X hits on this document

78 views

0 shares

0 downloads

0 comments

17 / 28

an MCPL task, but only when improper linear models fail to yield (nearly) correct predictions. And this in spite of the sub optimality of Outcome Feedback.

Following the learning task we gave the subjects in both conditions another block of 30 trials, now without OF, but after providing them with precise information about the task structure. In condition I subjects did slightly better than in block 3 of the learning phase (mean achievement = 0.82) and this was mainly the result of an increase in configural cue use: mean C = 0.36. In condition II mean achievement increased to 0.53, but this was mainly the result of an increase in linear task knowledge: mean G = 0.67, whereas mean C = 0.57.

As I am writing this abstract we are still a bit puzzled about these last results. It might be conceivable that a certain amount of cognitive laziness restrained the subjects in condition I from switching from a rather successful improper strategy to a configural strategy, but we expected the subjects in condition II to do much better. Or does this demonstrate once again that people find it very

difficult to work with start working on a experiment shortly.

negative linear relations? serious manuscript about

I'll this

Different kinds of confidence

Josh Klayman University of Chicago E-mail: joshk@uchicago.edu

I'm continuing my interest in subjective confidence, looking at different kinds thereof. Jack Soll (INSEAD) and I are studying confidence expressed in terms of subjective intervals ("I'm 80% sure that the planet Neptune was discovered

between 1760 and 1940"...). findings of interest.

We have two main

1.

Even using random sampling from whole

domains (to approximate representative sampling), subjective intervals show substantial overconfidence. That is, one's 80% intervals contain much less than 80% of the correct answers. Some of this is due to biasing effects of unbiased error in setting the endpoints of the interval. However, we demonstrate that subjective intervals are also

systematically too narrow.

2. How much too narrow they are depends on exactly how you ask for them. If, as above, you ask for a single interval, the intervals are barely half the size that a well-calibrated person would need. If you ask instead for two judgments--"I'm 90% sure it's greater than ... and I'm 90% sure it's less than ...", the implied 80% interval is larger--about 2/3 of the well-calibrated size. If you also ask for a median

estimate along with the two ends ("It's about equally likely to be more or less than ..."), then the interval is yet larger, almost the well-calibrated

size.

Why? Well... we're working on that.

We'd be very interested to hear the hypotheses of our fellow Brunswikians!

The second kind of confidence I'm working on has to do with judgments about where one's

performance stands relative to others'. with Katherine Burson (here at Chicago)

In work and Rick

Larrick debate.

(Duke U.), we first weigh It seems that the poorest

in on a recent performers on a

task are standing cognitive cognitive

also those who most overestimate their relative to others. Is this an interesting phenomenon (a correlation between ability and the metacognitive ability to

know how

skilled

estimates,

leading

you are) or the result of noisy to regression toward the mean,

plus

an

overall

upward

bias?

We

find

for

the

latter,

by

testing

tasks

in

which

there

is

no

overall

upward

bias.

Plus,

we

replicate

earlier

findings

that

people

are positively stand relative

biased in to others

judgments of where they only when they find the

task to be easy. When the actually negatively biased.

task So,

is difficult, they are now, we're looking

to create a egocentric

new debate: Does this represent anchoring on one's own ease

an of

performance,

with

insufficient

adjustment

for

what

one

knows

about

others'

ability

to

do

the

task?

Or

is it really a manifestation probabilistic cues to their impression of how hard it

of people's use of two standing, namely their was for them and their

impression of how people? We're still

hard it working

would be for other on that, but you can

probably guess that would tend toward the

a member of this latter explanation...

society Not that

we would let prior beliefs color our the data (not more than would appropriate, anyway!)

interpretation of be normatively

More on how he developed his ideas

Elke Kurz-Milcke, College of Computing, Georgia Institute of Technology E-mail: kurzmi@cc.gatech.edu

Volume 5 of a series titled Portraits of Pioneers in Psychology edited by G. A. Kimble and M. Wertheimer (published by Erlbaum) is soon to appear and will contain a chapter on Brunswik. The chapter is titled Egon Brunswik: Student of achievement and was written in a transatlantic collaboration between Nancy Innis from the University of Western Ontario and myself, then working at the University of Tübingen in Germany. The chapter gives an overview to Brunswik’s life and work, emphasizing the period of time just after he arrived in the United States, a period of

Newsletter 2002 page 17 of 28

Document info
Document views78
Page views78
Page last viewedSat Dec 10 19:11:19 UTC 2016
Pages28
Paragraphs1455
Words18198

Comments