• Question: Hi! My name is Rory. I just wanted to ask you 1 little question; you say you work with numbers to make people faster and more precise, how does that work? I never knew human behaviour had to do with mathematics, that's so cool!

    Asked by Rory@skerriescc to Greta on 11 Nov 2020.
    • Photo: Greta Monacelli

      Greta Monacelli answered on 11 Nov 2020:

      Hi Rory! Thank you for your question!

      We can use Mathematics (and Statistics) to describe many different aspects of our world. To do so, we use statistical models. They are used in physics, biology, and many other areas, including psychology and the study of human behaviour.

      Let’s say we want to study a particular phenomenon. First, you observe it and collect data about it. Then, you look for a suitable mathematical structure (model) that describes them and, you make sure to adapt it to your data (estimation of the parameters). In this way, you now have a mathematical description of the phenomenon.

      In our case, the phenomenon is human behaviour. To collect data on it we observe how people play a game (it is called the multi-armed bandit task).

      However, we may have to observe people play the game for a long time to choose a “good” mathematical description (models and parameters). Moreover, people may get bored of playing the game.

      My work aims to decrease the time people have to play the game for us to find the best mathematical description.

Comments

  • Photo: Rory@skerriescc

    Rory@skerriescc commented on :

    Hi! First of, thank you so much for your response. What kind of people do you study and what behaviors are you focusing on? And how long does it normally take you to build a model? It sounds like it needs a lot of data. Also, how can you gather people to experiment on? It sounds like it takes A LOT of time out of their lives. And what can this description tell you about the persons behavior? Would a more stubborn, narrow minded person behave differently to a thinker/adapter and be more prone to certain behavioral conditions? And would shortening the time and gathering less data inevitably give you a worse model? Sorry for the random questions in no order and thanks in advance

  • Photo: Greta

    Greta commented on :

    Hi Rory! Thank you again for your questions and sorry for the late answer.

    I study all kind of people and my work can be applied in a general context.

    The game takes around 15/20 minutes. It is not much, but it is a problem in certain situations. For example, if you want the same person to play more than once or if you ask a child to focus for that long on a tedious task. However, we can’t shorten too much the experiment. As a general rule in statistics, the more data you have, the better. My job is to obtain good results with the least data possible. Thus, I aim to obtain good estimations of the parameters with the least playing time.

    I focus on human decision-making. For example, I study how we learn, how we judge rewards and losses and how much we value past experiences. I use the parameters from the model to describe numerically these individual traits. All together they describe the behaviour of a person during the game.

    Let’s for example take a parameter called inverse temperature. It describes how much we are consistent with our choices during the game, how much we stick with our decision or explore different options. Another one measure how fast we learn, and it is called the learning rate.

    As for the adapter/ stubborn comparison, it is a difficult question. The parameters have a specific interpretation as the ones above. In my opinion, the examples I gave are linked to stubbornness and adaptability. Nevertheless, I am not a psychologist. 🙂

    Usually, you use models created by others. In this case, you need to research and understand their work and then apply it to your problem. But creating a new model is a long journey. You need to understand very well the phenomenon you are studying. Moreover, you need to prove it is useful with experiments and compare it with models already used in your field.

    I hope I answered all your questions. 🙂