Experimentation

Go/no-go experiment

Procedure

PsychDetect allows you to conduct psychoacoustic detection experiments. It employs the Go/no-go behavioural paradigm, which works as follows:

1. The subject signals that he is ready to start and initializes a trial.

2. After a random waiting interval, a stimulus that has to be detected will be emitted.

3. If the subject perceives the stimulus, he reacts on it and the trial will be recorded as a hit. If he does not, the trial will be recorded as a miss and a new trial will again follow after a random waiting interval.

4. A hit will be rewarded. Typically, there is no feedback on misses. A response during the waiting interval, that means that there was no stimulus, is an error.

   Basically, you have two options how to handle errors (for an extended explanation, see below):

   • Regard all errors as false alarms.
   • Regard only those errors as false alarms that happened during a predefined subinterval of certain waiting periods (reactions to catch trials).

   You have also different options with regard to feedback after errors:

   • No feedback for errors.
   • Punish all errors (a typical punishment is a time-out period in which no trials can be initialized).
   • Punish errors that are considered as false alarms.

5. After each subject's response, he needs again to signal his readiness by initializing the next trial.

An experimental session consists of multiple trials. If you need a lot of data from a single subject, you may need to conduct several sessions on different days.

Here is a UML activity diagram that visualizes a session.

UML activity diagram of the embedded activity present stimuli.

Implementation of the experimental procedure by PsychDetect

An instance of ExperimentController is used to control the operation of an experiment. You run an experimental session by calling the start() method. The session is then in idle mode. If the subject is ready, he can initialize the next trial by operating the I/O device. The experiment controller will then move on to the trial state by starting a stopwatch that counts down a random waiting interval, before the next Stimulus will be send to the AudioPlayer. Another stopwatch measures the time since the AudioPlayer has begun to play the Stimulus. The subject can respond to the stimulus by operating the I/O device. If he responds within the predefined maximum reaction time, he will earn a reward and the session goes to idle mode. If he does not react, the ExperimentController stays in trail state and will directly initialize the next trial, which again starts with a random waiting interval.

If the subject responds during the random waiting interval, the current trial is interrupted and the response is an error. Error may or may not be punished. In any case, after a response has been made, the experiment goes to idle mode and the subject needs again to operate the I/O device to proceed with the experiment. The next trial that follows will always be the interrupted trial. This prevents the subject from employing a strategy of responding with constant short reaction times, whereby he will earn rewards for all trials with a short waiting interval and skip all trials with longer ones.

Analysing experimental data

Go/no-go are commonly analysed by means of the Signal Detection Theory.

To make use of it, we need to measure four categories: hits, misses, correct rejections and false alarms. While the former two categories are easily identified in a Go/no-go experiment, the latter have to be defined. One possibility is to treat a Go/no-go experiment as if it were a two alternatives forced choice (2AFC) task. As the name suggests, during a 2AFC experiment, a subjects is prompted to respond with one of two choices. In the case of a signal detection task, these will be "signal is present" and "signal is not present".

The trick is then to have catch trials with no stimuli present next to the real trials, and to only look for responses within a predefined time window (maximum reaction time) directly after stimulus or non-stimulus presentation. The mapping between experimental events and detection theory categories is then as follows:

Trial category	Response	Detection theory category
trial	yes	hit
trial	no	miss
catch trial	yes	false alarm
catch trial	no	correct rejection