that prior study and word frequency affect the sub
ject's ability to discriminate between the target and foil
stimulus. In order to explain the enhanced discriminabil
ity effect for studied and HF words, McKoon and Ratcliff
(in press; Ratcliff & McKoon, 2000) proposed a modifi
cation of the counter model. The new version of the
counter model assumes that HF words as well as studied
LF words, apart from having an advantage due to bias,
also have a higher value ofps. The parameter ps denotes
the probability of detecting information that enables one
to discriminate between the target and the foil (e.g., the
first letter of the LIED-DIED pair). We will term this mod
ified model the counter model II.
Recently, two-alternative forced-choice procedure ad
vocated by Ratcliff and McKoon (e.g., Huber, Shiffrin,
Lyle, & Ruys, in press) has been adopted in several prim
ing studies. One ofthe main advantages of this procedure
is (...truncated)