Priming and Arielly data

[Bill_Powers1]

[From Bill Powers (2001.20.1115 MDT)]

Now that we seem to be reaching a consensus about the control model for the Arielly data, I think we might start re-evaluating the subject of priming. The explanation for the apparent effect of the social security numbers on the bid prices was simply that raising the SS# considered as an asking price resulted in an increase of the bid, as if increasing the SS# made the subjects want to pay more for the item. We may well be seeing an instance of population data being mistaken for individual characteristics, and leading to getting the relationship exactly backward.

This depends on the model, of course. The priming model is a simple stimulus-response model as I understand it. The magnitude of the independent variable is the size of the social security number, and the magnitude of the dependent variable is the bid offered for the item. The control-system model, on the other hand, proposes that raising the SS# acts as a disturbance of the price the subject feels called upon to pay, and the actual response of the subject (which we are calling O, the output) is to act in the opposite direction, attempting to bring the price to a lower reference level. So raising the SS# higher than the subject's reference level makes the subject want to pay less, not more. The bid price increases anyway because the action is based on the error, but without the action the payment offered (if the deal was not abandoned) would be much higher when the SS# increases. This is exactly analogous to the demonstration in my paper for the American Behavioral Scientist on this same subject.

This does not say we should prefer the control-system model over the S-R model; it merely shows that the two models are incompatible, since the actions they predict are exactly opposite. So now the problem becomes that of finding an experiment that would satisfy the premises of both models and also would involve an output that we could observe. That would allow us to reject the model that was incorrect, or both models if they were both incorrect.

We can do this with a computer. What we need is for the subject's alterations of the bid price to be made observable, which can be done just by having the subject move a mouse (which measures O) to adjust a bid price displayed on the screen. The proposed effect of the asked price on the bid price is harder, conceptually, to explain, but by actually giving the asked price an effect on the bid price we can satisfy the premises of the S-R model without having to explain the effect.

We almost have that worked out right now. The environmental part of the control-system model already says that B = A + O, so if A increases, B will increase. We can display numbers representing these three variables on the screen, with the subject's mouse altering O and thus causing a change in B adding to the change caused by A.

If the subject's output doesn't change when A changes, B will simply change proportionally to A. We could insert constant multipliers to vary the inherent amount of the effects of A and O, but let's keep it simple for now.

Now we can do the Arielly experiment on the computer screen instead of on paper. Each subject is asked his SS# and the experimenter enters it as A on the screen. The subject then moves the mouse to change the displayed bid price until it is whatever value the subject wants to see.

If a change in the SS# directly influences the price the subject wants to bid (still ignoring the use of population measures in place of testing specimens), we should observe that O and B both change in the same direction as the SS#. If the bid price is a controlled variable, we should see B change in the same direction as the SS# (by a smaller amount), while the output changes in the opposite direction. I predict that the latter is what we will observe.

We could also say that the effect of A on B occurs inside the subject, and remove the effect seen on the screen by presenting A but not adding its value to B on the screen. If there is an internal effect of the perception of A on the perception of B, inside the subject, we should see a change of O in the same direction as the change in SS# if the S-R model is right. If the control model is right, the subject will move the mouse to bring B to the subject's reference level for it, and the changes in SS# will have no effect.

It would be better, of course, to use a single subject and a series of different values of A. While this might alter the magnitudes of some effects, it is unlikely to reverse the relationship between A and O. Of course if someone wants to invent a model that would have that result without being a negative feedback control system, we could test it and see if it works. Telling the right background story as part of the instructions would, in my opinion, probably prevent any serious effects of repeated trials by the same subject, but again nobody has to take my word for that. We can just set up the experiment to satisfy the conditions of the model and test it. If it makes a difference, it makes a difference. If it doesn't, it doesn't.

Noticed that in both cases we assume a real effect of A on B as the subject perceives these variable. However, the "priming" theory predicts an action by the subject opposite to the direction of action predicted by the control-system model. If the control system model is favored, the term "priming" and all of its associated implications becomes inappropriate.

"Priming" is simply a term applied to an observed effect, an apparently direct effect. The word evokes images of certain possible mechanisms (I imagined a leakage between perceptual pathways). But those images are nonspecific and far from quantitative, and just saying the word priming doesn't prove that the images have anything to do with what is observed. Only testing models can allow us to distinguish one explanation from another or choose between them. I don't know if this is likely to be done. I have no facilities with which to do them. If anyone takes this challenge to the concept of priming seriously maybe the experiments will happen. Of course the best way to defend against the challenge is to ignore it, so I suspect that is what will happen.

Best,

Bill P.

[MartinT]

[Martin Taylor 2009.11.21.10.36]

[From Bill Powers (2001.20.1115 MDT)]

Now that we seem to be reaching a consensus about the control model for the Arielly data, I think we might start re-evaluating the subject of priming.

One possible control model, yes?

The explanation for the apparent effect of the social security numbers on the bid prices was simply that raising the SS# considered as an asking price resulted in an increase of the bid, as if increasing the SS# made the subjects want to pay more for the item.

That is a new suggestion that hasn't been mentioned hitherto, so far as I know.

....
"Priming" is simply a term applied to an observed effect, an apparently direct effect. The word evokes images of certain possible mechanisms (I imagined a leakage between perceptual pathways). But those images are nonspecific and far from quantitative, and just saying the word priming doesn't prove that the images have anything to do with what is observed. Only testing models can allow us to distinguish one explanation from another or choose between them.

Yes.

I don't know if this is likely to be done. I have no facilities with which to do them. If anyone takes this challenge to the concept of priming seriously maybe the experiments will happen. Of course the best way to defend against the challenge is to ignore it, so I suspect that is what will happen.

There is also the suggestion I made, that could also be tested in a control model. The reason I made it is that there is a lot of evidence, some of it from actual tracking studies, that prior perception of one value of an environmental variable affects the magnitude of the perception of another of the same kind. These often come under the heading of "figural after-effect" or "simultaneous contrast" or "adaptation". One of the most commonly observed is the motion after-effect you see after, say, watching a newscast with a continuous scrolling text bar below it. To anyone who has experienced it, the effect is on one's perception, not on one's control actions. One sees a portion of the visual field moving that one believes to be static. Bill's argument against "priming" would be equally valid in this situation, and could be used to say that this apparently perceptual effect is just a behavioural illusion.

Some years ago, I posted to CSGnet some tracking data on this effect, in which the subjects were asked to control the rotation velocity of a disc so that it was perceptually stationary, after having watched it rotating for some seconds or minutes. Several subjects (including me) described a strange logical inconsistency between the perceptions of location and of movement; the disc had clearly been held static for some seconds, but it was clear that the markings on it were now far from where they had been. It's a bit disconcerting, at first, to have these two conflicting perceptions. Logically, perceived rate of motion should be a function of the derivative of location, but it isn't. You might suggest that although the disc was perceived as static, yet there might have been imperceptible creep that would account for the movement, but that isn't the answer, because one can sometimes perceive the disc to rotate clockwise while perceiving the markings to move to their new locations counterclockwise.

I have no evidence as to whether such an effect exists on the perception of numerical magnitude as a function of prior consideration of another number. I suggested it as a possibility worth investigating, and showed that if there is such an effect and if the effect is to alter the scaling of the relation between number and perceived numerical magnitude, the Arielly data could be accounted for rather accurately -- more accurately than by the linear fit.

As for the term "priming", I don't know of it being used in this context outside the present discussion. In my experience, it is normally used in a linguistic context, such as this: I say "I need some cash, and I'm going to the bank", or I say "I like watching the river, and I'm going to the bank". In the first case, you would probably envisage the "bank" as a financial institution, and in the second case as a grassy slope with trees. That is what is called the "priming" effect of "cash" or of "river" on the category perception of "bank". "Priming" is also used to describe changes in the ability of people to detect patterns in noisy fields, after exposure to something that alters the likelihoods of different things being in the patterns -- for example, after a session of extracting woody nightshade vines from a hedgerow, I would see the remaining vines almost without searching, whereas initially I had to look carefully to see any of the then more abundant ones initially there. I don't much like the use of "priming" in that context, but it is used that way. It has also been called "perceptual enhancement", which I prefer. Both cases, however, deal with changes in the likelihood that certain patterns in the sensory data will correspond to particular perceptual categories.

Now back to your regularly scheduled programme.

Generically, two classes of models have been proposed. None of the suggestions are S-R. One suggests that the perception of what number corresponds to any specific value is affected (a proposal about a perceptual input function), whereas the other suggests that what is affected is the perceived value of the object. Within each class of model, different possibilities exist. In the perceptual input function proposal the possibilities are the different ways inwhich the SS# affects the perception (which need not be a simple scaling). In the perceived value class, we have Rick's original "bid/ratio" multiplicative model (which I modified to add a constant because simple bid-ratio did not fit the data) and Bill's additive disturbance model. Bill's model actually turns into the linearized bid-ratio model:

BP: Bid = SS#/(1+G) + Value*(G/1+G)

which is what I fitted to the data in my spreadsheet to examine whether the linearized value might be the controlled variable,

MT: Bid# = a*SS# + constant.

This model has a characteristic error in the form of the curve of bid versus SS#. The bids are too high in the central range of SS#s, and too low at the extremes by 10-15%. That might be fixed by altering the scaling of the SS#s, but to introduce this kind of scaling means to propose a particular perceptual input function for the perception of number. To do so increases the number of degrees of freedom using in describing the data.

Martin

[rsmarken]

[From Rick Marken (2009.11.21.1150)]

Martin Taylor (2009.11.21.10.36) –

There is also the suggestion I made, that could also be tested in a control model. The reason I made it is that there is a lot of evidence, some of it from actual tracking studies, that prior perception of one value of an environmental variable affects the magnitude of the perception of another of the same kind.

I think this could be incorporated into a control model as a different controlled variable. Instead of the controlled variable being O - A, for example, it could be something like O - (bAt-b2At-1), where now the effect of the asking bid on the controlled variable includes the effect of present (At) and prior (At-1) asks. But since the model sans the effects of prior perceptions picks up nearly all the variance in the observed outputs there probably isn’t much a prior perception model can do to improve things. But we’ll see. It certainly would be easy to replicate this study and capture the data for individuals that we need.

Best

Rick

···

–
Richard S. Marken PhD
rsmarken@gmail.com
www.mindreadings.com

[Bill_Powers1]

[From Bill Powers (2009.11.21.1035 MDT)]

Martin Taylor 2009.11.21.10.36 –

BP earlier: Now that we seem to
be reaching a consensus about the control model for the Arielly data, I
think we might start re-evaluating the subject of priming.

MT: One possible control model, yes?

BP: So far, one, though you seem to agree that a straight-line fit is
about the best we can do on the basis of available data. I’m leery of
introducing nonlinearities just to improve the fit without some
underlying model, because that opens the door to an infinite number of
kinds of nonlinearities – in this case, such as a polynomial with five
parameters that would fit perfectly at the five places where there are
data. If there’s no prospect of testing such a model (other than seeing
that it produces the desired fit) I think we should back off and settle
for what we have.

BP earlier: The explanation for
the apparent effect of the social security numbers on the bid prices was
simply that raising the SS# considered as an asking price resulted in an
increase of the bid, as if increasing the SS# made the subjects want to
pay more for the item.

MT: That is a new suggestion that hasn’t been mentioned hitherto, so far
as I know.

BP: I don’t understand. I thought I was just describing the appearance of
an effect. The SS# was explicitly presented as a price, with the subjects
being asked to indicate how much they would pay for the item. There
seemed to be an influence of the price on the bid. Are you referring to
the second clause, “as if increasing …”?

…

BP earlier: “Priming”
is simply a term applied to an observed effect, an apparently direct
effect. The word evokes images of certain possible mechanisms (I imagined
a leakage between perceptual pathways). But those images are nonspecific
and far from quantitative, and just saying the word priming doesn’t prove
that the images have anything to do with what is observed. Only testing
models can allow us to distinguish one explanation from another or choose
between them.

MT: Yes.

BP earlier: I don’t know if this
is likely to be done. I have no facilities with which to do them. If
anyone takes this challenge to the concept of priming seriously maybe the
experiments will happen. Of course the best way to defend against the
challenge is to ignore it, so I suspect that is what will
happen.

MT: There is also the suggestion I made, that could also be tested in a
control model. The reason I made it is that there is a lot of evidence,
some of it from actual tracking studies, that prior perception of one
value of an environmental variable affects the magnitude of the
perception of another of the same kind. These often come under the
heading of “figural after-effect” or “simultaneous
contrast” or “adaptation”.

BP: Those are three proposals that need a model behind them to be tested.
In some cases like the motion illusions you refer to, an actual change in
the perceptual signal might be occurring. I did an experiment with the
waterfall illusion very much like yours with the rotating disk, including
giving the subject a way to “stop” the after-effect by actually
causing the waterfall to move the opposite way. This gave very clear and
repeatable curves showing the illusion decaying with time. I would have
to accept a model in which the perceptual input function was actually
adapting to the continuous motion, although there might still be some
control-system model that would also work. I don’t know what it might be,
however. It would be a control system specifically connected with
perception, I would guess, but that might be indistinguishable from a
simple adaptive effect.

…

MT: I have no evidence as to
whether such an effect exists on the perception of numerical magnitude as
a function of prior consideration of another number. I suggested it as a
possibility worth investigating, and showed that if there is such an
effect and if the effect is to alter the scaling of the relation between
number and perceived numerical magnitude, the Arielly data could be
accounted for rather accurately – more accurately than by the linear
fit.

BP: Well, as I say, introducing nonlinearities opens the door to a lot of
fancy guesswork that can’t be verified.

MT: As for the term
“priming”, I don’t know of it being used in this context
outside the present discussion. In my experience, it is normally used in
a linguistic context, such as this: I say “I need some cash, and I’m
going to the bank”, or I say “I like watching the river, and
I’m going to the bank”. \

BP: OK, I admit not having much acquaintance with the way
“priming” is used. My description was consistent with what I
have seen, but obviously I haven’t seen enough.

MT: Now back to your regularly
scheduled programme.

Generically, two classes of models have been proposed. None of the
suggestions are S-R. One suggests that the perception of what number
corresponds to any specific value is affected (a proposal about a
perceptual input function), whereas the other suggests that what is
affected is the perceived value of the object.

BP: This is getting into the same problem as with your analysis of the
Shouten experiment, where the assumption was that after the initial
perceptual effect, the rest of the behavior was open-loop. If the
“bid” is considered the behavior of the subject, and a change
in the perception shows up as a change in the bid, that is a
stimulus-reponse or open-loop model. But this comment is easy to deal
with: just supply a diagram of the model showing how you propose that
it’s organized.

MT: Within each class of model,
different possibilities exist. In the perceptual input function proposal
the possibilities are the different ways inwhich the SS# affects the
perception (which need not be a simple scaling). In the perceived value
class, we have Rick’s original “bid/ratio” multiplicative model
(which I modified to add a constant because simple bid-ratio did not fit
the data) and Bill’s additive disturbance model. Bill’s model actually
turns into the linearized bid-ratio model:

BP: Bid = SS#/(1+G) + Value*(G/1+G)

MT: which is what I fitted to the data in my spreadsheet to examine
whether the linearized value might be the controlled variable,

MT: Bid# = a*SS# + constant.

This model has a characteristic error in the form of the curve of bid
versus SS#. The bids are too high in the central range of SS#s, and too
low at the extremes by 10-15%.

Here are the curves from my program with the percent deviations shown,
red being the real bid data and green being the model bids.

That might be fixed by
altering the scaling of the SS#s, but to introduce this kind of scaling
means to propose a particular perceptual input function for the
perception of number. To do so increases the number of degrees of freedom
using in describing the data.

Yes, I think this is really about as much analysis as the data can stand.
The RMS deviations of the model from the real data range from 1.5 to 4.4,
or as a fraction of the highest value of the model bid price for the
respective items, from 0.06 to 0.09.

Best,

Bill P.

[MartinT]

[Martin Taylor 2009.11.21.17.05]

[From Bill Powers (2009.11.21.1035 MDT)]

Generically, two classes of models have been proposed. None of the
suggestions are S-R. One suggests that the perception of what number
corresponds to any specific value is affected (a proposal about a
perceptual input function), whereas the other suggests that what is
affected is the perceived value of the object.

BP: This is getting into the same problem as with your analysis of the
Shouten experiment, where the assumption was that after the initial
perceptual effect, the rest of the behavior was open-loop.

May I quote from [From Bill Powers (2009.11.21.0645 MDT)]:
I therefore make a modest proposal: let’s all stop using
emotion-names,
and instead communicate what is wanted and what feelings go with the
cognitive reference level. So instead of saying “You’re annoying
me,” we would say “I do wish you wouldn’t talk that way, but
I’m not very upset about it.” That conveys much more clearly what
I’m feeling and what I want than does the phrase "mildly
annoyed."

I am mildly annoyed at that misrepresentation. It’s a complete
fabrication, in both cases.

Martin

[Bill_Powers1]

[From Bill Powers (2009.11.21.1535 MDT)]

Martin Taylor 2009.11.21.17.05 --

BP: This is getting into the same problem as with your analysis of the Shouten experiment, where the assumption was that after the initial perceptual effect, the rest of the behavior was open-loop.

I am mildly annoyed at that misrepresentation. It's a complete fabrication, in both cases.

I'm not going to try to find all that Schouten stuff again, but my memory says that you set up the model so it didn't use any feedback from pressing the button, and explicitly said, indeed insisted repeatedly, that this was an open-loop process because pressing the button couldn't affect the state of the light after it had turned on. In the present case, the parallel I see is that the effect of the SS# on the bid is open-loop because the bid can't affect the SS#. That's why I call it an S-R model.

Did I completely fabricate all that?

Best -- well, second best,

Bill P.

[rsmarken]

[From Rick Marken (2009.11.21.1530)]

Bill Powers (2001.20.1115 MDT)–

Now that we seem to be reaching a consensus about the control model for the Arielly data

Again, just for the heck of it, I tried a model with a different controlled variable: CV = O/A. This assumes that the subject is controlling for making a bid that is proportional to A (initial asking price). For the products that are more dear the reference for this ratio would be higher. The best fit version of this model does about 3 times worse than the simple difference model (CV = O - A). There are some others models I’ll try. But I think this Arielly experiment is a good way to demonstrate the difference between the PCT and S-R (priming theory, in this case) approach. And it should be easy to get individual data to test these models (that’s the part that’s attractive to me).

Best

Rick

···

–
Richard S. Marken PhD
rsmarken@gmail.com
www.mindreadings.com

[Richard_H_Pfau]

[From Richard Pfau (2001.11.23.1134 EST])

[From Bill Powers (2001.20.1115 MDT)]

[1] The priming model is a simple stimulus-response model as I understand it. The magnitude of the independent variable is the size of the social security number, and the magnitude of the dependent variable is the bid offered for the item.

[2] If the control system model is favored, the term “priming” and all of its associated implications becomes inappropriate.

[3] “Priming” is simply a term applied to an observed effect, an apparently direct effect. The word evokes images of certain possible mechanisms (I imagined a leakage between perceptual pathways). But those images are nonspecific and far from quantitative, and just saying the word priming doesn’t prove that the images have anything to do with what is observed. Only testing models can allow us to distinguish one explanation from another or choose between them. I

Regarding [1], [2], and [3] above (which are my numbers):

1. The dependent variable of priming isn’'t the bid, as I see it, but the sensitization of neural networks. The magnitude of the dependent variable would seem to be the degree to which related neural networks are sensitized by the prime.
2. I see no incompatability between the control system model and priming, as explained below.
3. I agree that priming is simply a term applied to an observed effect and that the possible mechanisms are speculative [but based upon some neural evidence]. However, I suggest that the priming effects on behavior be viewed as “indirect” rather than direct. Also, testing, as suggested in [3] above, may try to compare two different levels of explanation, which may not be productive.
   Let me explain:

PCT is a systems theory that explains relationships between perceptions and behavior using a number of components and processes (Input Function, Perceptual signal, etc). The idea of “priming” seems to be a concept on a different level of explanation than that of PCT (i.e., on s subsystem level), but a concept that may help to shed some light on how one or two PCT components operate.

That is, the concept of priming seems related to operations of the Input (Perceptual) Function and possibly the Output (Behavior) Function. To me at least, the priming idea indicates that certain sensory inputs may activate and thereby temporatily sensitize neural networks within the PCT Input Function and possibly within the Output Function (at time 1) such that the signals generated by these components may be affected shortly afterwards (at time 2) – as in the Ariely experiment when a person is asked to make a bid.

Any sensitization/priming effects that occur are viewed as being indirect in generating perceptual and later output signals, since the sensitization effects are only one of many factors affecting the perceptual and output signals generated.

I see no incompatibility between the idea of “priming” and that of PCT. The idea of priming, seems simply to caste light (via the sensitization of neural networks) on how the Input Function and Output Function possibly operate.

The testing being proposed seems to compare the entire PCT model against a possible explanation of how priming/sensitization may affect operations of the Input (Perceptual) Function and the Output (Behavior) Function. An analogy would be comparing how a heating system operates (model 1) to how the furnace of that system operates (model 2). In other words, comparing PCT (model 1) and priming (model 2) doesn’t seem to be valid – since the test seems to be comparing two different levels of explanation – one more macro (PCT) and one more micro (priming).

Since, like all of us, I’m in a learning mode, any feedback on these ideas will be appreciated.

With Regards,

Rich Pfau

[Bill_Powers1]

[From Bill Powers (2009.11.23.1112 MDT)]

Richard Pfau (2001.11.23.1134 EST])

···

–

1. RP: The dependent variable of priming isn’'t the bid, as I see it,
   but the sensitization of neural networks. The magnitude of the
   dependent variable would seem to be the degree to which related neural
   networks are sensitized by the prime.
   BP: We can’t measure the sensitization of neural networks; the bid
   is our only clue. That’s the dependent variable we can measure. You may
   propose that the sensitizing perception has some effect on the
   intervening neural network, but that’s not specific enough to build a
   model on.

With respect to the PCT model, if the perceptual input function is made
more sensitive, it will produce a larger perceptual signal for the same
value of the sensory stimulation as before. Is that the kind of change
needed to generate the priming effect? That would reduce the bid, not
increase it. So I don’t think so. I don’t think that increasing the
sensitivity (gain) of the output function would do that, either. So as
matters stand, I don’t think we can explain the priming phenomenon as an
increase of sensitivity of any part of a control system. In my model I
assume the effect is like that of an additive disturbance, with the
sensitivity of the input function being unaffected.

This says we need a more specific model or a different model if we’re
looking for a way to explain priming as sensitization. When you offer the
concept of sensitization, exactly what is it that becomes more sensitive?
So you have in mind a different model that would behave correctly if the
sensitization occurs?

RP: I see no incompatability
between the control system model and priming, as explained below.

That depends on what model you use to generate the priming effect. If you
use the control-system model, the effect of a high Asked price, AKA SS#,
on the output behavior of the model tends to reduce the bid price while
the implication of the term “priming” would lead us to think
that a high SS# would cause behavior that tends to make the bid price
higher. As far as I know, only the control-system model can explain how
the behavior could tend to reduce the bid but the net effect on the bid
would be an increase. The control-system model does predict correctly,
but by a non-intuitive mechanism.

1. RP: I agree that priming is simply a term applied to an observed
   effect and that the possible mechanisms are speculative [but based upon
   some neural evidence]. However, I suggest that the priming effects
   on behavior be viewed as “indirect” rather than direct.
   BP: The control-system model provides an explanation that’s about as
   indirect as you could wish for. We posit that with a high enough
   reference level, the control system would pay the asked price, but with a
   reference level below the asking price, the control system would lower
   the amount it is willing to pay, with the net effect that the actual
   increase in bid price is much less that it would have been without the
   feedback effect.
2. RP: Also, testing, as suggested in [3] above, may try to compare two
   different levels of explanation, which may not be
   productive.
   Let me explain:

PCT is a systems theory that explains relationships between
perceptions and behavior using a number of components and processes
(Input Function, Perceptual signal, etc). The idea of
“priming” seems to be a concept on a different level of
explanation than that of PCT (i.e., on s subsystem level), but a concept
that may help to shed some light on how one or two PCT components
operate.

BP: Not sure what you mean by a different level of explanation. However
you explain the effect, to show that your explanation is feasible you
have to lay out a model that will correctly predict the observations,
closely enough to be considered useful. I have been considering that
priming works by an effect of the Asked price on the perceptual input
function, which is a subsystem of the control system. That is, I have
assumed that the perceived (and controlled) bid price would be equal to
the asking price if the control system took no action to reduce the bid
price. The failure of the bid price to rise and fall by the same amount
as the asked price is due to negative feedback from the output of the
control system. In other words, I’m assuming 100% priming in the absence
of control (i.e., just pay the price on the price tag), with the control
system accounting for the much smaller priming effect actually
observed.

RP: That is, the concept of
priming seems related to operations of the Input (Perceptual) Function
and possibly the Output (Behavior) Function.

BP: Input function, yes. Output function no, because in a control system
changing the gain in the output function has little effect on the actual
output. Check that out with the Live Block Diagram, chapter 3.

RP: To me at least, the
priming idea indicates that certain sensory inputs may activate and
thereby temporarily sensitize neural networks within the PCT Input
Function and possibly within the Output Function (at time 1) such that
the signals generated by these components may be affected shortly
afterwards (at time 2) – as in the Ariely experiment when a person is
asked to make a bid.

BP: That’s not a specific enough statement of the mechanism to allow
making any predictions. The term “sensitize” has to be pinned
down to the actual effect being proposed, on the specific components of
the system you have in mind, in the whole model of which those components
are a part. Of course you’re free to adjust the parameters of the model
for best fit with the data, but before you can do that you have to
specify the model you’re testing.

RP: Any sensitization/priming
effects that occur are viewed as being indirect in generating perceptual
and later output signals, since the sensitization effects are only one of
many factors affecting the perceptual and output signals
generated.

BP: True, but not very useful unless you say specifically what is
affecting what, and what the consequences of that will be according to
your model.

RP: I see no incompatibility
between the idea of “priming” and that of PCT. The idea
of priming, seems simply to caste light (via the sensitization of neural
networks) on how the Input Function and Output Function possibly
operate.

BP: If you predict that priming will increase the behavior, the output,
of the system, that would be incompatible with the behavior of the
control-system model. The whole point I’m trying to get at here is that
we can decide issues such as incompatibility between two models only by
comparing the predictions of the two models in detail, considering not
just the overall effect but how that effect is produced.

RP: The testing being proposed
seems to compare the entire PCT model against a possible explanation of
how priming/sensitization may affect operations of the Input (Perceptual)
Function and the Output (Behavior) Function. An analogy would be
comparing how a heating system operates (model 1) to how the furnace of
that system operates (model 2). In other words, comparing PCT
(model 1) and priming (model 2) doesn’t seem to be valid – since the
test seems to be comparing two different levels of explanation – one
more macro (PCT) and one more micro (priming).

BP: In my opinion, no matter what kind of model you use, the bottom line
is that it must generate predictions of the values of observable
variables. If it does that, you’re free to use as much or as little
detail in the proposed model as you please. We have no way of comparing
internal details of one model with internal details or another: we have
to use observable variables. After we see that the model generates
reasonable accurate predictions, we can start testing it by varying the
conditions, which will lead to modifications of the model and further
testing. Where all models can be compared with all other models is at the
level of predicting what we will observe. If you can make correct
predictions with an abstract model that’s just as good as making them
with a boring low-level mechanical model. The only aspect of a model
that’s considered out of bounds is anything that makes use of the effect
that is to be explained so the model is forced to produce it: like saying
“Priming is caused by a modification of the system that causes
priming effects” – or any model that boils down to that
tautology.

RP: Since, like all of us, I’m
in a learning mode, any feedback on these ideas will be
appreciated.

BP: That is a correct use of the term feedback, which is an input to a
system that is produced by that system’s action on its
environment.

I learn mainly from people asking hard questions, so thanks.

Best,

Bill P.

[Richard_H_Pfau]

[From Richard Pfau (2009.11.23.1637 EST)]

[From Bill Powers (2009.11.23.1112 MDT)]

…

I learn mainly from people asking hard questions, so thanks.

And thanks to you for your detailed and patient response!

With Regards,

Rich Pfau