Dream Content Analysis

Introduction

This paper tries to show, in a friendly way, why any theorist doing systematic empirical work to test his or her theory would do well to use the Hall/Van de Castle system of content analysis, and to take established findings with this system seriously. This is because the system is proven when it comes to reliability and validity, and is theory-neutral as well. Conversely, there is much evidence that findings with newly developed scales must be treated cautiously because it is very difficult to construct adequate scales or content categories to study dreams. Unfortunately, most theorists go right on creating new scales that never receive the same amount of time and effort that the theorists put into developing ideas or writing up results.

In this particular case, the theorist, Ernest Hartmann (1998), has put forth a new theory of dreaming whose main tenet is that dreams, and especially the vivid or striking aspects of dreams, provide a context for expressing and dealing with emotional concerns. The late-night movies in the brain called a dream present “an explanatory metaphor for the dreamer’s emotional state of mind” (Hartmann, 1998, p. 4). He stresses that these explanatory metaphors “contextualize” emotions. Thus, metaphors and emotions are the key issues to be studied to test his theory.

Hartmann further claims that dreams can help people deal with traumatic emotional experiences by incorporating these experiences into dreams and giving them new connections. In effect, he is saying that dreams help resolve emotional problems. His evidence for this point is that dream content allegedly changes in a positive direction as the traumatic events are assimilated and overcome.

Hartmann’s approach is one that I share in several ways. We agree that the nightmares of posttraumatic stress disorder and other recurrent phenomena in dream life are a good starting point for developing a better understanding of dreams. Second, we both think that dreams can be studied systematically outside either the psychotherapy relationship or the sleep laboratory. Third, we share an emphasis on studying lengthy dream journals from a wide variety of individuals to test our hypotheses. Fourth, we are both intrigued by the idea that at least some dreaming is a figurative/metaphoric form of thinking during sleep.

However, the evidence for Hartmann’s theory is not convincing to me. First, I do not think he has a reliable method for studying emotions in dreams. He missed a good bet by not using the Hall/Van de Castle categories for emotions. Secondly, he has not provided convincing evidence that the dreams of the people he studied did change over time. Showing such change is extremely important because the evidence from many studies with the Hall/Van de Castle system is that the dreams of adults do not change much at all over months, years, or decades; this consistency includes repeated dream themes that seem to concern upsetting emotional experiences from the past (Domhoff, 1993; Domhoff, 1996; Domhoff & Schneider, 1998). Thirdly, as much as I like the idea that figurative thinking may be playing a role in dreaming, I believe that much research is needed to demonstrate the point. It cannot be taken as a starting point for theorizing, but must be treated as a hypothesis worth testing.

Fourth, there is little or no evidence that dreaming has any adaptive function, including the one ascribed to it by Hartmann. Most people recall less than 1% of their dreams, and even the best recallers are only remembering a few percent of their dreams, so dreams cannot be very important as a source of information. The few that are remembered rarely contain even a hint of a solution to a problem, contrary to the long list of questionable anecdotes relating dreams to artistic achievements. Moreover, there is no evidence that people who recall dreams, or make use of their dreams in some way, are any healthier or happier than those who do not recall dreams.

Despite the lack of evidence for recalled dreams having any function, it still might be argued that the process of dreaming has an adaptive purpose. However, there is no evidence from the sleep laboratory that the process of dreaming has any function in any evolutionary sense (Antrobus, 1993; Flanagan, 2000; Foulkes, 1993; Foulkes, 1999). There is even evidence that there are people who do not dream but suffer no adverse effects–pre-school children (Foulkes, 1999), adults with poor visuospatial skills (Butler & Watson, 1985), people who have suffered various kinds of head injuries (Solms, 1997), and schizophrenics who have been subjected to psychosurgery (Frank, 1946; Frank, 1950).

In the remainder of this paper I deal with two issues: how to study metaphoric thinking in dreams and how the Hall/Van de Castle could be used to test some of Hartmann’s hypotheses. I do not deal any further with the weighty issue of what, if any, adaptive function dreaming may have, instead referring readers to the citations in the previous paragraph.

Before I begin, I want to stress that this critique could be applied to virtually every modern-day dream theorist. Generally speaking, they have interesting ideas, but use weak and unreliable methods for testing their hypotheses (Domhoff, 1999; Domhoff, 2000; Domhoff & Schneider, 1998).

Studying Metaphoric Thinking in Dreams

The content-analysis tradition within dream research began with the thematic study of individual series of 10-25 dream reports provided by literally hundreds of college students (Hall, 1947; Hall, 1953) These studies of dream series included a search for metaphoric elements. This work provides several rules to thumb for identifying metaphoric elements within dreams.

First, the repetition of a theme or element in a dream series suggests that it might be metaphoric. To take an example from Hall’s later work, 15% or more of the dreams in a series of 904 dreams over a 50-year period contained the theme of not receiving enough to eat at meals with siblings (Domhoff, 1993; Smith & Hall, 1964). (Indeed, this was the theme of the last dream she sent Hall, about a week before she died unexpectedly in her mid-seventies, which might have seemed to be a prophetic dream in its barrenness if the whole series had not been available.) Comparing this subset of “deprivation/hunger” dreams with a random sample of dreams from the same series on characters, social interactions, and emotions would be one good way to approach the study of metaphor.

Second, impossible or unusual elements may well be metaphoric. Flying under one’s own power, for example, is clearly impossible, but we do say we are “flying” or “on cloud nine” or “high as a kite” when we are happy (Gibbs, 1994; Lakoff, 1997; Lakoff & Johnson, 1999). Similarly, we are not likely to appear in our pajamas or underwear in public, but 40-50% of the college students in two different samples reported this had happened to them at least once in a dream; perhaps such dreams relate to the conceptual metaphor “Embarrassment is Exposure,” which is expressed by phrases such as “caught with your pants down” (Domhoff, 1996; Holland & Kipnis, 1994).

Third, it is a useful to consider the possibility that any animal besides household pets, or any household pets acting in a highly unusual fashion, might have a metaphoric meaning. This hypothesis is based on the fact that animals are widely used in waking metaphors to characterize people or their attributes, as in he/she is a chicken, dog, fox, snake, worm, bug ,and/or cat. Moreover, there is reason to believe that children learn these metaphors of personification relatively early through the many nursery rhymes and animal stories that are read to them, where the animals enact the gender stereotypes of the Euro-American world (MacKay, 1986; MacKay & Konishi, 1980).

Once explicit criteria for identifying explanatory metaphors are established, the Hall and Van de Castle (1966) coding system can be used to make a wide variety of comparisons between groups of dreams. I advocate building on the Hall/Van de Castle system instead of developing a new rating scale for a number of strictly methodological reasons, all of which I will demonstrate with examples from Hartmann’s book.

The Advantages of Using the Hall/Van de Castle System

First, there is much evidence that the Hall/Van de Castle categories can be coded very reliably, whereas there are always big question marks about the reliability of any new rating scale unless it has been tested with great care. These questions loom especially large when it comes to the use of the scale by someone other than its originators. Hartmann, Rosen, and Rand (1998) provided a good recent demonstration of the reliability problems with new scales when they reported they could not achieve adequate reliability with three new scales of their own concerning “condensation,” “thick boundaries in dreams,” and “thin boundaries in dreams.” They also showed how difficult it is for a new research team to use a scale developed by another investigator when they conclude they could not obtain adequate reliability with a scale for “dreamlikeness” developed in Foulkes’s laboratory many years ago (Foulkes & Shepherd, 1971). Since many investigators may not be as frank and honest as Hartmann in reporting negative results, published examples such as these are probably only the tip of the iceberg on this problem.

In terms of studying “contextualizing emotions,” this point about reliability is first demonstrated by comparing Hartmann’s new rating scales with the Hall/Van de Castle scale for coding emotions. Hartmann begins with a list of 18 emotions derived from “a long list of human emotions” developed recently by another researcher on the basis of a cluster analysis of all emotional words and phrases. Hartmann chose these 18 from the longer list because they were found by him to be codeable on a sample of about 300 dreams. This is not an adequate reliability check in my experience. It is my prediction that it will not be possible to obtain good reliability for what are in effect 18 categories. This means there will be differences from coder to coder and study to study, and perhaps contradictory findings if researchers outside Hartmann’s group try to use the scale. It also means no one should abandon his or her own theory on the basis of published results with this scale until the findings are replicated by other investigators.

The Hall/Van de Castle system for coding emotions, on the other hand, consists of only five emotions categories: Happiness, Apprehension, Confusion, Anger, and Sadness. The system is limited to these five categories because it was impossible to develop a reliable one with more categories. Hall and Van de Castle (1966, p. 110) write that “The classification of emotions was one of our most difficult tasks.” There seem to be very good psychological reasons why it is hard to obtain reliable coding with more than these five categories. Several different kinds of psychological studies suggest that there are only five basic emotional categories–love, joy, anger, sadness, and fear.

These five categories from general psychological studies correspond very closely to the five in the Hall/Van de Castle system. There are two relatively minor, but theoretically interesting, differences. First, the “love” and “joy” categories are in effect collapsed into one category, “happiness,” in the Hall/Van de Castle system for the simple reason that the few positive emotions in dreams are difficult to distinguish in a reliable fashion. (Only 20% of all emotions in dreams fit into the happiness category.) So this category difference actually reflects a big psychological difference between waking life and dreams in terms of emotional states: the emotions expressed in dreams are overwhelmingly negative. (The high “negative emotions percent” in dreams has important implications for Hartmann’s theory that I return to shortly.)

Second, the Hall/Van de Castle system differs from the waking categories in that it includes a category for “confusion,” which is not considered an emotion in waking life. It is included in the Hall/Van de Castle system because it is a psychological state that occurs frequently in dreams, and does seem to have more of an emotional flavor in them. The highly empirical and dream-based nature of their coding system can be seen in Hall and Van de Castle’s discussion of this issue. They begin by noting their recognition of the fact that confusion is not usually thought of as an emotion because it “resides more in the head as a state of cognitive ambiguity than it does in the viscera as a gut-type reaction,” but they also conclude that it is a “feeling state” and therefore “emotionlike.” Second, and very importantly for my emphasis here, they argue that since confusion is “reported fairly frequently in dreams, mention of it seems to belong most appropriately in the classification of emotions” (Hall & Van de Castle, 1966, p. 110)

In other words, Hall and Van de Castle are guided by what they find in dreams, not by waking autonomic correlates or a theory of emotions. While this may seem misguided to some people, I think it is a mistake to use waking categories and implicit theories that make for less-than-useful content categories for the study of dreams. That is, the categories should respect and reflect what is in dreams if they are to reveal what is to be found in new samples of dreams. By taking this approach, Hall and Van de Castle end up with an important finding: there is less love and joy in dreams than in waking life, and much more confusion. There can be no better example of why the Hall/Van de Castle system is so useful.

Another big contrast between the Hall/Van de Castle and Hartmann approaches to studying emotions in dreams concerns the emotional states that Hartmann says are characterized by fear/terror and helplessness/vulnerability. While these feeling states are coded separately by Hartmann, they are all in the category called “Apprehension” in the Hall/Van de Castle system because the authors could not reliably differentiate among the emotion states that Hartmann has put into separate categories. Perhaps this is because these emotion states do share something in common, which is that “the person is uncomfortable because the threat of some potential danger exists,” whether it is the possibility of physical injury, punishment, social ridicule, or rejection (Hall & Van de Castle, 1966, p.111). Thus, Hartmann is courting unreliability in one of his areas of major concern, and probably for no good reason, conceptually speaking, given the common element of potential danger. And I repeat that the basic-level category in waking life is for fear, just as in the Hall/Van de Castle system, which should give pause to anyone who is thinking about using Hartmann’s distinction.

Hartmann also has made another coding decision related to emotions that contrasts greatly with the Hall/Van de Castle approach. He asks his coders to impute the most likely emotion to the element/theme they have identified as a contextualizing image when none is mentioned. This failure to mention emotions is not at all uncommon in the reporting of dreams. There are two problems with imputing emotions. The first is that it may be an important feature of dreams that they are not always accompanied by the emotions that seemingly should be present. Perhaps it is support for this point that people often give less negative ratings than might seem warranted by the content when they rate their emotions in the dreams they submit. Hall (1951) found this contrast between ratings by coders and dreamers in a study of 1000 dreams collected from Americans in the 1940s, and Schredl and Doll (1998) report it for 180 dreams collected in the 1990s from Germans.

Second, imputing motives is also a risky business methodologically because it becomes another possible source of unreliability in a type of research where reliability is an absolute bedrock necessity. Consider the following dream report for which Hartmann infers an emotion. Before reading what emotion he assigns, make your own choice among anxiety, fear, grief, and guilt: “A huge tree has fallen down right in front of our house. We’re all stunned ” (Hartmann, 1998, p.23). Hartmann reports the relevant emotion here is “grief.” But to an outsider reading the dream, it could just as easily have been anxiety or fear, or so it seems to me.

Reliable coding categories aside, there is a second good reason for using Hall and Van de Castle’s system for Hartmann’s purposes. Its normative findings can be used to make comparisons between samples of contextualizing-metaphor dreams and ordinary dreams. Consider again the Hall/Van de Castle emotions categories. An analysis of frequencies in those categories reveals that (1) women are far more likely to mention emotions in their dream reports than men; and (2) the most frequent emotions for both men and women are apprehension and confusion, followed by happiness, with sadness fourth on the list for women and anger fourth on the list for men. The norms also provide information on the emotions attributed to characters other than the dreamer. Overall, 80% of the emotions in the dreams of both men and women are in one of the four “negative” categories, which provides a very handy general indicator, the “negative emotions percent.” It is noteworthy that the same figure has been reported in four later studies using dreams from different eras, the sleep laboratory, and Canada (Hall, Domhoff, Blick, & Weesner, 1982; Merritt, Stickgold, Pace-Schott, Williams, & Hobson, 1994; Roussy, Raymond, & De Koninck, 2000; Tonay, 1990/1991)

By using the Hall/Van de Castle system, Hartmann and his co-workers could compare the negative emotions percent in any sample of dreams they studied with the norms, thereby providing them with the crucial “control group” they now lack. As things now stand, they may be overly impressed with the highly negative nature of the dreams they are studying because they do not realize that dreams in general are extremely negative. Moreover, if they used the Hall/Van de Castle system, they could see if contextualizing-emotion dreams have a higher proportion of any specific category, such as Apprehension, than might be expected from the norms.

Hartmann’s analysis also could benefit from normative comparisons on other Hall/Van de Castle categories that relate to the content in the type of dreams of interest to him. From my quick coding of several of the dreams in his book, they are filled with “physical aggressions” like chase, physical attack, and murder. They also contain many “misfortunes,” a general coding category for any “mishap, adversity, harm, danger, or threat which happens to characters as a result of circumstances over which they have no control,” and which are not due to the aggressive intent of some other dream character (Hall & Van de Castle, 1966, p. 103). This general classification has separate categories for being lost (M1), falling (M2), being threatened by an overwhelming environmental event (M3), being in an accident without injury (M4), suffering an injury or illness (M5), and death as a result of accident, illness, or some unknown cause (M6).

The dreams in Hartmann’s book are filled with aggressions and misfortunes, which may seem impressive in terms of his theory at first glance, but both the physical aggressions and misfortunes categories have fairly high frequencies in the Hall/Van de Castle norms: 23% of the men’s dreams and 15% of the women’s dreams have at least one chase, attack, or murder, and 33% of men’s dreams and 36% of women’s dreams have at least one misfortune. Even more to the point of Hartmann’s studies, 69%% of the men and 54% of the women had at least one chase, attack, or murder in the five dreams they contributed to the normative sample. Given these high normative figures, it is an open question as to whether his samples of five “most recently recalled dreams” from trauma sufferers would be higher on these figures than the norms are.

Reliability and norms aside, there is a third reason for using the Hall/Van de Castle system. It has ways of making distinctions among types of negative events in dreams that could be useful for Hartmann’s studies. In the area of aggressions, for example, it starts with a distinction between witnessing aggressions and being involved in them; about 20% of the aggressions in the dreams of both men and women are witnessed. The system then distinguishes between victim and aggressor status when the dreamer is involved, with men the aggressors 40% of the time and women 33% of the time. It next calls for determining whether or not dreamers “reciprocate” when they are victims, which is not often, 13% of the time for men and 10% for women. There are also norms for aggressions with specific types of characters, such as males, unfamiliar characters, and animals. In the case of misfortunes, there is a distinction between “dreamer-involved” misfortunes and misfortunes to others.

Many of these categories would be useful with the dreams in Hartmann’s book that I studied. There is a great deal of victimization of the dreamer, but there is also a witnessed aggression, and an aggression by an animal. Several misfortunes happen to others, including an animal. Using Hall/Van de Castle categories, many questions could be asked concerning individual differences, possible declines in victim status over time, or changes in the degree of dreamer involvement in misfortunes.

The fourth reason for using the Hall/Van de Castle system is that it contains procedures for dealing with sequences of events. These sequences are called “consequences” because they concern the aftermath of failures, successes, misfortunes, and good fortunes. A failure, for example, can be overcome by a helping hand from another character, which is a friendly interaction, by a success through renewed striving, or even by a miracle, which is a good fortune. Misfortunes, such as being lost, can be reversed by the good fortune of suddenly finding a familiar pathway, or by renewed striving that leads to success, or by receiving help from a dream character (a friendly interaction). Such “consequences” are rare in everyday dreams from average people, so they might be worth looking at in the types of dreams of concern to Hartmann to see if his dreamers seem even more resigned than average dreamers to their negative experiences in dreams.

Do Dreams Change Over Time?

Much of Hartmann’s theory, and especially the problem-resolving function he assigns to dreaming, stands or falls with the degree to which dreams change over time. He asserts that he has found positive changes in several dream series that he has studied, but he presents no systematic data on declines in particular negative themes, nor in any content categories that parallel the categories for aggressions, misfortunes, failures, and negative emotions in the Hall/Van de Castle system. Nor has he reported any comparisons with possible changes in the dreams of people who have not suffered traumas. The dreams are not available for others to study.

There is reason to be skeptical about his claims until systematic findings are presented. Several cross-sectional studies show few or no changes in dream content once early adulthood is reached (Cote, Lortie-Lussier, Roy, & DeKoninck, 1996; Hall & Domhoff, 1963; Hall & Domhoff, 1964; Winget, Kramer, & Whitman, 1972; Zepelin, 1980; Zepelin, 1981). There are two possible exceptions where the data are mixed, a decline in aggressions and in negative emotions (Brenneis, 1975; Cote et al., 1996; Howe & Blick, 1983). If it were to prove to be the case that one or both of these two categories does show normative declines, then it might be that the therapeutic results Hartmann is attributing to dreams are really due to declines that are part of the normal aging process.

Longitudinal studies show the same picture of consistency. Most of these studies are on continuous dream journals that stretch over years or decades (Bell & Hall, 1971; Domhoff, 1996; Smith & Hall, 1964). However, one study looked at the dreams of 21 women who kept dream diaries for a few weeks at intervals of 10, 15, or 17 years and found no changes (Lortie-Lussier, Cote, & Vachon, 2000). It would seem that the burden of proof would be on those who claim that dreams change over time, even though that would be the common-sense expectation.

However, there is a finding that could prove hopeful for Hartmann’s claims. There are people who report they stopped having their recurrent dreams, and there is evidence that their general dream content is more benign and that they are happier than those still experiencing recurrent dreams.(Zadra, 1996; Zadra, O’Brien, & Donderi, 1997-1998). Still, the acid test for Hartmann’s idea is the demonstration of positive changes in a dream journal in the weeks and months after a traumatic experience.

Conclusion

There is much more I could say about Hartmann’s theory, and also much more I could say about the uses of the Hall/Van de Castle system. But I hope I have said just enough about both to suggest that Hartmann’s theory will receive its best and most convincing test if those interested in testing the theory make judicious use of the theory-free nominal categories developed over many years by Hall, finalized by Hall and Van de Castle in two arduous years of work, and then used repeatedly for over three decades (Avila-White, 1999; Domhoff, 1999; Hall, 1951; Hall & Van de Castle, 1966; Lortie-Lussier et al., 2000; Strauch, 1999). It is unlikely that the Hall/Van de Castle system can be matched for reliability without a great deal of effort, and it is equally unlikely that any modern-day researchers are going to put in the kind of methodological effort needed to create new scales that are better.