Massive Questionnaires for Personality Capture

Social Science Computer Review, 2003, 21 (3): 267-280.

by William Sims Bainbridge

National Science Foundation*

*The views expressed in this essay do not necessarily represent the views of the National Science Foundation or the United States.

Keywords: personality capture, Survey2000, Survey2001, questionnaire, Internet, software

ABSTRACT

Contemporary information technology facilitates the creation and administration of much longer questionnaires than traditionally was feasible, and people may be motivated to respond to them as a means of capturing significant aspects of their personalities. This can be useful in designing sociable technology - computer avatars, software agents, and robots with simulated personalities - and in creating personality archives for research or memorial purposes. This article illustrates how personality capture can be accomplished through 20,000 questionnaire items culled from responses to open-ended online questions, content analysis of existing verbal or textual material, and using words from dictionaries, encyclopedias, and thesauri. This approach enables detailed idiographic study of a single individual, based on fresh measurement items and scales derived from the ambient culture.


Massive Questionnaires for Personality Capture

Personality capture is the process of entering substantial information about a person's mental and emotional functioning into a computer or information system, in principle sufficiently detailed to permit a somewhat realistic simulation. This term draws an analogy with the widely used technique called motion capture, in which the movements of a human being are entered into a computer, usually by some kind of machine vision system, so they can be used to program realistic images of people in movies and videogames. If motion capture records the motions of a person, personality capture records the emotions, attitudes, opinions, beliefs, values, habits, perceptions and preferences of a person.

The Leiden Institute of Advanced Computer Science uses personality capture in exactly the sense intended here, but the term has not yet become firmly rooted in the lexicons of either computer science or social science. Altiris, a software company, uses the term to refer to the process of migrating a person's files and software preference settings from one computer to another. The abstract of a computer science journal article about modeling a person's interpretations of images begins: "Personalizing web search engines, a crucial issue nowadays, would obviously benefit from the system's ability to capture such an important aspect of a user's personality as visual impressions and their communication" [emphasis mine] (Bianchi-Berthouze, 2002: 43). Clearly, computer science is on the verge of adopting the term personality capture, and I suggest that social science consider doing so as well.

Some recent work connects personality capture to motion capture. For example, researchers have been developing computer vision systems that can scan a person's facial expressions into a software system that performs emotion extraction to duplicate these expressions graphically in an electronic "clone" (Thalmann et al., 1998). Several kinds of conventional software already perform limited forms of personality capture. For example, a person who wants his or her word processor to handle speech to text dictation must train the speech recognition software by reading long samples of text aloud, thereby capturing the parameters of his or her own unique voice. Recognizing that human personality can express itself in many different modalities, this essay will explain how one traditional social science technique can be adapted for personality capture, in the process transcending some traditional limitations of that technique.

A Research Program

New information technologies can stimulate fresh developments in social science. Across the other sciences, a new approach for fundamental science has been emerging, in which terabyte datasets are used to explore complex systems dynamics. The same can be done for personality research, analyzing the complex connections among the thousands upon thousands of distinguishable memories and associations that make up a single human mind. It would be premature at this point to predict what discoveries might be gained, but one possible area of accomplishment would be uniting personality psychology with cognitive neuroscience (Gazzaniga, 1995).

New needs have also arisen in recent years, notably the growing realization that we must find ways to design computer and information systems that are optimized for use by human beings. This requires the development of data resources, tools, and conceptual approaches for designing sociable technology - computer avatars, software agents, and robots that possess personalities themselves, the better to serve our own personal needs (Turkle, 2002). The very first fragmentary applications of personable computing have begun to appear, for example adaptive interfaces that employ very primitive artificial intelligence techniques to adjust to the user's habits.

Another application area is the development of digital libraries and websites that preserve vast troves of information about individuals for historical or memorialization purposes. The Library of Congress pioneered historical digital libraries in the mid-1990s by posting nearly three thousand life histories on the web from the 1930s Folklore Project of the Work Projects Administration (WPA). The Survivors of the Shoah Visual History Foundation has carried out digital video interviews with more than 50,000 informants about the experience of enduring the Nazi holocaust. Several leading computer scientists have argued that it will some day be possible to enhance a rich archive of data about an individual with artificial intelligence to achieve a kind of immortality (Kurzweil, 1999; Bell and Gray, 2001; Robinett, 2002; Bainbridge, 2000a).

The chief function of most standard personality tests is to reduce the unique complexity of the individual to measurements along a small number of dimension, often as few as five (e.g. Zuckermann, et al., 1993). Much of the psychological research of the past century has been nomothetic, seeking comprehensive ways of understanding humanity in general and testing hypotheses about general tendencies. In contrast, personality capture is more idiographic, seeking to document the distinctive characteristics of a specific individual (Pelham, 1993; Shoda, et al., 1994). A vast number of psychological scales exist, measuring a multitude of concepts (Sweetland & Keyser, 1991; Goldman et al., 1995-1997). These will be useful for personality capture, but I suggest we also need to break new ground in the sociology and anthropology of personality, and develop a great diversity of culture-based measures to chart individual characteristics.

Individuals do not exist in isolation. Rather, they internalize or react against elements of the surrounding culture, for example speaking a shared language with only slightly distinctive pronunciation and vocabulary. Thus, one way to develop new measures that would be relevant for capturing the personality of a particular individual is to harvest questionnaire items from the ambient culture surrounding that individual. This essay offers examples of three ways of doing this:

To explore these potentials and develop some of the specific technical methods that would be needed for this work, I set the goal of using all three approaches to develop 20,000 questionnaire items. To facilitate creation and initial testing of such a massive corpus of items, it was necessary to employ many of the latest technologies. The hardware included desktop, laptop, and pocket computers. Data collection software was programmed in a conventional computer language (Pascal), as web pages (HTML forms), and as specialized spreadsheets (Excel), with text and data ported back and forth among these as well as to a standard statistical analysis package (SPSS). Thousands of people provided material for questionnaires, and eight adults volunteered to serve as intensive test subjects. The questionnaires were delivered to them as web pages, as Windows-based software downloaded from a web page or delivered via e-mail, on magnetic disk and CD, and transferred by wire between a desktop and pocket computer.

The 20,000 items were assembled into 10 Windows-based software modules, listed here in Table 1, and administered to one or more test subjects. Each item consisted of a stimulus, such as a statement, to which the subject would respond. As will be explained shortly, each module presented a pair of response scales for each stimulus, so in terms of the data collected there were actually 40,000 items, two for each of the 20,000 stimuli. We will look closely at five of these modules, beginning with one that asked the research subject to think about the future.

Table 1: The 20,000 Idiographic Questionnaire Items in Ten Modules
ModuleItemsTypeChief SourcesScale 1Scale 2
Year 21002,000predictions of the futureonline questionnaire Survey2000good-badlikely-unlikely
Beliefs2,000statementsonline questionnaires, social science literatureagree-disagreeimportant-unimportant
Beliefs II2,000statementsonline questionnaire Survey2001agree-disagreeimportant-unimportant
Wisdom2,000statementsBabylon 5 TV program and novelsagree-disagreeimportant-unimportant
Emotions2,000emotional stimulionline questionnaires, website searchgood-badmuch-little*
Experience2,000events, experiencesquestionnaire of a communal religiongood-badrecently-never
Taste2,000foodsonline questionnaire Survey2000like-dislikehealthy-unhealthy
Self1,600adjectives for a personsociology classes, dictionaries, thesaurigood-badmuch-little**
Association2,000pairs of wordsdictionaries, thesauristrong-weak***important-unimportant
Action2,400verbsdictionaries, thesaurilike-dislikeactive-passive
*How much or little the stimulus would make the respondent feel the given emotion. **How much or little the respondent judges he or she possesses the given quality. ***Mental connection between the two words.

The Year 2100

In May 1997, I launched a website, called The Question Factory, to administer surveys via Internet (Bainbridge, 2000b). One of these surveys pretested questionnaire items about the future, such as: "Imagine the future and try to predict how the world will change over the next century. Think about everyday life as well as major changes in society, culture, and technology." After successful preliminary work with The Question Factory, this item was included in the pioneering web-based questionnaire, Survey2000, organized by sociologist James Witte and sponsored by the National Geographic Society (Witte, et. al., 2000; Bainbridge, 2002; Weber, et al., 2003). About half of the roughly 46,500 adults who responded to Survey2000 gave thoughtful written responses to the item about the future, producing more than ten megabytes of text.

The method of analysis has been used many times before, for example in surveys about the space program (Bainbridge, 1991; cf. 1989; 1992). I read carefully through the text, copying out phrases and sentences that seemed to identify distinct ideas about the future. This process produced a new file with just over 5,000 text extracts, which were then combined and edited them into clear statements of single ideas. Iteratively, the ideas were categorized in many groups which were then combined, until there were 20 groups with 100 items in each, listed in Table 2 with data from Subject 1. For ease in remembering, the groups have simple mnemonic names. For example, the Domestic group not only has statements about people's homes but also includes ideas about urban and rural environments and about the food people will eat at home.

Table 2: Items about the Future, 100 in Each of 20 Categories, from Survey2000
MnemonicTopic AreasGood (6-8
on 1-8 Scale)
Likely (6-8
on 1-8 Scale)
Correlation of
Good & Likely
Artart, music, literature, culture, entertainment, sports, style19290.40
Businessbusiness, commerce, the economy, wealth, inequality13350.46
Conflictconflict between groups, including non-violent competition18290.37
Domestichome life, houses, foods, urban and rural communities.29370.54
Educationstudents, schools, academics, languages, education in society14300.26
Familymarriage, families, children, reproduction, sexuality.5260.10
Governmentgovernment, politics, politicians, political systems, ideologies24260.28
Healthhealth, medicine, sickness, genetics, drugs, specific diseases40390.30
Internationalinternational relations, nations, regions of the world17290.55
Justicecrime, justice, courts, law, police, morality, punishment7190.26
Knowledgeknowledge, science, beliefs, philosophies, world views45620.08
Laborjobs, labor relations, occupations, working conditions, careers20470.08
Miscellaneousmiscellaneous aspects of technology, culture, society, life27410.51
Natureenvironment, climate, natural resources, flora, fauna23370.13
Outer spacespace exploration, space technology, and human future in the universe.7723-0.36
Populationdemography, life span, fertility, mortality, migration, cloning1433-0.14
Quality of lifelifestyles, values, social problems, general quality of life13300.23
Religionreligion, spirituality, faith, secularization, denominations23280.61
Societyrelations between individual people and social classes7160.47
Technologytransportation, communications, computer technology19330.52

The items were embedded in an administration software module. One response scale asks the subject to say how good it would be if the particular statement came true, from 1=bad to 8=good. A second scale asks how likely it is that the statement will come true, from 1=unlikely to 8=likely. Figure 1 shows the area of the computer screen where the subject enters responses. Above this area, the particular item is displayed. For example, here is the first of the Domestic items: "There will be special rooms with three dimensional projectors set aside in homes for virtual reality entertainment." The respondent first thinks about this prediction and decides how bad or good it would be if this came true over the next century. Subject 1 rated the prediction "5" by using the computer's mouse to click the "5" button in the horizontal BAD-GOOD row. Then the respondent clicked the "6" button in the vertical UNLIKELY-LIKELY column, to indicate that the prediction was somewhat likely to come true. The computer displayed both clicked buttons in a lighter color, highlighting the subject's tentative choices. At this point, the subject could change either of the choices or click the OK button in the center to register the data.

Figure 1: The Input System for the Year 2100 Software Module

Input System

All eight subjects reported here employed this "cross" shape input method, which requires three clicks for each stimulus and pair of responses. However, the software also includes a "block" input method, where a single click on a checkerboard of 64 buttons registers both responses. Either way, after one pair of responses is registered, the next stimulus will appear. The [Back] and [Skip] buttons allow the subject to move backward or forward in the list of stimuli without responding to them, primarily to allow reconsideration of responses. The [Help] button leads to a screen that explains the input method and provides a demonstration of how it works. The [Return] button exits the input mode, for example allowing the subject to save data and quit the software.

It is essential to note that Table 2 is merely a summary that communicates a superficial overview of the items and the subject's responses. Personality capture really focuses on the full, undigested dataset. For example, one can output a text file of the predictions that the subject rated in any particular way. Subject 1 said that just two predictions were both very likely (7-8 on the 1-8 likely scale) and very bad (1-2 on the bad-good scale): "Humanity will not leave the Earth in meaningful numbers, because the technology required will be beyond its grasp." "Space exploration will stall, symbolizing the failed promises of technology." Coincidentally, the subject rated just two items at the extremes of Good (8) and Likely (8): "Human consciousness will be transmitted to advanced computers." "For the first time in human history, human-computer interfaces will permit development of technologies of the soul." Clearly, this particular subject seems to be pessimistic about the space program but optimistic that personality capture really can confer a kind of immortality.

Although this essay focuses on the personality capture itself, it is necessary to think ahead to how the data could be analyzed idiographically or employed in simulations. This requires exploration of ways in which patterns could be found in the data, whether by conventional social-scientific statistical analysis or by the pattern recognition and data mining techniques that have become prominent in contemporary computer science. Table 2 lets us see the differing levels of optimism the particular subject has concerning different areas of human life and culture. Note that Subject 1 rated 77 of the 100 "outer space" items high on the Good scale (6-8 on the 1-8 scale) but only 23 of them 100 high on the Likely scale. This is one indicator of the subject's mixture of enthusiasm and pessimism about the space program.

A good way of measuring a person's optimism is to calculate the correlation between Good and Likely ratings in a given area. As the last column in Table 2 shows, Subject 1 is most optimistic about religion (r = 0.61). This does not in itself say whether the subject is religious, but reveals that the predictions about religion the subject thinks would be good tend also to be likely. One would have to look at the particular religion items the individual thought were best: "Science will become the official state religion, with scientists as high priests." "The spiritual deadness affecting prosperous societies will lead to a proliferation of strange cults and fanatic religious movements." Clearly, this subject is not religiously conventional, although optimistic in the area of religion. Subject 1 is most pessimistic about the space program, as measured by a negative correlation (-0.36) between rating the space items Good and Likely.

This categorization in Table 2 is rather artificial, and the other modules described below used very different categorizing methods, beginning with one that also derived its items from an online survey.

Beliefs II

Three of the ten modules consisted of Likert-type agree-disagree statements: Beliefs, Beliefs II, and Wisdom. The material for 1,000 of the items in Beliefs II came from a second National Geographic web-based survey, Survey2001 (Bainbridge, 2003). A battery of 20 agree-disagree items measured people's beliefs about ten different issues at the borderland of science, often called pseudoscience (Frazier, 1981), primarily for nomothetic research on the cultural territory between religion and science (Bainbridge and Stark, 1980). These items were in pairs, one phrased positively, and the other, negatively.

For example, one pair concerned astrology: "There is much truth in astrology -- the theory that the stars, the planets, and our birthdays have a lot to do with our destiny in life." "Astrologers, palm readers, tarot card readers, fortune tellers, and psychics cannot really foresee the future." Another pair concerned spiritual development techniques: "Some techniques can increase an individual's spiritual awareness and power." "Yoga, meditation, mind control, and similar methods are really of no value for achieving mental or spiritual development." After this battery of items, subsets of the respondents were given pairs of statements like these again, and asked to write comments about their topics. Following the approach described above, 1,000 items were derived from the resulting verbiage, in ten categories, listed here in Table 3. Each category consisted of a pair of items from Survey2001 followed by 98 statements that came from the respondents' comments.

Table 3: Items about Pseudoscience, 100 in Each of 10 Categories, from Survey2001
Stimulus Statement from Online SurveyPercent Who Agree (N=3,909)Responses from Subject 2
Number of Items in 100 Supporting this ItemMean Rating on 1 to 8 False-True Scale
Items SupportingItems Not Supporting
There is much truth in astrology -- the theory that the stars, the planets, and our birthdays have a lot to do with our destiny in life.14.3%383.65.7
Every person's life is shaped by three precise biological rhythms - physical, emotional, and intellectual - that begin at birth and extend unaltered until death.28.1%323.35.5
Scientifically advanced civilizations, such as Atlantis, probably existed on Earth thousands of years ago.34.8%593.86.2
Dreams sometimes foretell the future or reveal hidden truths.55.4%394.45.4
Some people really experience telepathy, communication between minds without using the traditional five senses.48.0%542.75.7
Some UFOs (Unidentified Flying Objects) are probably spaceships from other worlds.22.2%122.85.2
Some scientific instruments (e.g., e-meters, psionic machines, and aura cameras) can measure the human spirit.9.1%423.35.4
Some techniques can increase an individual's spiritual awareness and power.57.3%553.55.1
Some people can hear from or communicate mentally with someone who has died.23.4%442.75.3
Some people can move or bend objects with their mental powers, what is called telekinesis.18.1%632.75.5

Using software similar to that just described, Subject 2 was asked to rate the 1,000 statements in terms of how true or false each was, as well as how important each was. Then Subject 2 was given a lap-top computer that had the 1,000 items in a spreadsheet file. The subject was permitted to take the lap-top for a few days, and whenever convenient to look though each group of 100 items and mark all the statements that supported the first one in the group. For the astrology items, Subject 2 marked 38 (including the first one) that in some way supported the idea that astrology might be true. The remaining 62 items either contradicted belief in astrology (like the second item) or were neutral. Thus, Table 3 begins with a categorization based on the origins of the items in an online survey, then sub-categorizes in terms of the subject's individual categorization habits.

Table 3 shows one way it is possible to locate the belief of a single subject in the surrounding culture. We see the percent of 3,909 respondents to Survey2001 who agreed with each of the 10 positive agree-disagree items, compared with the false-true ratings of Subject 2. Whereas fully 48 percent of the respondents to Survey2001 apparently believe in telepathy, only 9 percent believe scientific instruments can measure the human spirit. In contrast, Subject 2 rates 54 pro-telepathy items only 2.7 on the 1-8 false-true scale, compared with 3.3 for 42 items supporting the idea that the spirit can be measured.

Emotions

The Emotions module consists of 2,000 items measuring what stimuli make the respondent have twenty feelings: Love, Fear, Joy, Sadness, Gratitude, Anger, Pleasure, Pain, Pride, Shame, Desire, Hate, Satisfaction, Frustration, Surprise, Boredom, Lust, Disgust, Excitement, and Indifference. One thousand stimuli came from a pair of questionnaires administered through The Question Factory. Each questionnaire listed ten emotions, each followed by a space in which to write, and explained: "For each of these ten emotions, we will ask you to think of something that makes you have that particular feeling. By 'things' we mean anything at all - actions, places, kinds of person, moods, physical sensations, sights, sounds, thoughts, words, memories - whatever might elicit this emotion. We will also ask you to think of what makes someone else - a person very different from you - have the same feelings."

The other thousand stimuli came from 20 searches of the World Wide Web using search engines (Google, Alta Vista, Metacrawler) to find texts describing situations that elicited each of the emotions. By this means, a large number of works of literature and online essays were located that used the words in context. Each of the stimuli in the set was written on the basis of the entire context around the quotation, although in many cases the phrase is a direct quotation. Thus, 1,000 or these items were collected by means of a web-based survey, whereas the remaining 1,000 we culled from existing expressions of the culture on the web. Table 4 shows how Subject 3 responded to these 2,000 stimuli.

Table 4: Stimuli Eliciting 20 Emotions
Category Defining Words in Ten Near Antonym PairsMean Rating of 100 Stimuli in Each of 20 Categories on 1-8 Bad-Good ScaleCorrelation between Saying 100 Stimuli are Good and They Elicit the Given Emotion
First Cat.Second Cat.First Cat.Second Cat.
LoveFear5.074.320.59-0.72
JoySadness5.093.590.79-0.56
GratitudeAnger5.343.800.60-0.34
PleasurePain4.783.830.66-0.53
PrideShame5.533.880.75-0.50
DesireHate4.564.130.44-0.66
SatisfactionFrustration5.003.740.73-0.53
SurpriseBoredom4.514.62-0.03-0.26
LustDisgust4.613.900.55-0.60
ExcitementIndifference4.304.270.05-0.01

For example, one of the stimuli in the Fear category was "not being able to breathe." Subject 3, who was asthmatic as a child, said that this would be extremely bad (1 on a 1-8 bad-good scale) and would very strongly tend to elicit the given emotion of Fear (8 on a 1-8 scale of how much or little the stimulus would make the respondent feel the given emotion). The 20 emotions were naturally arranged in ten pairs of opposites, as shown in Table 4, and the subject generally prefers the stimuli in the first category of each pair, with the exception of Surprise and Excitement, on which the subject appears to be neutral or ambivalent. The last two columns of the table show the correlations between the two scales, within each of the 20 categories, again showing a connection between the stimuli in a category and goodness or badness, with the notable exceptions of Surprise, Excitement, and Indifference.

Wisdom

The next module, Wisdom, shows how questionnaire items can be derived entirely from a particular exemplar of the ambient culture. Material for it came from content analysis of 120 hours of the science-fiction television program, Babylon 5 (B5), guidebooks to its complex mythos, and B5 fiction (up to but not including the Technomage trilogy of novels). Traditionally, social scientists have often culled potential questionnaire items from the writings of a great thinker, as Richard Christie and Florence Geis (1970) did when they created the influential Mach Scale from the writings of political philosopher Nicolo Machiavelli. I have merely done the same thing with a contemporary source that addressed some of the same issues of power in human relationships as did Machiavelli.

Created by J. Michael Straczynski, B5 draws deeply from the traditions of science fiction literature, thereby reflecting a major genre of popular culture (Bassom, 1997). B5 is a city in space, where humans and aliens meet, unaware that two vast powers are battling for dominance of the universe, on a level of technical sophistication far beyond human understanding. On one side are the Vorlons, who value order and ask "Who are you?" On the other side are the Shadows, who value chaos and ask, "What do you want?" The challenge of the TV series concerns whether humans can unite the other aliens against both of these forces, and establish a cosmopolitan culture valuing liberty and diversity.

The items are statements derived from sentences spoken in an episode of the program, limited to 10 per hour of television, or published in a book, limited to 20 from each source. In many cases, the item is a verbatim quotation, but in other cases the original was edited minimally to transform it into a statement about people or life in general. Table 5 shows how Subject 4 responded to the 2,000 items, categorized by the B5 character who spoke the original words, arranged in ascending order of mean "true" rating. Subject 4 did not know the identities of the characters while rating their statements, and they were administered in random order.

Table 5: Wisdom Module Popular Culture Items Drawn from Babylon 5
CharacterNumber of
Statements
Mean Rating on
1-8 True Scale
Mean Rating on
1-8 Important Scale
Lennier324.224.69
Delenn1474.445.08
Jeffrey Sinclair864.494.85
Lyta Alexander344.505.32
Byron304.605.37
G'Kar1314.605.08
Alfred Bester724.614.69
Vir Cotto374.624.76
Michael Garibaldi1004.624.88
minor characters6364.635.04
Marcus364.644.97
Londo Mollari1474.645.01
Susan Ivanova724.824.93
Dr. Stephen Franklin684.855.12
John Sheridan1904.905.05
anonymous1595.015.33
Kosh235.095.48

The first two characters in the list, Delenn and Lennier, are aliens from a haughty species that once tried to exterminate humans, and Subject 4 tends to rate their statements lowest on the "true" scale. One of the statements spoken by the character Delenn well communicates the central principle of their caste-ridden society: "Understanding is not required, only obedience." At the opposite end of the list, Subject 4 gives the highest "true" rating to statements by Kosh, the enigmatic Vorlon. Kosh is famous throughout the science fiction subculture for making inscrutable pronouncements hinting at profound wisdom, such as: "The avalanche has already started; it is too late for the pebbles to vote."

Subject 4 also give somewhat high "true" ratings to anonymous statements and those from the commander of Babylon 5, John Sheridan. The "anonymous" category consists of statements from television characters who are so minor they lack names and from authors writing about B5 without taking the voices of characters, so in a sense these statements lack personality. Sheridan, the central character of the series, is a Christ-like figure who dies but is reborn. His statements express both optimism and stoicism: "If you're falling off a cliff, you might as well try to fly." "The way to deal with pain is to turn it into something positive."

Self

Having seen several examples of how items could be collected by means of web-based questionnaires, or extracted from published exemplars of the ambient culture, we must now conclude with an example that focuses on the language itself. This is also the only example here of comparing across individuals, which can be a valid part of understanding the individual in a social context.

The Self software module consists of 1,600 adjectives that could describe a person. They came from a line of research that began a decade ago with a project exploring the "semantic differential." This is a commonly-used kind of questionnaire scale, developed back in the 1950s, that asks the respondent to judge something in terms of several pairs of opposite adjectives (Osgood, et al., 1957; Bainbridge, 1994). The items were developed with the help of 36 students in classes on the Sociology of Organizations and on Small Group Processes. Students were asked to think about the qualities they would like to see in people they were working with. Each student wrote down as many as twenty of these terms, then wrote down the antonym of each. Four standard thesauri were then used to check these antonyms and to generate pairs of opposites that described personal qualities relevant outside the context of work, without reusing any of the words or employing any obscure terms. Fully 800 pairs of antonym adjectives were incorporated in the Self software, but each item was just a single word, and the software unobtrusively kept track of antonym linkages that connected the 1,600 words into pairs.

Table 6 summarizes responses from four subjects, numbers 5 through 8 in this study. A respondent judged how bad or good it was for a person to have the quality described by each word, and how little or much he or she actually possessed the quality. Because we have so many data points for each individual, it is possible to correlate people with each other, to see how similar or different their ratings are. For example, Subject 5 and Subject 6 correlate 0.67 with each other of their bad-good ratings, and 0.52 on the little-much ratings of the 1,600 qualities. The averages for the 6 coefficients linking the 4 subjects are 0.67 again (ranging from 0.61 to 0.74) for bad-good and 0.47 (ranging from 0.37 to 0.56) for little-much. The difference between 0.67 and 0.47 is actually quite interesting. Apparently the four subjects share cultural assumptions about how good or bad the qualities are, but they have different self images, each stressing a somewhat different collection of personal qualities.

Table 6: Adjectives Describing a Person's Character, Categorized by Subject 5
Most "Good" Items in Subject 5's CategoryN of ItemsSelf-Esteem (correlation GOOD and MUCH)*
Subject 5Subject 6Subject 7Subject 8
1. alert, alive, motivated720.570.770.950.83
2. clear, dedicated, focused1100.630.780.940.89
3. unique, credible, exceptional1020.710.780.890.82
4. healthy, complete, durable820.120.580.920.58
5. enlightened, innovative, aware900.870.920.950.95
6. future, real, instinctive880.590.700.930.71
7. courageous, hopeful, inquisitive1500.130.440.870.77
8. constructive, inspiring, true2840.610.820.910.85
9. spiritual, affectionate, loveable114-0.210.370.870.75
10. resourceful, best, energetic1780.390.800.810.83
11. able, capable, honest1960.870.910.900.93
12. celestial, cosmic, eternal560.380.800.870.87
13. good-natured, initiating, approachable780.330.730.950.90
*Self-esteem is defined as saying qualities are "good" and having them "much."

The 13 categories of qualities that define the row of Table 6 were developed by Subject 5. We gave the subject a pocket computer loaded with a spreadsheet listing the 800 pairs and asked the subject to categorize them in about a dozen groups, using any principles the subject wished. Over a period of several days, the subject carried the pocket computer and from time to time worked on the categorization task, which in itself was yet another way of capturing aspects of the subject's personality. The labels of the 13 categories are the three words that garnered the highest total "Good" score from all four subjects.

For each subject, Table 6 gives the correlations between the Good and Much scales in each of the categories, which is a plausible measure of self-esteem. For example, category 7 includes qualities like courageous and hopeful (and their antonyms, fearful and despairing). Subject 5 placed 75 pairs of items in this category. Among the ratings given these 150 items by Subject 5, the correlation between the Good and Much scales is only 0.13, which means essentially no correlation between rating a quality good and feeling that one possesses it. This is much lower than the self-esteem coefficients for the three other subjects: 0.44, 0.77, 0.87.

However, it may not be appropriate to say that Subject 5 has abnormally low self-esteem, because we do not have population norms for the coefficients. In addition, it is important to remember that self-esteem can be abnormally high, as well as abnormally low. This can occur, for example, during a clinically manic episode, as was in fact the case for Subject 7. More importantly, we can compare the self-esteem coefficients within the data for a given respondent. Subject 5's self-esteem is lowest for qualities like "spiritual, affectionate, loveable" (-0.21), and highest for qualities like "able, capable, honest" (0.87). Indeed, the tables in this paper are only the most superficial sketch of the patterns that can be seen looking closely at extremely rich data concerning one individual.

Conclusion

Tens of millions of people work and play daily on computers, and a few million already carry laptops, pocket computers, or PDAs. They could, if they wished, respond to very long questionnaires by doing a few items at a time whenever they had a few spare moments. For example, this paragraph was written with a pocket computer while riding on a subway. Archiving one's own personality could become a pleasurable hobby in which a few people invest hundreds of hours over a period of years.

Obviously, this vision has little to do with the traditional use of questionnaires as tools for surveying random samples of the population. But the new information technology may enable a very wide of new social science applications and research methods, that enrich both science and human life.

In a sense, this article has turned questionnaire methodology upside down. Instead of having one person write a questionnaire for a thousand people to answer, thousands of people created questionnaires for one individual respondent. Instead of calculating the correlation between two items across a thousand respondents, we calculated the correlation between two responses across two thousand items within one person.

Personality capture may be carried out in a variety of ways for a variety of purposes. Thus a great number and diversity of scientific studies will be needed to determine which applications will be valuable and how to create them. Massive questionnaires created from the ambient culture are one viable approach for idiographic social-science study of an individual personality.

References

Bainbridge, W. S. (1989). Survey Research: A Computer-assisted Introduction. Belmont, California: Wadsworth.

Bainbridge, W. S. (1991). Goals in Space. Albany, New York: SUNY Press.

Bainbridge, W. S. (1992). Social Research Methods and Statistics. Belmont, California: Wadsworth.

Bainbridge, W. S. (1994). Semantic differential. In R. E. Asher and J. M. Y. Simpson (Eds.), The Encyclopedia of Language and Linguistics (pp. 3800-3801). Oxford: Pergamon.

Bainbridge, W. S. (2000a). New technologies for the social sciences. In M. Renaud (Ed.), Social Sciences for a Digital World (pp. 111-126). Paris: Organisation for Economic Co-Operation and Development.

Bainbridge, W. S. (2000b). Religious ethnography on the World Wide Web. In J. K. Hadden and D. E. Cowan (Eds.), Religion on the Internet (pp. 55-80). New York: Elsevier.

Bainbridge, W. S. (2002). Validity of Web-based surveys. In O. V. Burton (Ed.), Computing in the Social Sciences and Humanities (51-66). Urbana: University of Illinois Press.

Bainbridge, W. S. (2003). Public attitudes toward nanotechnology, Journal of Nanoparticle Research, in press.

Bainbridge, W., & Stark, R.. (1980). Client and audience cults in America, Sociological Analysis, 41:199-214.

Bassom, D. (1997). Creating Babylon 5. New York: Ballantine.

Bell, G., & Gray, J. (2001). Digital immortality. Communications of the Association for Computing Machinery, 44(3):29-31.

Bianchi-Berthouze, N. (2002). Mining multimedia subjective feedback. Journal of Intelligent Information Systems, 19(1)43-59.

Christie, R., & Geis, F. L.. (1970). Studies in Machiavellianism. New York: Academic Press.

Frazier, K. (Ed.). (1981). Paranormal Borderlands of Science. Buffalo, New York: Prometheus.

Gazzaniga, M. S. (Ed.). (1995). The Cognitive Neurosciences. Cambridge, Massachusetts: MIT Press.

Goldman, B. A., et al. (1995-1997). Directory of Unpublished Experimental Mental Measures. Washington, DC: American Psychological Association, 7 volumes.

Kurzweil, R. (1999). The Age of Spiritual Machines. New York: Penguin.

Osgood, C. E., Suci, G. J., & and Tannenbaum, P. H. (1957). The Measurement of Meaning. Urbana: University of Illinois Press.

Pelham, B. W. (1993). The idiographic nature of human personality: Examples of the idiographic self-concept," Journal of Personality and Social Psychology, 64(4):665-677.

Robinett, W. (2002). The consequences of fully understanding the brain. In M. C. Roco & W. S. Bainbridge (Eds.), Converging Technologies for Improving Human Performance (pp. 148-151). Washington, DC: National Science Foundation.

Shoda, Y., Mischel, W, & Wright, J. C. (1994). Intraindividual stability in the organization and patterning of behavior: Incorporating psychological situations into the idiographic analysis of personality. Journal of Personality and Social Psychology, 67(4):674-687.

Sweetland, R. C., & Keyser, D. J. (Eds.). (1991). Tests: A Comprehensive Reference for Assessments in Psychology, Education, and Business. Austin, Texas: Pro-Ed.

Thalmann, N. M., Kalra, P., & Escher, M. (1998). Face to virtual face. Proceedings of the IEEE, 86(5):870-883.

Turkle, S. (2002). Sociable tTechnologies: Human performance when the computer is not a tool but a companion." In M. C. Roco & W. S. Bainbridge (Eds.), Converging Technologies for Improving Human Performance (pp. 148-151). Washington, DC: National Science Foundation.

Weber, L. M., Loumakis, A., & Bergman, J. (2003). Who participates and why? Social Science Computer Review, 21(1), 26-42.

Witte, J. C., Amoroso, L. M., & Howard, P. E. N. (2000). Method and representation in Internet-based survey tools: Mobility, community, and cultural identity in Survey2000. Social Science Computer Review, vol. 18, no. 2, pp. 179-195.

Zuckermann, M., Kuhlman, D. M., Joireman, J., Teta, P., & Kraft, M. (1993). "A comparison of three structural models for personality," Journal of Personality and Social Psychology, 65, (4): 757-768.

URLs:

Altiris Corporation: www.altiris.com

American Memory, WPA Federal Writers' Project: http://memory.loc.gov/ammem/wpaintro/wpahome.html

Leiden Institute of Advance Computer Science, Digital Life Technologies: www.liacs.nl/research/DLT/

Survivors of the Shoah Visual History Foundation: http://www.vhf.org/