Hard-nosed personnel managers swear by graphologists' usefulness in selecting employees. Some do. Most do not. Regardless, there are many reasons, other than the validity of graphology, that could account for these relatively rare endorsements . First, there is ample reason to believe that, even if they are not aware of it, graphologists use other, non- graphological clues that could highlight the better candidates. For instance, the contents of handwritten application letters are rich in useful biographical information. Although graphologists claim to ignore these leads, there is evidence to the contrary . Also, graphologists often chat up the managers who consult them to see which candidates the employers are already leaning toward. Thus the graphologist is often privy to conventional information about the applicants and, in many cases, merely reinforces the managers' intuitions. Employers are often interested, as much as anything else, in this kind of reassurance that their hunches are correct. This helps soothe the unease that surrounds the inherently error-prone practice of hiring and the high costs of a mistake. Graphologists can supply this peace at mind because they make comforting but highly inflated claims that ethical personnel experts would not and could not make . And, finally, in a corporate hierarchy, where covering one's backside is a fundamental imperative, it is also prudent to have someone like a graphologist to blame if the risky procedure of selecting an employee turns out badly.

Another unearned source of satisfaction with graphology stems from the fact that employers rarely give the scripts of all applicants to a graphologist -- that would be too expensive. The graphologist usually sees only the scripts of short-listed applicants, those already selected on the basis of superior education, work experience, supervisors' recommendations, etc. Thus it is likely that everyone in this much-reduced pool would be at least adequate for the job. Because the rejects are not given a chance to show what they could do if hired, we have no way of knowing whether they would have performed as well as or better than the applicant recommended by the graphologist. And, of course, the mere fact that a graphologist has anointed the successful candidate may affect later appraisal of his or her job performance. Much research on so-called "halo effects" shows that a recommendation from a trusted source can make average performance seem better than it is and can also make supervisors more apt to excuse less than adequate performance as a temporary aberration. The vast literature on "cognitive dissonance" shows that people who have staked their reputations or significant amounts of money on a course of action, especially if others have questioned its advisability, have strong psychological motives to interpret the outcome as favorable, even in the face of contrary evidence .

In scientific tests of the ability of graphologists to recognize job-relevant traits, it is possible to control for these spurious sources of consumer satisfaction. Klimoski contrasts the methods of scientifically-based personnel selection with those of graphologists. He conducted many studies designed and carried out with the collaboration of eminent graphologists who approved all procedures in advance. In controlled tests in the workplace, handwriting analysis has fared very poorly .

Graphologists must have noticed over the centuries that certain kinds of people write in certain ways. They might have, but they didn't. Systematically tabulating any relationships between personality and writing is the way a scientific investigator would have proceeded but, as shown above, there is overwhelming evidence that graphology has always followed the rules of divination rather than those of modern personality research . In fact, as Dean, Kelly, Saklofske, and Furnham  forcefully argue, the founders of graphology couldn't possibly have kept track of the huge number of independently varying combinations of writing and personality traits necessary to be able to extract any such patterns, had they existed in the first place. As they also point out, that is because:
Graphological effects are too small to have been reliably observed.
Graphological features are too numerous to be reliably combined.
Assessment of the match between graphology and the person suffers from too many biases to allow valid

Psychologists have shown that, without sophisticated aids, human cognitive abilities are not capable of tracking the interrelationships of that many variables simultaneously. As it turns out, modern mathematical techniques that would reveal such patterns find none, but even if they had been there, graphologists did not really go about looking for them systematically. The intuitive approach they did adopt would have been incapable of extracting any possible signals from the noise.

Research shows that when the ascriptions of a tea leaf reader, palmist, astrologer, or graphologist turn out by chance to fit, this will count far more than it should in supporting the belief that "it works" (see the "the subjective validation effect," discussed below). If someone asserts that potato- eating improves your tennis game and you find that your next five superior opponents recently ate potatoes, it hardly clinches the case, does it? But clients and graphologists alike tend to be impressed by this kind of "hit". They rarely look to see how many other people have the written sign but not the trait, and how many have the personal attribute without its graphological indicator. Graphologists rely almost exclusively on anecdotal reports and personal testimonials that lack these essential comparisons. For reasons that will become apparent later in this chapter, such affirmations are worthless as scientific evidence. Research into so-called heuristic biases shows how common judgmental shortcuts often lead us to espouse bogus ideas and commodities . When competent, impartial outsiders tally up both the hits and the misses, the seemingly impressive track record of graphology evaporates. Let us now turn to that evidence.

My task in summarizing the extensive scientific research on graphology has been made quite easy by the superb efforts of Geoffrey Dean , who has carried out an exhaustive review of the literature. In any area of scientific controversy, a single study practically never decides the issue. It is only through the patient accumulation of many experiments, replicated by different investigators with converging methodologies, that a dependable pattern will emerge. Until recently, the most common way of trying to settle disputes in contentious areas was essentially to take a "box score" -- i.e., so many studies for conclusion X and so many against -- the preponderance carrying the day. Not all empirical findings should count equally in such a tally, however. Those studies with larger sample sizes, better methodology, and less noisy data ought to carry heavier weight in the grand adjudication. Fortunately, there has emerged a way of factoring such considerations into the overall assessment, and thereby drawing more reliable conclusions from multiple studies on a given topic. It is called "meta-analysis." Dean's review of the empirical research on graphology applied this mathematical technique to assess the cumulative effect of over 200 published studies from numerous countries and in several languages .

In deciding whether graphology really works, Dean addressed questions about both its reliability and validity. In the case of reliability, we are asking about the consistency or repeatability with a given measurement technique. I.e., if an operator repeats a measurement, or different operators use it on the same object, will the results concur? The former is called "test-retest reliability," the latter, "inter-rater reliability." Imagine a rubber yardstick that gave variable results on each attempt -- how useful would such an implement be? In technical terms, we would say that it had low reliability. Reliability is an essential, but not sufficient, condition for acceptance of a measurement method.

Unless a measuring instrument is reliable, it cannot have validity which is defined as the ability of the technique, test, etc., to measure what its proponents say it measures. A mercury thermometer provides a valid measure of mean kinetic energy, for instance, but it would lack validity as a measure of gravitational pull. A thermometer is reliable in that, all things being equal, repeated observations usually produce very close to the same result. That reliability, by itself, is no guarantee of validity can be seen from the following. If I assert that counting the number of moles on your back is a good way to estimate your intelligence, I could probably get roughly the same total on successive counts (i.e., the measure has reliability), but you would be right to ask me for evidence of its validity as an index of intelligence. To satisfy you, I would need to present independent confirmation that variability in mole density in the population at large correlates well with accepted criteria of intelligence. Obviously, this it would not do, so the measure lacks validity.

With respect to graphology, reliability within and across practitioners trained by the same school has been tolerable in some, but not all, studies. I.e., sometimes when the same sample of writing was submitted twice it came back with more or less the same profile after two perusals by the same graphologist or from both tries by two different graphologists  but see Goldberg  and Dean  for examples where even this minimum requirement was not met). Since the various graphological schools often disagree, one would not expect the same result from followers of different systems. But even if graphology in the hands of well-practiced disciples of the same school gives the same answer on repeated assessments of the same script, is that sufficient reason to believe that it will be accurate when it is used to predict your degree of friendliness, honesty, creativity, or devotion to an organization?

Computing scientists constantly warn us about the GIGO problem: "Garbage In, Garbage Out." In other words, no matter how accurately a computer might follow its program, if you feed it meaningless input, it will methodically grind out equally inane results. Consistently-processed rubbish is still rubbish. Increasingly, graphologists are appealing to the unwary by advertising that they now use computers, hoping by adopting these trappings of science to acquire a patina of respectability. Computerization may increase the reliability of graphological attributions, but if the raw materials of an analysis (slants, pressures, flourishes, i-dots, etc.) are not valid indicators of personality traits, then the fact that the computer derives a similar portrait of the client on multiple tries is of little comfort. And that, in a nutshell, is the question: "Are graphological 'signs' valid indicators of their supposedly correlated personality traits or aptitudes?"

In order to answer questions about validity, one must have a criterion for the trait that is supposedly indicated by the measure. If we are evaluating a test that claims to predict superior sales ability, for instance, the criterion might be the agent's total annual sales or the number of deals closed per number of contacts. An acceptable test would have to show not only that those who ace the test tend to be high on such criteria but also that those who do poorly end up at the bottom of the sales charts . In his worldwide search for empirical evaluations of graphology, Dean  unearthed more than 200 studies that had unambiguous criteria of this sort and were acceptable with respect to sample sizes, experimental controls, statistical analyses, etc. After subjecting these studies to a meta-analysis, Dean showed that graphologists have failed unequivocally to demonstrate the validity or reliability of their art for predicting work performance, aptitudes, or personality. Graphology thus fails according to the standards a genuine psychological test must pass before it can ethically be released for use on an unsuspecting public.

Dean found that no particular school of graphology fared better than any other, belying the smug claims of Graphoanalysis that it is scientifically superior to its rivals. In fact, no graphologist of any stripe was able to show reliably better performance than untrained amateurs making guesses from the same materials. In the vast majority of studies, neither group exceeded chance expectancy.

Perusing Dean's accumulated corpus of studies, an interesting relationship emerges. The better a given study is, methodologically, and the more stringent the peer review process of the journal in which it is published, the more likely it is that the results will be unfavorable to graphology. For this reason, it is not surprising that the majority of studies that find any merit whatever in graphology are published by graphologists themselves -- in promotional pamphlets, their own proprietary journals, or the for-profit popular press. When pro-graphology pieces occasionally make it into scientific journals they are typically the organs that have the lowest rejection rates and charge the authors for the privilege of publishing.

Of course, graphologists hotly contest the foregoing conclusions, claiming that the tests that belittle their abilities are unfair and irrelevant. The fact remains, however, that, in many of the best studies, graphologists gave prior approval to the tasks they would be asked to perform and the assessment criteria; i.e., they were willing participants until the negative results became known. Often graphological societies nominated their best to represent them in these tests. In one rigorous series of studies, by Klimoski and his colleagues, the graphologists were so confident they would excel that they even funded the projects. They agreed at the outset that the assigned tasks were a fair approximation of what they do in their everyday practices. Only when the results turned out disastrously for them did the graphologists begin to quibble about the fairness of the tests, at one point even going so far as to threaten legal action to suppress publication of the results. Summarizing his own research and that of many others, Klimoski concludes, "...a manager receiving solicitations for graphological services or seeing assistance in personnel decision making would be wise to heed the American credo, 'Caveat Emptor' -- let the buyer beware."

Faced with the consistently poor showing of handwriting analysis in scientific tests, the typical response from graphologists is, "I don't need to prove anything to you. I know it works and I have hundreds of satisfied customers to prove it." Of course, this is the same rejoinder I have received from every tea leaf and tarot card reader I've debated as well. People pay good money and come back for more -- they must be receiving good value, mustn't they? Not necessarily.

If graphology's track record in large-scale, carefully controlled tests is as poor as critics say it is, how could so many intelligent, well-educated people still believe it has merit? As mentioned earlier, the power of personal experience often overshadows reams of tables and graphs when people try to make complex judgements about the world . Hope and uncertainty evoke powerful psychological processes that keep all occult and pseudoscientific character readers in business. In everyday settings, their pronouncements can seem remarkably specific and telling, even though they are not. The spurious feeling that something deeply informative has been revealed in an astrological, graphological, or psychic reading arises from a kind of cognitive slippage that has come to be known as "the Barnum effect." Its other names are the "subjective validation effect" or the "personal validation effect." The more colorful appellation recalls the famous American showman, P.T. Barnum, who advertised, "I have a little something for everyone."

As many studies have demonstrated, people invariably interpret vague, positive generalizations that are true, in some form, of nearly everyone as if they applied specifically to the particulars of their own lives . Have you ever opened a fortune cookie that didn't somehow apply to you? The fascinating thing is that we "read in" the specifics with practically no awareness that they arise from our own associative processes, rather than the character reader's insights. This is not mere gullibility. It stems, instead, from the overapplication of one of our most useful cognitive skills -- the ability to make sense out of the barrage of disconnected information we face daily. In fact, we become so good at filling in to make a reasonable scenario out of disjointed input that we sometimes make sense out of nonsense. Human nature is so complex and individual behavior so varied, there is almost always something in our background to fit a reader's pronouncement. Psychologists have learned a great deal about the social and cognitive variables that make Barnum-type generalities seem so penetrating and personally relevant .

The Barnum effect is so powerful that an informal demonstration of any personality test, fringe or orthodox, is all but useless. Our enquiring minds will automatically embellish the bare bones of such output to make it seem self-referential. Once again, this is not feeblemindedness; in fact, more intelligent people are more facile at inserting these extrapolations. For that reason, a proper test of any character reading scheme will need to control for this false sense of accuracy. Thus, instead of simply asking clients if the palm reader or astrologer has accurately portrayed them, a proper test would first have readings done for a large number of clients and then remove the names from the profiles (coding them so they could later be matched to their rightful owners). After all clients had read all of the anonymous personality sketches, each would be asked to pick the one that described him or her best. If the reader has actually included enough uniquely-pertinent material, members of the group, on average, should be able to exceed chance in choosing their own from the pile. No occult or pseudoscientific character reading method, graphology included, has successfully passed such a test.

Additional evidence that the apparent accuracy of nonscientific character readings is, like beauty, in the eye of the beholder can be found in many studies (see references above) that led people to think they were receiving a reading done specifically for them. When experimental subjects are asked to rate how well the resulting profile describes them, they overwhelmingly endorse its contents although, unbeknownst to them, they are all given the identical astrologer or graphologist's report. In one recent study, subjects read statements about other people produced by a certified Graphoanalyst and an number of "Barnum statements," intentionally written to be so vague as to be applicable to virtually everyone . The subjects rated the Graphoanalyst's descriptions of strangers as being just as good descriptors of themselves as the intentionally-vague Barnum statements. When a group is given random profiles from valid psychological tests under the same conditions, they do not rate them as good a match to themselves because legitimate diagnostic tools do produce profiles that are not equally applicable to everyone.