How Do We Deal With Failed Predictions?

Just occasionally, my kids won’t eat what I put in front of them. Sometimes, I can guess when that will happen (e.g. spaghetti with creamed broccoli, green curry) but sometimes it hits me out of the blue. The times I don’t see it coming are the best because they remind me that, like any psychologist, I am an imperfect predictor. But how can I live with my imperfect prediction?

Last week, I argued that partial prediction is the reality of psychology. This week, I want to look at how different psychologists, including some famous ones, would advise me on my failure to always predict when my kids won’t want their dinner.

Sigmund Freud would say “My hypothesis is that this…is not left to arbitrary psychical choice but follows paths which can be predicted."₁ I imagine he’d want to ask my children several questions, let them talk and afterwards would be able to give me a decent explanation. I’d happy for a while until I realized that Freud had worked, like he always did, after the disaster. There was no prediction ahead of time, just explanation. This is not an atypical move: One method to ensure complete prediction is to pretend post-hoc explanations are predictions and hope that no one notices.

B. F. Skinner would bemoan my lack of household control, the paucity of experimental conditions. He would say “[if] a researcher is able to establish sufficient control over environmental influences, then orderly behavior will occur and can be easily observed (i.e. without statistical analysis).”₂ There’s a chance he could be correct—that I don’t control things enough for that level of order to emerge. But even if I wanted to control things more, what guarantee is there that behavior will be totally orderly? Skinner never found perfect prediction even with his highly controlled rat cages. And, while I can understand he wants to believe that perfect prediction is possible from environmental conditions alone, I’d like proof that the car works before I buy it, please.

Kurt Lewin would say that “the actual behavior of the child depends in every case both upon their individual characteristics and upon the momentary structure of the environment.”₃ He would tell me to work out my children’s perceptions of the possibilities in the situation and work from there. I’d be OK with this for a while because Lewin was excellent at predicting things (like the end date of WWII₄), and because he considers both the person (unlike Skinner) and also the environment (unlike Freud), but then I’d remember that he and his students used statistical levels of prediction in all their papers. The cases of error are fundamentally left unaddressed.

Someone like David Funder might encourage me to not be too dismayed. He would point to the reasonably high level of accuracy in guessing the outcome, looking at such things as binomial effect size displays₅, R²% or “area under the curve.” In essence, he’d say to quantify where I was on the scale, from guessing to perfect. I can live with that, but it still sidesteps the issue of what to do with my errors.

Lastly, I think the most current personality psychologists (such as Will Fleeson) would ask me to include some within-person variability in my model. Fleeson would tell me that my kids’ eating pattern will show random fluctuations each day but also some long-term consistencies that are unique to them₆. In other words, to allow my kids space to be different at different times but accept that they have some overall preferences. Instead of thinking I could ever achieve perfect prediction, live with the reality of partial prediction.

We will never know why psychological prediction is partial. The only way to demonstrate partial prediction’s cause is to find something that perfectly predicts the partial prediction. And, if we did that, we’d actually move to the point of perfect prediction. Hence, there must always, logically, be a level of uncertainty about why we can’t predict things perfectly.

Partial prediction is not a big problem. Science loves to predict, humans love to relate. Partial prediction brings opportunities to interact and to care. Semi-randomness allows for relationship. Specifically, when I believe that kids aren’t entirely predictable, the dreary eggplant dhal can be rescued by fried-egg-on-rice, and the smiles on their faces when I change plans is more enjoyable than predicting the correct meal in the first place.