[Online] Effects of Confidence Scale Length on Recognition Accuracy


Previous studies have found that regardless of the condition, the more confident an individual is, and the faster their responses are, and the more accurate their identification is. Previous research has also found that increasing the length of the confidence rating scale led to poorer recognition performance. This research aims to find out if the same effects of recognition can be found when participants have enough practice on using the response keys while being motivated to use all keys strategically in order to maximise their gains. Although there exists many models that aim to explain the relationship between confidence, response time and accuracy in decision making (e.g., Ratcliff & Starns, 2009; Voskuilen & Ratcliff, 2015, Wixted & Mickes, 2014), those models were inadequate to provide a unified account that relates confidence and RT to the accuracy of choices in recognition. Therefore we proposed a new model, the Multiple Threshold Race model (MTR) that has the potential to provide such unified account.

Research Questions / Hypotheses

The aim of the present research is to conduct a visual recognition experiment that aim to test and validate the use of MTR in predicting relationship between confidence, RT and accuracy. The peripheral aim of the present research is to test what effect of varying confidence scale length would have on recognition performance. It is hypothesised that we will observe longer RT for larger confidence scale length.


A total of 35 REP participants completed the study.


Participants first went through a response key practice block where they were presented with a series of confidence options, one at a time. Their task was to press the corresponding key as quickly and accurately as possible. Then they completed 3 practice and 15 experimental cycles of computer-based visual recognition tasks. During each trial, they were presented with a large square which consists of 400 small grids of two different colours (orange and blue) that flashed on the screen. The task was to decide whether the colour orange or blue appeared more frequently. They were also asked to report how confident they were in that decision on either a 2-point, or a 4-point, or a 6-point confidence scale.


When the length of the confidence scale is increased, there is a slowdown in RT, which is consistent with the findings of Benjamin et al. (2013). We plan to fit the MTR model with the data using hierarchical bayesian techniques, and see what parameter variations are needed to improve model fit. By fitting MTR with different combination of parameter variations, the validity of MTR could possibly be established.


The findings of the experiment suggest that longer confidence rating scales could induce decision noise that contributes to recognition decisions. The MTR model has not been fitted to this dataset. Yet based on our modelling results from previous recognition memory dataset, it is possible for us to conclude that the validity of MTR could be established. This gives support to the single-staged confidence theory using Balance of Evidence hypothesis, as it could simultaneously account for the relationships between confidence, accuracy and RT. The results from this study will be written up as a journal article for publication, and will be presented at various conferences on psychological research.