Adding feedback

Adding feedback

Let’s try and create a ‘Feedback’ Routine for the Posner task we want to:

  • Add trial by trial feedback on response times

  • Adjust the colour of the feedback based on RT

  • Give feedback at the end on average RT overall, on valid trials and on invalid trials.

What can we give feedback on?

We can explore the variables available to us by examining the data output of our experiment. If we used a mouse component for example, we can see that there is usually:

  • mouse.time - the time(s) of the mouse click(s)

  • mouse.clicked_name - the name(s) of the object(s) clicked by the mouse.

We can also store other things from our mouse component by adding parameters to the Store params for clicked field.

We can use any of these variables to then provide trial-by trial feedback. Imagine our trial has a Mouse Component called resp.

We would need to add a routine called “feedback”, in this feedback routine add a text component and in the text field we could write:

$'RT was ' + 'str(resp.time)' + ' ms'

Here we are concatinating strings using the + operator and we are also converting our resp.time variable to a string using the str() method (we can’t concatinate strings and numbers!).

The problem here is that 1) the resp.time value is actually a list, so we may want to index either the first or last element 2) we probably want to round the value to be a bit prettier:

$'RT was ' + 'str(round(resp.time[0], 3))' + ' ms'

An alternative way of doing this in python would be to use a “formatted string” (this is better practice for python, but it might not translate so smoothly online):

$f'RT was {resp.time[0] : .3f} ms'

feedback dependant stimuli

We can make feedback response dependant by using simple if statements. To adjust feedback colour based on response time we need a code component:

if resp.time[0] < 0.5:
    feedbackCol = 'green'
    feedbackCol = 'red'

Providing overall feedback

To give feedback at the end for each condition let’s learn about lists. We want three lists to keep track of RTs:


Some useful Python methods

  • .append() - adds to a list

  • np.average() - returns average of a list using the numpy (np) library.

We can use these to give feedback at the end of our experiment to summarise performance.

On each trial we add to the list:

if label == 'valid':
elif label == 'invalid':

At the end of the experiment we can average these lists:

validAv = np.average(validRTList)


  1. Add a feedback tone that varies in frequency depending on if the RT is fast (e.g. <.05) or slow.

  2. Add a text component to the end routine to tell participants if they showed a Posner cueing effect.

  3. IF participants show a posner cueing effect, tell them how large their effect was in ms.

What next?

Code components allow us to extend mouse responses in some fun ways. So let’s talk about Making the most of mouse inputs.

Back to top