Background
A detailed introduction to the analysis of recognition memory can be found in Neath & Surprenant (2003).
Operation
In standard recognition tests, subjects are shown a test item and are asked to decide whether the item was seen previously. The two possible responses are YES (also known as OLD) and NO (also known as NEW). If you calculate the proportion of times that the subject said YES and the item was on the list, you get the hit rate. Just looking at the hit rate tells you nothing about the ability of the subject to discriminate between old and new items. You must also look at the false alarm rate, the proportion of times that the subject said YES but the test item was not on the list.
There are several ways to do this.
Signal Detection Theory
SDT uses the hit rate and the false alarm rate to estimate two measures. The measure of discriminability is called d' (pronounced dee prime) and the measure of bias is called C (an older measure was beta). The Java applet uses a method suggested by Brophy (1986) to calculate these values. The larger d', the better the subject's ability to truly discriminate between old and new items. Values of C above 0 indicate a conservative bias (less willing to guess OLD) whereas values of C below 0 indicate a liberal bias (more willing to guess OLD). d' cannot be calculated when there are hit or false alarm rates of 1 or 0.
Non-Parametric Analog: A'
To use signal detection theory, you need to assume various things about your data. A' (pronounced a prime) is a measure of discriminability like d' but it can be calculated when subjects have hit or false alarm rates of 1 or 0. In addition, a' does not require homogeneous variance. A' varies from 0 to 1 with 0.5 indicating chance performance. The corresponding measure of bias is called B"D (pronounced B double prime D). Values greater than 0 indicate conservative bias, values less than 0 indicate liberal bias. The Java applet uses a method suggested by Snodgrass, Levy-Berger, & Haydon (1985) to calculate A' and a method suggested by Donaldson (1992) to calculate B"D.
Two-High Threshold Model
This model's name comes from the idea that there exists a threshold for old items and a threshold for new items and only items that exceed these thresholds will be recognized. Pr is the discrimination measure (it is also sometimes called the corrected recognition score). Br is the bias measure; values greater than 0.5 indicate a liberal bias, values less than 0.5 indicate a conservative bias. The Java applet follows the suggestions of Feenan & Snodgrass (1990) to calculate both Pr and Br.
Instructions
Click once on the `Run the Program' button. You will be prompted to enter two values. First, you will be asked for the hit rate. Type in a value between 0 and 1. (If you type and nothing appears on the screen, click once in the middle of the window and try again.) Second, you will be asked for the false alarm rate. Again, enter a value between 0 and 1. You will see three measures of discriminability (d', A', and Pr) and three measures of bias (C, B"D, and Br).
The values calculated are generally very accurate except when the hit or false alarm rates are close to 1 or close to 0.
