The "Process" of John Edward

[BillHoyt]

First off, provide evidence for your assertion that the homicide rate data did not "suffer from the same problem". You made the claim, you provide the evidence.

That was an assumption on my part based on the Uniform Crime Reporting program first established by the FBI in 1929. It gives uniform definitions to crimes and specifies how they are to be tallied and reported. All cities and states report to this system.

" To ensure these data are uniformly reported, the FBI provides contributing law enforcement agencies with a handbook that explains how to classify and score offenses and provides uniform crime offense definitions. Acknowledging that offense definitions may vary from state to state, the FBI cautions agencies to report offenses not according to local or state statutes but according to those guidelines provided in the handbook. Most agencies make a good faith effort to comply with established guidelines."
UCR

Now it is possible that the data you referred to was not UCR data or it is possible that something went awry with these data such that the assumption that any random or systematic biases are not the same for all the data. If that is so, then they also erred in the comparisons.

Second, where did I mention significance of my data? I merely compared means.

That would be found in your post:

"While this analysis is anecdotal I think it clearly shows that using only a sample size of 78 people to define a population's first initial to their first name is woefully inadequate. At least if you want to have a reasonable precision to your numbers."

The bolded part translates to "significant".

Third, you said there was something wrong mathematically with my formula. I have yet to see you provide evidence of this.

I did not say something was wrong with the formula! I said you had an arithmetic error and a statistical error. The arithmetic error is using different denominator units, which I have explained ad nauseum. The statistical error overlaps that one in one sense, because you need to analyze the sample and the population statistically to understand whether or not the denominator units are truly the same. It then goes beyond that in not using statistical tools to compare populations. Ask yourself this: if you could really do what you did, why can't we simply set a "differences in percentages" criterion and use this simple test to compare populations. Why on earth do we go through all the fuss of statistics? Chi-square tests, one-tailed, two-tailed, binomial, fisher's exact, student's? Why all these clap-trap terms? Mean, variance, standard deviation, skew, kurtosis, moment-generating functions? Why not just subtract, divide and see if the percentage difference is greater than some pre-set criterion?

 

[Thanz]

These other factors are other hypotheses that might also be tested. I didn't say it says anything about the rare letters. In order for the frequent ones to be used excessively, other letters must be used less frequently. We can focus on the upper tail of the frequent letters to look for the skew. We do not need to see the full histogram.

Wouldn't the full histogram be more accurate? Wouldn't it make more sense to at least look at more than one letter? I am shocked that you attach any meaningful significance to your simple analysis of one letter.

I know the test. I commented on why it would not work for these data. Jeff Corey then suggested Fisher's exact test. Please refer to those posts.

What you said was:

Chi square won't work with frequencies less than 5. That makes it useless for this case.

In other words, we don't have the data to do the test. The sample size is TOO SMALL. Just like I said.

You have not answered my question in the previous post, which was:

Do you agree that such an analysis would be much more meaningful, in terms of demonstrating whether JE consistently uses more frequent letters at the expense of less frequent letters (as we would expect a cold reader to do)?

There are 26 letters and a 1 in 20 chance of getting a significant answer. We'd expect to stumble onto one. That is why my a prior selection of "J" is important. There was a rational basis for it, and it absolutely fit the hypothesis. If we speculate there may be skewing in favor of the most frequent initials, we'd expect to see that by looking at the absolutely most frequent initial.

Yes, we would expect to see that. But how can we have any confidence, considering that we have not looked into any other letters, that out of the 26 letters "J" is not just a significant answer we "stumbled onto"? How do we know that "J" isn't just a fluke? Is it simply because it confirms your hypothesis? That is not what I would consider great critical thinking.

Given what you have just said, looking at "J" in isolation of the other letters is a fool's errand. We can not know with any certainty whether the significant result we find is significant due to the cold reading hypothesis or if it is just a fluke.

 

[Lurker]

Originally posted by BillHoyt

That was an assumption on my part based on the Uniform Crime Reporting program first established by the FBI...


That still does not resolve the POP A vs POP B question. How can someone compare a specific city's homicide rate to the states'? Clearly theyare two different populations and the denominators would be different. Do you agree or disagree that the denominators are different? Further, they then compare the percentages and say how different they are. They did not mention significance. They did not mention confidence. They only compared the means. Yet you accept their statements. It strikes me as a bit hypocritical, on your part.



"While this analysis is anecdotal I think it clearly shows that using only a sample size of 78 people to define a population's first initial to their first name is woefully inadequate. At least if you want to have a reasonable precision to your numbers."

The bolded part translates to "significant".



I am going to go over this very slowly, Bill, so you understand. If 78 people fail to adequately define the population due to lack of significance, it only proves my contention that 78 was far too small a sample size. You see, if each letter is +/- 10% then all letters seem pretty much the same. Thus, the sample size is far too small. It seems you are in full agreement with my contention. I am glad to see you are coming around.


I did not say something was wrong with the formula! I said you had an arithmetic error and a statistical error. The arithmetic error is using different denominator units, which I have explained ad nauseum. The statistical error overlaps that one in one sense, because you need to analyze the sample and the population statistically to understand whether or not the denominator units are truly the same. It then goes beyond that in not using statistical tools to compare populations. Ask yourself this: if you could really do what you did, why can't we simply set a "differences in percentages" criterion and use this simple test to compare populations. Why on earth do we go through all the fuss of statistics? Chi-square tests, one-tailed, two-tailed, binomial, fisher's exact, student's? Why all these clap-trap terms? Mean, variance, standard deviation, skew, kurtosis, moment-generating functions? Why not just subtract, divide and see if the percentage difference is greater than some pre-set criterion? [/QUOTE]

Clearly a quick comaprison of means tells can indicate if we have tested enough to arrive at the proper distribution. I never meant it as an absolute test. Once we think we are getting better numbers (subjectively) we can then perform teh exhaustive statistical analysis. At least, that is how I would approach it for a problem like this. Frankly, that is probably how YOU would approach it also since you have for many days now refused to comment on how big a sample is required to create a proper histogram of the US. The real reason you won't supply an answer is you don't know how to figure it out. Welcome to the club.

Lurker

 

[Thanz]

Bump for BillHoyt. You have some unanswered queries here, sir.

 

[BillHoyt]

Bump for BillHoyt. You have some unanswered queries here, sir.

Thanz,

1. I am repeating myself and the points are still not getting across to either you or Lurker. I can't say whether this is deliberate or not, but it is quite tiresome. The specific points you last raised, for instance, I addressed in at least three different posts!

2. I stand by the ooodles of posts I have made on this subject.

3. Did you not notice there is nobody else on this thread anymore?

4. Bye.

 

[Thanz]

Mr. Hoyt -

Perhaps your points do not get through because you consistently fail to make them in clear and concise language. Instead, you just play some game of socratic method an oblique answers, and complain when we don't make your points for you.

I am simply asking for a clear, direct answer to my clear, direct questions. I don't think that is too much to ask.

Here they are again. They can be dealt with by you in one post, in the same time it took you to type your previous non-answer.

1. Do you agree that such an analysis would be much more meaningful, in terms of demonstrating whether JE consistently uses more frequent letters at the expense of less frequent letters (as we would expect a cold reader to do)?

2. How can we have any confidence, considering that we have not looked into any other letters, that out of the 26 letters "J" is not just a significant answer we "stumbled onto"?

3. How do we know that "J" isn't just a fluke? Is it simply because it confirms your hypothesis?

 

[Lurker]

Thanz:

It would appear that we can put Mr. Hoyt in with Claus as people that refuse to answer questions when the going gets tough. A shame, really.

Lurker

 

[BillHoyt]

Mr. Hoyt -

Perhaps your points do not get through because you consistently fail to make them in clear and concise language. Instead, you just play some game of socratic method an oblique answers, and complain when we don't make your points for you.

I am simply asking for a clear, direct answer to my clear, direct questions. I don't think that is too much to ask.

Here they are again. They can be dealt with by you in one post, in the same time it took you to type your previous non-answer.

1. Do you agree that such an analysis would be much more meaningful, in terms of demonstrating whether JE consistently uses more frequent letters at the expense of less frequent letters (as we would expect a cold reader to do)?

2. How can we have any confidence, considering that we have not looked into any other letters, that out of the 26 letters "J" is not just a significant answer we "stumbled onto"?

3. How do we know that "J" isn't just a fluke? Is it simply because it confirms your hypothesis?

1. That is not the question. The question is: what can be learned from the available data. I have made this clear going back pages, in post after post.

2. We would have LESS confidence by dipping into the same data set. The significance level is .05. Twenty six dips in, we would expect to get a "significant" hit by chance alone. You cannot do that. I chose the most frequently seen letter based on the control population. I saw excessively reliance on that letter. I stopped dipping in NOT because I got what I was looking for, but because I had completed the test and because further such testing would be statistically invalid. I addressed this previously and I will not repeat this again.

3. We don't. Your response to this ought to be rich.

 

[Thanz]

Thanz:

It would appear that we can put Mr. Hoyt in with Claus as people that refuse to answer questions when the going gets tough. A shame, really.

Lurker

I am afraid that you are correct. It bothers me when I see people who can "dish it" but can't "take it". They are behaving like bullies, nothing more.

BTW, do you ever check your PM's?

 

[Thanz]

1. That is not the question. The question is: what can be learned from the available data. I have made this clear going back pages, in post after post.

It is MY question, and I note that you have evaded it again. This started when I complained that the data was insufficient to learn anything meaningful. A question about whether a test that would have meaning can be conducted with this data is certainly relevant. Do we have enough data to learn anything? If the only test we can do is on one letter, is it worth anything at all?

Isn't it better to design a test that will best answer our question, and then try to seek data for that test? You are coming at the problem backwards. Or, at least, from a different direction.

I am saying - the data is insufficient for the kind of test we need to do. You say, I can do X test with this data, without regard for whether that test actually tells us anything meaningful.

So, please answer my question.

2. We would have LESS confidence by dipping into the same data set. The significance level is .05. Twenty six dips in, we would expect to get a "significant" hit by chance alone. You cannot do that. I chose the most frequently seen letter based on the control population. I saw excessively reliance on that letter. I stopped dipping in NOT because I got what I was looking for, but because I had completed the test and because further such testing would be statistically invalid. I addressed this previously and I will not repeat this again.

3. We don't. Your response to this ought to be rich.

All that your test has shown is that the number of J guesses out of a sample of 85 are higher than random. We don't know if this is a fluke, if it is related to the way the callers are chosen, if it is related to the demographics of JE's audience, or if it is due to cold reading by JE. So, we really don't know anything. It provides very weak support, at best, to the cold reading hypothesis.

Do you not agree that at least some of these concerns would be addressed by the analysis proposed by Tai Chi? That comparing the entirety of his guesses (all letters) with an adequate sample size will reveal MUCH MUCH more than your simple J comparison?

I do not consider your J comparison to be meaningful or helpful in determining whether JE is a cold reader. If this is the best analysis that you can do with this data, then I stand by my assertion that the sample size is too small for a meaningful analysis and thank you for helping to demonstrate this fact.

 

[BillHoyt]

It is MY question, and I note that you have evaded it again. This started when I complained that the data was insufficient to learn anything meaningful.

No, Thanz, it is your question NOW, and it is changed from your original claim. Let me refresh your memory. From 8/11 4:29:

Next, 78 guesses is a woefully small sample size. I don't know if we can really glean anything significant from such a small sample.

Now you've slipped in "meaningful", a weaseled substitution for "significant". I demonstrated a perfectly valid approach is significant. The null hypothesis failed.

A question about whether a test that would have meaning can be conducted with this data is certainly relevant. Do we have enough data to learn anything? If the only test we can do is on one letter, is it worth anything at all?

Yes, we have more than enough data to say JE appears to have over-selected the most frequent forename initial. The number of Js in his readings were significantly higher than would have been expected by chance.

Isn't it better to design a test that will best answer our question, and then try to seek data for that test? You are coming at the problem backwards. Or, at least, from a different direction.

You are trying to make the data irrelevant. It won't work. Would I like more data? Most certainly. That, however, is a different question. What we have can be analyzed and profitably. What I analyzed is very suggestive. Sorry it doesn't agree with you. But argue it on honest grounds, not this pap that is so insulting to the audience. Not with this weaseling.

I am saying - the data is insufficient for the kind of test we need to do. You say, I can do X test with this data, without regard for whether that test actually tells us anything meaningful.

So, please answer my question

Please shut up long enough to listen. This has gotten so tiresome, you nitwit, that everybody has disappeared from the thread. Congratulations on yet another attempt to thwart JREF!

THE DATA ARE SUFFICIENT TO DO A TEST. THEY DEMONSTRATE A SKEWING OF FORENAME INITIAL CHOICES. THEY SHOW THAT JE'S CHOICE OF THE MOST COMMON FIRST INITIAL WAS CHOSEN FAR TOO FREQUENTLY.

All that your test has shown is that the number of J guesses out of a sample of 85 are higher than random. We don't know if this is a fluke, if it is related to the way the callers are chosen, if it is related to the demographics of JE's audience, or if it is due to cold reading by JE. So, we really don't know anything. It provides very weak support, at best, to the cold reading hypothesis.

This demonstrates an astound lack of knowledge of science. Yes, it can be a fluke. It would be a fluke 1 time in 20. I have stated that before, several times.

Do you not agree that at least some of these concerns would be addressed by the analysis proposed by Tai Chi? That comparing the entirety of his guesses (all letters) with an adequate sample size will reveal MUCH MUCH more than your simple J comparison?

It might. I would welcome more testing. I would welcome more data. It is you who wish to sweep these data under the rug.

I do not consider your J comparison to be meaningful or helpful in determining whether JE is a cold reader. If this is the best analysis that you can do with this data, then I stand by my assertion that the sample size is too small for a meaningful analysis and thank you for helping to demonstrate this fact.

The test is significant. The hypothesis is valid. It addresses the essential question. And the test rejects the null hypothesis.

 

[Thanz]

Now you've slipped in "meaningful", a weaseled substitution for "significant". I demonstrated a perfectly valid approach is significant. The null hypothesis failed.

I apologize for loose language. I did not mean "significant" in the solely statistical sense. I meant "Significant" as a synonym for "meaningful"

Yes, we have more than enough data to say JE appears to have over-selected the most frequent forename initial. The number of Js in his readings were significantly higher than would have been expected by chance.

There were a grand total of 7 extra J guesses. Again, we have no idea why. One letter does not cold reading make.

Please shut up long enough to listen. This has gotten so tiresome, you nitwit, that everybody has disappeared from the thread. Congratulations on yet another attempt to thwart JREF!

I am not trying to thwart the JREF at all, sir. If you are looking for the reason others have left the thread, I would suggest you look at yourself.

THE DATA ARE SUFFICIENT TO DO A TEST. THEY DEMONSTRATE A SKEWING OF FORENAME INITIAL CHOICES. THEY SHOW THAT JE'S CHOICE OF THE MOST COMMON FIRST INITIAL WAS CHOSEN FAR TOO FREQUENTLY.

Plese be quiet long enough to listen. There is no need to shout. The only test that the data is sufficient to do does not meaningfully illuminate the question as to whether JE is cold reading. The analysis of any one initial, whether it be the most common, least common, or whatever is in the middle, is not sufficient to provide meaningful evidence for the cold reading hypothesis.

This demonstrates an astound lack of knowledge of science. Yes, it can be a fluke. It would be a fluke 1 time in 20. I have stated that before, several times.

It could also be any of the other things I mention. There is nothing inherent in your test to isolate the cold reading effect. An analysis of just one letter is simply insufficient to do this. I don't think that anyone besides you has said this.

Let me turn this around. If the sample showed, for example, 12 J guesses and was therefore not statistically significantly different from random guesses, would you conclude that JE was likely NOT cold reading?

It might. I would welcome more testing. I would welcome more data. It is you who wish to sweep these data under the rug.

I do not wish to sweep the data under the rug. I applaud the idea behind the analysis - that is, a comparison of JE guesses to the general population. I even went and found a possible source for the population name data. I just don't think that we have ENOUGH of the data yet. I am not saying we ignore what we have - I am saying we need more to add to what we have.

The test is significant. The hypothesis is valid. It addresses the essential question. And the test rejects the null hypothesis.

Not to put too fine a point on it, the test is crap. It's results would neither confirm nor deny the essential question of whether JE cold reads. It simply does not provide enough information.

It is like someone walked up to us with a spoon and asked us to build a swimming pool for tomorrow. I say that the tools are insufficient. You say Nonsense, grab the spoon and dig a small hole. You then fill it with water, dip your leg in, and move it around. You then try and convince me you are swimming.

You are not swimming. You are just kicking your leg around in a puddle of muddy water.

 

[T'ai Chi]

2. We would have LESS confidence by dipping into the same data set. The significance level is .05. Twenty six dips in, we would expect to get a "significant" hit by chance alone. You cannot do that. I chose the most frequently seen letter based on the control population. I saw excessively reliance on that letter. I stopped dipping in NOT because I got what I was looking for, but because I had completed the test and because further such testing would be statistically invalid.

This is why we need to test all the high frequency letters at the same time in one test, for example.

(I defined high frequency letter as a letter where 78*frequency of that letter > 5)

letter, frequency from Census Bureau combined name list (1990)
a, .064861
c, .072108
d, .074201
j, .133615
m, .100377
r, .08003

And if anyone has the actual observed counts of these letters from JE's readings, we could try another test.

 

[BillHoyt]

It could also be any of the other things I mention. There is nothing inherent in your test to isolate the cold reading effect.

The test rejects the null hypothesis. If you do not understand either the import or the limits of this statement I can elaborate. Or you can consult some good basic experimental design textbooks or websites.

An analysis of just one letter is simply insufficient to do this.

An analysis of this particular letter IS sufficient to reject the null hypothesis. This is all I have said. This is all I am claiming. You seem to not understand what an experimental result means.

Let me turn this around. If the sample showed, for example, 12 J guesses and was therefore not statistically significantly different from random guesses, would you conclude that JE was likely NOT cold reading?

No, I would conclude that I cannot reject the null hypothesis. If I were writing this as a scientific paper I would say "The results do not support refuting the null hypothesis."

Now, Thanz, let us turn the tables back and look at what I said way back when:

"I used the Poisson function to model the population. With an expected mean of 11.05, a count as high as (or higher than) 18 is expected to happen around 3% of the time. (Now remember we're rejecting the null hypothesis at 5%.) Based on that, I would reject the null hypothesis and say that this analysis refutes the hypothesis that JE's guesses are indistinguishable from purely random."

My language was circumspect in the ways that scientific papers are circumspect. One of the other woos whose understanding of science, like yours, rivals that of a tablespoon of peanut butter, castigated me for using "lawyer-like" language. She was, of course, ignored.

 

[Lurker]

The results were a revised total guess count of 85. I then tallied the "J"s separately. I picked the "J"s because they are the most frequent initial. According to the US census data presented earlier, "J" surnames are 13.36% of the total population. In this analysis of 85 JE name guesses, I counted 18 "J" names. I calculated the expected number of "J"s (formally, the "expectation function") as 11.05.

I used the Poisson function to model the population. With an expected mean of 11.05, a count as high as (or higher than) 18 is expected to happen around 3% of the time. (Now remember we're rejecting the null hypothesis at 5%.) Based on that, I would reject the null hypothesis and say that this analysis refutes the hypothesis that JE's guesses are indistinguishable from purely random.

I'm double-checking my counts and calculations.

I have double checked them for you Bill and you were pretty close to the actual answer.

First off, your expected mean is off as it should be 11.357. Using Poisson, I found the probability of getting 18 or more to be equal to 1-P(17) = 1-0.958476 = 0.041524 or just over 4%. Still rejected by the null hypothesis of 5%.

Lurker

 

[Lurker]

Thanz:

Finally got around to my PM. I have looked at your "A" analysis and it is correct (I ran the numbers and came out with the same).

Bill objects to you looking for you predetermined answer. Yet didn't he notice that the "J" seemed overrepresented in the JE sample and then ran the stats on them? Isn't that data mining? Why doesn't Bill recognize that he did it himself?

Funny how he dismissed your "A" analysis even though the significance is the same as his analysis. It appears only Bill's stats are valid (and I caught some minor erros in THAT too!).

Lurker

 

[Thanz]

BillHoyt -

How can you read my post and not understand that the problem lies in your hypothesis? Your hypothesis, in that it focusses solely on the one letter, is insufficient to tell us anything meaningful about the broader question of JE and cold reading.

The data is insufficient to test a hypothesis that would be meaningful to the broader question of JE and cold reading. Which is what I have been saying.

 

[BillHoyt]

I have double checked them for you Bill and you were pretty close to the actual answer.

First off, your expected mean is off as it should be 11.357. Using Poisson, I found the probability of getting 18 or more to be equal to 1-P(17) = 1-0.958476 = 0.041524 or just over 4%. Still rejected by the null hypothesis of 5%.

Lurker

Thank you for the corrections to my reported results.

 

[BillHoyt]

Thanz:

Finally got around to my PM. I have looked at your "A" analysis and it is correct (I ran the numbers and came out with the same).

Bill objects to you looking for you predetermined answer. Yet didn't he notice that the "J" seemed overrepresented in the JE sample and then ran the stats on them? Isn't that data mining? Why doesn't Bill recognize that he did it himself?

Funny how he dismissed your "A" analysis even though the significance is the same as his analysis. It appears only Bill's stats are valid (and I caught some minor erros in THAT too!).

Lurker

Lurker,

That would be because I did not do what you said. I chose "J" because it was the most frequent letter in the Census data. I have said this several times! I then did the count and the analysis. Please read my posts and report my posts accurately.

Here is my post, once again, with a pertinent section in bold:

"I re-worked the transcripts and came up with different results and a different method. Here it is, in a nutshell:

1. I used the census data figures orginally presented, although these may need tweaking.
2. I excluded the CO show data, and concentrated solely on the available, unedited transcripts from LKL, etc.
3. I looked at JE's style and adjusted the counting procedure as follows:
o I counted all of his name guesses
o Whether he stated them as names or initials, I counted them
o I excluded impossible-to-deal-with things such as "a B softened by a vowel," and chalked that up to a "B" guess.
o I included even bizarre names such as "pepper", "salt", "brooklyn" and other nickname guesses, except that
o I only counted "Liz", "Elizabeth" type guesses as the full given name, and did not also count an "L". but
o When JE recited a littany of names, I counted each one, whether they had the same initial or differing initials (again excluding the "Liz/Elizabeth, Ronny/Ronald, and Bill/William" type guesses, where I only counted the intial of the full given name.

Sound a bit complicated? You should read the transcripts. I could not see another way to approach things fairly given that sometimes he was all over the board. My hypothesis was, that, if there is a JE mediumship process, i should honor as much of it as I could figure in making the counting rules.

The results were a revised total guess count of 85. I then tallied the "J"s separately. I picked the "J"s because they are the most frequent initial. According to the US census data presented earlier, "J" surnames are 13.36% of the total population. In this analysis of 85 JE name guesses, I counted 18 "J" names. I calculated the expected number of "J"s (formally, the "expectation function") as 11.05.

I used the Poisson function to model the population. With an expected mean of 11.05, a count as high as (or higher than) 18 is expected to happen around 3% of the time. (Now remember we're rejecting the null hypothesis at 5%.) Based on that, I would reject the null hypothesis and say that this analysis refutes the hypothesis that JE's guesses are indistinguishable from purely random."

 

[neofight]

My language was circumspect in the ways that scientific papers are circumspect. One of the other woos whose understanding of science, like yours, rivals that of a tablespoon of peanut butter, castigated me for using "lawyer-like" language. She was, of course, ignored.

Ah, yes. Ignored at the time, but eventually acknowledged. Oh, was I supposed to be insulted by that comment, Hoyt? LOL Not bloody likely, considering the arrogant boor that delivered the 'insult'.

And certainly, by "One of the other woos", you were not implying that Thanz is a *woo*, are you? Because that's how your statement reads..........neo

P.S. And Bill, Thanz is right. If anyone killed this thread, it was you, and not the two honest and likeable skeptics known as Lurker and Thanz.