Sunday, 16 August 2009

The psychology of grit

I've been browsing Jonah Lehrer's blog recently, and an article on grit caught my attention ( here ).

Grit is broadly defined as the resolve to continue in pursuing an aim, no matter the obstacles or distractions.

A quick afterthought.
How do you model "success" depending on grit?

A toy-model could comprise a space of "choices" at discrete time-points. There is some reward-function which is parametrized for "grit" and it determines the probability of changing a "choice" or sticking to it. Choices lead to "outcomes" at the end of the simulation.

To me it seems - the more grit you have, the less likely you'll be to optimize the reward function by exploring the space by jumping around a little.

How one defines the trajectory from choice to outcome should factor in that success does depend on spending some time on a project, so the reward function is increasing in time for fixed value of choice.

But still, if grit is very high, it seems that the initial (random?) choice will be the one trajectory followed till the end. Whereas for some "optimal" value, initially the space can be explored by jumping between trajectories.

Versions of this kind of reasoning permeate mundane things like deciding how much to work and how much to surf for fun during a typical work day, but also i guess long-term choices such as choice of career, home city, maybe in some version also the life partner, etc.


  1. could you link the blog please?


    (the post from 3 August)

  3. dah! i wrote tom lehrer above, when i meant jonah lehrer. tom lehrer is a mathematician turned comedian (check the song "lobachevsky" for example :))


    (it's too late for posting anything more intelligent)

  5. You can edit the post directly to correct the name.

  6. For posterity: the Jonah Lehrer article mentioned above, "Grit", is at

  7. I believe this model needs to account for a couple of more things.

    Jumping around different states assumes
    that you have stepped on each state and learned its value:
    how much time does one need to get enough good of an estimate
    of the value of the current state when talking about a way of life/research area/activity/person?

    It looks like grit is necessary on a very basic level.

    But let's say our agent with less grit has enough grit to overcome this obstacle.
    Still, there is another issue that could work against it.

    It looks like a path that one with more grit would follow is less of a turbulent, more conventional path,
    where you smoothly transition from state to state and it is implied that time to figure out the next
    state to move to is not very long.

    On the other hand, an agent that follows a more creative path of exploring various states
    faces a danger of spending too much time evaluating where to go next.
    There is no simple predefined evaluation function in real life, and it tends to change over time - seems more so for an agent with less grit.

    It is clearly possible even under these circumstances for an agent with less grit
    to reach the optimum faster than an agent with substantial grit. But only if they are able
    to complete small-scale tasks quickly enough.

  8. That's a very good point. I guess you'd introduce that into the model as a hopping penalty. Currently I can't think of a physical analogy but I'll consider it a bit more.

    Having said that I'd wager good money that the time taken to learn the most important details of a position is considerably less time than your average employee would tend to spend in that job.

  9. It seems to me that Dzejla's definition of trying more things despite etc etc, which he/she describes as less grit, is in fact the same as more grit. Perhaps I'm confused.

  10. Dzejla - Having considered this further I now think i understand the problem. I couldn't really figure out why i couldn't think of a model that had a hopping penalty that you suggest. But I think it's a mis-understanding because I didn't describe the problem very well.

    I think the key is that the algorithm doesn't necessarily find the optimum path - it can take backwards steps as well moving forwards.

    What this means is that an algorithm that has less grit will bounce around more - the penalty (for moving too often) that you suggest will effectively come in because as the gritless algorithm bounces around it will go back as easily as forwards.

    I think that the algorithm with more grit will make more transitions in the right direction. It will just be prone to getting caught in a local minimum.

    Do you think that this makes sense?