Start probe trials and learn how to retrieve data from the experiment.
Over the course of the last week, your starling was trained in a basic song discrimination task. Once your bird reached asymptotic performance, the reward rate was reduced to 80% in preparation for probe trials.
In a probe experiment, we’ll continue giving the bird trials in which it hears the S+ and S- training stimuli, but we will add in probe trials. Probe trials differ from S+ and S- trials in that they are nondifferentially reinforced, which means that the probability of reinforcement and punishment are fixed and don’t depend on the stimulus. The rates of reinforcement and punishment are matched to the base rate of reinforcement, which is now 80%. Probe stimuli will therefore be reinforced 40% of the time, punished 40% of the time, and result in no consequences 20% of the time. We also match the frequency of probe trials to the base rate of reinforcement, so that 80% of trials are S+ or S-, and 20% of trials are probe trials.
In order to get good data, we need to present the stimuli in isolation, so that the birds aren’t influenced by stimuli being presented to other birds or by other events in the environment. This means that you’ll monitor your bird remotely.
Navigate your browser to https://aplonis.psyc.virginia.edu/decide/. You should see a table of subjects organized by id. Find your subject by locating the experiment with your group number in it. Click on the name of the controller for your subject, and you should see the interface you’ve been using over the past two weeks.
Try running a probe trial or two:
Once you’ve run a couple of trials, ask the instructor to set up the automated testing software. You can watch your bird run trials via the interface.
You’ll access the data for your probe experiment in the same way you accessed
data for today’s computer exercise. Use the same URL, but change
group1-probe or whatever matches your group:
For your report, you will need to calculate the proportion of trials in which the bird responded for each of your probe stimuli and the training stimuli. You should be able to do this easily by adapting the code you used for generating the bar plot and running χ-squared tests at the end of the in-class exercise.