Specs for FadeTime1

This is quite a simple experiment. We are asking what conditions help people learn to make a fine perceptual discrimination faster. The discrimination concerns duration of a tone.

Task: You hear 600 hz tones that play for a longer or shorter interval. Decide whether the tone was short or long. The boundary is 500 msec.

Basic Details. The experiment consists of 20 blocks of 40 trials/block. The first 16 of these are training blocks; the final 4 are test blocks. On each trial in either type of block, you randomly select whether the tone is LONG or SHORT. On each subject, the subject simply hears the tone and decides whether he thinks it is LONG or SHORT (this is an accuracy experiment not a speed experiment). The response should be to press the M key for LONG and the N key for SHORT (mouse takes too long). They can use the mouse to start each block, however. Put up a visual fixation point for 500 msec, followed by 500 msec blank period, followed by the tone. After the subject responds, put up “Correct” or “Error” on the screen for 1000 msec, then wait another 1500 msec until you put up the plus sign.

Training Conditions: This is a between-subject experiment with one independent variable and two conditions (i.e., half the subjects get one set of conditions, and half get the other). The independent variable is training procedure. At the beginning ask the experimenter to enter a Subject Number, Initials, and Condition (1 or 2). If they enter 1, the subject is trained with Fading Procedure. If they enter 2, the subject is trained on the Final-Discrimination Procedure. This applies for the first 16 blocks.

1. Fading Procedure. For subjects in Condition1, you start with tone durations that are widely separated (making the task very easy) and adjust them on every trial using a procedure that will insure that subjects make relatively few errors. Start with 600 and 400 msec for long and short, respectively. Every time the subject is wrong, increase the long and decrease the short by 10 msec. Every time the subject is right three times in a row, decrease the short and increase the long by 10 msec. You can look at an algorithm I wrote for doing this incorporated in some code that Ryan Brunton made called Keller1. Do not let the gap become shorter than 20 (long=510 and short=490).

What this is doing is making the task easier when they make a mistake, and making it harder when they get it right. But it tends to keep it pretty easy because it only makes it harder when they get it right three times in a row.

2. Difficult-Training Procedure. Here the subject is trained with long=520 and short=490 on every trial. This is a difficult discrimination but it is the same as what subjects will be tested on at the end.

Final Test Blocks. In the final four test blocks, all subjects (regardless of whether they are in condition 1 or 2) are tested on the same difficult discrimination. They should receive NO FEEDBACK during the test – this is just for assessment, not training. Subjects get tested with long=520 and short=490. (I may have to adjust this, but it is my best guess about what might produce the desired difficulty level).

Data Storage. Save three files per subject: raw data file, training summary data file, and test summary data file.

• In the raw data file, there is one row per trial for all 20 blocks, with the block number, subject cond (1 or 2), trial number, exact duration in msec, long vs. short, and subject’s response (long vs. short). [I know: subject cond is just the same number on every row, but this will be convenient for data analyses]

• In the training summary data file there should be one row per block containing 4 columns: the block number (1-16 only), subject cond, the average gap between long and short tones on that block, and the average percent correct the subject achieved.

• In the test summary data file there should be one row per block for blocks 17-20, containing 4 columns: the block number, subject cond, and the average percent correct the subject achieved.