stimuli of pictures with a random prob sequence

[cynthiayu316]

Hi, I am working on designing an online experiment. There are three stimuli and the probability of winning the reward from the stimuli is fixed. But for each trial, the probability of which ice hold corresponds to which fixed probability will be randomly assigned. So is there a way when designing the experiment, I could randomly generate sequences for different prob and assign them to different stimuli? For example: for the first trial, 30% for left stimuli, 50% for the middle one, 80% for the right one. Then the next trial, 30% for middle, etc) I check the documentation but did not find a parameter that could solve the question mentioned above. I was wondering id there any way I could achieve that? Thank you!

[Lucy - FF Researcher Support]

Hello! Could you explain a little more about what you are trying to achieve? In your design, when a participant clicks on the image that has 80% probability assigned to it, there's an 80% chance the next trial will tell them they got a reward and a 20% chance the next trial will tell them they didn't get a reward?

[cynthiayu316]

Hello! Could you explain a little more about what you are trying to achieve? In your design, when a participant clicks on the image that has 80% probability assigned to it, there's an 80% chance the next trial will tell them they got a reward and a 20% chance the next trial will tell them they didn't get a reward?

Hi,  thank you for your reply.  It means that we have two rounds of games for each participant while each round contains 100 trials. Then inside the 100 trials, we will have a fixed probability for three stimuli(three pictures) such as for the left it is 80% for the 100 trials, the middle is 50%, and the right is 20%, etc. And the next round(next 100 trials), the left will be maybe 50%, the middle would be 20%, the right would be 80%. The fixed probability will be designed at the beginning of the experiment. what I need is more like the random sequence assign to the stimuli. When choosing an 80 % percent reward it means that they will have a bigger chance to get a reward than other stimuli, but still 20% a chance they will get nothing. Then they will receive the response of stimuli(images such as tick and cross) to indicate whether they get the reward. Thank you!

---

Hello! Could you explain a little more about what you are trying to achieve? In your design, when a participant clicks on the image that has 80% probability assigned to it, there's an 80% chance the next trial will tell them they got a reward and a 20% chance the next trial will tell them they didn't get a reward?

Hi,  thank you for your reply.  It means that we have two rounds of games for each participant while each round contains 100 trials. Then inside the 100 trials, we will have a fixed probability for three stimuli(three pictures) such as for the left it is 80% for the 100 trials, the middle is 50%, and the right is 20%, etc. And the next round(next 100 trials), the left will be maybe 50%, the middle would be 20%, the right would be 80%. The fixed probability will be designed at the beginning of the experiment. what I need is more like the random sequence assign to the stimuli. When choosing an 80 % percent reward it means that they will have a bigger chance to get a reward than other stimuli, but still 20% a chance they will get nothing. Then they will receive the response of stimuli(images such as tick and cross) to indicate whether they get the reward. Thank you!

[Ting]

Hi Cynthia,

Are you trying to implement some type of a multi-armed bandit experiment? I think I mostly understand what you are after, but it'd be helpful to get an explicit confirmation to my interpretation below:

Let's assume that in your block 1, you are using a choice response with three options. The left option has a 80% reward rate, the middle 50%, and the right 20%. Then, what you are looking for is a way for FF to tell participants "you are correct!" 80% of the time if they choose the left option, 50% for the middle, and 20% for the right.

Is this interpretation correct? If not, can you explain what you need in a similarly concrete example? Thanks!

[cynthiayu316]

Hi Cynthia,

Are you trying to implement some type of a multi-armed bandit experiment? I think I mostly understand what you are after, but it'd be helpful to get an explicit confirmation to my interpretation below:

Let's assume that in your block 1, you are using a choice response with three options. The left option has a 80% reward rate, the middle 50%, and the right 20%.  Then, what you are looking for is a way for FF to tell participants "you are correct!" 80% of the time if they choose the left option, 50% for the middle, and 20% for the right.

Is this interpretation correct? If not, can you explain what you need in a similarly concrete example? Thanks!

Hi, Ting. Indeed it is a multi-armed bandit experiment! Thank you for your reply. Yes, the interpretation is correct! Is there any way I could achieve that? And for the next round game, the three fixed probability will be assigned to different options randomly.

[Ting]

Awesome! I'm glad that we are on the same page!

The first aspect of your design, where you want the correctness feedback to be probabilistic, is not currently possible. However, we are very interested in developing this feature for you. Is there any specific requirements other than the probabilistic nature? For example, if FF tells the participant "you are correct" 80% of the time for a .80 reward option, what should FF do in the rest 20%? Should FF say "you are incorrect", as if the participant chose the incorrect option, or should FF just remain silent?

The second aspect of your design, where you want the reward options to be randomly placed, should be readily achievable by adding "locations": "random" to the choice response. Is that not the case? Check out the details in the documentation: https://help.findingfive.com/api/responses.html#choice

[cynthiayu316]

Awesome! I'm glad that we are on the same page!

The first aspect of your design, where you want the correctness feedback to be probabilistic, is not currently possible. However, we are very interested in developing this feature for you. Is there any specific requirements other than the probabilistic nature? For example, if FF tells the participant "you are correct" 80% of the time for a .80 reward option, what should FF do in the rest 20%? Should FF say "you are incorrect", as if the participant chose the incorrect option, or should FF just remain silent?

The second aspect of your design, where you want the reward options to be randomly placed, should be readily achievable by adding "locations": "random" to the choice response. Is that not the case? Check out the details in the documentation: https://help.findingfive.com/api/responses.html#choice

Hi, Ting. Sry for the late reply. So basically, 80% of the time, the participant would get the reward (image stimuli) or 20% chance they would get another stimulus: cross image(indicating that he/she did not get the reward). There is a total of 100 trail in each round, and there is a total of 2 rounds. Also, two participants are playing and taking turns. 
Thank you for the information about the reward option place. 
In addition, I have already had python version of the game code. Is there any tool that could convert the python code into the language we are using on finding five? That would make implementation much easier. Thank you so much for your help!

[Ting]

Hi Cynthia,

We don't have tools for converting Python code to FF grammar. FF grammar is designed to be as accessible as possible so it operates more like LaTeX rather than a programming language. The good news is that it probably takes you very little effort to learn the FF grammar given your experience in Python.

I think the best way to go forward is for you to try implementing your study on FF so that it resembles your Python version as closely as possible, given the current features of FF. Once you are at that stage, ping us back again and we'll develop the probabilistic target matching feature for you. If there are other nitty gritty features you'd like to achieve then, we can then deal with them afterwards.

What do you think?