Validity of Graphology

Cecil's Columns/Staff Reports
21

Well that’s a mighty scientific and objective setup now, ain’t it?

The shaman gets to pick the people, and the samples, and set all the conditions. Hey, why bother with handwriting? Anybody with zero “talent” for “graphology” can go through a few dozen samples of a few people’s writings and match them up by writing style (including vocabulary, common errors of spelling, punctuation or grammar, and commonly chosen voice, for example). No handwriting is needed, and no mystical “graphology” folderol. Might as well just have somebody pick ten typewritten posts from the SDMB at random, delete the names, and then submit them along with a separate list of the names of the ten posters, and say “Okay, match 'em up!” Then, when the claimant takes a little time to go through the board reading previous posts by those ten posters and matches them up, wowee, the “science” really truly ackshully works! Gawsh! (Of course an “acceptable” success rate would have to be negotiated - say 70%, right? I mean, any parlor trick that scores a 70% “success” has to be for real, doesn’t it?)

But what’s with the concern about “false samples”? Any experiment has to have controls, and any science worthy of the name can detect fakery without much sweat. Why not just throw in some samples not from one of the pre-named list of submitters, and let the mystical “graphological science” weed them out with a result of “No, this one doesn’t belong to anybody on the list.” Why would that be any more difficult?

As for measuring “personality traits”: you can’t measure what you can’t define, and taking a poll of purely subjective opinions to compare against another purely subjective opinion doesn’t translate into objective measurement.

I don’t think anybody needs to worry about trying to make this proposed flubdubbery “double-blind”. That’s a term usually applied to scientific experiments, and therefore it isn’t applicable here.

Maybe next we’ll all write down the names of SDMB posters, and match them up in pairs and start crossing out matching letters, to see what letters are left over. I bet then we can get a good objective scientific notion of who likes whom. Gee, but how an “experiment” like that would be kept double-blind I don’t know. . . .

22

hmmm… I’d really like this to be as objective as possible, and voting by random members is bad. Even assessment of personalities by an individual tester or testing committee throws in a whole bunch of subjectivity.

For what it’s worth, if I match personalities to handwriting samples, I will also give the explanation of why I matched each to each, explaining what traits seemed to match which.

zut wrote

Well, here’s how the theory goes: you can’t know for certain just from looking at someone’s handwriting if they’re male or female. As it turns out, males (on average) do have a lot of personality traits in common, and so do females, and so you can have a good guess, but you can often be wrong. And in fact, one of the things you ask for before you analyze something is the subject’s sex, so you can take that into account.
As far as athletic interest goes, that is a personality trait and comes out in hand-writing, and in fact that’s one of the things that I’d ask via email of each of the subjects.
Left-handed people do tend to write a bit differently then right-handed, and some analyzers consider this but I personally don’t, so it’s not something I’d ask about.
To the best of my knowledge, height doesn’t impact handwriting.

23

oh, sorry, I simulposted with some noise. I was of course responding to zut. zut’s point (that this doesn’t exactly measure handwriting as a predictor of personality) is true. Is it a close enough test? You’ll have to judge. It’s certainly objective, and it’s tough to do anything ojectively when measuring personality.

24

Why don’t the subjects just submit our IPIP-NEO scores, and have Bill H. match those with the handwriting samples? That will eliminate the question of using non-personality information to come to a conclusion, and also eliminate the problem of losing information via email or whatever. We wouldn’t even have to tell him who’s participating!

25

You know, don’t you- that the test doesn’t prove much? Let us say that indeed you can match 8of 10 handwriting samples to dudes you know- so? Can you tell me whether or not- after meeting them, AND reading their handwriting- whether or not they are “honest” or “a good salesman”?

I can tell by meeting someone quite a bit from their dress. However, unless they are wearing gang colors, Biker threads or prison clothes- it tells me nothing about their “honesty”- and even the “gang colors” is at best a “tends to”.

I don’t need any handwriting samples. Give me a dozen of a posters most revealing posts- and a 15 minute interview- and I’ll spot 10 of 10. But- I 'll be damned if I can tell you whether or not any of these posters are more likely to steal from an employer.

So-

Bill- “extroverted” “passionate” and “quick thinking”- even if you COULD reliably judge those traits- are not traits you could use to hire someone- at least not in most jobs. And at best- you’d get a “tends to” or a “maybe”. But I can get a much better read on those 3 traits in a 30 minute interview- but again, I’ll have no idea whether or not the potential employee is “honest” or a “good security risk”.

They aren’t using graphology to find out if some dude is “outgoing”- they don’t need a special test for that. No- they are using graphology to find of if that dude will steal from them- and the test is completely worthless for this- and so are polygraphs.

So Bill- riddle me this. Pick any ten posters, with more than a thousand posts, and whom you do not know. Read all the posts you want. Then tell me which of them are more likely to steal from their employer, or are a “poor security risk”. No fair taking someone who admitted such a thing here, however. THEN PROVE IT.

26

DrDeth: I think you may be getting your threads confused. We’re not talking about hiring practices here. We’re talking about Cecil’s claim that “More than 200 objective scientific studies have demonstrated that graphology is worthless as a predictor of personality.”

27

Now, now, DrDeth, it’s not fair to put any of those burdens on Bill H. He’s not claiming to be able to determine honesty, salesmanship, or any other employable trait. We should only test him on what he claims to be able to do. Of course, if he does pass the test, then that might indicate that further tests are in order, to determine whether such traits can be determined from handwriting. But we’ll cross that bridge when and if we come to it.

As for sample size: If we can get this thread Threadspotted, we’ll get a lot more eyes reading, and a lot more potential volunteers. Ideally, of course, we should have as many people in our sample as possible.

As for myself as tester: I have no objection, but we really can’t have the same person as both tester and sample. The various samples will all have to be labelled in some manner, for comparison, prior to being sent to Bill H., and if I write a number, for instance, on each slip of paper, then he’ll have a sample of handwriting known to be mine (the numbers) to which he can compare the specimen I turn in.

As for control: It would be possible to have each person submit two samples, of the same text, and then randomly send one of each pair to Bill H. and one to someone who claims to have no graphological ability, such as C K Dexter Haven. Both would then evaluate the samples, and give their best guesses.

As for scoring: All test subjects (Bill H. and any controls) would return some permutation of the correct list of names in the sample. A test result could then be scored based on the minimum number of transpositions to obtain the tested permutation from the correct permutation. I (presumably backed up by the better statisticians on the board) could determine, given the number of participants, how many transpositions could be expected by random chance, to perhaps a 95% confidence level. If Bill H. achieves a better result than chance, to that confidence level, then he is considered to have passed the test.

As for the writing sample to be used: Bill H., how long of a sample do you need? A single sentence, a paragraph? The sample used should probably contain a minimum of punctuation, and fairly simple words, so that punctuation use and characteristic misspellings will not provide any information.

28

I also think the control evaluator should put a lot of effort into finding out stuff like age, sex, and location of primary education.

29

Chronos, I’m not clear about the value of having a non-expert (not to imply I’m an expert) have a go at this and compare my results with his. If anything, that person getting a better then average score could indicate that even a layman can get a feel for say “angry handwriting” and attribute it to someone they know fits the bill. I’m not opposed to going against someone else, just debating about whether they are an actual sugar-pill.

Going against random chance seems to be more reasonable. After all, the claim I’m contradicting is that better then 50% is impossible.

I prefer a full page. I prefer text that is not artificially minimized or improperly punctuated. The ideal is a letter with an addressee and a signature, but of course signatures wouldn’t be appropriate here.

Also, I need to re-iterate my concerns that a) I’ll get a number of samples that are too similar for me to differentiate, and b) I won’t be able to discern via email and post-viewing certain personality traits of subjects. I’m particularly nervous that the typical reader of GQ is extremely similar in key personality traits as compared to an average writer of the world. I need the ability to discard subjects and writings. This can be done leaving the remaining test pool still completely blind. I never claimed this was akin to DNA matching technology.

Also, in the off chance this becomes very large, I don’t have the actual cycles to do 100 analyses. I’m not sure what a top limit is, but I’m thinking a max of 30 people who enter of which I accept and match 10-20.

30

OK, I think we need one person to set forth the rules. Clearly, all participants (and I am glad to be one) need to copy the same paragraph. I suggest the first paragraph of Cecil’s response to the handwriting analysis question: “At first this question might seem like a great opportunity to lay out the difference between science and pseudoscience. …”

We write it out, nothing else on the paper. We should probably try to use the same type of paper and pen/pencil (for instance, writing it on your personal stationary would be a clue) and mail it to Chronos. Chronos will put an identifying number on each one, and keep the code secret. the code should be randomly assigned, like a three digit number.

Chronos will then mail all the handwriting samples to Bill H, along with an alphabetical list of the people who submitted samples. (Query: should there not be a few additional names on the list of people who did not submit samples? Otherwise, it’s multiple choice, bound to get a few right by random guessing.)

Bill H is then to try to identify who wrote each sample. He posts his list here. Then Chronos posts the correct list.

Interesting experiment, and I’m in.

BTW, I thought it was “impractical” to spend a day with each person (as the original suggestion) to “get to know them.” The amount of time – for Bill and everyone else – would be prohibitive.

31

C K Dexter Haven wrote

Copied text is not ideal, but that’s fine. I prefer a page of text.

I prefer that the paper and pen/pencil be of the writers choosing and be what they normally use. Obviously personalized stationary is out. The theory says people choose pens in the same way they choose writing styles, and it’s one of the things that are analyzed.

I’d prefer there be no fakes. In fact, I’ll be very disappointed if it turns out someone does give me something they scribbled using their foot or some nonsense.

As far as “bound to get a few right,” there are plenty of statistical wizards here that can tell what random luck should result in, and how much better (or worse I suppose) I did.

32

If we don’t use a copied text, then you’ll get clues from the vocabulary, grammar, and writing style that would be independent of the handwriting. Remember, this is supposed to be about the handwriting, not about the literary style.

I’ll give you the pens and pencils and paper, I suppose, just because it will be too difficult (although we could specify a number 2 pencil, and standard paper that goes in your printer) … although I think those may provide clues independent of handwriting.

33

Perhaps I wasn’t clear. Though I don’t prefer copied text, I’m fine to accept it. I need a page of text though.

34

See, I suspected this would come up, and this is exactly what I had in mind when I asked about a control. That person getting a better than average score implies no such thing. The claim we’re supposed to be testing is that handwriting indicates personality, not handwriting indicated who wrote it. But if you control with a random matching, it’s the latter that you’re testing, not the former.

35

Achernar, what other features do you propose I may be using intentionally or inadvertantly to match them other then handwriting analysis?

If others get a higher then random score, what other features do you propose they used?

I propose that if someone gets a higher then average result, it’s because they intentionally or unintentionally matched traits of the persons handwriting to what they know about them. Which is the exact thing we’re trying to quantify.

36

I listed age, sex, and location of primary education before. There could be others, but I think these three are enough to cause concern.

37

I use a fountain pen with ink in a custom blended color and unruled paper (or else one with lines faint enough for me to ignore). How much would that info by itself tell you about my personality?
(once saved from ugly legal mess by handwriting analysis expert)

38
39

Well, now it tells me that you’ve sorta disqualified yourself.

40

I volunteer. I am happy to answer any of your questions via email or telephone. I will pay for the phone call.

Haj