Notes on Paul Meehl’s “Philosophical Psychology Session” #10

These are the notes I made whilst watching the video recording of Paul Meehl’s philosophy of science lectures. This is the tenth episode (a list of all the videos can he found here). Please note that these posts are not designed to replace or be used instead of the actual videos (I highly recommend you watch them). They are to be read alongside to help you understand what was said. I also do not include everything that he said (just the main/most complex points).

Clinical vs statistical prediction

Must distinguish between kind of data and mode of combining data.

Important to use prior/wider evidence/theory when diagnosing.

Phenotypic trait: collection of lower order dispositions which have certain common properties (needs semantic resemblance) and empirically covariant. If it lacks common properties it’s a candidate for genotypic theoretical inference. If it lacks covariance then it’s a logical trait (in the dictionary but not the real world). Covariance can refer to time correlation (p) or individual correlation (r).

In prediction, the broad meaning is mechanical and the narrow meaning is actuarial. All actuarial prediction is mechanical but not vice versa.

If a medical professional/life insurance salesperson refuses to give you information about the long run probability of complications/cashing out because they state “you’re only doing this once and you are a unique individual, you shouldn’t think about the group”, you would be suspicious.

Problem of applying a statistical frequency number to an individual case which belongs to a wider class. Some want to narrow down the class by adding my information, necessarily making a subset.

Reichenbach: take the smallest reference class for which you have stable frequencies.

Many believe you can better predict outcomes for a person by better understanding the individual.

Psychometric/actuarial/statistical predictions are more accurate than clinical interviews then prediction. Human mind isn’t an efficient user and assigner of beta weights.

Examples in criminal justice sentencing where large reports by professionals reduce the predictive accuracy of judges.

Factors predictive of recidivism: how many crimes the person has committed; severity of crime (some studies); age when crime was committed; school educational; horizontal mobility; substance abuse history; social group; length of time in employment in the private sector; IQ (some studies).

With greater information, clinician’s predictive power went down (overload of irrelevant information so less beta weights are assigned to the relevant predictors).

Statistical prediction is atheoretical. Equation deals with things at the first level, without reference to any theory. Theory might dictate what variables you try out but there is no theoretical underpinning.

To make predictions by means of a theory, you need to have a theory with high verisimilitude (takes account of most, if not all, of the powerful variables/best predictors), accurate instrumentation to assess those variables. Most subfields of psychology don’t meet these standards.

Any field without suitably detailed theory or the ability to procure accurate measurements in time at the state of the system will struggle with theoretical prediction.

When the actuarial system omits a factor which, when present, is a potent predictor troubles those who try to use it. This predictor is so potent it countervails all the other factors in the equation.

Broken leg factor: a factor that wasn’t in the initial actuarial calculation due to inadvertence or rarity. Some argue because it’s so rare it shouldn’t be considered in the calculation, but this is a mistake as it could have a large effect on the outcome.

If the clinician can accurately identify the broken leg case, they will necessarily outpredict the statistician with their actuarial method who ignores the rare broken leg case. This is because they will have the same accuracy when they agree about the factors but if the clinician accurately factors in the broken leg factor (which has a significant impact on predictive accuracy) over the long run, they will better predict. But that’s not the case, so they must not accurately identify these cases.

Some argue clinical and statistical are complementary but they aren’t as they give opposite answers (yes or no, they will reoffend or they won’t).

References

Yonce, J. L., 2016. Philosophical Psychology Seminar (1989) Videos & Audio, [online] (Last updated 05/25/2016) Available at: http://meehl.umn.edu/video [Accessed on: 04/02/2018]

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: