A Primal Difference (Part 3 of Creating a Baseline for Studying Patterns in Dream Content)

A Primal Difference (Part 3 of Creating a Baseline for Studying Patterns in Dream Content) by Kelly BulkeleyWhat makes unusually memorable dreams different from average, ordinary dreams?  Putting it in Jungian terms, how are big dreams different from little dreams?


The SDDb baselines for most recent dreams (MRDs) and memorable dreams (MemDs) give a very precise and empirically based answer to this question.


MRDs tend to use more words relating to perception, emotion, cognition, social interactions, and culture.


MemD tend to use more words relating to flying, air, family, animals, fantastic beings, Christianity, and death.

Overall, MRDs are more anchored in present waking circumstances, while MemDs seem to have less connection to current social reality.   MRDs reflect more of daily life, while MemDs express deeper existential themes.

These results derive from 828 female MRDs and 691 male MRDs, compared to 801 female MemDs and 504 male MemDs.  You can see the spreadsheet here.

With the help of Dominic Luscinci at Far West Research, I analyzed these four sets of dreams in terms of their similarities and differences, adjusting the levels of statistical significance to account for multiple tests in each word class to protect against type 1 errors.  Fisher’s Exact Test was used in cases where the criteria for chi-square testing were not present.

Before getting into the MRD vs. MemD comparison, I wanted to know what gender differences were most significant. I found that female MRDs and MemDs are more likely than male reports of both types to include references to emotions (especially fear), characters (especially family), speech, and friendly social interactions.  Female MRDs have more perception and cognition, while male MRDs have more physical aggression and sexuality.

There are fewer gender differences in the MemDs than the MRDs.

Then I looked at the MRDs and the MemDs to see what differences show up for both males and females.  I found that MRDs for both genders have more references to emotion, cognition, social interactions, and culture.  MemDs have more references to nature (especially air and flying), characters (family, animals, fantastic beings), Christianity, and death.  The female MemDs have more fire, falling, and physical aggression words.  The male MemDs have more chromatic and achromatic colors.

There are many more differences between MRDs and MemDs than between the males and females.  The comparison with the fewest differences was female MemDs vs. male MemDs; these two sets of dreams were the most like each other.

My first reaction to these findings was surprise that the MRDs had more words relating to perception and emotion, since I expected these indices of intensity and vividness would be more frequent in highly memorable dreams.

But I also felt good because these results basically replicate a 2011 study I did with Ernest Hartmann on big dreams.  In the conclusion of that article we wrote “people’s big dreams are distinguished by a tendency toward ‘primal’ qualities of form and content: more intense imagery, more imagery picturing nightmarish emotions, more nature references, more physical aggression, more family characters, more fantastic/imaginary beings, and more magical happenings, along with less high-order cognition and less connection to ordinary daily surroundings.” (p. 165)

These findings are very similar to the SDDb baseline results.  They give me confidence that these differences between MRDs and MemDs are real and not the result of random variations in the data.

The comparison with the 2011 study is not perfect, since a) that project did not adjust the dream reports for word length, including reports of less than 50 and more than 300 words, unlike the SDDb baselines, b) some aspects of the conclusion (e.g., intense imagery, nightmarish emotions) were derived from Hartmann’s Central Image scoring system and did not emerge from the word search analysis, and c) the participant pool had a big gender imbalance (147 female, 15 male).  However, the mostly female composition of the 2011 study actually points to an even closer alignment with the SDDb female results because the female MemDs (but not the male MemDs) have higher frequencies of fire, falling, and physical aggression, all of which seem consistent with the 2011 study’s conclusion.

In a future post I will look at the SDDb’s high-frequency MemD elements–flying, air, family, animals, fantastic beings, Christianity, death–to try and discern what each of them adds to the dream’s memorability and impact on the dreamer.


SDDb Baselines for Recent Dreams and Memorable Dreams (Part 2 of Creating a Baseline for Studying Patterns in Dream Content)

SDDb Baselines for Recent Dreams and Memorable Dreams (Part 2 of Creating a Baseline for Studying Patterns in Dream Content) by Kelly Bulkeley“Dreams are not mysterious, supernatural, or esoteric phenomena.  They are not messages from the gods nor are they prophecies of the future.”  That’s what Calvin Hall said in his 1966 book The Meaning of Dreams (New York: McGraw-Hill, revised edition, p. 120).  Hall’s secular beliefs may or may not be justified, but what’s certain is that his baseline of “norm dreams” was designed to explain normal, average, ordinary types of dreams.  He was not interested in unusual, extraordinary types of dreams involving “esoteric phenomena.”  As a result, the baseline he developed gives what I call a homogenized view of dreams, privileging the theoretical significance of common, recently remembered dreams and denying the scientific relevance of rare but extremely memorable types of dreams from earlier times of life.

This is why I’ve created not one SDDb baseline, but two–one for most recent dreams (MRDs), and one for highly memorable dreams (MemDs).  You can find a spreadsheet with the baseline word usage frequencies here.  As always, I offer the caveat that this is a work in progress and will surely grow and change in the future.  My focus for now is to clarify some of the basic features of different types of dreams.  I’d like to know how MRDs and MemDs are similar, because that could tell us something interesting about how the sleeping mind operates consistently across varying dream types.  I’d also like to know how MRDs and MemDs are different, because that could tell us something interesting about the complexity of the mind in sleep and the creative potentials of the nocturnal imagination.  Setting up two baselines will, I hope, help the cause of answering these questions and provide a more sophisticated resource for the comparative analysis of other collections of dreams.

MRD Baseline: This includes 828 female dream reports and 691 male dream reports, all from the USA, all between 50 and 300 words in length, drawn from three sources.  The Hall and Van de Castle norm dreams (491 male, 490 female) form one component of the baseline.  This enables future analyses to maintain a solid “backwards compatibility” with the traditional standard of measurement in the dream research field, even as we continue trying to expand and improve beyond it. The additional dreams come from two SDDb sources: The Demographic Survey 2010, which included a “most recent dream” question, and the SCU Sleep/Wake Study 2008, which asked each participant to keep a dream journal and provide their two most recent dreams.  The SCU participants were college students like the HVDC norm dreams participants.  Those in the Demographic Survey were considerably older; I don’t yet have a detailed analysis of the age data, but I’m pretty sure the majority of participants were 50+ years of age.

MemD Baseline: This includes 801 female reports and 504 male reports, all from the USA, all between 50 and 300 words in length, drawn from four SDDb sources.  One is a question asking participants in the Demographic Survey 2010 to describe the earliest dream from childhood they can still remember.  Second is a question asking those same participants to describe the worst nightmare they can recall from any time in their life.  Third is a survey of children ages 8-18 asking them to describe the most memorable dream they’ve ever had.  Fourth is a survey of adults asking them to describe the most memorable dream they’ve ever had.  Unlike the MRD baseline, this one includes reports from children and reports answering different types of questions.  I have grouped these sources into a single baseline because they all fit comfortably under the heading “highly memorable dreams.”  Two of the sources use exactly that phrase in their questions, and the other two asked questions implicitly seeking reports of dreams with unusual memorability.  The inclusion of children’s reports is justified, I believe, because so many highly memorable dreams come from childhood, and thus children themselves may be in an especially good position to recall these dreams and describe them in detail.

At the far right of the spreadsheet you can see the word usage frequencies for each of these constituent sources of the two baselines.  As I said in the previous post, two important principles for creating a useful baseline are transparency and flexibility.  The baselines I’ve created have their limits, but they offer a great deal of transparency–you can see exactly where the reports are coming from–and flexibility–you can change or revise the baselines to suit your own purposes.

In the next post of this series, I’ll talk about some of the initial patterns I see in comparing the MRD and MemD baselines.  I invite your thoughts and observations! And corrections where I’ve gotten something wrong…



Creating a Baseline for Studying Patterns in Dream Content (Part 1)

Creating a Baseline for Studying Patterns in Dream Content (Part 1) by Kelly BulkeleyCompared to what?


That’s a question I’ve learned from Tracey Kahan to ask whenever I study a set or series of dreams.  If I find, for example, that 13% of a given collection of dreams include words related to fire, I can only assess the significance of that number in comparison to some other collection of dreams.  Maybe 13% is unusually high, maybe it’s unusually low; we can’t say for sure unless we have some kind of standard or baseline against which to compare it.


For the past half century, the Hall and Van de Castle (HVDC) Norm Dreams have been used as a general baseline to compare the content analysis findings of other sets or series of dreams.   No disrespect to Hall or Van de Castle, but I’ve always thought it would be good for the field of dream studies to develop a baseline that includes input from more than just 200 college students from 1950’s Ohio. We have indeed learned a great deal from that set of dreams, and now it’s time to widen our perspective.  One of my goals with the SDDb is to expand the HVDC approach by creating a bigger and better baseline for studying patterns in dream content.

Any dream research baseline, short of a total collection of all human dreams ever experienced, will inevitably be partial and limited, a tiny fraction of the totality of human dreaming.  This fact imposes an obligation of humility on those who pursue this kind of research.  A baseline is a pragmatic tool we create and use to help answer our questions, not a perfect representation of objective reality.

That said, it is not only possible but extremely important to make reasonable distinctions between better and worse baselines.

The bigger and more broadly based, the better.  The larger the database, the more likely the patterns in content are genuine and not just statistical noise (though we can never be absolutely sure).

Always, always, quality of data is essential–garbage in, garbage out, no matter how big your N.

The baseline’s sources should be very transparent, so researchers can make informed decisions about how much weight to give the results comparing their data with the baseline.

The HVDC Norm Dreams are divided by gender, and I think this is a good practice to continue for a couple of reasons.  First, there do seem to be significant differences between male and female dreaming, so creating a baseline for each gender offers a more precise tool for comparative research.  Second, many studies have a drastic imbalance in the gender of their participants, specifically a much higher proportion of female than male dream reports.  Hence the practical importance of offering a baseline for each gender, to facilitate the analysis of these kinds of imbalanced sets. (Why it’s easier to gather female than male dreams is a separate topic of discussion.)

Baseline frequencies for dream content will be sensitive to the word counts of the reports.  A collection of extremely long dreams will likely have higher frequencies of ALL categories of content, while a collection of extremely short dreams will likely have lower frequencies across the board.  The HVDC set draws the line at 50 words minimum and 300 words maximum.  I’m willing for now to go along with that policy, though eventually I want to return to consider what we may be losing by excluding shorter and longer dream reports.

What types of dreams should be included in a general baseline for dream research?  That’s a trickier question.  Should it blend together many different types of dreams, or should it concentrate on a single generic type of dream?

Many researchers have opted for the latter approach. The HVDC Norm Dreams include five dream reports from each participant, presumably recent dreams from the previous few nights, although several of the dreams are recurrent and/or come from an earlier time of life.  It’s not a “pure” set, but it purports to be a reasonable selection of the average dreams of this group of people.

Sleep laboratory researchers like David Foulkes have argued that dreams gathered in a home setting are too unreliable and only dream reports gathered in a controlled laboratory setting with accompanying sleep stage data should be considered when assessing basic patterns in dream content.  However, Bill Domhoff has made the case that dream reports gathered outside the lab setting can also be a valid source of insight, especially questionnaires asking people to describe their “most recent dreams.”

The difficulty in defining what counts as the most generic type of dream makes this approach problematic.  Another drawback is the under-reporting of the incidence of rare but intense and highly memorable types of dreams–nightmares, lucid dreams, visitation dreams, recurrent childhood dreams, etc.  These exceptional types of dreams may not occur as frequently as ordinary dreams, and thus they do not appear as often when people are asked to describe their most recent dreams.  But these unusual dream types are widely experienced and reflect important features of the dreaming mind that we need to account for in any general theory of dream psychology.  We lose sight of those features when we focus only on allegedly “average” dreams.

The advent of database technology makes it easier than ever to try the former approach: Creating a baseline that accepts rather than denies the “multiplicity of dreams” (in Harry Hunt’s terms), a baseline that blends together many different types of dreams and seeks a dynamic balance representing the varied phenomenology of dreaming across the widest possible range of its occurrence.

In Part 2 I’ll describe how I’m trying to develop this kind of blended baseline using data in the SDDb.


Hall and Van de Castle Norm Dreams Now in the SDDb

Hall and Van de Castle Norm Dreams Now in the SDDb by Kelly BulkeleyThanks to the help of Bill Domhoff and Adam Schneider (and of course Kurt Bollacker), the set of 981 Hall and Van de Castle male and female “Norm Dreams” are now in the SDDb and available for study using the database tools.  Long available on the Dreambank.net website, the Norm Dreams have been widely cited in research literature for many decades, and it’s a big boost to the SDDb to include this historically significant dream collection.

Calvin Hall gathered these dreams from 100 female and 100 male college students from two colleges near Cleveland, Ohio, from 1947-1950.  Each student provided five dream reports of no less than 50 words and no more than 300 words in length.  The complete set of 1000 dreams served as the foundation for Hall’s book with Robert Van de Castle, The Content Analysis of Dreams in 1966.  Hall and Van de Castle called them the Norm Dreams because their content frequencies could be used as a basis for comparison with other groups, as a measuring stick to determine what counts as normal or abnormal proportions of dream content.

That’s a strong claim, of course, too strong perhaps, but only because Hall and Van de Castle’s data were relatively limited.  The goal of trying to identify large-scale, widely distributed patterns in dreaming remains a worthwhile pursuit, and now we have much more data and much better tools than Hall and Van de Castle had to seek them out.

The first thing I did once the Norm Dreams were in the SDDb was to try a series of identical word searches in the Dreambank and the SDDb.  I wanted to insure that the original texts (981 remain, 19 were lost some time ago) were exactly the same in both databases and that their search results were directly comparable.

Phew!  Every word I searched for in the Norm Dreams in the SDDb yielded the same results as a search for the same word in the Norm Dreams in the Dreambank. (Individual words being searched in the Dreambank have to be framed with^ ^.  For example, to search for the word anger, the term must be typed ^anger^.)

Next, I wanted to check the Norm Dreams for their frequencies on the SDDb 40-category template and compare these results to the frequencies I found using an earlier prototype of this template in my 2009 paper in Consciousness and Cognition, where I reported word search findings on the Norm Dreams in the Dreambank.  I have made several minor changes and additions to the 40 categories since 2009, so I expected the results now to be slightly higher but essentially the same.

Again, the results were reassuring (although I didn’t have the counts from 2009, just the percentages).  When I searched the Norm Dreams for each of the SDDb’s 40 word categories, the frequencies were the same or slightly higher as the frequencies I found in 2009 applying similar categories to the Norm Dreams in the Dreambank.  The Earth and Transportation categories had the biggest increase between the two analyses, due to the addition of several new terms to these two categories when I originally programmed the SDDb’s template.

The one exception was the Weather category, which initially showed a lower frequency in the SDDb analysis compared to the earlier Dreambank analysis.  When I investigated the differing results more closely, I found I had not done a very good job translating all the weather-related words into the SDDb template.  Several words were missing from Weather category in the SDDb template that I had used in the Dreambank analysis.


When I performed an adjusted SDDb search including these previously excluded words, the results were back in line with the expected similarity between the two databases. (This makes me think I’ll need to re-check all the categories when I next get a chance to upgrade the template.)

These initial findings have given me confidence that the Hall and Van de Castle Norm Dreams can be studied using the word search tools of the SDDb in a way that’s consistent, reliable, and open to comparison with analyses from the Dreambank or any other research project making use of the Norm Dreams.

All of this means it’s getting easier and easier to make apples-to-apples comparisons of dream content using word search technology.

I doubt the dreams of 200 college students from 1940’s Ohio can give us a complete representation of all human dreaming (though there are actually many intriguing “big dream” experiences in the set).  But I share Hall and Van de Castle’s goal of identifying broad patterns of dream content.  I’m hopeful that word search methods, applied to larger collections of data from more diverse groups of people, will help us move closer to that goal.

Note: the statistical table I created with the frequencies for the 40 categories can be found here.













Word Searching as a Tool in the Study of Dreams, or, Dream Research in the Era of Big Data

Word Searching as a Tool in the Study of Dreams, or, Dream Research in the Era of Big Data by Kelly BulkeleyI’m giving a presentation with that title on Saturday, June 23, at the annual conference of the International Association for the Study of Dreams, held in Berkeley, California.  The presentation is part of a panel session, “What’s New in the Scientific Study of Dreams.”  I’m giving an overview of the word searching method I’ve been developing over the past several years, with a special focus on four “blind analysis” studies I’ve performed with the help of Bill Domhoff.  A youtube video preview of the presentation can be found here.

Here’s how I define blind analysis in the paper:

A blind analysis involves an exclusive focus on word usage frequencies, bracketing out the narrative reports and personal details of the dreamer’s life and making inferences based solely on statistical patterns in word usage—not reading the dreams at all, and basing one’s analysis strictly on numerical data.  The aim is to assess the patterns of dream content with the fewest possible preconceptions, as objectively as possible, before reading through the narratives and learning about the individual’s waking activities and concerns.

Zeo Sleep Data and the Ur-Patterns of Dream Content

Zeo Sleep Data and the Ur-Patterns of Dream Content by Kelly BulkeleySo far I’ve done word search analyses on 20 series of dreams from individuals and 9 sets of dreams from large groups of people, a total of more than 18,000 dream reports. It’s too early to say anything definite about the patterns that have emerged from this data. More reports need to be gathered from a wider variety of people, and more improvements need to be made in the SDDb word search template.

Still, a few basic patterns have appeared in nearly all the collections I’ve studied. I’m calling them ur-patterns because they seem to represent deep structural elements of dream content (ur- as in “original” or “primal”). That’s my general hypothesis, anyway, and each new set of dreams is another chance to test and refine it.

Here are the ur-patterns I’ve identified so far:

  1. Of the five senses, sight words are used most often, smell and taste the least.
  2. Of the five major emotions (fear, anger, sadness, confusion, happiness), fear words are used most often.
  3. Of all the categories of cognitive activity, speech words are used most often.
  4. Of the four natural elements, water words are used most often.
  5. Falling words are used more often than flying words.
  6. There are more references to family characters than animal characters, and more to animals than to fantastic beings.
  7. There are more references to friendliness than physical aggression.

Looking at the KB DJ 2009-2010 series with Zeo sleep data (available at google docs), a scan for these patterns finds good but not perfect evidence for each one.

Vision-related words are used more frequently across all the Zeo measurements, with smell and taste words almost entirely absent. Fear words are used more frequently than other emotion words. Speech words are the most used among the cognition categories, and water is the highest among the natural elements, though earth is a consistently high second. The usage of falling words is always higher than, or equal to, flying words.

The family > animals pattern > fantastic beings was not as clear-cut. Fantastic beings always had the lowest word usage, but animals were not always lower than family. When the names of the dreamer’s immediate family were added to the search for characters, the total frequency of family-related words rose higher than the usage of animal words in 15 of the 17 subgroups.

The friendliness > physical aggression pattern was not perfectly evident either. In part this is due to a “false positive” problem in the SDDb template. The word search category for physical aggression includes the word “bit,” which the dreamer used in almost 10% of all the reports as a term meaning “small amount,” not a physical bite. I’ll provide revised numbers once I’ve fixed this. For now, looking at how often the word “bit” is used in each Zeo subgroup, it appears the physical aggression frequencies will drop below the friendliness frequencies in most, but not all, subgroups.

In sum, the ur-patterns appear across virtually all the subgroups of Zeo sleep measurement. No matter what aspect of sleep was measured, the dream reports used the same basic frequencies of words in several major categories. High or low proportions of sleep did not correlate with any major change of dream content, at least at this level of analysis.

In future posts I’ll look at the few variations from these patterns (high physical aggression, animal, flying, and earth references) in relation to the dreamer’s waking life concerns, taking the possibility of metaphorical meaning into account.

I will also look at each of the five types of Zeo data and see if I can identify any particular variations that rise to the level of statistically significant correlation. If any such correlations emerge, they may guide us toward specific areas where a measurable aspect of sleep does interact with basic patterns of dream content.