Cold Reading Demos at TAM2

Thanz said:
Yes, I have. Look at page 26 of the thread - the first post by me neatly summarizes the problems with his method.

Basically, he counts "I'm getting a J connection, like John or James" as three guesses - as if he had said "J, like J or J" in one reading or "J" connection once in three separate readings. This is an inaccurate count of both the number of J guesses and the total number of guesses. You cannot get a reliable result from flawed data.

You have?

What statistical method are you using here?

Don't even bother with anything else but the name of the statistical method you used.

I am not interested in your filibuster.
 
Thanz said:

Yes, I have. Look at page 26 of the thread - the first post by me neatly summarizes the problems with his method.

Basically, he counts "I'm getting a J connection, like John or James" as three guesses - as if he had said "J, like J or J" in one reading or "J" connection once in three separate readings. This is an inaccurate count of both the number of J guesses and the total number of guesses. You cannot get a reliable result from flawed data.
Thanz,

These claims didn't work before and they won't work now unless and until you can demonstrate the flaw. You claim it is an inaccurate count of both the J and total number of guesses. Let us assume, for the moment, that that is correct. Why, then, do the "J" counts soar so far above all the others? I posed this before, you couldn't answer then. Your situation worsened when Lurker applied my counting methods to an entirely different data set and got the same result: his "J" counts soared far above all the others.

What you want us to believe is that I concocted a technique to make this so. That, somehow, my technique favors "J"s. How? Why? Why did Lurker achieve the same result with different data?

You now have two separate people making the same finding with two separate data sets. Did Lurker commit fraud? How? Walk us through it, count by count.
 
Hoyt,

It is imperative that Thanz point out the statistical flaws. That's what it comes down to.

If you want to complain about methods, you should be able to point out where the problems are.

Whatever other reason you want to bring into it is completely irrelevant. If you want to be taken seriously, you got to stay focused.

Filibuster does not substitute content. Ridicule does not substitute content.

If Thanz has a problem with your use of statistics, he should be able to pinpoint what the problem is.
 
CFLarsen said:

Back up a bit here. If I have not understood what you are saying, then let's get that cleared up now.


Ok..


We have one dataset (A) that comes from the "paranormal" group.

We have one dataset (B) that comes from the "normal" group.


What?? :) Sorry if I was unclear.

No, we have a dataset. This dataset has data from one group that doesn't receive the treatment, and data from a group who has received the treatment (drug, or whatever).


The statistician knows which group is which.


If he's unblinded, yes, he'd have the labels for each group, say Drug and Control. If he's blinded, then no, he'd only have the labels Group1 and Group2.
 
BillHoyt said:

Jeff was talking about your use of 2-tail on the one hand and 1-tail on the other hand. Now you start talking about your feet. Stop displaying your ignorance and diverting from the topic, sir.


Still waiting for you or him to address the updated example...


If a priori, you are claiming a paranormal situation.

Nope. Unconscious bias is not paranormal, it is normal.
 
Jeff Corey said:
...just like I analyzed TC's data and found his huge error, why would someone risk their rep by being careless (as TC was) or by cheating?

:rolleyes:

Please address the updated example. Or you can remain focusing on the small error and avoid the relevant issue.
 
CFLarsen said:

I see T'ai Chi try to counter Hoyt, but is rebutted by Jeff Corey.


Oh please. :D

I made a minor error, corrected it, yet Jeff focuses on the minor error and avoids addressing the corrected version.

That is being rebutted?
 
CFLarsen said:
...and still you are not able to show - by pointing to statistic methods - where Hoyt is wrong. A simple request, crucial to your claim.

Irrelevant. Hoyt, using statistical methods, has not provided evidence his approach is anywhere near correct in the first place!

If J counts are large, but other commonly used letter counts are not large, then what can we say?

I don't see what, if anything, can be gained by focusing on counts of one letter. It seems overly simplistic. Maybe Bill could explain to us all why he thinks it is a good idea?
 
You know what puzzles me most about this whole harangue of T'ai Chi, is that it only makes sense if the "skeptics" arguing with him think that there is absolutely -no chance- of unconscious bias influencing the selection of methods for analyzing data.

So...Bill et al. ...is that your position? That people charged with analyzing or interpreting data never, ever could have an unconscious bias that would lead them to favor one method over another?

Iyo, does bias just never become an issue? Or are you saying that it does, but it is never unconscious bias....that it is always intentional?
 
T'ai Chi said:
No, we have a dataset. This dataset has data from one group that doesn't receive the treatment, and data from a group who has received the treatment (drug, or whatever).

Sorry, but how is this considered one dataset? We have to keep them apart, to tell the difference!

How can you determine which data belongs to which group, if you put both groups together?

Two datasets. "Drugs". "Control". You say so yourself:

T'ai Chi said:
If he's unblinded, yes, he'd have the labels for each group, say Drug and Control. If he's blinded, then no, he'd only have the labels Group1 and Group2.

It doesn't matter if he is blinded or not. If he applies the same method to each group, how can he get a biased result, before he has analyzed the data?

How does he know, in advance, which method to use?

I really don't see why we have to go through all this. It seems very simple enough, but - for some reason - you have a lot of trouble explaining it.

Clancie said:
You know what puzzles me most about this whole harangue of T'ai Chi, is that it only makes sense if the "skeptics" arguing with him think that there is absolutely -no chance- of unconscious bias influencing the selection of methods for analyzing data.

It is not "harangue". If it were, you should file a complaint to the moderators. Yet, you do not.

Nobody is saying that this is impossible. All that is needed is evidence that it is possible.

That's what you are so intent on missing.
 
T'ai Chi said:
No, we have a dataset. This dataset has data from one group that doesn't receive the treatment, and data from a group who has received the treatment (drug, or whatever).
No, we have two datasets. Each dataset represents a distinct population whose characteristics are to be contrasted by the statistical test. Each population can be independently compared with any other population to determine a degree of "sameness" or "differentness." This is absolutely fundamental to an understanding of statistical inference testing.
 
T'ai Chi said:
.Nope. Unconscious bias is not paranormal, it is normal.
A priori is a basic term that any first-year stat student would be familiar with. It means "beforehand." What you claim here is that somebody, beforehand, before taking a measurement, before seeing the data, would know what test will best suit their bias. This is not statisitical intuition. This is not experience. This is psychic nonsense.

If, instead, you claim this person first, does the calculations and then doesn't like the outcome and tortures the data or test to force an outcome, you are no longer speaking of bias, but of fraud.
 
CFLarsen said:

Sorry, but how is this considered one dataset? We have to keep them apart, to tell the difference!


It really is one dataset.

Imagine a spreadsheet where each row represents an individual person, with say 7 columns which are variables relating to those peopele: AGE, GENDER, Sys. BLOOD PRESSURE, Dias. BLOOD PRESSURE, HEIGHT, WEIGHT, and GROUP. The first 6 are self explanatory.

If the evaluator is unblinded, the GROUP column entries (if we are just looking at two groups) would be either 'drug' or 'no drug', for example. If the evaluator is blinded to group membership, the GROUP column entries would be either 'group 1' or 'group 2', for example. This way the evaluator would not be aware of which group received the drug.


How can you determine which data belongs to which group, if you put both groups together?


See above. There is simply a column in the spreadsheet that records group membership information.


Two datasets. "Drugs". "Control". You say so yourself:


Two, for example, choices for the groups involved. In this case, two groups because we have one treatment (one drug).


It doesn't matter if he is blinded or not.


Yes, it does, for reasons I've already stated. But I'll state them again: If not blinded the evaluator could impart concious or unconscious bias in the analysis and/or interpretation of the analysis of the data.


I really don't see why we have to go through all this. It seems very simple enough, but - for some reason - you have a lot of trouble explaining it.


That's an interesting spin on it.
 
BillHoyt said:

A priori is a basic term that any first-year stat student would be familiar with. It means "beforehand."


Yeah, thanks. :rolleyes:


What you claim here is that somebody, beforehand, before taking a measurement, before seeing the data, would know what test will best suit their bias. This is not statisitical intuition. This is not experience. This is psychic nonsense.


What in the world are you talking about?? Are you reading the same thread? Where do you get all of this 'before taking a measurement, before seeing the data' nonsense?
 
T'ai Chi said:
What in the world are you talking about?? Are you reading the same thread? Where do you get all of this 'before taking a measurement, before seeing the data' nonsense?

Very well, then: Time to show your evidence.

Here are two groups of data:

Coffee
19,07,02,17,98,60,94,37,02,77,05,39,21,71,76,29,31,76,75,23,84,67,48,18,46,76,69,40,51,32,26,19,31,18,81,71,01,00,03,13,78,38,75,28,86,58,75,33,20,83,81,42,06,17,17,76,69,14,73,03,59,82,53,49,04,28,75,54,68,73,11,59,56,28,63,88,23,53,78,75,93,75,19,57,78,18,57,78,05,32

Not coffee
17,32,17,21,47,72,35,01,41,44,19,73,56,85,48,16,13,61,15,73,52,55,21,33,47,57,41,84,94,68,43,85,23,32,39,07,39,41,43,33,45,47,76,24,16,86,25,18,98,35,69,48,55,62,09,92,19,22,21,84,27,25,50,25,42,56,88,76,71,79,04,94,60,16,53,46,68,04,98,86,27,23,27,91,78,60,85,78,43,83

Please analyze these data, using a method that will show bias towards the "coffee" group.

Show the result. Explain, step by step, how you got it.
 
CFLarsen said:

Very well, then: Time to show your evidence.


Um, you replied for Bill, and on top of that, your reply was a non-sequitor in response to my question to Bill.

My question to Bill:

---
What in the world are you talking about?? Are you reading the same thread? Where do you get all of this 'before taking a measurement, before seeing the data' nonsense?
---


Here are two groups of data:

Coffee
19,07,02,17,98,60,94,37,02,77,05,39,21,71,76,29,31,76,75,23,84,67,48,18,46,76,69,40,51,32,26,19,31,18,81,71,01,00,03,13,78,38,75,28,86,58,75,33,20,83,81,42,06,17,17,76,69,14,73,03,59,82,53,49,04,28,75,54,68,73,11,59,56,28,63,88,23,53,78,75,93,75,19,57,78,18,57,78,05,32

Not coffee
17,32,17,21,47,72,35,01,41,44,19,73,56,85,48,16,13,61,15,73,52,55,21,33,47,57,41,84,94,68,43,85,23,32,39,07,39,41,43,33,45,47,76,24,16,86,25,18,98,35,69,48,55,62,09,92,19,22,21,84,27,25,50,25,42,56,88,76,71,79,04,94,60,16,53,46,68,04,98,86,27,23,27,91,78,60,85,78,43,83

Please analyze these data, using a method that will show bias towards the "coffee" group.

Show the result. Explain, step by step, how you got it.

First, you implicitly assume that I claimed that one could always find areas where unconscious bias plays a role in the analysis for certain. Sure, one can cook up examples where you can't or it isn't too easy; I never said otherwise. Bias, in general, is just an area to be aware of when doing an experiment, and hopefulyl experimenters can prevent as much of it as possible.

Second, not even graphing these data, I could see that it might possibly come down to a Mann-Whitney U-test vs. a T-test. In this case, since our sample size is relatively large, that probably isn't much of an issue though. It would be something to be aware of when working with smaller samples (samples sizes around 20 or fewer).

The main area unconscious bias could creep in here in your example as I see it would probably be in the interpretation of the analysis, if the evaluator (me) loves or hates coffee, and/or in the graphical displaying of the results. Maybe I love coffee so much I'd say the a large p-value (ie. no statistical difference between the Coffee and No Coffee group, in say intelligence levels) isn't that big of deal and not a lot of weight should be put on it due to various factors the study didn't mention, because I'm biased to believe that coffee drinkers are smarter than non-coffee drinkers. Perhaps I'd also make the title of the Coffee histogram in a pretty blue, and make the No Coffee title in a harsh red? Perhaps I'd unconsciously be less critical in general because I am aware that coffee, something I love, is involved? Maybe one of my relatives owns a coffee store? I wouldn't want to possibly decrease his sales. Gees, what if I know that my department gets major funding from Starbux? I wouldn't want to disappoint them.

Forget coffee for a second and change the 'treatment' to AIDS, blacks, cigarettes, polluted rivers, alcohol, abortion, especially anything that humans can have a large emotional attachment with and ego invested in (which is about anything), and that is the main issue.

Yeah, I could go through the entire process of analyzing the data step by step, but I'm out of school, so I don't have to do homework. :) You could analyze it, and report back to us. It might be instructive.

You also could read the Fisher's Exact vs. Chi square example (some pages back now) for an example already discussed where the choice of the actual statistical test mattered more. But gees, I made a silly error when I posted that example, so better to forever ignore to corrected version that I followed up with. ;)
 
T'ai Chi said:
First, you implicitly assume that I claimed that one could always find areas where unconscious bias plays a role in the analysis for certain. Sure, one can cook up examples where you can't or it isn't too easy; I never said otherwise. Bias, in general, is just an area to be aware of when doing an experiment, and hopefulyl experimenters can prevent as much of it as possible.

Second, not even graphing these data, I could see that it might possibly come down to a Mann-Whitney U-test vs. a T-test. In this case, since our sample size is relatively large, that probably isn't much of an issue though. It would be something to be aware of when working with smaller samples (samples sizes around 20 or fewer).

The main area unconscious bias could creep in here in your example as I see it would probably be in the interpretation of the analysis, if the evaluator (me) loves or hates coffee, and/or in the graphical displaying of the results. Maybe I love coffee so much I'd say the a large p-value (ie. no statistical difference between the Coffee and No Coffee group, in say intelligence levels) isn't that big of deal and not a lot of weight should be put on it due to various factors the study didn't mention, because I'm biased to believe that coffee drinkers are smarter than non-coffee drinkers. Perhaps I'd also make the title of the Coffee histogram in a pretty blue, and make the No Coffee title in a harsh red? Perhaps I'd unconsciously be less critical in general because I am aware that coffee, something I love, is involved? Maybe one of my relatives owns a coffee store? I wouldn't want to possibly decrease his sales. Gees, what if I know that my department gets major funding from Starbux? I wouldn't want to disappoint them.

Forget coffee for a second and change the 'treatment' to AIDS, blacks, cigarettes, polluted rivers, alcohol, abortion, especially anything that humans can have a large emotional attachment with and ego invested in (which is about anything), and that is the main issue.

Yeah, I could go through the entire process of analyzing the data step by step, but I'm out of school, so I don't have to do homework. :) You could analyze it, and report back to us. It might be instructive.

You also could read the Fisher's Exact vs. Chi square example (some pages back now) for an example already discussed where the choice of the actual statistical test mattered more. But gees, I made a silly error when I posted that example, so better to forever ignore to corrected version that I followed up with. ;)

Can't do it, huh?
 
T'ai Chi said:
Oh please. :DI made a minor error, corrected it, yet Jeff focuses on the minor error and avoids addressing the corrected version.[/B]
A minor error? The p value was twice what you represented. The point is that your huge error was made to prove your point that stat rats could be biased.
Don't you get it?
 

ISF - Join now!

Every member here is approved by hand. No bots, no spam, just people who care about evidence and honest debate.

Membership is free!

Create your free account

Back
Top Bottom